Are you sure?
This action might not be possible to undo. Are you sure you want to continue?
European Equity Derivatives Research
J.P. Morgan Securities Ltd.
London, 17 November, 2006
Peter Allen
(4420) 73254114
peter.allen@jpmorgan.com
Stephen Einchcomb
(4420) 73259064
stephen.einchcomb@jpmorgan.com
Nicolas Granger
AC
(4420) 73257033
nicolas.m.granger@jpmorgan.com
www.morganmarkets.com
The certifying analyst is indicated by an
AC
. See page 102 for analyst
certification and important legal and regulatory disclosures.
I
N
V
E
S
T
M
E
N
T
S
T
R
A
T
E
G
I
E
S
:
N
O
.
2
8
Variance swap cashflows
• Variance swaps offer straightforward and direct exposure to
the volatility of an underlying
• Returns from variance swaps can act as a diversifying asset
• Variance swaps can be used for hedging volatility exposures
or generating alpha
Overview
In this note we discuss the variance swap market, mechanics, pricing and
uses.
Variance swaps offer straightforward and direct exposure to the volatility of
an underlying asset. They are liquid across major equity indices and large cap
stocks, and increasingly across emerging market indices and other asset
classes.
Major uses include taking a volatility view, diversifying returns, hedging and
relative value trading. Variance swaps can also be used to trade forward
volatility and correlation.
Variance swaps can be replicated by a deltahedged portfolio of vanilla options,
so that pricing reflects volatilities across the entire skew surface. In practice
this means that variance swaps trade at a small premium to ATM implied
volatilities.
Source: JPMorgan
Buyer of
variance
swap
Buyer pays
variance swap
strike
Seller of
variance
swap
Seller pays
realised variance
at expiry
Buyer of
variance
swap
Buyer pays
variance swap
strike
Seller of
variance
swap
Seller pays
realised variance
at expiry
Table of Contents
Part 1: Variance Swap Mechanics........................................................................................................................ 8
1.1: Realised volatility.......................................................................................................................................... 8
1.2: The variance swap contract ....................................................................................................................... 11
1.3: Vega notional and variance notional .......................................................................................................... 12
1.4: Variance swap convexity............................................................................................................................ 13
1.5: Variance swap example – accruing realised volatility................................................................................ 14
1.6: Variance swap marktomarket .................................................................................................................. 15
1.7: Forward variance........................................................................................................................................ 17
1.8: Variance swap contract specifications ....................................................................................................... 19
1.9: Example Variance Swap Term Sheet ........................................................................................................ 22
Part 2: The Variance Swap Market ..................................................................................................................... 24
2.1: Market development................................................................................................................................... 24
2.2: Historical prices.......................................................................................................................................... 26
2.3: Variance swaps and option volatilities ....................................................................................................... 28
2.4: Pricing rules of thumb ................................................................................................................................ 30
2.5: What drives variance swap levels?............................................................................................................ 33
2.6: The volatility risk premium.......................................................................................................................... 34
2.7: Variance swaps as predictor of future volatility.......................................................................................... 36
2.8: Is variance swap convexity fairly priced?................................................................................................... 38
2.9: Variance term structure.............................................................................................................................. 41
Part 3: Uses of Variance Swaps......................................................................................................................... 44
3.1: Exploiting a volatility view........................................................................................................................... 45
3.2: Specific hedging purposes......................................................................................................................... 46
3.3: Rolling short variance................................................................................................................................. 47
3.4: Diversification............................................................................................................................................. 50
3.5: Index variance spreads.............................................................................................................................. 51
3.6: Relative value singlestock volatility........................................................................................................... 54
3.7: Variance dispersion and correlation trading............................................................................................... 56
3.8: Forward variance and volatility spikes ....................................................................................................... 58
3.9: Trading the variance swap term structure.................................................................................................. 64
3.10: Skew and convexity trades ...................................................................................................................... 66
3.11: Cross asset class trades: trading equity volatility against credit .............................................................. 68
Part 4: Replication and Hedging........................................................................................................................ 72
4.1: Delta hedging and dollar gamma ............................................................................................................... 73
4.2: Theta  the cost of gamma ......................................................................................................................... 76
4.3: Options pathdependency: can volatility be captured by deltahedging? .................................................. 77
4.4: From options to variance swaps ................................................................................................................ 79
4.5: Variance swap replication in one page ...................................................................................................... 81
4.6: Sensitivity to skew and convexity............................................................................................................... 82
4.7: Variance swap Greeks............................................................................................................................... 84
4.8: Setting up a replicating portfolio................................................................................................................. 87
4.9: Replication and hedging in practice........................................................................................................... 89
4.10: Effects of variance swap hedging ............................................................................................................ 90
4.11: Why not volatility swaps?......................................................................................................................... 92
Part 5: Future developments.............................................................................................................................. 94
References ........................................................................................................................................................... 98
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
2
Introduction
Variance swaps are instruments which offer investors straightforward and direct exposure to the volatility of an
underlying asset such as a stock or index. They are swap contracts where the parties agree to exchange a preagreed
variance level for the actual amount of variance realised over a period. Variance swaps offer investors a means of achieving
direct exposure to realised variance without the pathdependency issues associated with deltahedged options.
Buying a variance swap is like being long volatility at the strike level; if the market delivers more than implied by the strike
of the option, you are in profit, and if the market delivers less, you are in loss. Similarly, selling a variance swap is like
being short volatility. However variance swaps are convex in volatility: a long position profits more from an increase in
volatility than it loses from a corresponding decrease. For this reason variance swaps normally trade above ATM volatility.
Figure 1 : Variance swap payoffs are convex in volatility
p/l for 100K vega notional, variance strike of 20
1,500,000
1,000,000
500,000
0
500,000
1,000,000
1,500,000
2,000,000
5% 10% 15% 20% 25% 30% 35%
variance swap p/l
payout linear in volatility
realised volatility
Source : JPMorgan
Figure 2 : Euro Stoxx 50 implied and realised variance levels
Euro Stoxx variance
5%
10%
15%
20%
25%
Oct04 Apr05 Oct05 Apr06 Oct06
3M Implied Variance
3M Realised Variance
Source : JPMorgan
Market development
Variance swap contracts were first mentioned in the 1990’s, but like vanilla options only really took off following the
development of robust pricing models through replication arguments. The directness of the exposure to volatility and the
relative ease of replication through a static portfolio of options make variance swaps attractive instruments for investors and
marketmakers alike.
The variance swap market has grown steadily in recent years, driven by investor demand to take direct volatility exposure
without the cost and complexity of managing and deltahedging a vanilla options position. Although it is possible to
achieve variance swap payoffs using a portfolio of options, the variance swap contract offers a convenient package bundled
with the necessary deltahedging, offering investors a simple and direct exposure to volatility, without any of the path
dependency issues associated with deltahedging an option.
Variance swaps were initially developed on index underlyings. In Europe, variance swaps on the Euro Stoxx 50 index are
by far the most liquid, but DAX and FTSE are also frequently traded. Variance swaps are also tradable on the more liquid
stock underlyings – especially Euro Stoxx 50 constituents, allowing for the construction of variance dispersion trades.
Variance swaps are tradable on a range of indices across developed markets and increasingly also on developing markets.
Bid/offer spreads have come in significantly over recent years and in Europe they are now typically in the region of 0.5
vegas for indices and 1–2 vegas for singlestocks – although the latter vary according to liquidity factors. We expect
spreads to come in further in future as liquidity continues to improve. Spreads for indices in the US and Japan are similar
with single stock bid/offers in the 2 – 2.5 vegas range. Spreads are naturally higher in emerging markets although these too
are becoming more liquid. With volatilities often much higher and less stable than in developed markets, emerging market
variance can offer interesting opportunities.
3M Realised Volatility
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Introduction
3
The most liquid variance swap maturities are generally from 3 months to around 2 years, although indices and more liquid
stocks have variance swaps trading out to 3 or even 5 years and beyond. The most liquid maturities generally coincide with
the quarterly options expiry dates, meaning that they can be efficiently hedged with exchangetraded options of the same
maturity. Good liquidity is also seen in the front 3 months for shortdated index variance. The VIX, VSTOXX and VDAX
indices represent the theoretical prices of 1month variance swaps on the S&P500, Euro Stoxx and DAX indices
respectively, and are calculated by the exchanges from listed option prices, interpolating to get 1month maturity. These
volatility indices are widely used as benchmark measures of equity market risk, even though they are only shortdated
measures and are not directly tradable.
Uses of variance swaps
Increasingly, investors have come to view volatility itself as an asset class, one that can diversify investment returns or
hedge unwelcome investment scenarios. For example, volatility typically rises in a bear market, so holding a long volatility
position (e.g. being long variance swaps) can help to hedge an equity portfolio. However, as with any asset, the path to
riches is to buy when cheap and sell when expensive. And here investors can look to exploit the meanreverting nature of
volatility to add alpha to portfolios. For example, systematically selling volatility has historically been profitable, with
returns akin to selling credit protection.
Variance swaps can be traded outright or spread against other products to create a wide variety of exposures. Below we list
some common uses.
• Exploiting a volatility view: Variance swaps are ideal for taking a direct view on the volatility of an underlying
without the pathdependency issues of a deltahedged option.
• Specific hedging purposes: Variance swaps can be used for macrohedging and also for hedging specific
volatility exposures, such as that resulting from structured products or life assurance policies.
• Rolling short variance: Short variance swaps can be used to capture the observed equity index volatility risk
premium. Rolling short index variance is an attractive systematic volatility strategy from a riskreturn perspective.
• Diversification: Volatility can be thought of as an asset class in its own right, and as such can act to diversify
returns within a portfolio.
• Index variance spreads: Variance swaps can be used to trade the spread of volatilities between two indices. Such
trades can be thought of as either relative value volatility trades or as ‘volatilitybeta’ trades aiming to profit from a
spread of volatilities widening as volatility increases.
• Relative value singlestock volatility: Use volatility pairs, or crosssectional regression volatility models to find
rich/cheap singlestock volatilities
• Variance dispersion and correlation trading: Trading variance swaps on an index against variance swaps on its
constituents provides exposure to equity correlation.
• Forward variance and volatility spikes: Long forward volatility can avoid potentially negative carry at the cost
of slide down the term structure, and can be a useful way of positioning for volatility spikes.
• Trading the variance term structure: Variance swaps can be used to trade the shape of the variance term
structure, analogous to the use of spreads and butterflies in fixed income to trade the shape of the yield curve.
• Skew and convexity trades: Variance swaps are long skew and convexity. Trading variance against (delta
hedged) vanilla options provides interesting exposures to skew and/or convexity.
• Cross asset class trades: Equity Volatility and credit spreads are correlated, both being measures of corporate
risk. Variance swaps are useful instruments in debt/equity trades, either at the index or single name level.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Introduction
4
Outline of the document
Part 1: Variance Swap Mechanics
We consider the mechanics of variance swaps. We begin by defining realised volatility and consider some of its
commonly observed properties. We go on to define a variance swap, including a sample term sheet and specifications, and
look at the different measures of notional: vega and variance notionals. Variance swap marktomarkets are easily
calculated, since variance is additive. This leads to the simple valuation of forwardstarting variance swaps.
Part 2: The Variance Swap Market
We look at the variance swap market development, the pricing of variance swaps (including rules of thumb) and variance
swap drivers. Variance swap strikes are well correlated with standard BlackScholes implied volatility derived from options
prices. This should not be surprising, as both can be interpreted as market estimates of future volatility. Variance swaps
trade slightly above ATM volatility levels, essentially due to the fact that variance swaps are long convexity (or volatility of
volatility).
Realised volatility and riskaversion are major drivers of variance swap levels, with structured product flows impacting
the long end of the maturity curve. Shortdated variance swaps tend to do a relatively good job of predicting future
realised volatility, at least better than using previous realised.
Part 3: Uses of Variance Swaps
We discuss the uses of variance swaps, ranging from simple expressions of a volatility view, to relative value volatility
trades, to more complex trades using variance to trade correlation, skew, convexity etc. One important use of variance
swaps, at least at the index level, is to exploit the volatility risk premium by systematically rolling a short variance swap
position. The volatility risk premium reflects the observed tendency of implied volatility to trade above realised volatility.
Even though returns from a short variance or volatility position have more downside than upside risk, historically investors
have been more than compensated for this risk. Such strategies are now being incorporated into structured products such as
JPMorgan’s Yield Alpha product.
Part 4: Replication and Hedging
Part 4 is more technical in nature. We outline the mechanism whereby a variance swap can be replicated by a portfolio
of options. The main drawback of a deltahedged option as a vehicle for trading volatility is that the dollargamma, which
characterises the exposure to realised volatility, is path dependent. It is possible to remove this path dependency by
trading a carefully chosen portfolio of options of different strikes. A variance swap consists of just this portfolio, bundled
up with the necessary deltahedging to give direct exposure to realised volatility. We give a concrete example of such a
portfolio and consider some practical issues arising from variance swap hedging.
In particular we discuss the potential impact of variance swap hedging on the underlying market, driven by the need of
replicating option portfolios to be deltahedged on the close (unlike other options). Depending on market positioning, such
hedging activity may exacerbate large moves into the close, potentially adding to market volatility. Hedging flows, and
anticipation thereof, have been much talked about, although in an efficient market such flows should be propagated back
throughout the trading day.
Part 5: Future Developments
We discuss 3rd generation variance swap products and future developments. Commonly watched volatility indices
such as VIX and VSTOXX are in fact calculated as variance swap strikes, and futures exchanges have listed futures on
these indices. As yet there is no official variance swap fixing mirroring deposit or swap rate fixings. Conditional
variance swaps have also increased in liquidity, and variance swaps on other asset classes are starting to be traded. We
expect variance swap markets to continue to gain liquidity, fuelled by investor demand to trade pure volatility and by the
desire of investment banks to recycle their exotic and structured product risks.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Introduction
5
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
6
7
1. Variance Swap Mechanics
• 1.1 Realised volatility
• 1.2 The variance swap contract
• 1.3 Vega notional and variance notional
• 1.4 Variance swap convexity
• 1.5 Variance swap example
• 1.6 Variance swap marktomarket
• 1.7 Forward variance
• 1.8 Variance swap contract specifications
• 1.9 Example variance swap term sheet
P
a
r
t
1
:
V
a
r
i
a
n
c
e
S
w
a
p
M
e
c
h
a
n
i
c
s
7
Part 1: Variance Swap Mechanics
Variance swaps are instruments which offer investors straightforward and direct exposure to the volatility of an
underlying asset such as a stock or index. They are swap contracts where the parties agree to exchange a preagreed
variance level for the actual amount of variance realised over a period (variance is the square of volatility). Variance swaps
offer investors a means of achieving direct exposure to realised variance without the pathdependency issues associated with
deltahedged options.
Before discussing the mechanics of the variance swap contract we begin by defining the notion of realised volatility, and
look at some of its characteristics. We then define a variance swap and describe some common conventions such as the
means of expressing trade notionals. A variance swap pays out linearly in terms of the realised variance, but when
considered in terms of volatility (as it more commonly is) the payout is convex. We give a concrete example of a variance
swap trade to illustrate how the p/l from realised variance accrues over time. This naturally leads into a discussion of how
variance swaps are markedtomarket, through changes both to implied and realised variance. The linearity of variance
which makes markingtomarket so straightforward also allows us to construct exposure to forward variance, through
forward starting variance swaps which offer an interesting alternative to the usual spot variance exposure. Finally we
consider some technicalities associated with variance swap contracts, such as the treatment of dividends and M&A events,
and conclude Part 1 by giving an example of a variance swap term sheet.
1.1: Realised volatility
Volatility measures the variability of returns of an underlying asset and in some sense provides a measure of the risk of
holding that underlying. In this note we are concerned with the volatility of equities and equity indices, although much of
the discussion could apply to the volatility of other underlying assets such as credit, fixedincome, FX and commodities.
Figure 3 shows the history of realised volatility on the Dow Jones Industrial Average over the last 100 years. Periods of
higher volatility can be observed, e.g. in the early 1930’s as a result of the Great Depression, and to a lesser extent around
2000 with the buildup and unwind of the dotcom bubble. Also noticeable is the effect of the 1987 crash, mostly due to an
exceptionally large single day move, as well as numerous smaller volatility spikes.
Figure 3 : Long term history of realised volatility
3month realised volatility (Dow Jones Industrial Average)
0%
10%
20%
30%
40%
50%
60%
70%
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Source: JPMorgan
The following are some of the commonly observed properties of (equity market) volatility:
• Volatility tends to be anticorrelated with the underlying over short time periods (Figure 4)
• Volatility can increase suddenly in ‘spikes’
• Volatility can be observed to experience different regimes (Figure 5)
• Volatility tends to be mean reverting (within regimes)
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
8
This list suggests some of the reasons why investors may wish to trade volatility: as a partial hedge against the underlying –
especially for a volatility spike caused by a sudden market selloff; as a diversifying asset class; to take a macro view e.g.
for a potential change in volatility regime; or to trade a spread of volatility between related instruments.
Figure 4 : in the short term volatility is anticorrelated with index level
Euro Stoxx 50 realised volatility index level (inverted)
0%
10%
20%
30%
Jan06 Mar06 May06 Jul06 Sep06
3400
3500
3600
3700
3800
3900
4000
4100
1m realised volatility (lhs)
index level (inverted, rhs)
Source : JPMorgan
Figure 5 : 20 years of Euro Stoxx realised volatility: the index has
experienced both high and low volatility regimes
realised volatility
0%
10%
20%
30%
40%
50%
1987 1990 1993 1996 1999 2002 2005
Euro Stox x 50 3m realised v olatility
Source : JPMorgan
Measuring realised volatility and variance
Realised (or historical) volatility can be measured in a variety of ways. One of the simplest measures is the annualised
standard deviation of daily (log) returns of an underlying, calculated from a data set spanning some fixed period of time –
for example 1month historical volatility is calculated as the annualised standard deviation of daily returns occurring in one
month. This type of measure has the advantage of being very easy to understand and calculate and it is this which we used
throughout this note.
One issue with using this fixed timeperiod volatility is that the realised volatility calculated in this way can suddenly drop
as a large magnitude return falls out of the window of observation. Other, more sophisticated, measures of realised
volatility exist, which seek to alleviate this problem and provide a potentially better means of forecasting future volatility –
for example by giving more weight to recent returns (exponentially weighted measures) and/or attempting to model the
meanreverting and regime changing nature of volatility (GARCH models). See “Options, Futures and Other Derivatives”,
J. Hull, for further details.
Volatility can also be calculated using returns of various frequencies (weekly, daily, intraday returns etc). If the returns are
truly independent, then, over a long enough timeframe, the realised volatility should be independent of the frequency of the
returns used to calculate it. However, in practice, volatility is usually calculated using daily closetoclose returns. This is
almost always the type of volatility which dictates the payoff of a variance swap contract.
Volatility is usually quoted as an annualised volatility and expressed as a percentage. The annualised volatility represents
the expected annual standard deviation in distribution of the underlying, assuming returns are independent. To move from
the standard deviation of daily log returns to annualised volatility, simply multiply by the squareroot of the number of
trading days in the year, usually taken to be 252. Note that since the squareroot of 252 is about 16, a volatility of 16%
equates to a standard deviation of daily moves of around 1%.
Volatility can also be defined by an RMS (rootmeansquared) measure, which is like a standard deviation but assuming a
zero mean (Box 1). This simplifies calculations, and in practice, the difference from the traditional standard deviation is
very small, for all but the shortestdated volatility (Figure 6, Figure 7). The average return is the drift, which should be near
zero on a daily basis. Furthermore, it is this RMS volatility measure which is used to define the payout of the variance
swap contract.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
9
Box 1: Definition of realised (RMS) volatility
Volatility, σ, is defined by:
2
1 1
2
ln
252
∑
= −


.

\

=
T
i i
i
S
S
T
σ
where S
i
is the stock price on day i, and
T is the number of days
Table 1: Calculating realised volatility (RMS)
Index
level
Log
return
Log return
squared
13Oct05 3331.4
14Oct05 3349.6 0.5% 0.000030
17Oct05 3356.8 0.2% 0.000005
18Oct05 3334.8 0.7% 0.000043
19Oct05 3279.6 1.7% 0.000278
20Oct05 3284.8 0.2% 0.000002
21Oct05 3271.0 0.4% 0.000018
24Oct05 3310.1 1.2% 0.000141
25Oct05 3292.5 0.5% 0.000028
26Oct05 3304.3 0.4% 0.000013
27Oct05 3241.1 1.9% 0.000372
Variance of log returns 0.000093
Annualised variance 0.0234394
Annualised volatility (RMS) 15.3%
Source : JPMorgan
Volatility conveys information about the usual size of moves in the underlying in relation to its absolute price. Typical
stocks have (annualised) volatilities of around 1535%, with equity indices generally somewhat lower. Maturities
considered usually range from a few days to a few years. Clearly shortdated realised volatility will be much ‘noisier’ than
longer dated, but will convey more uptodate information. Longer dated volatility represents a smoother ‘average’
volatility but will include older data possibly outside of the current regime – e.g. 5year Euro Stoxx volatility currently
remains some 9% above 1year volatility due to the inclusion of data from the high volatility regime in 20022003.
Figure 6 : The difference between RMS volatility and standard
realised volatility can be noticeable for shortdated volatility…
5day realised volatility (Euro Stoxx 50)
0%
5%
10%
15%
20%
25%
30%
35%
40%
Jul05 Oct05 Jan06 Apr06 Jul06
Standard v olatility
RMS (zeromean) v olatility
Source : JPMorgan
Figure 7 : …but the difference is negligible for longerdated volatility
6month realised volatility (Euro Stoxx 50)
10%
11%
12%
13%
14%
15%
16%
17%
Jul05 Oct05 Jan06 Apr06 Jul06
Standard v olatility
RMS (zeromean) v olatility
Source : JPMorgan
Although it is volatility which is the more intuitive concept, especially since it is measured in the same units as the
underlying, variance (volatility squared) is in some sense more fundamental. Most importantly variance is additive,
whereas volatility is not. It is realised variance which the deltahedging of options seeks to capture, although the exposure
is complicated by path dependency issues. Variance swaps payout purely based on realised variance.
Average the
squared log
returns
Annualise (x 252)
Take square root
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
10
1.2: The variance swap contract
Variance swaps are instruments which offer investors straightforward and direct exposure to the volatility of an
underlying asset such as a stock or index. They are swap contracts where the parties agree to exchange a preagreed
variance level for the actual amount of variance realised over a specified period (Figure 8). Variance swaps offer investors
a means of achieving direct exposure to realised variance without the pathdependency issues associated with deltahedging
options.
The strike of a variance swap, not to be confused with the strike of an option, represents the level of volatility bought
of sold and is set at trade inception. The strike is set according to prevailing market levels so that the swap initially has zero
value. If the subsequent realised volatility is above the level set by the strike, the buyer of a variance swap will be in profit;
and if realised volatility is below, the buyer will be in loss. A buyer of a variance swap is therefore long volatility.
Similarly, a seller of a variance swap is short volatility and profits if the level of variance sold (the variance swap strike)
exceeds that realised.
Figure 8: P/L of a variance swap
By convention, volatility is scaled by a factor of 100, for example a strike of 20 represents a volatility of 20%. Note that
variance swap strikes are quoted in terms of volatility, not variance; but pay out based on the difference between the level of
variance implied by the strike (in fact the strike squared) and the subsequent realised variance. This means the p/l of a
variance swap is nonlinear (convex) with volatility, although of course it is linear in terms of variance.
Box 2: Variance swap p/l
The p/l for a (long) variance swap is given by:
( )
2 2
/ K N l p
Variance
− × = σ
where K is the variance swap strike,
2
σ is realised variance, and
Variance
N is the variance notional.
Example 1: Variance swap p/l
An investor wishes to gain exposure to the volatility of an underlying asset (e.g. Euro Stoxx 50) over the next year.
The investor buys a 1year variance swap, and will be delivered the difference between the realised variance over the next
year and the current level of implied variance, multiplied by the variance notional.
Suppose the trade size is €2,500 variance notional, representing a p/l of €2,500 per point difference between realised and
implied variance.
If the variance swap strike is 20 (implied variance is 400) and the subsequent variance realised over the course of the year is
15%
2
= 0.0225 (quoted as 225), the investor will make a loss because realised variance is below the level bought.
Overall loss to the long = €437,500 = €2,500 x (400 – 225) . The short will profit by the same amount.
Buyer of
variance swap
Seller pays
realised variance
at expiry σ
2
Buyer pays
variance swap
strike K
2
Seller of
variance swap
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
11
1.3: Vega notional and variance notional
The notional for a variance swap can be expressed either as a variance notional or a vega notional. The variance notional
is in some sense the ‘true’ notional – representing the p/l per point difference between the strike squared (implied variance)
and the subsequent realised variance.
However, since variance is a somewhat unintuitive quantity, and most market participants are used to thinking in terms of
volatility, trade size is typically expressed in vega notional. Vega notional represents the average profit or loss for a 1%
(1 vega) change in volatility, and thus gives a more economically meaningful idea of the variance swap’s exposure to
volatility. The vega notional is equal to the variance notional multiplied by twice the variance strike.
The p/l of a (long) variance swap can be calculated by multiplying the vega notional by the difference between the realised
variance and the strike squared, and then dividing by twice the strike. This ensures that when realised volatility is close to
the variance strike, the p/l is close to the difference between implied and realised volatility multiplied by the vega notional.
That is, the variance swap payout, expressed in vega notional, is locally linear around the strike.
Box 3: Vega and variance notionals
l p /


.

\
 −
× =
K
K
N
Vega
2
2 2
σ
( )
2 2
K N
Variance
− × = σ
where N
Vega
is the veganotional, and
N
Variance
is the variance notional.
N
Vega
= N
Variance
× 2K
Example 2: Vega notional
Suppose a 1year variance swap is stuck at 20 with a vega notional of €100,000.
If the index realises 25% volatility over the next year, the long will receive €562,500 = €100,000 x (25
2
– 20
2
) / (2 x 20).
However if the index only realises 15%, the long will pay €437,500 = €100,000 x (15
2
–20
2
) / (2 x 20).
Therefore the average exposure for a realised volatility being 5% away from the strike is €500,000 or 5 times the vega
notional, as expected.
Note that the variance notional is €100,000 / (2 x 20) = €2,500, giving the same calculation as that used in Example 1.
The p/l of a variance swap is often expressed in terms of vega notional. In the example above, a gain of €562,500 is
expressed as a profit of 5.625 vegas (i.e. 5.625 times the vega notional). Similarly a loss of €437,500 represents a loss of
4.365 vegas. The average exposure to the 5% move in realised volatility is therefore 5 vegas, or 5 times the vega notional.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
12
1.4: Variance swap convexity
Although variance swap payoffs are linear with variance they are convex with volatility. The vega notional represents only
the average p/l for a 1% change in volatility. In fact a long variance swap position will always profit more from an
increase in volatility than it will lose for a corresponding decrease in volatility (as in Example 2). This difference
between the magnitude of the gain and the loss increases with the change in volatility. This is the convexity of the variance
swap.
Ultimately this convexity is the reason that variance swaps strikes trade above atthemoney volatility. Because of the
convexity, a variance swap will always outperform a contract linear in volatility of the same strike (e.g. Figure 9). The
price paid for this convexity should depend on the expected variability of the (realised) volatility. The higher the variability
of volatility, the larger the likely discrepancy between the variance strike and the realised volatility, and hence the more
beneficial the convexity becomes (Figure 9, Figure 10). In the limiting case, with volatility of volatility zero, the p/l of the
variance swap will always be zero (volatility is constant) and hence the variance swap strike and hypothetical volatility
swap strike must coincide. See Section 2.8 for a detailed discussion of the value of the variance swap’s convex payoff.
Figure 9 : When volatility remains close to the variance swap strike,
the variance swap payout is similar to a linear “volatility swap”…
p/l for 100K vega notional, strike of 20
1,000,000
500,000
0
500,000
1,000,000
1,500,000
10% 15% 20% 25% 30%
v ariance swap p/l
hypothetical "volatility swap" p/l
realised volatility
Source : JPMorgan
Figure 10 : … but for large swings in volatility, the convexity of the
variance swap payout becomes important
p/l for 100K vega notional, strike of 20
2,000,000
0
2,000,000
4,000,000
6,000,000
8,000,000
10,000,000
12,000,000
0% 10% 20% 30% 40% 50% 60% 70%
v ariance swap p/l
hypothetical "volatility swap" p/l
realised volatility
Source : JPMorgan
In common with long options positions, the loss from a long variance swap is limited, whereas the loss from a short
variance swap is not (unless capped). In the case of options, the loss from the long is limited to the premium paid. In the
case of variance swaps the maximum loss occurs when realised volatility is zero, but in this case the convexity of the payout
works in favour of the long with the result that the maximum vega loss on a long variance swap is half the strike
(Example 3). For the short, losses are potentially unlimited, unless the variance swap payout is capped. The standard cap,
if applied, is 2.5x the variance strike (see Section 1.8 for details of caps). In this case, it turns out that the maximum vega
loss on a short 2.5x capped variance swap is 2.625 times the strike (see Example 3).
Example 3: Maximum loss of a variance swap
Suppose a 1year variance swap is stuck at 20 with a vega notional of €100,000.
For a long variance swap the maximum loss will occur when the realised volatility is zero.
In this case p/l = €100,000 x (0
2
– 20
2
) / (2 x 20) = €100,000 x 10 =  €1M
This is equal to a loss of ½ times the 20 strike or 10 vegas, as expected.
For a 2.5x capped short variance swap the volatility on which the variance swap pays out will be capped at 50 = 2.5 x
strike. The maximum loss will occur when realised volatility is at least 50.
This gives a maximum loss of €5.2M = €100,000 x (50
2
– 20
2
) / (2 x 20) = 2.625 x strike x vega notional.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
13
1.5: Variance swap example – accruing realised volatility
In this example we show the evolution of the p/l of a short Euro Stoxx 50 variance swap through the accrual of realised
variance, using prevailing market prices. The variance swap was sold on 14 October 2005, with a strike of 16.5 for
€100,000 of vega notional (equivalent to €3,030 variance notional). The variance swap maturity was 20 (business) days,
expiring on the 10th November 2005.
Over the 20day period realised volatility was 14.3%, 2.2% lower than the level of implied sold. Since the vega notional
gives the expected p/l for a 1 point (1%) move in volatility, we would expect a p/l from the short position of €220,000 (=2.2
times the vega notional). However, due to the convexity of the payout working in favour of the long, and against the short,
the actual p/l is slightly less: €206,714.
Table 2: Daily p/l and cumulative realised volatility from a 20day short variance swap struck at 16.5
Day Date Index Level return log return squared
daily volatility
(annualised) daily p/l (€)
accrued realised
volatility (annualised)
accrued
p/l (€)
13Oct05 3331.4
1 14Oct05 3349.6 0.5% 0.000030 8.6% 29,962 8.6% 29,962
2 17Oct05 3356.8 0.2% 0.000005 3.4% 39,494 6.6% 69,457
3 18Oct05 3334.8 0.7% 0.000043 10.4% 24,777 8.1% 94,234
4 19Oct05 3279.6 1.7% 0.000278 26.5% 65,070 15.0% 29,165
5 20Oct05 3284.8 0.2% 0.000002 2.5% 40,304 13.4% 69,469
6 21Oct05 3271.0 0.4% 0.000018 6.6% 34,554 12.6% 104,023
7 24Oct05 3310.1 1.2% 0.000141 18.9% 12,622 13.6% 91,401
8 25Oct05 3292.5 0.5% 0.000028 8.5% 30,377 13.1% 121,778
9 26Oct05 3304.3 0.4% 0.000013 5.7% 36,403 12.5% 158,181
10 27Oct05 3241.1 1.9% 0.000372 30.6% 100,824 15.3% 57,357
11 28Oct05 3246.0 0.1% 0.000002 2.4% 40,400 14.6% 97,757
12 31Oct05 3320.1 2.3% 0.000510 35.9% 153,615 17.4% 55,858
13 01Nov05 3312.5 0.2% 0.000005 3.7% 39,194 16.8% 16,664
14 02Nov05 3320.6 0.2% 0.000006 3.9% 38,935 16.2% 22,271
15 03Nov05 3361.6 1.2% 0.000151 19.5% 16,312 16.4% 5,959
16 04Nov05 3355.3 0.2% 0.000004 3.0% 39,880 15.9% 45,839
17 07Nov05 3362.8 0.2% 0.000005 3.6% 39,318 15.5% 85,157
18 08Nov05 3361.7 0.0% 0.000000 0.5% 41,210 15.0% 126,367
19 09Nov05 3355.8 0.2% 0.000003 2.8% 40,040 14.6% 166,407
20 10Nov05 3361.0 0.2% 0.000002 2.5% 40,307 14.3% 206,714
206,714
Note how most days were low volatility and produced modest positive returns for the short variance position. However, a
few days with relatively large moves caused significant daily losses (Table 2, Figure 12). This pattern is common: short
variance will profit modestly most of the time, but will lose heavily on large moves.
Figure 11 : Index performance over the lifetime of the variance swap
Euro Stoxx level
3200
3250
3300
3350
3400
13Oct 20Oct 27Oct 03Nov 10Nov
index level
Source : JPMorgan
Figure 12 : Accrued volatility vs. daily returns
Accrued realised volatility daily return
0%
5%
10%
15%
20%
13Oct 20Oct 27Oct 03Nov 10Nov
0%
1%
2%
3% absolute daily return (rhs)
realised v olatility to date (lhs)
Source : JPMorgan
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
14
1.6: Variance swap marktomarket
Although the example on page 14 illustrates how the variance swap accrues realised volatility over time, if an investor
wishes to close out a variance swap position before maturity they must also consider any change in value of the
exposure to volatility over the remainder of the variance swap term. Indeed for variance swaps valued relatively soon
after inception, this change in value of future expected variance can be the main driver of p/l, an especially important
consideration for long dated variance swaps.
Marking to market of variance swaps is easy: variance is additive. At an intermediate point in the lifetime of a variance
swap, the expected variance at maturity is simply the timeweighted sum of the variance realised over the time elapsed, and
the implied variance over the remaining time to maturity.
All that is needed to compute the marktomarket of a variance swap is:
• The realised variance since the start of the swap; and
• The implied variance (variance strike) from the present time until expiry
Since the variance swap is usually settled at maturity, a discount factor between the present time and expiry is also required.
Example 4: Variance swap marktomarket
Suppose a 1year variance swap is stuck at 20 with a vega notional of €100,000 (variance notional of €2,500).
If the volatility realised over the first 3 months is 15%, but the volatility realised over the following 9 months is 25%, then,
since variance is additive, the variance realised over the year is:
Variance = [ ¼ x 15
2
] + [ ¾ x 25
2
] = 525 (≈22.9 volatility).
At expiry the p/l would be €2,500 x (22.9
2
– 20
2
) = €312,500
Now, suppose again that realised volatility was 15% over the first 3 months. In order to value the variance swap after 3
months we need to know both the (accrued) realised volatility to date (15%) and the fair value of the expected variance
between now and maturity. This is simply the prevailing strike of a 9month variance swap. If this is currently trading at
25, then the same calculation as above gives a fair value at maturity for the 1year variance swap of €312,500.
Although the fair value at maturity (now 9 months in the future) is €312,500, we wish to realise this p/l now (after 3
months). It is therefore necessary to apply an appropriate interest rate discount factor.
If, after 3months, 9month LIBOR is 4%, the discount factor applied would be 1/(1+¾ x 0.04) = 0.97. Hence the markto
market p/l after 3 months would be equal to about €303,400.
Box 4: Variance swap marktomarket
0 T (1Y)
σ
t
(=15%)
Knew (=25)
Kold (=20)
Variance
swap
inception
t (3m)
At time t (3m), valuation is a function of
realised variance before t, and implied
variance from t to T
Variance
swap expiry
(1Y)
p/l = N
var
x ( [σ
t
2
–K
old
2
] x %time elapsed + [K
new
2
– K
old
2
] x %time remaining )
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
15
Box 5: Calculating the variance swap marktomarket
Let K
s,t
represent the strike for a variance swap at time s maturity at time t.
Let σ
s,t
be the realised volatility between time s and time t.
Then the p/l (in terms of variance notional) at time t, of a variance swap maturing at time T, is given by:
Variance (forward) p/l at time t (per unit variance notional): = ( ) ( )
2
, 0
2
,
2
, 0
2
, 0 T T t T t
K K
T
t T
K
T
t
−
−
+ − σ
( )  
2
, 0
2
,
2
, 0
1
T T t t
K K t T t
T
− − + = σ
Example 5: Variance swap marktomarket continued
Returning to Example 4 we can calculate the marktomarket p/l (at maturity) directly from the formula above as follows:
p/l = €2,500 x ( [15
2
– 20
2
] x ¼ + [25
2
– 20
2
] x ¾)
= €2,500 x ( 175 x ¼ + 225 x ¾ )
= €2,500 x 125
= €312,500
Applying the interest rate discount factor of 0.97 as explained in Example 4 gives a marktomarket p/l of €303,400
Box 6: Variance swap marktomarket through daily accruals
An alternative approach to calculating marktomarket is to consider a variance swap as a sequence of daily variance swaps.
On each day the p/l accrued by a long variance swap is given by:
Daily accrual on day i (per unit of variance notional) = ) (
1
2
, 0
2
1 , T i i
K
T
−
−
σ ,
where
1 , − i i
σ is the daily annualised realised volatility equal to


.

\

−1
2
ln 252
i
i
P
P
and
2
, 0 T
K is the initial strike.
Each daily p/l is given by a daily variance exposure equal to the total variance notional divided by the total number of days
between the start and end dates of the swap. It is important to notice that the total remaining exposure to variance
decreases linearly with time. Calculating a marktomarket or closing out an existing position requires entering into a new
but opposing variance swap position in the remaining variance notional, not the initial notional. This is done in order to
cancel out the remaining exposure to the daily realised volatility component. The remaining variance notional size will be
equal to (Tt)/T times the original notional, proportional to the time remaining to maturity on the swap. Thus p/l of the
trade at time t will be equal to:
) (
1
) (
2
, 0
2
,
1
2
, 0
2
, 0 T T t
T
t
T t
K K
T
K
T
t
− + −
∑
+
σ
The first term represents the realised variance accrued by the variance swap up to time t, whilst the second term represents a
fixed stream of daily accruals up to expiry at time T. This is exactly equivalent to the marktomarket equation given above
and presents a useful framework for thinking about forward variance, the subject of the next section.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
16
1.7: Forward variance
The reasoning used to compute the variance swap marktomarket in the preceding section can also be used to calculate the
expected variance over a forward starting window. Suppose we know the strikes for a shortmaturity variance swap
expiring at time t, and a longer maturity variance swap expiring at time T. We want to find the expected realised variance,
F, between time t and time T. Since variance is additive, the longer maturity variance swap is simply the timeweighted
sum of the short maturity variance and the expected variance over the forward period, thus enabling us to compute this
expected forward variance. This equates to the fair strike of the forwardstarting variance swap.
Example 6: Calculating forward variance
Suppose that one year variance is struck at 20, but 3month variance is struck at 15. What is the fair strike for a variance
swap starting in 3 months time, and expiring 9 months later?
Since variance is additive, the variance expected over the year is the time weighted sum of the variance expected in the first
three months, and that expected in the remaining nine months. This second value is the forward strike F. Then, using the
additivity of variance as for the marktomarket calculation:
20
2
= [¼ × 15
2
] + [¾ × F
2
]
Solving for F, gives a strike for the forward variance of 21.4.
Box 7: Calculating forward variance through daily accruals
Using the approach that a variance swap can be thought of as a strip of daily variance swaps (see p16), a combination of
long and short variance swaps can be used to cancel out exposure to realised volatility over a particular time period. This
shows how a forward variance swap can be constructed, giving exposure to realised volatility starting after a future date.
To illustrate this we being by considering the p/l of a long variance swap, which because of the additivity of variance can be
spilt into two periods (1 t and t+1 T) as follows
) ( ) (
2 2
1 ,
1
2 2
1 ,
1
T i i
T
t
T
T i i
t
T
K
T
N
K
T
N
− + −
−
+
− ∑ ∑
σ σ where N
T
is the variance notional and
1 , − i i
σ is realised daily volatility
Consider the p/l for a short variance swap, maturity t, of variance notional
T
N
T
t
(with delayed payment). This is equal to:
) (
2 2
1 ,
1
t i i
t
T
K
t
N
T
t
− −
− ∑
σ which will offset the realised variance from the second term in the p/l from the long above.
Combining the long and short variance positions gives an overall p/l of: ) ( ) (
2 2
1 ,
1
2 2
1
T i i
T
t
T
T t
t
T
K
T
N
K K
T
N
− + −
−
+
∑ ∑
σ
The first p/l term represents a fixed quantity which can be subtracted from the fixed strike in the second p/l term. We then
define a forward variance to be equal to this combined strike, such that
2 2 2 2 2
1
2 2 2 2
,
) ( ) (
1
t T T t T
t
T t T T t
K
t T
t
K
t T
T
K K
t T
t
K K K
t T
K F
−
−
−
= −
−
− = −
−
− =
∑
The overall p/l becomes ) (
2
,
2
1 ,
1
T t i i
T
t
T
F
t T
N
T
t T
−
−
−
−
+
∑
σ
This just represents the p/l of a variance swap starting a time t, ending at time T of notional size:
T
N
T
t T −
.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
17
The calculations in Box 7 show that a long position in forward starting variance (of unit variance) can be constructed from:
• Long T/(Tt) variance notional of spot variance maturity T; and
• Short t/(Tt) variance notional of spot variance maturity t, but with payment delayed until maturity T.
Box 8: Calculating forward starting variance
Suppose K
t
is the strike of the shortmaturity variance swap (of maturity t),
K
T
is the strike of the longermaturity variance swap (maturity T), and
F
t,T
is the strike of the forward starting variance swap (with exposure to variance between t and T), then:
2
,
2 2
T t t T
F
T
t T
K
T
t
K
−
+ = (as in the marktomarket calculation in Box 4)
Solving for F gives a value for the forward variance:
2 2 2
, t T T t
K
t T
t
K
t T
T
F
−
−
−
= or
2 2
, t T T t
K
t T
t
K
t T
T
F
−
−
−
=
Example 7: Forward variance construction
As in Example 5, suppose a 3month variance swap is struck at 15, and a 1year variance swap is struck at 20.
The 9month variance swap, 3month forward variance will be struck at 21.4.
Suppose we wish to trade €100,000 of forward variance swap notional.
As in the case of a spot variance swap, this equates to €100,000 / (2 x 21.4) = €2,336 variance notional.
Hence we require 3/9 x €2,336 = €778 variance notional of the 3month variance swap (with payment delayed); and
12/9 x €2,336 = 3,115 variance notional of the 12month variance swap.
This in turn equates to short 2 x 15 x €778 = €24,000 vega notional of the 3month variance swap; and
long 2 x 20 x €3,115 = €125,000 of the 12month variance.
One of the things worth noting on forward starting variance, is that more variance is need on the longer leg (which tends to
be less liquid) than the shorter leg and that the total notional of the two legs will be greater than the notional of the
forward. This can have a practical impact of increasing the bid/offer of forward starting variance swaps. However, as the
variance swap market evolves and matures, forward starting variance is increasingly quoted as a single level rather than a
weighted spread with the consequence that bid/offers are converging to similar levels as seen for spot variance.
The ability to reexpress the spot variance term structure into a strip of forward starting variances raises natural analogies
with fixed income – in particular the swap curve and a strip of forwards/futures. Forward variance swaps are useful for
taking a direct view on the future value of implied variance and/or the future shape of the variance term structure curve in
the same way that Eurodollar and Euribor futures are used to take view on the future level of interest rates. Furthermore the
fixed income concepts of slide and carry are useful concepts to apply to both spot and forward starting variance swaps. See
Sections 3.8 and 3.9 for examples of the uses of forward variance swaps.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
18
1.8: Variance swap contract specifications
Variance swap conditions are set out on term sheets such as the one shown in Section 1.9. Strikes are typically expressed as
100 times the annualised volatility represented – for example 25 will represent a breakeven realised volatility of 25%. For
this reason the calculated value of the realised variance (using an RMS calculation, see p10) must also be multiplied by 100.
Margins and collateral
Variance swaps are usually margined in a similar manner to options, with an initial amount to be posted as collateral (3
times the vega amount in the example termsheet below). Further margin calls will be made during the course of the trade
as necessary. Settlement is calculated at maturity and cashflows exchanged shortly afterwards (T+2 in the example below).
Disrupted days
The realised variance used in a variance swap payout is calculated from closing (or official observed) prices on observation
dates over a specified period. Observation dates are defined as all those scheduled trading days which are not disrupted.
The occurrence of a disrupted day could potentially work for or against a long variance position. Suppose a 5% loss is
followed by a 6% gain over 2 consecutive trading days. If the first day is declared as disrupted, its closing price will not be
used in the calculation of the realised volatility and only the single combined return of 0.7% will be counted (see Example
8). Conversely, if a 5% loss on a day declared disrupted is followed by a 6% loss the next day which is nondisrupted, this
will count as a single, compounded 1day loss of 10.7%. Such a move will act to increase realised volatility more than by
the separate 5% and 6% moves, and will work in the favour of the long.
Example 8: Disrupted days
On Wednesday, 18 January 2006 the Tokyo Stock Exchange (TSE) closed early following revelations of the investigation
into Livedoor. At the point when trading was suspended on the TSE the Nikkei was at 15341, down 2.9% on the previous
day’s close of 15806. However, the day was declared disrupted, and the provisional closing price was not accepted for
variance swap calculations.
The next day, Thursday, January 19th, was considered a normal trading day, and the Nikkei recovered most of the previous
day’s losses, closing at 15696, up 2.3% on the last level on the disrupted Wednesday.
For the purposes of the variance swap calculation, only the change from the close on the Tuesday to the close on the
Thursday, a return of 0.7% counted towards the calculation of realised volatility. A long variance swap position would
therefore have lost out due to the Wednesday being declared a disrupted day.
Index reconstitution risk
Variance swaps on indices are defined to pay out on the returns of the index and not on the weighted returns of the basket of
current constituents. This means that index variance swaps (especially longdated ones) will be exposed to the risk of index
reconstitution, and the variance swap may end with an exposure to a very different set of stocks, potentially with different
volatility characteristics than when it was originally traded. For example, in the Hang Seng, the cap on the weight of
HSBC, a low volatility stock, and the forthcoming addition of a number of potentially higher volatility Chinese HShares
may act to increase the overall volatility of the index.
Dividend adjustments
Variance swaps on single names are typically adjusted for dividends, both special and regular. This means that the return
on the exdividend date is calculated after adjusting for the dividend. For example, if a stock is worth €100 on the day
before the dividend, pays a dividend of €5 and closes at €94 on its exdividend day, the return used in calculating the
realised volatility for the variance swap payout will be 94/951 = 1.05% (not 94/1001 = 6%). This means that changes
to the underlying due to dividend payments do not count towards the realised volatility calculation for the variance
swap. This can be important for stocks with high dividend yields which pay entire annual dividend in a single payment –
e.g. Italian stocks such as Telecom Italia and ENEL (Figure 13).
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
19
By convention index variance swaps are not usually adjusted for dividends – since the payments are usually more spread
out and the impact of a few index points worth of dividend is generally small in comparison to the average daily move.
However, there are cases of indices containing high dividend paying stocks some of which pay on the same day – e.g. the
SPMIB and the SXKE (the Eurozone Telecoms index) – where index moves as a result of dividend payments can be
significant. In these cases the extra volatility resulting from dividend payments will count in the calculation for the variance
swap payout, potentially giving an artificial boost to the realised volatility.
Variance swap caps
Variance swaps, especially on singlestocks (and sector indices), are usually sold with caps. These are often set at 2.5 times
the strike of the swap capping realised volatility above this level. Variance swap caps are useful for short variance
positions, where investors are then able to quantify their maximum possible loss (to 2.625 times the strike, see Box 9).
Box 9: Variance swap caps
Suppose a variance swap is traded with a cap set at 2.5 times the strike. Then the p/l (for the long) is given by:
{ } ( )
K
K K Min
N l p
Vega
2
5 . 2 ,
/
2 2
−
× =
σ
Note that this sets the maximum loss for the short to
( )
K N
K
K K
N
Vega Vega
625 . 2
2
5 . 2
2 2
× =
−
×
In practice caps are rarely hit – especially on index underlyings and on longerdated variance swaps. When caps are hit, it is
often due to a single large move – e.g. due to an M&A event or major earning surprise on an individual name, or possibly
from a dramatic selloff on an index. Singleday moves needed to cause a variance swap cap to be hit are large and increase
with maturity. For example a 3month variance swap struck at 20 and realising 20% on all days except for one day which
has a oneoff 24% move, will hit its cap. A similar 1month maturity swap would need only a 14% move to hit the cap,
whereas the required 1day move on a 1year swap would be 46%. For lower strikes the required moves are also lower.
Closing out variance swaps
Investors wanting to realise marktomarket p/l by closing out a variance swap before expiry generally have two choices.
They can either enter into an opposing offsetting contract, potentially with a different counterparty, or they can agree with
the original counterparty on a level at which to close out the contract. In theory this second alternative is simpler as it
removes all collateral obligations, frees up capital and avoids any technical issues with changes in cap level (see Example
9). However, in practice it is possible to negotiate bespoke contracts specifically to close out existing positions in a process
known as novation which allows for the different counterparties to close out the contract fully.
Example 9: Offsetting capped variance swaps
Suppose that in an investor buys a 6month variance swap with a strike of 20%.
This has the standard 2.5x cap meaning the exposure to realised volatility will be capped at 50%.
1month later, the (5month) variance swap is now trading at a strike of 30% leading to a significant marktomarket p/l –
the exact amount depending on the level of volatility realised over the month. The investor wants to lock in this profit.
With the strike now at 30%, the cap on an new variance swap contract will by default be set at 2.5 x 30% = 75%
Then if the investors sells this 30strike variance swap in an attempt to close out his position the difference in caps will
mean he takes on a short volatility exposure if the subsequent realised volatility is above 50% (although capped at 75%)
In effect, in the course of trying to close out his position, he will have sold a ‘50%/75% call spread’ on volatility. Whilst
the price he gets for selling the variance swap will reflect this higher cap, the residual volatility exposure is presumably
unwanted, and the investor would be best either trading directly with the original counterparty or negotiating a bespoke
contract with another counterparty in order to fully close out his outstanding contract.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
20
M&A events
M&A events present particular issues for variance swaps, as an underlying being acquired can effectively cease to exist.
The various options exchanges have rules for how such an event is dealt with, and variance swap contracts generally follow
these rules. Investors should be aware in which exchange the variance swap is based since M&A rules are exchange
specific. For share offers, or composed offers consisting of less than 66% cash, the options, and hence the variance swap,
will usually transfer into a variance swap on the new underlying after the transaction date.
In the case of cash bids, options are unwound immediately after the offer has been declared unconditional. In Europe, some
bourses unwind vanillas at intrinsic and others at the theoretical fair value (generally taking the implied volatility to be the
average over the 5 days before the bid). The Euronext exchange unwinds at fair value; Eurex unwinds at intrinsic but is
switching to fair value from the beginning of 2007; IDM in Italy unwinds at intrinsic and MEFF in Spain works on a case
by case basis. In any case documentation should either specify the exact rules in the case of an M&A event or should refer
to the rules of the appropriate exchange.
One further issue with cash bids, is that the underlying will usually trade at close to the bid level after the offer has been
accepted, often with close to zero volatility. However, the options are not unwound until the offer is declared
unconditional. For example when Telefonica bid for O2 on 31 October 2005, it took until 3 January 2006 for the offer to go
unconditional (Figure 14). Although outstanding options were settled at the implied volatility prevailing before the October
bid, they were not unwound until the beginning of January, leading to two months of virtually zero accrued volatility on any
outstanding variance swaps. In this case, given the size of the daily move on the day of the bid (+ 25%) the 3month
realised volatility from the beginning of October up until the unconditional date on January 3rd, was around 50%, well in
excess of levels of implied variance trading before the bid. However, in cases where the move on the bid is smaller, and the
time up until unconditional acceptance is longer, the low realised volatility over this period could hurt long variance swaps.
Figure 13 : The apparent drop in spot and consequent spike in
volatility was entirely due to the payment of a single 6.3% dividend.
Share price (€) 1m realised volatility
6.5
7.0
7.5
8.0
Nov05 Feb06 May06 Aug06 Nov06
0%
5%
10%
15%
20%
25%
ENEL IM Equity
1M realised v olatility (incl. div pay ment)
1M realised v olatility (ex cl. div pay ment)
dividend
paid
Source : JPMorgan , Bloomberg
Figure 14 : O2 realised volatility collapsed following the bid. But the
large 1day move from the bid was large enough to keep realised
volatility high
Share price (p) 1m realised volatility
100
120
140
160
180
200
220
Aug05 Sep05 Oct05 Nov05 Dec05 Jan06 Feb06Mar06
0%
20%
40%
60%
80%
100%
OOM LN Equity
1M realised v olatility
3M realised v olatility
cash bid
offer goes
unconditional
Source : JPMorgan, Bloomberg
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
21
1.9: Example variance swap term sheet
Sep06 OTC Variance Swap: FTSE Index Variance
Instrument: Variance Swap
Trade Date: 10th August 2006
Maturity Date: 15th September 2006
Variance Buyer: JPMORGAN
Variance Seller: *******
Underlying: FTSE100 INDEX (Bloomberg: UKX Index)
Denominated Currency: Great British Pounds (“GBP”)
Vega Amount: GBP 100,000
Strike Price: 16.95
Cap Level: 36.95
Equity Payment: T+2 after the Observation End Date, the Equity Amount will be calculated and paid in
accordance with the following formula:
( )  
2 2 2
Price Strike , FRV Min
Strike 2
Amount Vega
− × 
.

\

×
CapLevel
If Equity Amount is positive the Variance Seller will pay the Variance Buyer the Equity Amount.
If Equity Amount is negative the Variance Buyer will pay the Variance Seller an amount equal to
the absolute value of the Equity Amount.
Where FRV =
100
_
ln 252
1
2
1
×


.

\

×
∑
= −
N Expected
P
P
N
t t
t
t means each Observation Day.
N means number of Observation Days.
Expected_N means 25, the number of days which, as of the Trade Date, are expected to be Scheduled Trading Days in the
Observation Period.
ln means the natural logarithm.
P
t
means in respect of: (i) the final Observation Day, the Official Settlement Price of the Exchangetraded Contract on the
Observation End Date; and (ii) each other Observation Day, the official level of the Index at the Valuation Time on such
Observation Day.
P
t1
means in respect of: (i) the first Observation Day, the Settlement Price per Underlying on the Observation Start Date and;
(ii) each subsequent Observation Day, the Settlement Price per Underlying on the Observation Day immediately preceding
such Observation Day.
Observation Day Each Scheduled Trading Day that is not a Disrupted Day during the Observation Period.
Observation Period The period from, but excluding, the Observation Start Date to, and including, the Observation End Date.
Observation Start Date Trade Date.
Observation End Date Maturity Date.
Settlement Price means the official price or, if there is no official price, the midmarket price per Underlying on the Exchange
at the scheduled weekday closing time (without regard to after hours or any other trading outside of the regular trading session
hours) on an Observation Day.
Calculation Agent: JPMorgan Securities Ltd.
Documentation: ISDA
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 1: Variance Swap Mechanics
22
23
2. The Variance Swap Market
• 2.1 Market development
• 2.2 Historical prices
• 2.3 Variance swaps and option volatilities
• 2.4 Pricing rules of thumb
• 2.5 What drives variance swap levels?
• 2.6 The volatility risk premium
• 2.7 Variance as a predictor of future volatility
• 2.8 Is variance swap convexity fairly priced?
• 2.9 Variance term structure
P
a
r
t
2
:
T
h
e
V
a
r
i
a
n
c
e
S
w
a
p
M
a
r
k
e
t
Part 2: The Variance Swap Market
Part 2 begins with an overview of the development and current state of the variance swap market. We then look at where
variance swap levels have traded historically, concentrating particularly on the Euro Stoxx 50 and S&P500 indices. We
leave the detailed theoretical pricing argument until later (see Part 4), but discuss briefly the relationship between variance
levels and traded option volatilities and consider some pricing rules of thumb.
Variance swaps can be thought of as the market’s best estimate of future realised variance and we consider how good a
job they do of predicting this. We discuss the convexity premium for variance swaps in terms of the spread of variance
swap prices over ATM options implied volatility.
We also consider the drivers of variance swaps prices: principally realised volatility and risk aversion at shorter
maturities and structured product flow at the longer end of the curve. This leads in to a discussion of the variance swap
term structure and curve dynamics.
2.1: Market development
Variance swap contracts were first mentioned in the 1990’s, but like vanilla options only really took off following the
development of robust pricing models through replication arguments. (See e.g. Derman, 1999 or “Towards a Theory of
Volatility Trading, Carr and Madan’, 2001). The directness of the exposure to volatility and the relative ease of replication
through a static portfolio of options make variance swaps attractive instruments for investors and marketmakers alike.
The variance swap market has grown steadily in recent years, driven by investor demand to take direct volatility exposure
without the cost and complexity of managing and deltahedging a vanilla options position. Although it is possible to
achieve the variance swap payoff through a portfolio of options, the variance swap contract offers a convenient package
bundled with the necessary deltahedging, offering investors a simple and direct exposure to volatility, without any of the
path dependency issues associated with deltahedged options.
As liquidity in the variance swap market has increased, to a certain extent levels now deviate from the theoretical price
obtained from a replicating portfolio of options. This is partly because the construction of the replicating portfolio includes
a relatively large contribution from the deep outofthemoney options and the reduced liquidity in these options gives rise
to some discrepancy in the overall pricing of variance swaps. We discuss practical replication issues in more detail in
Section 4.9.
Variance swaps were initially developed on index underlyings. In Europe, variance swaps on the Euro Stoxx 50 index are
by far the most liquid, but DAX and FTSE are also frequently traded. Variance swaps are also tradable on the more liquid
equity underlyings – especially Euro Stoxx 50 constituents, allowing for the construction of variance dispersion trades.
Variance swaps are tradable on a range of indices across developed markets and increasingly also on emerging markets.
Table 3: Global variance swap liquidity
Major indices on which variance is
traded
Typical index bid offers ,vegas
(most liquid maturities)
Typical singlestock bid offers, vegas
(most liquid stocks and maturities)
Developed Europe Euro Stoxx 50, FTSE, DAX, CAC 0.4 – 0.6 1 – 2
U.S. S&P500, Nasdaq, Russell 2000 0.4 – 1 1 – 2
Japan Nikkei 225, Topix 0.4 – 0.6 2 – 2.5
Asia Ex. HSI, HSCEI, Kospi 1 – 2 2  4
Emerging Markets CECE, Wig20, RDX, Top40, Bovespa 3 – 5 5 – 6
Source: JPMorgan estimates, November 2006
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
24
Bid/offer spreads have come in significantly over recent years and in Europe they are now typically in the region of 0.5
vegas for indices and 1 – 2 vegas for singlestocks – although the latter especially vary according to liquidity factors. We
expect spreads to come in further in future as liquidity continues to improve. Spreads for indices in the US and Japan are
similar with single stock bid/offers in the 2 – 2.5 vegas range. Spreads are naturally higher in emerging markets although
these too are becoming much more liquid. With volatilities often much higher and less stable than in developed markets,
emerging market variance can offer interesting opportunities.
The most liquid variance swap maturities are generally from 3 months to around 2 years, although indices and more liquid
stocks have variance swaps trading out to 3 or even 5 years and beyond. Maturities generally coincide with the quarterly
options expiry dates, meaning that they can be efficiently hedged with exchangetraded options of the same maturity. Good
liquidity is also seen in the front 3 months for shortdated index variance. The VIX, VSTOXX and VDAX indices
represent the theoretical prices of 1month variance swaps on the S&P500, Euro Stoxx and DAX indices respectively, and
are calculated by the exchanges from listed option prices, interpolating to get 1month maturity. These volatility indices are
widely used as benchmark measures of equity market risk, even though they are only shortdated measures and are not
directly tradable. See p94 for further details.
Typical sizes of variance swap trades are around € 100,000 – 200,000 of vega for index underlyings and € 50,000 vega
notional for singlestock underlyings. As with any OTC market, estimating total traded volumes is problematic, with recent
estimates for daily traded volumes on indices in the region of $35 million vega (source: Risk Magazine). This represents
an estimated 5 or 10fold increase on volumes three years ago. Traders in Europe and Japan estimate that approximately
1030% of the vega traded in the market is done through variance swaps, with a slightly lower proportion in the U.S.
Forward variance and index variance spreads (e.g. Euro Stoxx 50 vs. S&P500) are increasingly quoted with a single
bid/offer spread, often comparable to, or only slightly higher than, the spread on a single leg, meaning the second
component of the spread can effectively be traded at midmarket levels. These developments make forward variance and
index spreads much more attractive for investors who may previously have found the transaction costs prohibitive.
JPMorgan publishes its variance swap runs, including prices for forwards and spreads, and conditional variance prices, on
Bloomberg JPVS.
The advent of the variance swap market has allowed the investor community to trade variance directly, without the onerous
human costs (and transaction costs) of deltahedging a portfolio of options. Of course, the marketmaking community can
offset variance swap exposure with the advantage of economy of scale. As dedicated ‘volatilityarbitrage’ funds have
grown, variance swaps have allowed them to directly express their views, encouraged in part by the attractive returns
available from shorting variance in 20032004 as equity volatility fell dramatically. In addition, selling stock correlation
using variance swaps (socalled ‘variance dispersion’ trades) has also been a popular trade in the hedge fund community.
More recently, the investor base has widened further with the advent of structured products which embed variance swap
returns. Such products may be leveraged and/or offer capital guarantees, making them attractive to a broader range of
institutions including private banks and sophisticated private investors. For example, JPMorgan’s recently launched
Yield Alpha and Variance Alpha indices incorporate variance swap exposure, with a variety of structured products
offering access to these indices. See Section 3.3 and JPMorgan’s Yield Alpha Strategy Guide for further details.
Variance swap liquidity has so far been concentrated on the equity markets, in part fuelled by the need for investment banks
to recycle volatility and correlation exposures that are embedded in structured products. However variance swaps do trade
on nonequity underlyings – for example FX, interest rates, commodities and credit – albeit with less liquidity. In theory
there is no reason why variance swaps should not gain liquidity in these asset classes, depending on investor demand and
providing that the underlying option markets are liquid enough.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
25
2.2: Historical prices
Figure 15 shows Euro Stoxx 50 implied variance levels between 2000 and 2006 for maturities of 1month, 6months and 3
years. Shortdated variance swap prices tend to be more variable than those for longer maturities, with 1month variance
experiencing highs above 60 and lows of almost 10 over the six year period. Variance swap prices can be observed to
follow high and low regimes in a similar manner to realised variance (Figure 5), and within this period different volatility
regimes are clearly discernible:
• A ‘medium’ volatility regime up until mid 2002, with implied variance priced in the 2030 range, interrupted by a
large spike in September 2001;
• A ‘high’ volatility regime from mid 2002 – mid 2003 with shortdated variance priced consistently above 40;
• A ‘low’ volatility regime since mid 2004 with shortdated variance below 20 (excepting the May 2006 spike).
Longer maturity variance swap levels vary less, with 3year implied variance remaining in the 1535 range, and in particular
reacting less to spikes. A sudden unexpected event likely to dramatically increase shortterm volatility dramatically, is
perhaps less likely to cause the same level of elevated volatility over the next few years.
Figure 15 : Euro Stoxx implied variance levels since 2000
Implied variance
10%
20%
30%
40%
50%
60%
Sep00 Sep01 Sep02 Sep03 Sep04 Sep05 Sep06
1month implied v ariance
6month implied v ariance
3y ear implied v ariance
Source : JPMorgan
Figure 16 : VIX (shortdated S&P500 variance) since 1990
VIX level realised volatility
0
10
20
30
40
50
May90 May93 May96 May99 May02 May05
0%
10%
20%
30%
40%
50%
VIX Index
SPX 3m realised v olatility
Source : JPMorgan
Singlestock variance swaps generally trade at higher levels than index variance, due to the fact that index volatility
will be less that the average of its constituents because of diversification effects. Single stock variance can increase
substantially in times of high company uncertainty, for example ahead of an M&A event or following a profit warning or
unexpected earnings announcements. See Table 4 below for some examples of index and singlestock variance prices.
Figure 16 shows VIX levels since 1990. The VIX gives a proxy for rolling 1month maturity variance swap levels on the
S&P 500 (see p94). Again variance swap levels can be seen to be well correlated with realised volatility both in the
short term (including spikes) and in the long run, following the underlying pattern of volatility regimes.
Note also that variance swaps tend to trade somewhat above levels of realised volatility – in part reflecting the volatility
risk premium, with investor risk aversion and hedging programmes in the market supporting demand for index options
leading to a persistent overpricing of (index) variance. Below we discuss the two principal reasons why variance swaps
tend to trade above realised volatility.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
26
In general, index variance swap strikes tend to trade above the level of prevailing realised volatility, for two distinct reasons.
1. Variance swaps are convex in volatility and the variance swap buyer should fairly pay for this convexity,
meaning that equity variance swaps trade above implied ATM volatility.
2. The volatility risk premium mentioned above, which is particularly noticeable at the index level, and is manifest
in the premium of implied ATM volatility to realised volatility.
Of these two factors, the second applies both to variance swaps and vanilla options, whereas the first applies only to
variance swaps. Thus a simple framework has (ATM) vanilla option implied volatilities trading above prevailing levels of
realised volatility due to the volatility risk premium; and then variance swap levels trading above ATM options implied
volatility due to the additional premium required to pay for the convexity. Typically, at least for shortdated index variance,
the convexity premium is worth perhaps 12 vegas with the volatility risk premium around 23 vegas, representing the
spread of ATM implied volatility over realised volatility.
The following sections consider the drivers of variance swap prices, with the discussion centring around attempts to
quantify, and to some extent value, the contributions of these two factors – convexity and volatility risk premium –
presented above. We begin by considering the relationship between ATM options prices and variance swap levels for
various underlyings, demonstrating that they are well correlated (both across underlyings and over time). The difference
comes down to value of the convex payout which is represented by the convexity of the volatility skew. This leads into a
discussion of rules of thumb for pricing variance swaps based on vanilla options prices – using assumptions about the skew
to produce an estimate for the variance swap price which in theory is priced using the entire skew curve.
We go on to explicitly consider the drivers of variance swap prices: principally realised volatility/variance and the
convexity and riskpremium discussed above as the main reasons for variance swaps to trade above these levels. Risk
aversion (driving the risk premium) can perhaps be measured in the market as the excess demand for put options or its
natural effect – increasing put skews. Note that this kind of increase in risk aversion also increases skew convexity, leading
to a higher implied volatility of volatility and more value being assigned to the convex variance swap payout.
In Section 2.6 we explicitly discuss and attempt to measure the volatility risk premium. This risk premium can partly be
explained by the asymmetry of returns of volatility (and variance) products: longs occasionally make large gains but mostly
suffer small losses, whereas a short position can be likened to selling insurance (similar to selling credit protection) with a
steady income punctuated with occasional large drawdowns. Indeed, the volatility risk premium can be viewed as a
potential source of alpha, exploited by strategies which systematically short volatility, for example selling index variance
swaps or traditional index covered call strategies.
Although convexity and the risk premium cause variance swaps to trade above prevailing levels of realised volatility, a
simple regression demonstrates that variance swaps tend to give a better estimate of future volatility (after adjusting for the
bias) than that given by prevailing levels of realised. This is especially true at short maturities where market information
priced into the variance swap market, but not represented in realised volatility can give variance swaps some real forward
looking attributes.
In Section 2.8 we turn our attention to convexity, the other cause of variance swaps trading above realised volatility, and the
reason why they trade above ATM implied volatility. The value of this convex payout depends on how much future
realised volatility is likely to deviate from the variance swap strike, with large potential moves in volatility increasing the
value of the variance swap. The convexity premium should thus be linked to the ‘volatility of volatility’ and by comparing
the spread of variance over ATM implied volatility with realised volatility of volatility we find that this is the case.
Nevertheless, at least for S&P500 variance, we find like volatility, that ‘volatility of volatility’ also tends to be overpriced:
perhaps of ‘volatility of volatility risk premium’?
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
27
2.3: Variance swaps and option volatilities
Variance swap strikes are well correlated with standard BlackScholes implied volatility derived from options prices.
This should not be surprising, since both can be interpreted as market estimates of future volatility. However, ATM option
volatility reflects the market estimate of future volatility realised around the current level, whereas variance swaps represent
the market estimate of variance, independent of future market level. As explained below this means that variance swaps
tend to trade somewhat above levels of ATM volatility.
Table 4 shows ATM volatility and variance swap levels for a number of European stocks and indices. We have used
indicative midmarket levels as of 24th October 2006. Note that these variance swap levels are typically about 12 vegas
above ATM volatility. Figure 17 and Figure 18 show the relationship between ATM volatility and variance swap strikes
for the members of the Euro Stoxx 50. Whilst all variance swaps price above their corresponding ATM volatility, there is
variation in the spread, in part due to the differing skews. This spread of variance above ATM volatility becomes larger and
more unpredictable at longer maturities, as the effect of the particular shape of the skew becomes more important.
Table 4: ATM volatility and variance levels (as of COB 24 Oct 2006)
Ref Dec06 Mar07 Dec07 Dec09 Dec11
ATM
implied
volatility
Variance
swap
strike
ATM
implied
volatility
Variance
swap
strike
ATM
implied
volatility
Variance
swap
strike
ATM
implied
volatility
Variance
swap
strike
ATM
implied
volatility
Variance
swap
strike
Euro Stoxx 50 4,014.0 12.9% 14.2% 14.7% 15.9% 16.2% 17.3% 18.3% 19.1% 19.8% 20.4%
FTSE 6,182.5 11.3% 12.4% 12.6% 13.8% 14.3% 15.1% 16.5% 16.9% 18.4% 18.4%
SMI 8,662.6 11.9% 12.8% 12.8% 14.0% 13.8% 15.0% 15.3% 16.1% 16.5% 17.1%
DAX 6,247.5 13.8% 14.5% 15.8% 16.6% 17.7% 18.1% 20.5% 20.1% 22.6% 21.6%
Nestle 424.0 13.9% 14.3% 14.3% 14.8% 15.4% 16.3% 16.7% 17.9% 17.0% 18.4%
BP 6.08 17.2% 17.9% 18.3% 19.2% 19.9% 21.2% 21.3% 23.2% 22.0% 24.5%
Axa 30.44 21.9% 23.0% 23.2% 24.5% 24.3% 26.3% 25.7% 29.0% 27.7% 32.4%
Ahold 8.26 27.7% 28.5% 26.4% 27.2% 27.1% 28.1% 27.3% 28.2% 28.2% 29.1%
Rio Tinto 28.50 32.9% 34.6% 32.9% 34.8% 33.2% 35.5% 34.1% 37.0% 34.7% 38.2%
Source : JPMorgan
For example note that at 3year maturity (Figure 18) the variance strikes of Ahold and ING are similar (around 27) even
though their ATM implied volatilities differ by more (24% vs. 27%). Whilst this is partly due to the different forwards
(ING has a dividend yield of 3.6%; Ahold doesn’t pay a dividend) the skew surface of Ahold is relatively flat, whereas that
of ING is more convex, due to much steeper put skews. Note also that at 6month maturity the index variance appears
somewhat rich in comparison to its ATM volatility. This is caused by the index having higher put skews than the single
stocks. However, at 3year maturities, the greater upside convexity of the singlestocks becomes more important and the
index variance appears to be priced in line with its ATM volatility in comparison to the single stocks.
Figure 17 : 6m variance swaps trade above 6m ATM implied volatility
6M implied variance
14%
16%
18%
20%
22%
24%
26%
28%
30%
14% 16% 18% 20% 22% 24% 26% 28%
Euro Stox x Members
Euro Stox x Index
CGE FP
AH NA
ENEL IM
BN FP
6M ATM implied volatility
Source : JPMorgan
Figure 18 : ... and longer maturities trade at a higher premium
3Y Implied variance
16%
20%
24%
28%
32%
16% 18% 20% 22% 24% 26% 28% 30%
Euro Stox x Members
Euro Stox x Index
CGE FP
AH NA
ALBK ID
TEF SQ
BN FP
IBE SQ
INGA NA
3Y ATM implied volatility
Source : JPMorgan
line y = x
line y = x
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
28
Although they move in close step with changes in ATM volatility, as noted above, variance swaps invariably trade above
the level of ATM implied volatility of the corresponding maturity. Why is this? There are two ways of looking at this
which both turn out to be the same explanation.
• Variance swaps trade above ATM volatility because you pay extra for the convexity of the variance swap: the gain
from an increase in volatility is more than the corresponding loss from a decrease in volatility (Figure 9,
Figure 10). This does not come free.
• The theoretical price of the variance swap is calculated from the prices of a replicating portfolio of options, so the
variance swap strike can be thought of as a weighted average of vanilla option volatilities (see e.g. p82). In the
presence of skew and skew convexity, average volatilities will usually be above ATM volatility, making the
variance swap more expensive.
Ultimately both of these explanations come down to the fact that a long variance swap is also long volatility of volatility,
due to the convex nature of the payoff discussed previously. This convex payoff is paid for through a replicating portfolio
of options, across the whole range of strikes, which in effect price in this volatility of volatility (see Part 4 for details).
Variance swap prices arise from a kind of weighted average of volatilities across the skew surface. For short dates where
the skew is relatively linear around ATM volatility, and put skews are more significant than call skews, the price can be
thought of as a function of ATM volatility level and the slope of skew. This assumption lies behind Derman’s
approximation to the variance swap level, which is particularly useful for estimating variance swaps where a limited
number of option volatilities are available. See the next section for more details.
In practice, the contribution of the skew component means that variance swap strikes tend to trade at similar levels to OTM
puts – usually around 95100 strike for maturities of around 13 months (Figure 19), and 9095 strikes for maturities of
about a year (Figure 20). The skew, and convexity (in terms of the overall shape of the skew surface) become more
important factors for pricing at longer dates, where the probability of reaching more OTM strikes increases (Figure 18).
Figure 19 : The VSTOXX (theoretical interpolated mid variance price)
generally trades just below the EuroStoxx 1M 95% put volatility
Implied volatility/variance
8%
12%
16%
20%
24%
28%
Jul04 Jan05 Jul05 Jan06 Jul06
Euro Stox x 1M ATM implied v olatility
VSTOXX index
Euro Stox x 1M 95strike implied v olatility
Source : JPMorgan
Figure 20 : 1Y EuroStoxx variance swaps trade closer to 90strike
put volatility
Implied volatility/variance
12%
16%
20%
24%
Jul04 Jan05 Jul05 Jan06 Jul06
Euro Stox x 1Y ATM implied v olatility
Euro Stox x 1Y Variance sw ap strike (estimated)
Euro Stox x 1Y 90strike implied v olatility
Source : JPMorgan
Variance swap pricing is explored in much greater detail in the Part 4, where we outline how the theoretical variance swap
price is arrived at by considering a replication argument using a static portfolio of vanilla options. This enables the variance
swap price to be expressed as a weighted average of option volatilities across the skew surface, with the weighting relating
to the probability of the corresponding option expiring in the money. This makes explicit the precise impact of skew and
convexity on variance levels. See p82 for details.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
29
2.4: Pricing rules of thumb
Whilst it is necessary to have prices available for the entire strip of (OTM) options in order to calculate the true theoretical
price of a variance swap, reasonable approximations for variance swap prices can be made under certain assumptions
about the skew. In this section we introduce some commonly used approximations obtained by assuming linear, log
linear and quadratic skews curves. We discuss their effectiveness and applicability and give examples of their use.
Flat skew
In the hypothetical case where the skew surface is flat meaning that all strikes trade at identical implied volatilities – the
variance swap level will be the (constant) implied volatility level. However, this is not a particularly useful observation
because a flat skew implies zero volatility of volatility (i.e. constant volatility) in which case volatility cannot change and
the p/l of a variance swap could only be zero.
Linear skew
If the skew is assumed to be linear, at least for strikes relatively close to the money, then Derman’s approximation can be
used (See “More than you ever wanted to know about volatility swaps”, Derman, 1999). This presupposes a linear (put)
skew, effectively assuming the call skew is flat, and calculates the variance swap strike as a function of three variables:
ATM (forward) volatility, the slope of the skew and the maturity of the swap.
Box 10: Derman’s approximation
2
3 1 skew T K
ATMF VAR
× + ≈ σ
where K
VAR
is the strike of the variance swap
σ
ATMF
is the implied volatility of the forward strike
T is the maturity in years
skew is generally taken to be the slope of 90/100 skew
In practice, this approximation tends to work best for shortdated index variance (up to about 1year), where put skews
are often relatively linear and call skews relatively flat, at least close to the money. As maturity increases and the OTM
strikes have a greater effect on the variance swap price, the contribution of the skew becomes more important (Figure 21),
but the inability of the approximation to account for the skew convexity can make it less accurate. Similarly, for single
stocks, where the convexity can be much more significant, even at shorter dates, the approximation can be less successful.
Where the estimate is accurate it tends to be closer to the traded bid than the offer (Figure 22).
Figure 21: Variance swap prices are more sensitive to skew at longer
maturities
Variance swap price, Derman’s approximation (ATM volatility = 20%)
10%
20%
30%
40%
50%
0% 2% 4% 6% 8% 10%
1M Variance Sw ap
3M Variance Sw ap
1Y Variance Sw ap
2Y Variance Sw ap
90/100 skew
Source : JPMorgan
Figure 22: Derman’s approximation has been relatively accurate for
index variance, but can underestimate the variance swap mid.
Indicative Euro Stoxx 6month variance swap prices
12%
14%
16%
18%
20%
22%
24%
Jul05 Oct05 Jan06 Apr06 Jul06
v arsw ap estimate
v arsw ap bid
v arsw ap ask
Source : JPMorgan
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
30
Example 10: Derman’s approximation
An index is trading at 100, with the 6month forward price at 102.5. The 90strike volatility is 26%, 100strike volatility is
22% and the ATM forward volatility (102.5strike volatility) is 21%.
6month volatility surface: 90strike: 26%
100strike: 22%
102.5strike: 21%
Slope of the skew = (26%  22%)/10% = 0.4.
Maturity, T=0.5,
1 + 3T x skew
2
= 1.24
Therefore K
VAR
= 21 x sqrt(1.24)
= 23.38, which is 2.4 vegas above the ATM forward volatility
Table 5Table 8 show volatilities, variance swap levels and their approximations for a number of European stocks and
indices as of 23 October 2006. Note that consistent with the discussion above, Derman’s approximation is relatively
accurate for indices in the shorter maturities. For singlestocks and longerdated index estimates, the approximation is less
successful.
Table 5: Dec06 (2month) volatility, skew and variance
ATM
volatility
ATMF
volatility
Skew
(90/100)
Variance
estimate
(Derman)
Actual
Variance
Swap
SX5E 12.9% 12.8% 5.8% 13.7% 14.2%
FTSE 11.3% 11.2% 5.9% 12.0% 12.4%
SMI 11.9% 11.8% 5.1% 12.5% 12.8%
DAX 13.8% 13.5% 5.3% 14.3% 14.5%
NESN VX 13.9% 13.8% 3.1% 14.1% 14.3%
BP/ LN 17.2% 17.2% 2.4% 17.4% 17.9%
CS FP 21.9% 21.9% 3.1% 22.3% 23.0%
AH NA 27.7% 27.7% 1.8% 27.8% 28.5%
RIO LN 32.9% 32.8% 1.9% 33.1% 34.6%
Source : JPMorgan; Data from 23 October 2006.
Table 6: Mar07 volatility (5month), skew and variance
ATM
volatility
ATMF
volatility
Skew
(90/100)
Variance
estimate
(Derman)
Actual
Variance
Swap
SX5E 14.7% 14.4% 3.7% 15.5% 15.9%
FTSE 12.6% 12.4% 3.9% 13.5% 13.8%
SMI 12.8% 12.7% 3.6% 13.6% 14.0%
DAX 15.8% 15.3% 3.7% 16.4% 16.6%
NESN VX 14.3% 14.2% 1.9% 14.5% 14.8%
BP/ LN 18.3% 18.2% 1.6% 18.5% 19.2%
CS FP 23.2% 23.0% 2.1% 23.6% 24.5%
AH NA 26.4% 26.3% 1.1% 26.5% 27.2%
RIO LN 32.9% 32.8% 1.3% 33.1% 34.8%
Source : JPMorgan; Data from 23 October 2006.
Table 7: Dec07 (c. 1year) volatility, skew and variance
ATM
volatility
ATMF
volatility
Skew
(90/100)
Variance
estimate
(Derman)
Actual
Variance
Swap
SX5E 16.2% 16.0% 2.2% 17.3% 17.3%
FTSE 14.3% 13.8% 2.5% 15.2% 15.1%
SMI 13.8% 13.7% 2.1% 14.6% 15.0%
DAX 17.7% 16.7% 2.3% 18.1% 18.1%
NESN VX 15.4% 15.4% 1.1% 15.7% 16.3%
BP/ LN 19.9% 19.7% 1.2% 20.2% 21.2%
CS FP 24.3% 24.2% 1.4% 25.0% 26.3%
AH NA 27.1% 26.9% 0.8% 27.2% 28.1%
RIO LN 33.2% 32.9% 0.9% 33.4% 35.5%
Source : JPMorgan; Data from 23 October 2006.
Table 8: Dec09 (c. 3year) volatility, skew and variance
ATM
volatility
ATMF
volatility
Skew
(90/100)
Variance
estimate
(Derman)
Actual
Variance
Swap
SX5E 18.3% 18.1% 1.4% 19.6% 19.1%
FTSE 16.5% 15.7% 1.6% 17.5% 16.9%
SMI 15.3% 15.1% 1.3% 16.3% 16.1%
DAX 20.5% 18.8% 1.4% 20.4% 20.1%
NESN VX 16.7% 16.6% 0.7% 17.0% 17.9%
BP/ LN 21.3% 21.0% 0.9% 21.9% 23.2%
CS FP 25.7% 25.7% 1.0% 26.8% 29.0%
AH NA 27.3% 26.9% 0.5% 27.2% 28.2%
RIO LN 34.1% 33.6% 0.6% 34.2% 37.0%
Source : JPMorgan; Data from 23 October 2006.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
31
Loglinear skew
In reality the volatility skew is not linear across all option strikes – and more accurate variance swap approximations can be
used. The loglinear model described below is one possibility.
Box 11: Loglinear skew
Suppose that the skew curve is loglinear of the form:
( ) ( ) F K K
ATMF
ln β σ σ − = ,
where F is the forward price and K is the strike
and β represents the slope of the skew
Then: ( )
2 4 2
2
3 2
5 12
4
T T K
ATMF ATMF ATMF ATMF VAR
σ σ
β
βσ σ + + + ≈
See “Just what you need to know about variance swaps”, JPMorgan, February 2005 for further details
Example 11: Loglinear skew
An underlying is trading at 100, with 6month forward at 102.5. 90strike volatility is 26%, 100strike volatility is 22% and
the ATM forward volatility (102.5strike volatility) is 21%.
6month volatility surface: 90strike: 26%
100strike: 22%
102.5strike: 21%
( )
38 . 0
9 . 0 ln
% 4
≈
−
= β , T = 0.5
Then the loglinear approximation gives
K
VAR
= 23.55%, in this case slightly higher than the value given by Derman’s approximation in Example 10
Gatheral’s formula
A further approximation to the variance swap price can be obtained from Gatheral’s formula which expresses the variance
swap strike as an integral (or summation) of the implied volatilities across the entire range of strikes. This formula
characterises the skew curve in terms of the BlackScholes d2 parameter, which measures the ‘moneyness’ of the associated
OTM option. Implied volatilities of traded options can be used to parameterise the skew curve in this form, by converting
the strikes into their d2 equivalents. This leads to potentially powerful methods of variance swap approximation: by fitting
a quadratic, or higher order polynomial to the skew surface parameterised in terms of d2, it is then possible to directly
calculate a theoretical variance swap price from this parameterisation.
For example, if the skew curve is quadratic in the variable z = d2 : ( )
2 2
0
2
z z z β α σ σ + + = then it can be shown that the
theoretical variance swap strike is T T K
VAR
β σ + =
2
0
, i.e. in ‘d2space’ the linear component of the skew, α, has no
effect on the variance swap price: only on the base level of volatility σ
0
, and the convexity β affect the swap strike See p82
and “The Volatility Surface”, Gatheral, 2006 for further details
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
32
2.5: What drives variance swap levels?
Since variance swap levels move in close step with options implied volatility, they are in large part driven by the same
factors which drive these volatilities, as well as by the shape of the skew curve. Thus the drivers of variance swap levels (at
least theoretical levels) are essentially the same drivers as for options volatilities and associated skews. However, supply
and demand issues particular to the variance swap market can cause levels to differ slightly from their theoretical values,
usually within the arbitrage corridor defined by bid/offer spreads.
In common with options volatilities, one of the most important drivers of variance swap levels, at least at shorter
maturities, is historical realised volatility (e.g. see Figure 16). The correlation between strikes of 1month variance swaps
and the volatility realised over the previous month is 0.91 (R
2
= 0.83, Figure 23). However, as maturity increases, the
correlation (and beta) between variance swap prices and realised volatility decreases (Figure 24). Even at short maturities,
realised volatility is not the only driver; evidenced by the fact that shortdated variance is superior to realised volatility as a
predictor of future volatility (see p37).
Figure 23: Shortdated implied variance is driven by shortdated
realised volatility
1M Implied/variance
y = 0.848x + 0.057
R
2
= 0.829
0%
10%
20%
30%
40%
50%
60%
70%
0% 10% 20% 30% 40% 50% 60% 70%
trailing 1M realised volatility
Source : JPMorgan (Euro Stoxx 50 data since August 2000)
Figure 24 : Shortdated variance is most sensitive to realised
volatility
Correlation / beta
0.00
0.20
0.40
0.60
0.80
1.00
0 1 2 3 4 5
Rsquared of implied v ariance v s 1M realised v olatility
beta of implied v ariance v s 1M realised v olatility
variance swap maturity
Source : JPMorgan (Euro Stoxx 50 data since August 2000)
Clearly riskaversion is also an important driver, dictating supply and demand factors for volatility products, variance
swaps included. Since variance swap prices in effect carry a significant skew component, demand for protection in the form
of put options will clearly drive up put skews, as well as volatilities, and thus have a feedthrough into variance swap prices.
The skew (in fact more correctly the skew convexity) directly influences variance swap levels, with its effect becoming
more pronounced at longer maturities (Figure 21). In addition, at longer dates structured product flow and the longer
term macroeconomic outlook also become important considerations.
Ultimately volatility itself is directional, at least over short time frames, with volatility tending to increase if the
market sells off. In this sense, variance swaps can be driven in the short term by the (inverse of) market level (Figure 25,
Figure 26). This is most clearly seen during market shocks when volatility tends to spike as the underlying sells off.
Finally, it should be added that variance swap flows can have a feedback effect into the rest of the market. With variance
swaps currently accounting for perhaps 1030% of the traded vega demand, variance swap hedging by market makers can
influence implied volatilities and particularly skews. Indeed under certain conditions the effects of variance swap hedging
can potentially influence the underlying market. See p90 for a more detailed explanation.
Whatever the drivers of variance swap levels, we do observe that they tend to trade above realised volatility, most notably at
the index level. Whilst this variance premium is partly due to the convexity of the variance swap payout, the remainder is
the socalled volatility risk premium, which is the subject of the next section.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
33
Figure 25 : Over short time periods, variance swap levels are
inversely correlated with market level ...
Implied variance index level (inverted)
12%
14%
16%
18%
20%
22%
24%
Sep05 Dec05 Mar06 Jun06 Sep06
3000
3200
3400
3600
3800
4000
Euro Stox x 6m implied
v ariance (lhs)
Euro Stox x index lev el,
inv erted (rhs)
Source : JPMorgan
Figure 26 : ... with the relationship changing as volatility switches
regimes
Euro Stoxx 6m implied variance
10%
15%
20%
25%
30%
35%
40%
45%
50%
2000 2500 3000 3500 4000
Jan 2002  Apr 2003
May 2003  Dec 2004
Jan 2005  Apr 2006
Apr 2006  Sep 2006
Euro Stoxx 50
Source : JPMorgan
2.6: The volatility risk premium
The variance swap strike represents the market’s fair price for exposure to realised variance over a specified period of time.
However variance swap levels tend to trade consistently above comparable realised volatility (Figure 23), with only
part of the spread attributable to the convexity of the variance swap payout. Backtesting shows that long volatility
positions, at least on index underlyings, are biased to make a loss (Figure 27). Conversely, short index volatility
positions are, on average, profitable. This bias is referred to as the volatility risk premium. In this section we
investigate why this bias exists, what it represents and how it changes over time.
Figure 27 : Average returns from long variance are negative…
Variance swap p/l (vegas)
20%
10%
0%
10%
20%
30%
40%
Aug00 Aug01 Aug02 Aug03 Aug04 Aug05 Aug06
long Euro Stox x 1month v ariance sw ap p/l
long Euro Stox x 1y ear v ariance sw ap p/l
Source : JPMorgan
Figure 28 : …but the tail is heavily skewed to the upside
Frequency of long variance swap p/l
0%
5%
10%
15%
20%
25%
30%

2
0
%

1
6
%

1
2
%

8
%

4
%
0
%
4
%
8
%
1
2
%
1
6
%
2
0
%
2
4
%
2
8
%
3
2
%
3
6
%
Source : JPMorgan (p/l from 1m Euro Stoxx variance since Aug 2000)
Investors in the real world are not riskneutral and, in general, are prepared to pay slightly higher than the theoretical fair
value for exposure to long volatility, often in the form of protection. This is in some sense understandable, since returns
from short volatility positions are limited to the premium received in the case of options, or half the strike multiplied by the
veganotional in the case of short variance swaps. On the other hand, returns from long volatility positions can be
potentially unlimited. Thus although expected returns from long volatility positions are, on average, negative, the
potential rewards can be high if timed correctly. In this sense long volatility is like buying insurance: paying a relatively
small premium for a potentially large payout if things go wrong, but expecting to forfeit some, or all, of the premium on
most occasions. Conversely selling volatility can be likened to selling insurance (or credit protection): potentially very
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
34
risky, but in the long run likely to make lucrative returns, providing enough capital is available to absorb the occasional
inevitable losses.
Example 12: The insurancelike nature of long volatility positions
A variance swap is struck at 10% for €100K of vega notional.
Realised volatility 0%: loss for the long is €0.5M (this is the maximum loss, not €1M, due to the favourable
effect of convexity for the long, see Example 3)
Realised volatility 20%: profit from the long is €1.5M
Realised volatility 40%: profit from the long is €7.5M
Example 12 above shows why sellers often fail to drive variance swap prices below certain thresholds (e.g. around 12% on
the Euro Stoxx), even when shortdated realised volatility drops to very low levels. For example, the yintercept on Figure
23 suggests an absolute floor of 6% for implied variance in the hypothetical case that realised volatility drops to zero.
The above discussion illustrates that the returns from long variance positions, although on average negative, have long
upside tails (Figure 28). That is, returns are not normally distributed: long positions tend to lose a little and often, but
can occasionally make large profits. Short positions usually make modest gains, but can be exposed to very large losses if
volatility spikes up significantly (see Section 3.3).
From the point of view of a short volatility position, the observed premium of implied over realised volatility is often
referred to as the carry of the short position and can be a good source for generating alpha, especially in the form of
strategies continuously selling variance through variance swaps (see p47). This strategy can work best as a rolling short
maturity strategy, with the occasional (large) losses from being short volatility diversified away over time. This is
somewhat analogous to the vanilla options equivalents: straddle overlays or covered call overwrite strategies (See e.g.
“Calling All Overwriters” JPMorgan, August 2002), or to selling credit protection, which usually generates positive carry
but has a small chance probability of large losses.
We can attempt to quantify the volatility risk premium by computing the spread of implied variance over contemporaneous
realised. This is complicated by a number of factors.
Firstly, real variance swap data is limited, though indicative variance swap levels back in the 1990’s can be inferred from
levels of the corresponding volatility indices (VIX for the S&P, VSTOXX for the Euro Stoxx – see page 94 for more
details). These volatility indices represent rolling 1month mid implied variance levels on their respective indices but tend
to be slightly higher (around 0.5 – 2 vegas) than real variance swap prices.
Secondly, part of the risk premium is attributable to the convexity of the variance swap payout. Whilst we make no attempt
here to remove this from this characterisation of the risk premium, any backtest of variance swap payoffs will implicitly
take this into account (e.g. Figure 28). We have estimated the convexity premium for 1month variance swaps at
somewhere around 1% (1 vega) under normal market conditions (see p 38). Finally, since volatility can spike from time to
time, an average spread of implied to realised volatility needs to be taken over a suitable time period to smooth out some of
the noise.
Given these caveats, we characterise the risk premium as the 1year rolling average of the spread between (estimated)
1month variance swap strikes and 1month realised volatility. Values seem to be relatively stable over time with the S&P
risk premium varying between 2% and 9% since 1990, with an average value of 5%. Euro Stoxx since 1999 is similar with
an average riskpremium of about 4%. The recent peak in risk premiums for both indices occurred in 2004, when volatility
had subsided from its highs but riskaversion remained elevated. Since then the risk premium has fallen from about 7% to
2.4% over 2 years on the Euro Stoxx (Figure 29), and from 5.8% to 2.8% on the S&P (Figure 30). Note that these measures
include a convexity premium compensating for the negative convexity of a short variance swap as mentioned above.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
35
Figure 29 : Variance riskpremium on the Euro Stoxx 50
Volatility/ points
20%
10%
0%
10%
20%
30%
40%
50%
60%
Jan99 Jan00 Jan01 Jan02 Jan03 Jan04 Jan05 Jan06
1M realised
v olatility
1M implied 
realised v ariance
v ariance risk
premium
Source : JPMorgan
Figure 30 : Variance riskpremium on the S&P500
Volatility points
10%
0%
10%
20%
30%
40%
Jan90 Jan93 Jan96 Jan99 Jan02 Jan05
1M realised
v olatility
1M implied 
realised v ariance
v ariance risk
premium
Source : JPMorgan
2.7: Variance swaps as predictor of future volatility
In theory the variance swap strike represents the market prediction of realised variance over the term of the swap.
This is complicated somewhat by supply and demand factors, especially at longer maturities where prices can be driven by
structured product flow. At shorter maturities riskaversion and protection buying can play a role in biasing the variance
swap strike to be above a ‘fair’ expectation of future realised variance as explained in the section above.
Nevertheless, at least at shortdates, implied variance tends to do a relatively good job of forecasting future realised
variance. Using data since August 2000, the R
2
between 1month Euro Stoxx variance swap strikes and subsequent
1month realised volatility is 0.71 (Figure 32). This compares with an R
2
of 0.61 between 1month realised volatility and
that realised volatility 1month later (Figure 31). This suggests that implied variance better predicts changes in future
realised volatility in comparison with previous realised volatility. Note that here we are correlating the square roots of
implied and realised variance, not least because units are more tractable, although results using variance itself are similar.
As discussed in the previous section, variance is itself a biased estimator, as it generally overestimates realised volatility.
Given historical data, a simple model of future realised volatility as 0.9x variance strike provides the best fit with the
data (Figure 32). We find similar results (implied variance a good predictor, with a slight edge over trailing realised) using
VIX, VSTOXX and VDAX levels as predictors for future realised volatility on the S&P, Euro Stoxx and DAX respectively,
using data back to the inception of the relevant volatility indices.
Given that implied variance appears to have better predictive power than prevailing realised volatility, what extra
information could the variance swap and option markets be pricing in?
• Mean reversion of volatility: the implied variance can take account of the current medium term “average” of
volatility and the fact that volatility tends to be meanreverting within a regime. If volatility over the last year has
been say 30%, but over the last month 10%, 1month implied variance will likely be priced well above 10%.
• Knowledge of forthcoming events: volatilities (especially shortdated) will tend to be marked up ahead of
earnings seasons or other events likely to increase volatility. In this sense implied variance can be truly forward
looking, whereas past realised volatility is not.
• Risk aversion in the market: If investors are nervous and willing to pay extra for protection, they are perhaps also
prepared to liquidate their positions quickly in a downturn and increase market volatility. Thus an increase in
implied variance could precede an increase in realised volatility if the market sells off.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
36
Figure 31 : 1month realised volatility is relatively well correlated
with realised volatility over the next month…
subsequent realised 1month volatility
y = 0.7843x + 0.0452
R
2
= 0.6157
0%
10%
20%
30%
40%
50%
60%
70%
0% 10% 20% 30% 40% 50% 60% 70%
1month realised volatility
Source : JPMorgan (Euro Stoxx 50 data since Aug 2000)
Figure 32 : … but the correlation of 1month implied variance with
subsequent realised variance is somewhat stronger
subsequent realised 1month volatility
y = 0.9037x  0.0022
R
2
= 0.7087
0%
10%
20%
30%
40%
50%
60%
70%
0% 10% 20% 30% 40% 50% 60% 70%
1month implied variance
Source : JPMorgan (Euro Stoxx 50 data since Aug 2000)
Instead of trying to predict future 1month realised volatility using volatility realised over the past month, we could use a
combination of realised volatilities of different maturities (e.g. 5day, 1month, 3month and 1year). Even choosing the
weightings of these by performing an insample regression still does not give as good a correlation with future realised as
using the implied variance. That is, variance swap prices really do seem to be adding additional information to that
encapsulated in backward looking volatility measures.
Although shortdated implied variance is a relatively good predictor of subsequent realised volatility, the same is not
true for longer maturities. Further in the future is harder to predict. Moreover, any edge of implied over realised
volatility as a predictor for future realised effectively disappears for maturities longer than 3months (Figure 34).
Figure 33 : Correlation between implied variance and past and future
realised variance
Source : JPMorgan (Euro Stoxx 50 data since Aug 2000)
Figure 34 : Correlation of implied variance with subsequent realised
of similar tenor decreases with maturity.
rsquared vs. future realised volatility
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0 3 6 9 12
correlation of implied v ariance v s.
subsequent realised v olatility
correlation of realised v olatility v s.
subsequent realised v olatility
maturity (months)
Source : JPMorgan (Euro Stoxx 50 data since Aug 2000)
Current 1M
implied
variance
Previous
1M realised
variance
Future 1M
realised
variance
R
2
=0.83
R
2
=0.61
R
2
=0.71
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
37
2.8: Is variance swap convexity fairly priced?
As previously discussed (see p29) variance swap strikes usually trade above corresponding atthemoney implied
volatilities. This is essentially due to convexity, which can be thought of in the form of the extra value of the convex
variance swap payout. Equivalently, convexity can be thought of as the extra volatility paid for the OTM options needed to
replicate the variance swap, as represented by a convex skew. In this section we attempt to quantify the value of the convex
variance swap payout in order to account for the observed premium of variance over ATM volatility.
The convex payout means that a variance swap must be worth more than a hypothetical ‘volatility swap’ struck at the same
level. How much more the variance swap is worth depends on how much volatility is likely to change (volatility of
volatility) and hence how much effect variance swap convexity will have. If volatility is unlikely to change much, say a
vega or two either way (e.g. Figure 9), then the convexity of the variance swap will be worth little and the variance swap
should trade only just above ATM volatility. In contrast if the likely range of volatility is high, then the convexity is
likely to be valuable (e.g. Figure 10), and the variance swap should price this in by trading significantly above ATM
volatility.
We can attempt to quantify the price paid for the convexity by computing the difference between the variance swap
strike and the ATM volatility, implicitly assuming that we could trade a “volatility swap” linear in volatility, struck at
ATM implied volatility. We can then compare the (linear) payoff from the volatility swap with the (convex) payoff of the
variance swap to estimate the value of convexity.
Figure 35 : Measuring the value for variance swap convexity
Extra vega p/l due to convexity (var swap – vol swap p/l)
0%
2%
4%
6%
8%
10%
12%
14%
0% 5% 10% 15% 20%
strike = 15%
strike = 20%
strike = 25%
strike = 30%
change in volatility from strike level
Source : JPMorgan
Figure 36 : 1month implied variance has some predictive power for
future realised variance
subsequent realised 1month volatility
y = 0.9037x  0.0022
R
2
= 0.7087
0%
10%
20%
30%
40%
50%
60%
70%
0% 10% 20% 30% 40% 50% 60% 70%
1month implied variance
Source : JPMorgan (Euro Stoxx 50 data since Aug 2000)
As demonstrated above, shortdated variance swaps have some value as a predictor of future volatility. As well as
looking at the regression R
2
, we can calculate the standard error of the future volatility as estimated by the variance swap
strike (Figure 36). This tells us the average RMS error in the 1month variance strike as a predictor of realised volatility
over the coming month. The idea is that this ‘average error’ is giving information about the historical variability of the
future realised volatility around the estimate provided by (0.9x) the variance swap strike. Taking data from the last 6years,
this standard error is around 6% (6 vegas), that is, the standard deviation of the difference between implied variance and
future realised variance is 6%.
This 6 vegas represents an ‘expected’ range of values for future volatility, thus determining the magnitude of the convexity
of the variance swap payoff. How much is the 6% volatility of volatility worth? It depends on the strike of the variance
swap, but for strikes in the region of 1520% the convexity is worth an extra 11.5% (Figure 35), which corresponds
fairly closely to what is observed in practice for shortdated Euro Stoxx variance.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
38
We can also calculate an ‘implied’ variability of volatility by considering the spread between short dated ATM volatility
and variance strikes. By assuming that a hypothetical ‘volatility swap’ would be struck at ATM volatility, and the only
difference between the price of this instrument and a standard variance swap of the same maturity is driven by the expected
volatility of volatility, we can back out a value for this ‘volatility of volatility’. Over the last six years, the difference
between Euro Stoxx ATM volatility and variance swap strikes suggests an average implied ‘volatility of volatility’ of
about 7%  fairly close to the 6% average error of the variance swap in predicting volatility.
Although, this figure of 6% volatility of volatility for the Euro Stoxx is a long term average (Aug 2000 – Sep 2006), like
volatility, it changes over time. By estimating this standard error, on a rolling 1year basis, we see that it has a minimum of
just below 2% and a maximum around 10%, the changes roughly reflecting ambient levels of realised volatility (Figure 39).
We observe a similar pattern, but slightly lower range for the S&P (Figure 40). Notice also that over the last few years,
Euro Stoxx implied volatility of volatility has been more or less in line with the actual variability of volatility (variance
appears fairly priced with respect to ATM volatility), but in the US, S&P implied volatility of volatility has consistently
overestimated the actual variability of volatility (variance is overpriced with respect to ATM volatility; skew/convexity is
too high)
Figure 37 : Variance swaps trade at a premium to ATM volatility
Implied volatility / variance
10%
20%
30%
40%
50%
60%
70%
Aug00 Aug01 Aug02 Aug03 Aug04 Aug05 Aug06
Variance sw ap strike
ATM implied v olatility
Source : JPMorgan (Euro Stoxx 50 data since Aug 2000)
Figure 38 : The spread between variance and ATM volatility reflects
the implied volatility of volatility
Spread / implied volatility of volatility
0%
5%
10%
15%
20%
25%
30%
Aug00 Aug01 Aug02 Aug03 Aug04 Aug05 Aug06
Variance strike  ATM
implied v olatility
implied v olatility of
v olatility
Source : JPMorgan (Euro Stoxx 50 data since Aug 2000)
Figure 39 : Euro Stoxx 50: implied and realised volatility of volatility
Implied/realised volatility of volatility
0%
5%
10%
15%
20%
Aug00 Aug01 Aug02 Aug03 Aug04 Aug05 Aug06
SX5E Implied v olatility of
v olatility (from Varw sap v s
ATM spread)
SX5E realised v olatility of
v olatility (std error of 12m
regression)
Source : JPMorgan
Figure 40 : S&P 500: implied and realised volatility of volatility
Implied/realised volatility of volatility
0%
2%
4%
6%
8%
10%
12%
14%
Aug00 Aug01 Aug02 Aug03 Aug04 Aug05 Aug06
S&P Implied v olatility of
v olatility (from Varw sap v s
ATM spread)
S&P realised v olatility of
v olatility (std error of 12m
regression)
Source : JPMorgan
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
39
Note that there is a strong correlation between volatility and the spread between variance swap levels and ATM
volatility. This makes sense, if we believe that the volatility of volatility is correlated with volatility (Figure 41, Figure 42).
That is, large (absolute) changes in volatility are more likely when volatility is high that when volatility is low. This is not
surprising: when volatility is low, it is generally stable (give or take the odd spike); when volatility is high it tends to swing
about more wildly.
Figure 41 : Volatility is correlated with volatility of volatility
Volatility / volatility of volatility
0%
10%
20%
30%
40%
50%
60%
70%
Jan87 Jan90 Jan93 Jan96 Jan99 Jan02 Jan05
1month realised v olatility
stdev of 1month realised v olaitlity
(last 12 months, annualised)
Source : JPMorgan (Euro Stoxx 50 data since 1987)
Figure 42 : Correlation of 1year realised volatility, with the variability
of the volatility realised over each of the 12 months
Stddev of monthly realised volatilities over the year
2002
1987
2003
1998
2001
1990 1991
1989
1992
1999
1988
1997
2000
2006
2004
1994
1996
1993
1995
2005
y = 1.69x  0.07
R
2
= 0.79
0%
10%
20%
30%
40%
50%
60%
0% 10% 20% 30% 40%
1year realised volatility
Source : JPMorgan (Euro Stoxx 50 data since 1987)
Estimates for the expected volatility of volatility are used for pricing options on variance. However in the case of variance
swaps they are also of use for pricing variance swap caps. A long capped variance swap can be thought of as a standard
variance swap plus a short call on variance, stuck at the cap level (Figure 43). A standard cap of 2.5x current implied
variance strike is relatively far outofthemoney, assuming that the volatility of volatility is not too large. This means that
the value of the cap should be relatively small compared to the variance swap strike and should not have a major effect on
pricing. However, for a long position, a variance swap with a cap will always be worth less than an uncapped variance
swap of the same strike. Therefore capped variance swaps must trade with strikes slightly below their uncapped equivalents
– the difference, in theory, representing the current value of the call on variance.
Figure 43 : A capped variance swap is equivalent to a covered call on
variance
p/l (vegas)
20%
0%
20%
40%
60%
80%
100%
0% 20% 40% 60% 80%
capped variance swap p/l
(cap at 2.5x strike)
uncapped variance swap p/l
(strike = 20)
realised volatility
Source : JPMorgan
Table 9: P/L from a long variance swap @ 20 strike
Realised
volatility
Realised
minus
implied
volatility
Uncapped
variance swap
p/l (vegas)
Capped
variance
swap p/l
(vegas)
0
10%
20%
30%
40%
50%
60%
80%
100%
20%
10%
0%
10%
20%
30%
40%
60%
80%
10.0%
7.5%
0.0%
12.5%
30.0%
52.5%
80.0%
150.0%
240.0%
10.0%
7.5%
0.0%
12.5%
30.0%
52.5%
52.5%
52.5%
52.5%
Source : JPMorgan
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
40
2.9: Variance term structure
The variety of drivers of short and longdated variance can lead to interesting dynamics of the variance swap term structure
curve. The shape and potential movement of the implied variance curve is important in determining the most
promising parts of the curve to buy or sell variance, including positioning for volatility exposure through forward
variance.
Variance swap term structures (in common with ATM volatility term structures) are usually upwards sloping (Figure 44)
and have a tendency to flatten, or even invert at the short end, following increases in volatility. Term structures for
variance are generally steeper than the equivalent ATM volatility curves due to the increasing effect of the skew at longer
maturities (Figure 44).
The variance swap curve can also be thought of representing the meanreverting nature of volatility. The short end is
most sensitive to prevailing levels of realised volatility (Figure 45), whereas the long end (5years+) tends to be more
anchored to some longterm estimate of average volatility. However, longerdated volatility is also driven by structured
product flows and tends to be more susceptible to supply/demand dynamics.
Figure 44 : Euro Stoxx term structure COB 23 October 2006
Implied volatility/variance
12%
14%
16%
18%
20%
22%
0 1 2 3 4 5
ATM i mplied volatility
I mplied variance
12%
14%
16%
18%
20%
22%
0 1 2 3 4 5
ATM i mplied volatility
I mplied variance
Maturity, years
Source : JPMorgan
Figure 45 : Shortdated volatility is driven by shortdated realised
Implied volatility/variance
4%
8%
12%
16%
20%
24%
28%
Jul04 Jan05 Jul05 Jan06 Jul06
VSTOXX Index
Euro Stox x 1M realised v olatility
Source : JPMorgan
An upward sloping curve in a lowtomedium volatility environment can also be thought of pricing in the chance of a
change of volatility regime sometime in the future. Volatility regimes represent a shift in the medium/longterm average
of realised volatility and are generally considered to last a number of years, sometimes correlated to economic cycles. Over
a short time period the chance of a switch to a higher volatility regime may be negligible, whereas even if volatility has
been low for 23 years, the chance of a change in volatility regime of the course of a 5year trade could be significant.
Another important point to consider, especially for longer maturity trades, is the effect of the marktomarket p/l. Whilst a
shortdated variance swap is principally exposed to realised variance, longer dated variance swaps take on significant
exposure to changes in implied variance before expiry. Therefore, if an investor has just bought or sold a 5year
variance swap, their principal exposure will initially be to 5year implied variance. As observed above, this is driven by
factors not necessarily correlated with current realised variance (structured product flows etc.) and the p/l can be
unpredictable.
These observations help to explain why investors can be reluctant to sell longdated variance at seemingly high levels (e.g.
25% for Euro Stoxx 5years) when the relevant indices have almost never realised these levels over any 5 year period
(Figure 46). For example, in 1998 LTCM experienced crippling marktomarket losses on short positions on longdated
volatility. As it happened, 19982003 was probably the worst 5year period in the last halfcentury to be short realised
volatility, so this particular trade may well have lost even if it had been held to maturity.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
41
Figure 46 : Realised index volatility has rarely exceed 25% over 5 years
5year realised volatility
0%
5%
10%
15%
20%
25%
30%
Jan50 Jan55 Jan60 Jan65 Jan70 Jan75 Jan80 Jan85 Jan90 Jan95 Jan00 Jan05
S&P500
Euro Stox x 50
Dow Jones
DAX Nikkei
FTSE
Source : JPMorgan
Investors looking to buy (or sell) volatility through variance swaps will clearly need to consider the best placed part of the
term structure curve to target. In general shortdated variance will have the greatest exposure to realised volatility
(gamma), whereas longer data variance will initially be more exposed to changes in implied variance (vega), and
movements in the curve. See Section 3.8 for details of using forward variance to position for an increase in volatility.
Typical movements of term structures can be explained in part by the “roottime” rule sometimes used in the market for
characterising curve changes. That is for a ‘normal’ move in volatility the change in implied variance at a given point on
the curve will be proportional to the inverse square root of the maturity. For example, if 1year variance increases by 1%,
3month variance moves up by 2% and 4year variance moves up by 0.5%. These kind of movements will have the effect
of flattening a normal upwardssloping term structure as volatility increases, and steepening it if volatility decreases, and
changes will be most visible at the shorter end of the curve. Since the relative price for an ATM option is approximately
0.4σ sqrt(T) a roottime move in volatility ensures that prices for (ATM) vanilla options will change in the same
proportion across the maturity curve.
Analysis of term structure movements is very important especially for long forward volatility trades. Here, the loss due the
slide from moving down a normal termstructure is tradedoff against the savings in carry due to the elimination of exposure
to realised volatility. Direct trades on the predicted evolution of the variance maturity curve are also possible (Section 3.9).
Figure 47 : Variance term structure moves more at the short end…
Implied variance
12%
14%
16%
18%
20%
22%
24%
0 1 2 3 4 5
07Sep06
22May 06
01Jan06
Maturity (years)
Source : JPMorgan (Euro Stoxx 50 data)
Figure 48 : …as demonstrated by the May 2006 volatility spike when
the short end inverted and the longer end flattened
Variance spread
5%
0%
5%
10%
Sep04 Mar05 Sep05 Mar06 Sep06
5Y  1Y
1Y  1M
Variance swap maturity
Source : JPMorgan (Euro Stoxx 50 data)
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 2: The Variance Swap Market
42
3. Uses of Variance Swaps
• 3.1 Exploiting a volatility view
• 3.2 Specific hedging purposes
• 3.3 Rolling short variance
• 3.4 Diversification
• 3.5 Index variance spreads
• 3.6 Relative value singlestock volatility
• 3.7 Variance dispersion and correlation trading
• 3.8 Forward variance and volatility spikes
• 3.9 Trading the variance swap term structure
• 3.10 Skew and convexity trades
• 3.11 Cross asset class trades
P
a
r
t
3
:
U
s
e
s
o
f
V
a
r
i
a
n
c
e
S
w
a
p
s
Part 3: Uses of Variance Swaps
Overview
In Part 3 we consider uses of variance swaps, outlined briefly in the list below. Taken at face value, a variance swap is just
a contract providing direct exposure to (realised) volatility. But why is it useful to have the ability to trade such an
instrument? Variance swaps can be used for trading macro volatility views, diversifying returns, hedging specific
exposures, and exploiting market inefficiencies. By trading combinations of variance swaps either of different maturities
(e.g. forward variance) or on different underlyings (variance spread trades and dispersion trades) it is possible to use
variance swaps to create exposure to factors other than realised volatility – e.g. future implied volatility, moves in the
volatility term structure or correlation. By combining variance swaps with other (volatility) instruments trading on the same
underlying, such as straddles, deltahedged options, or conditional variance swaps, it is possible to take exposures to other
factors such as skew and convexity. Finally, variance swaps can be used together with credit default swaps (CDS) to trade
equity volatility against credit.
• Exploiting a volatility view: Variance swaps are ideal for taking a direct view on the volatility of an underlying
without the pathdependency issues of a deltahedged option.
• Specific hedging purposes: Variance swaps can be used for macrohedging and also for hedging specific
volatility exposures, such as that resulting from structured products or life assurance policies.
• Rolling short variance: Short variance swaps can be used to capture the observed equity index volatility risk
premium. Rolling short index variance is an attractive systematic volatility strategy from a riskreturn perspective.
• Diversification: Volatility can be thought of as an asset class in its own right, and as such can act to diversify
returns within a portfolio.
• Index variance spreads: Variance swaps can be used to trade the spread of volatilities between two indices. Such
trades can be thought of as either relative value volatility trades or as ‘volatilitybeta’ trades aiming to profit from a
spread of volatilities widening as volatility increases.
• Relative value singlestock volatility: Use volatility pairs, or crosssectional regression volatility models to find
rich/cheap singlestock volatilities
• Variance dispersion and correlation trading: Trading variance swaps on an index against variance swaps on its
constituents provides exposure to equity correlation.
• Forward variance and volatility spikes: Long forward volatility can avoid potentially negative carry at the cost
of slide down the term structure, and can be a useful way of positioning for volatility spikes.
• Trading the variance term structure: Variance swaps can be used to trade the shape of the variance term
structure, analogous to the use of spreads and butterflies in fixed income to trade the shape of the yield curve.
• Skew and convexity trades: Variance swaps are long skew and convexity. Trading variance against (delta
hedged) vanilla options provides interesting exposures to skew and/or convexity.
• Cross asset class trades: Equity Volatility and credit spreads are correlated, both being measures of corporate
risk. Variance swaps are useful instruments in debt/equity trades, either at the index or single name level.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
44
3.1: Exploiting a volatility view
The most obvious use for variance swaps is to express a direct view on the volatility of the underlying. Deltahedged
options can be used for this purpose but have two principal drawbacks:
• The deltahedging requires active management (and therefore incurs human and transaction costs).
• The volatility exposure is pathdependent: the p/l of a deltahedged option depends not only on the amount of
volatility realised by the underlying, but where and when this volatility is realised.
Variance swaps have none of these issues and provide a pure exposure to volatility.
At the index level investors may use variance swaps to express a macro view – for example to buy volatility ahead of
possible recession, or sell volatility into a temporary spike. Note that the lack of path dependency can make this a more
attractive alternative than buying/selling the underlying (or options thereon) since the p/l depends on the outcome over the
entire lifetime of the trade and not just where the underlying ends up at maturity.
For example, in a bull market, suppose an investor feels there will be a sharp downwards correction at some stage in the
future, say within the next year. The investor does not want to sell his position in the underlying as the market may rally
further. Put options are a possibility to use for hedging, but will need to be restruck in order to roll up the protection as the
market rallies. Buying variance can prove profitable if the correction is relatively sudden (volatile) and occurs any time
within the maturity of the variance swap. However the investor must be careful to maintain the required volatility exposure
and must also consider whether an exposure to realised volatility is desirable, where the usually negative carry from holding
a long variance position can be compared to the put premium. To deal with these issues, it is sometimes preferable to use
forward starting variance swaps (Section 3.8) to position for a sudden increase in volatility.
Similarly, investors who expect a quiet market, which may gradually trend up, may want to find ways of boosting their
alpha based on this view. Straddles are not perfect since a low volatility market often displays trending behaviour; even
deltahedging the straddles will prove suboptimal as the market moves away from the strikes and the (favourable) exposure
to volatility is reduced through the decreased gamma. However, selling variance is an efficient, and nonpathdependent,
way of capitalising on this low volatility view. Indeed backtesting shows that shortvariance swap strategies have
outperformed short straddle strategies (both deltahedged and not) in the low volatility, rising market environment
since 2004. See European Equity Derivatives Weekly Outlook, 6 November 2006 for more details.
On singlename underlyings, variance swaps could be used to take a view on the uncertainty of a company. This can
be particularly valuable if an event timeline for a company is known, but the likely outcome of the event is unknown,
especially if an investor feels that the options market has not correctly priced in the risk associated with such an event. For
example, suppose a pharmaceutical company is due to announce the results of a drug trial, but the result of the trial is in the
balance. This could be a catalyst for large moves in the stock price in either direction depending on the outcome, which is
difficult to capitalise on using the underlying alone. Whilst straddles could be attractive, they may fail to capitalise e.g. if
the underlying trends up before the event and then afterwards sellsoff suddenly back towards the straddle strike. In this
instance variance swaps can be used to implement a view on the uncertain outcome, but otherwise known timing, of an
event.
Another use of variance swaps is to take advantage of an expected structural change in a company. For example,
suppose an investor expects an industrial conglomerate to sell a more volatile subsidiary. If this part of the business was
sold at a fair level, the spot price of the company may remain relatively unchanged, but the volatility characteristics of the
company may alter significantly. Hence, whilst it may be far from obvious how to play such an event using the underlying
alone, the likely long term volatility expected should fall and selling long dated variance may be profitable.
Variance swaps can also be used by investors to express their view on the likelihood of an M&A event. The nature of
variance swap payouts means that large singleday stock moves will generate large p/l for long variance positions, even in
comparison to vanilla options. Depending on what is priced in, and the investor’s view, long or short variance swap
positions may be attractive, either outright or traded against vanilla options such as straddles.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
45
3.2: Specific hedging purposes
Variance swaps can be useful for hedging purposes. Since volatility is directional (see Section 2.5, Figure 25) at least over
short time horizons, variance swaps will likely profit if markets selloff significantly and especially if this happens
suddenly. Being long a variance swap can be compared to owning a put option: a variance swap will likely profit in a
market downturn, as will a long put position. But if the market is benign, a variance swap will likely suffer negative carry
as realised volatility is below the level bought; likewise a long put will lose its premium.
There are problems however with this approach of attempting to hedge an underlying using variance swaps. Firstly it is not
clear how much variance should be used to hedge a fixed quantity of the underlying. If the selloff is rapid, the variance
protection will be effective and only a small notional amount will be required. However, if the underlying slowly drifts
down, without any large moves, the long variance swap may fail to provide protection altogether. Sometimes variance
swaps do offer more attractive riskreturn characteristics for hedging, depending on both hedging need and relative value.
In Section 3.10 we analyse the valuation and payoffs of long variance swaps combined with short put positions, effectively
analysing the relative attractiveness of long variance and long put positions. In practice, other types of protection, which
include a delta exposure, can be more efficient, particularly using combinations of vanilla options to provide protection
within a targeted range – e.g. putspreads, putladders, collars etc.
Forward variance swaps can offer a useful alternative to holding spot variance. Forward variance swaps are not directly
exposed to realised volatility so do not suffer negative carry in a benign market environment. They do however suffer slide
as the position ages (assuming term structure is positively sloped). The p/l on forward variance is determined by changes
in the implied volatility term structure, and to the extent that implied volatility reacts to market downturns, they can provide
a hedge against an equity portfolio. Forward variance can also be useful for hedging against volatility spikes. See Section
3.8 for a discussion of spot and forward variance swaps including carry and slide analysis.
Variance swaps can also be useful for hedging out specific volatility exposures. For example, life assurance companies
offer many products which offer some form of guaranteed benefits (e.g. variable annuities, withprofits funds), returns for
which they attempt to generate by holding equities. These companies have essentially sold equity put options to their policy
holders and hence have taken on a short volatility exposure. Changes to accounting standards are encouraging markto
market valuation of these embedded liabilities, whilst capital adequacy requirements (Solvency II) require portfolios to be
reserved against changes in market level, volatility, interest rates etc.
Although assurers may hedge embedded optionality by deltahedging, this will not hedge marktomarket volatility risk or
shifts in implied volatility. Increasingly, insurers are looking to use variance swaps (either spot or forward) to hedge
volatility exposure, increasing the demand for longdated variance – and potentially steepening the volatility term structure.
Note also that longdated variance swaps offer particular exposure to changes in implied volatility, which is often what
insurance companies need to hedge. For any long variance swap position, realised volatility losses will accrue if the market
is relatively benign, although longerdated maturities accrue realised volatility losses at a relatively slow rate. Alternatives
such as forward starting variance and conditional variance can also be explored as ways of targeting hedge exposure. See
Section 2.10 for a discussion of forward starting variance swaps and Conditional Variance Swaps, April 2006 for a
discussion of conditional variance swaps.
Similarly, structured products desks provide investors with a wide variety of equitybased investments which incorporate
volatility (and correlation) exposures, ranging from simple capital guaranteed products to more complex multiasset
payoffs. The aggregate risks of their positions needs to be managed, hedged or soldon into the market, and variance swaps
provide a very useful way for desks to manage their volatility exposures, which will itself change as the market moves.
Such hedging flows, or anticipation thereof, can have a large effect on the volatility surface, particularly at longer dates.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
46
3.3: Rolling short variance
Historically one of the most successful volatility strategies has been the systematic selling of shortdated index
variance. This strategy takes advantage of the volatility risk premium described in Section 2.6. Returns from this strategy
are on average positive, but with a large negative tail from volatility shocks. By implementing a longterm rolling strategy
we attempt to diversify away these occasional large losses over time.
A diversification argument also argues for using shortdated variance swaps, even though term structures are generally
upward sloping, meaning the shorterterm trades earn less carry. From a riskreturn perspective, it may be better to
regularly take in a relatively small premium (e.g. 2.3 vegas) every month than to take in a larger premium (e.g. 5 vegas)
once a year. In practice 1month maturity variance swaps appear to be a good maturity to target – shorter dated options are
available but are liable to incur relatively higher transaction costs. In addition 1month trades correspond to
VIX/VDAX/VSTOXX levels allowing for greater visibility and ease of backtesting.
The backtest presented here comprises the short variance component of JPMorgan’s “Yield Alpha” product (see “Yield
Alpha Strategy Guide”, JPMorgan). The various forms of the Yield Alpha product aim to take advantage of different types
of carry trade across a spectrum of asset classes, by choosing a basket of strategies with attractive ‘carrytorisk’
characteristics. Short variance presents a good example of such a carry trade, by taking advantage of the observed equity
volatility risk premium (Section 2.6).
We have backtested the strategy of selling 1month S&P variance on the first day of every month since March 1990. As
variance swap prices were not available for the whole period, we have estimated the variance swap strike from VIX levels.
Recall that the VIX represents the theoretical 1month variance swap price interpolated from exchangetraded S&P options
(see p94). Since the VIX tends to trade somewhat above real variance swap prices (typically 0.5–1 vega) we have estimated
variance swap mid levels at 75 bps below the VIX. Then to get a realistic variance swap bid level we have deducted a
further 50bps. Finally, to limit the risk of the strategy, we have only sold capped variance swaps (capped at 2.5x the strike)
and assumed that these trade 50 bps below uncapped variance swaps. To summarise, we have modelled the selling price for
capped variance swaps at 1.75% below prevailing VIX levels, a level we believe is somewhat conservative.
Results from backtesting are very encouraging (Figure 49). Average returns are 2.3 vegas per month, with 161 of the
196 months yielding positive returns (Table 10). Nevertheless, the maximum gain (14.8%) is much less than the
maximum loss (22.6%) and there were four months with losses greater than 10%, compared with only two months with
gains greater than 10% – indicating the negative tails of the return distribution (see e.g. Figure 28). This is also seen when
comparing the standard deviation of negative returns (5.2%) with that of the positive returns (only 2.5%). Although
positive returns dominate, losses are relatively infrequent but can be large and unpredictable.
Figure 49 : Performance of rolling short variance on the S&P
Vega points
20%
10%
0%
10%
20%
30%
40%
Mar90 Mar92 Mar94 Mar96 Mar98 Mar00 Mar02 Mar04 Mar06
S&P 1month v ariance bid
subsequent S&P 1month realised v olatility
short 1month S&P v ariance p/l
Source : JPMorgan
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
47
The annualised return is 27.3% (uncompounded), which when compared to the annualised standard deviation of returns of
15.3% gives an information ratio of 1.8. Incidentally, over the course of the backtest, the 2.5x cap was never even close
to being hit. In fact the greatest level of realised compared to the level of implied bought at was in July 2002 when the
realised volatility of 42% was about 1.7x the 25% level of variance sold. Note that if we assume we can actually trade at
prevailing VIX levels, the average monthly return rises to 4% (48% annualised) and the information ratio to about 3.1.
Given this pattern of returns we can consider creating an index based on the performance of shorting variance
swaps. We assume that a fixed amount of capital, say $100M, can be risked (with the 2.5x cap allowing the maximum
losses to be known). Each month we sell a 1month variance swap with vega notional equal to 0.5% of the available capital
(e.g. $500K). This means that as long as the variance swap strike does not exceed 76, the maximum loss cannot be greater
than the capital held due to the cap. At the end of each month when the variance swap expires, the capital will increase or
decrease depending on the p/l of the short variance swap, and a new short variance swap position for the next month is
initiated using 0.5% of the new capital amount.
We consider such an index, starting in 1990 with a value of 100 and each month entering into a short 1month variance
swap with vega notional equal to 0.5% of the current index level (Example 13). By July 2006, 16 years since inception, the
index level has reached 741 (Figure 50).
Example 13: Calculation of the ‘Variance Alpha’ index
Suppose the index is at 100, and the 1month variance swap strike is 20.
The veganotional will be 0.5 index points (ip),
Given the 2.5x cap, the maximum loss will occur if realised volatility is 50% or more
In this case the monthly loss will be capped at 0.5 x ( 20
2
 50
2
) / ( 2 x 20 ) = 26.25ip, and if this loss occurred, the
subsequent veganotional will be 0.36875 index points.
In contrast if realised volatility over the next month was only 10%, the p/l would be: 0.5 x (20
2
10
2
)/(2x20) = 0.5 x 7.5 =
3.75ip, the new index level would be 103.75, and the new vega notional equal to 0.51875 index points.
Figure 50 : Index performance from rolling short variance strategy
Index performance
0
200
400
600
800
1000
1990 1993 1996 1999 2003 2006
80%
60%
40%
20%
0%
20%
index performance
monthly short v ariance p/l (rhs)
Source : JPMorgan, results from Mar 1990 – Jul 2006
Table 10: Performance statistics for rolling short variance strategy
Average Return (monthly) 2.3%
Average Return (annualised) 27.3%
Annualised Stdev 15.3%
Riskreturn 1.78
Index performance (1990 = 100) 741
All Positive Negative
Number of months 195 161 34
Percentage 100% 82.6% 17.4%
Standard deviations 4.4% 2.5% 5.2%
Returns In excess of +/ 10% 6 2 4
Largest return 14.8% 22.6%
Source : JPMorgan, results from Mar 1990 – Jul 2006
Similar results are obtained from this strategy using the Euro Stoxx 50 or DAX indices, with the VSTOXX and VDAX as
the basis for variance swap prices. Using slightly longer maturity variance swaps, e.g. 3months, works well although the
timing of the index rolldates then becomes more important.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
48
It is an interesting exercise to compare the performance of rolling short volatility strategies using variance swaps with
similar strategies using short straddles, both deltahedged or not (see European Equity Derivatives Weekly Outlook, 6
November 2006). Whilst all three strategies perform well, short variance swaps appear to deliver the most consistent alpha
across different volatility environments, and most effectively capture the premium of implied to realised volatility. Rolling
short 1month naked straddle strategies perform well, profiting from the somewhat rangebound behaviour of equity
indices, and avoiding the negatively convex payoff which can hurt short variance positions in times of high volatility.
However, their p/l is extremely path dependent and they can suffer in times of lowvolatility trending markets (e.g. 2004
2006). Deltahedged straddles do a better job than unhedged straddles of capturing the premium of implied to realised
volatility, but also suffer in trending markets, as otherwise positive returns from selling volatility above the level
subsequently realised are eroded as the market trends away from the straddle strike, and the gamma exposure of the straddle
is reduced.
With returns from short volatility trades somewhat uncorrelated with the underlying, these types of strategy work well as
overlay strategies aimed at boosting alpha and diversifying returns.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
49
3.4: Diversification
One of the notable features of the returns of a rolling short volatility index is its similarity to returns of a bond index (Figure
51, Figure 52). In rolling short variance, regular periods of positive p/l resulting from the spread between implied and
realised are punctuated by large losses arising from spikes in volatility. In bond indices, the regular p/l resulting from
coupon payments and accrued interest is punctuated by capital losses caused by rising yields, or, in the case of credit
portfolios, defaults. In the same way as interest is the payment required for providing financial capital, the implied 
realised volatility premium can be thought of as the payment required for providing “equityinsurance” capital. It is this
“equityinsurance” capital that can be considered an alternative asset class that can be invested in and the payment
received is captured through a rolling short variance swap strategy.
Figure 51 : Index performance from rolling short variance strategy
0
200
400
600
800
1000
1990 1993 1996 1999 2003 2006
80%
60%
40%
20%
0%
20%
index performance
monthly short v ariance p/l (rhs)
Source : JPMorgan
Figure 52 : Bond index performance – JPMorgan USD traded bonds
50
100
150
200
250
300
350
400
1990 1993 1996 1999 2003 2006
20%
15%
10%
5%
0%
5%
bond index performance
monthly p/l (rhs)
Source : JPMorgan
Hence, volatility (or more correctly the implied – realised spread) can be thought of as an asset class in its own right.
Moreover since there is relatively low correlation between the p/l from short variance and the underlying equity market, it
can be used within a portfolio to diversify returns (Figure 53, Figure 54).
Figure 53 : Efficient frontier of stock plus bond moves further out by
adding short variance…
Portfolio return
4
5
6
7
8
9
10
0 5 10 15 20 25
Bond
Equity
Short equity variance
Portfolio risk
Source : JPMorgan
Figure 54 : … and variance generally replaces bonds in asset
allocation along the efficient frontier.
Asset allocation, white dashed line is optimal allocation split for bond + equity only
0%
20%
40%
60%
80%
100%
5.1 6.0 6.8 7.7 8.6
Short equity
variance
Bond Equity
Required portfolio return
Source : JPMorgan
In Figure 53 we show the risk/return for each individual strategy – equity (long Euro Stoxx), bonds (long Euro
Governments), short Euro Stoxx 50 variance – overlaid with an efficient frontier for a twoasset bond/equity mix (grey line)
and an efficient frontier for all three asset classes (blue line). Because variance returns are not fully correlated to any
combination of bonds and equities, including returns from short variance swaps helps to push out the efficient frontier.
Moreover when used in conjunction with bonds and equities, short variance swaps generally replace bonds within an
efficiently allocated portfolio (Figure 54), due to somewhat bondlike nature of short variance swap returns.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
50
3.5: Index variance spreads
One use of variance swaps is to trade the spread of volatility between two indices. These type of trades fall into two main
categories:
1. Relative value volatility trades, where the variance of one index appears to be trading at an unwarranted premium
or discount to another.
2. ‘Volatilitybeta’ trades which are long variance on a higher beta index and short variance on a wellcorrelated but
lower beta index. These trades aim to be net long volatility, particularly as volatility increases, with the short
variance leg hedging out some or all of the otherwise negative carry if volatility remains at prevailing levels.
Opportunities of the first type can be evaluated by looking at current spreads and comparing these to historic spreads of
realised (and implied) variance, at least in similar volatility environments, to see if an opportunity exists. The volatility
pairs regression analysis discussed in the following section can also be used to screen for these kind of opportunities. See
also “Volatility Pairs, JPMorgan, August 2004”. Some index variances have a tendency to appear either structurally rich or
cheap as a result of local activity, such as protection buying, or other supply and demand dynamics.
Note though that some indices have local ‘regimechanges’ as their composition or behaviour of the members changes. For
example the Nasdaq was a very high volatility index in 20002001 realising well over 50% volatility. However, following
the unwind of the dotcom bubble, many companies of the companies in the index have now ‘matured’ and the Nasdaq
behaves much more like other developed market indices – realising around 1020% volatility in 20052006.
Volatilitybeta trades
The second type of spread trading involves taking opposing volatility positions in two wellcorrelated indices, but where
one index tends to be more sensitive to increases in volatility than the other. We have seen previously that there exists a
premium of implied to realised volatility, which will probably exist for both indices. Thus a spread trade may to some
degree cancel this extra cost, whilst maintaining some overall exposure to increased volatility through the higher sensitivity
of the buy side index. Such trades, as previously stated, are often referred to as volatilitybeta trades.
One of the characteristics of a low market volatility regime is the compression of volatility spreads across stocks and
indices. This is highlighted by the contraction of crosssectional deviation between different index realised volatilities
whenever overall market volatility becomes more subdued (Figure 55).
Figure 55: Spreads between different indices have compressed as
implied volatility has fallen…
6M realised volatility 6M realised volatility std.dev
0%
10%
20%
30%
40%
50%
60%
1986 1990 1994 1998 2002 2006
0%
5%
10%
15%
20%
25%
Euro Stoxx 50 6M realised
volatility (lhs)
crosssectional stddev of 6M
index realised volatilites (rhs)
Source: JPMorgan
Figure 56: …while there has been a general trend of increasing
correlation between indices
IndexIndex, 6M Average pairwise correlation
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
1991 1993 1995 1997 1999 2001 2003 2005
Top 4 indices
Developed indices
Developed + EM indices
Source: JPMorgan
Moreover, in the current period of low volatility (20042006), this deviation of index volatility is now close to its alltime
lows, though this may also be a consequence of the trend of increasing correlation between indices globally over the last
few years (Figure 56).
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
51
The high correlation between indices, coupled with the likelihood that index volatilities will probably diverge in a rising
volatility market, gives opportunity to investors to position themselves more efficiently for an increase in volatility using
volatility spread trades rather than outright long volatility positions.
As an example of this type of trade, we look at the behaviour of DAX volatility compared to that of Euro Stoxx 50. Given
that the Euro Stoxx 50 is a panregion index with a greater number of members and a more balanced sector composition
than the DAX, it should be expected to realise less volatility than the DAX on a purely correlation effect alone. This is
indeed the case, with the average spread of realised volatility at around 2% over the last fifteen years (Figure 57).
Another important feature of the DAX / Euro Stoxx volatility spread is that it tends to widen as volatility increases. This is
a consequence of the volatility beta effect, which sees DAX realised volatility increase (and then decrease) by more than
Euro Stoxx realised volatility. We can quantify this effect by looking at the beta of a regression of DAX on Euro Stoxx 6
month realised volatility, which has been about 1.25 over the last three years and 1.05 over the last fifteen years (Figure 58).
Figure 57: DAX tends to realise more volatility than the Euro Stoxx …
6M realised volatility
10%
0%
10%
20%
30%
40%
50%
60%
1991 1993 1995 1997 1999 2001 2003 2005
SX5E
DAX
SX5E  DAX
Source: JPMorgan
Figure 58: … and spread tends to expand as volatility increases
6M DAX realised volatility
y = 1.0493x + 0.0142
R
2
= 0.9721
0%
10%
20%
30%
40%
50%
60%
0% 10% 20% 30% 40% 50% 60%
6M Euro Stoxx 50 realised volatility
Source: JPMorgan, data since 1991
Given the directionality shown in Figure 58, it makes sense to compare a long variance strategy on the DAX with a spread
trade which is long DAX / short Euro Stoxx 50 variance. We can backtest this directly by comparing the prevailing levels
of implied variance of the two indices, and the spread between them, with the historical realised volatility over recent years.
Example 14: DAX vs. Euro Stoxx volatility beta trade
On 25 September 2006, 6month DAX and Euro Stoxx implied variance were trading at 18.6% and 17.2% respectively,
leading to a (mid) spread of 1.4% of DAX over Euro Stoxx.
Suppose it had been possible to own DAX variance at the 18.9% level (mid + 30bps) throughout the last 15 years, then the
average profit would have been 3.6% (3.6 vegas) with a standard deviation of 13%. However, the majority of profits would
have been made in the high volatility period between 1997 and 2003, and outside of this period the average p/l would have
been 3.8%, reflecting the usually negative carry of a long volatility position.
Suppose instead we consider the p/l of being long the DAX/Euro Stoxx variance spread. This would have lead to an
average p/l of about 0.25%, but with a standard deviation of returns of only 2%.
To compare these two alternatives fairly, we scale the p/l of the spread trade by its standard deviation of returns. In this
case, this leads to a p/l of 1.7%, less than the 3.6% of the long DAX variance trade. However, much of this difference is
due to the higher assumed bid offer costs for executing the spread trade. Also note that the returns for the spread trade are
more normally distributed than those of the long DAX trade (Figure 60).
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
52
Figure 59: Long DAX variance p/l is similar to the spread trade
against Euro Stoxx …
6M DAX variance p/l 6M DAXEuro Stoxx 50 variance p/l (6.5X)
10%
0%
10%
20%
30%
40%
50%
60%
1991 1993 1995 1997 1999 2001 2003 2005
SX5E
DAX
SX5E  DAX
Source: JPMorgan; assuming prevailing variance levels as of 25 Sep 2006
Figure 60: … but the spread trade does have a more normally
distributed p/l
Population of p/l scenarios
0%
5%
10%
15%
20%
25%
30%
35%
40%

1
5
%

1
0
%

5
%
0
%
5
%
1
0
%
1
5
%
2
0
%
2
5
%
3
0
%
3
5
%
4
0
%
4
5
%
5
0
%
DAX p/l
SX5E  DAX p/l (SD Scaled)
6M variance p/l
Source: JPMorgan; assuming prevailing variance levels as of 25 Sep 2006
The above example shows that in this case the spread trade had no real benefit compared to just holding long DAX
variance. However, in some sense the indices were too correlated in this case. Scaling up the spread trade by 6.5 times also
necessitated effectively scaling the 60bps transaction costs up to 3.9 vegas in comparison with the long DAX trade
(transaction cost 30bps). Without these transaction costs the average long DAX p/l would have been 3.9 vegas and the
spread traded p/l, 5.6% for the same level of risk.
In general we can look for these kinds of spread trades across a wide range of indices, to find index pairs where the
volatilities are well correlated and which display the volatility beta effect described above. We can asses the attractiveness
of these pairs by backtesting the variance spread trade against going long variance on the higher beta index, and scaling the
spread p/l to give a similar level of risk.
One point worth making in this analysis of volatilitybeta trades is that it does not attempt to directly measure the relative
value between index volatilities. What we are attempting to achieve is to get a long volatility exposure, but additionally
using a correlated index with lower beta to mitigate the carry of such positions. Hence, indices with a low correlation
between their realised volatilities are not considered as a direct volatilitybeta trade.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
53
3.6: Relative value singlestock volatility
Variance swaps can also be used to trade spreads between singlestock volatilities – either as a way of partially hedging
the variance of a singlename against wider market/sector moves, or in order to directly trade relative value singlestock
volatility. Two possible methods for assessing relative value single stock variances are temporal regression and cross
sectional regression.
Volatility pairs analysis
In this case, temporal regressions work by looking at the evolution of the ratio (or spread) of the implied variances between
two stocks. A zscore can be computed to express how far the ratio is from its long term average. Assuming these ratios of
implied variance have meanreverting tendencies, then pairs with zscores of large magnitude, expressing a large deviation
from their long term average, could potentially represent opportunities for volatility pairs trades. This applies particularly to
well correlated stocks, typically from the same sector with similar volatility characteristics.
Of course sometimes a divergence in implied variances of two stocks whose variances were previously highly correlated
may simply reflect real changes in the underlying market. For example, consider the case of two similar companies with
similar volatility characteristics where one company issued an unexpected profit warning raising concerns about its future
viability. This could lead to a large change in ratio of implied variances as the variance of one company spikes but the other
remains relatively stable. Clearly in this case, the ratio of implied volatilities would not necessarily be expected to mean
revert and selling the ratio in the hope that it would close up would not necessarily be a sensible trading strategy.
To some extent we can screen out these type of scenarios by also considering the zscore arising from the ratios of realised
volatilities. If the realised volatility ratio has recently diverged from its longterm average in the same way as the ratio of
implied variances, then the change in the implied variance ratio may be justified, and no trading opportunity exists. Thus
the trick is to look for pairs with a high zscore in their implied volatility ratio, which is not reflected in a corresponding
high zscore for their realised volatility ratio (Figure 61, Figure 62).
Figure 61: Alcatel implied volatility increased in relation to Cap Gem...
1Y Implied volatility implied volatility ratio
20%
25%
30%
35%
Jun05 Aug05 Oct05 Dec05 Feb06 Apr06 Jun06
0.90
1.00
1.10
1.20
1.30
Alcatel 1Y implied v olatility
Cap Gem 1Y implied v olatility
ratio of implied v olatilities (rhs)
Source: JPMorgan
Figure 62:…but this is not reflected in the ratio of realised volatilities
3M realised volatility realised volatility ratio
10%
15%
20%
25%
30%
35%
Jun05 Aug05 Oct05 Dec05 Feb06 Apr06 Jun06
0.00
1.00
2.00
3.00
4.00
Alcatel 3M realised v olatility
Cap Gem 3M realised v olatility
ratio of realised v olatilities (rhs)
Source: JPMorgan
JPMorgan produces a daily Volatility Pairs Report considering all possible pairs within 11 broadbased sectors and ranks
them according to the difference between the implied and realised zscores. We have backtested the results of trading
volatility on the top 3 ranked pairs in sector, with encouraging results. See Volatility Pairs, August 2004 and “Identifying
and exploiting dislocations in singlestock volatility”, June 9 2006 for further details. Note that variance swaps are
particularly suited to this type of trade due to the direct nature of the exposure to volatility they offer.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
54
Crosssectional regression
An alternative means of screening for rich/cheap singlestock variance is to use a crosssectional regression methodology.
The idea here is to compare the current implied volatilities (or variances) of a universe of stocks against a number of other
factors which may have some bearing on their volatilities – for example their historic volatilities and credit spreads. It is
important to note that unlike the temporal regression method explained above, this regression is not trying to model the
historical relationship of stock volatilities, but is simply trying to model all current stock implied volatilities as a multi
linear function of a number of various other stock properties. The following variables are used in our regression model:
• Historic volatility (should usually be the main factor determining a stock’s implied volatility)
• Stock beta vs. the market (high beta should translate into high volatility)
• Dividend yield (high yielding should perhaps be less volatile)
• 3month return (poor performers may tend to have higher volatility)
• CDS spread (reflects a company’s credit risk and leverage, and should correlate with implied volatility)
Figure 63 and Table 11 show the results of the crosssectional regression model for 112 major European names as of 9
November 2006. The implied volatilities of Centrica, EADS, BG Group, Scottish Power and Rio Tinto appear rich, all with
actual implied volatilities well above those suggested by the regression model. Nestle, UBS, Ciba, Iberdrola and MAN
appear below the regression line, having apparently cheap volatilities.
We have backtested this method by trading selected stock variances from a universe of major European names  buying
variance on the 10 stocks with the cheapest looking variance according to this metric and selling variance on the 10 richest
looking names. Results are good, with around 6 vegas average p/l (prebid/offer) since 2003 from trading the spread
between the long and short variance baskets. See European Equity Derivatives Weekly, 21 November 2005 and
“Introducing JPMorgan’s CrossSectional Volatility Report”, November 2006 for further details.
Figure 63: Crosssectional model: stocks with rich volatility appear
above line; those with cheap volatility below it
3m implied volatility
12%
16%
20%
24%
28%
32%
12% 16% 20% 24% 28% 32%
NESN VX
UBSN VX
MAN GY
IBE SQ
CIBN VX
CNA LN
EAD FP RIO LN
BG/ LN
SPW LN
crosssectional model implied volatility
Source: JPMorgan
Table 11: The top 5 richest and cheapest European volatilities
according to the crosssectional model (9 November 2006)
Bloomberg
Ticker
3m
realised
volatility
3m ATM
implied
volatility
model 3m
implied
volatility zscore
1 CNA LN 16.0% 24.9% 18.9% 2.2
2 EAD FP 30.8% 29.7% 25.4% 1.6
3 BG/ LN 22.8% 25.9% 21.8% 1.6
4 SPW LN 23.5% 24.5% 20.7% 1.4
5 RIO LN 33.4% 30.7% 26.9% 1.4
…
108 NESN VX 11.2% 12.6% 15.2% 1.0
109 IBE SQ 33.9% 21.9% 25.2% 1.2
110 CIBN VX 25.5% 19.6% 23.0% 1.2
111 MAN GY 34.3% 25.8% 29.3% 1.3
112 UBSN VX 20.4% 17.3% 21.8% 1.7
Source: JPMorgan
This method can also be used to select stocks with cheap implied volatility/variance to trade against index variance in a
bespoke dispersion trade. See e.g. European Equity Derivatives Weekly, 12 June 1006.
On both the temporal and crosssectional regressions investors should consider whether they wish to be exposed to the
relative realised or implied variance of the stocks. Investors believing that the current spread between the implied
volatilities of a pair of stocks represents an unreasonable view of the difference between their future realised volatilities,
should trade this spread through relatively short dated variance swaps with the intention of holding them to expiry.
Investors who believe that the spread between the implied volatilities may meanrevert in the short term, may be better
trading longdated variance with the intention of closing out their positions early, or even using forward starting variance.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
55
3.7: Variance dispersion and correlation trading
An extension to trading volatility pairs is to trade the volatility of an index against a the volatility of a basket of stocks.
This naturally leads to the concept of trading correlation. Index variance should always be less than the average variance of
the constituents, due to the diversifying effects of the index, the size of this discount effectively being related to the
correlation between the constituents of the index. However, backtesting shows that the premium of index implied over
realised variance tends to be greater than the corresponding premium for singlestock variance. That is, selling variance on
an index and buying variance on its members has generally been a profitable strategy. This can be partly explained by
demand for protection at the index level, whilst fund managers seek to sell options on singlestocks on which they have a
view in order to raise premium.
Realised index volatility can be thought of as arising from two factors: (a) the volatilities of the constituents, together with
(b) the correlation between them. Greater correlation means less diversification and hence higher index volatility.
Realised index variance is described by the equation:
ij j i j i j i i i i I
ρ σ σ ω ω σ ω σ
<
∑ + ∑ = 2
2 2 2
where
i
ω and
i
σ are respectively the weight and volatility of the i
th
stock in the index and
ij
ρ is the pairwise correlation of
the i
th
and j
th
stocks.
Since these are realised quantities, they can be observed in the market and the pairwise correlations directly calculated, and
averaged to give an overall correlation figure for the index. Another way of expressing this ‘average correlation’ is to
assume that the correlation is the same for all distinct pairs of stocks and then rewrite the above equation to solve for this
average correlation.
( )
∑ ∑
∑
−
−
=
i
i i
i
i i
i
i i I
w w
w
2 2
2
2 2 2
σ σ
σ σ
ρ or approximately (for a well diversified index):
( )
2
2
∑
≈
i
i i
I
w σ
σ
ρ
The advantage of this approach is that the average implied correlation, which we cannot directly measure, can be backed
out in exactly the same way from the implied variance of the index and its constituents. As in the case of realised
correlation, for a well diversified index, this turns out to be approximately equal to the square of the ratio of index volatility
to average singlestock volatility.
This implied correlation can be traded by selling index variance and buying singlestock variance, the resulting
position being short correlation. To hedge out the exposure to volatility, at least initially, the index and singlestock legs of
the trade must be weighted with more index veganotional than singlestock vega notional – the exact amounts depending
on the level of implied correlation. Such a trade is known as a variance dispersion trade.
Like volatility, correlation usually trades at a premium to realised levels (Figure 64), making it another attractive candidate
for carrystyle trades: rolling short correlation through variance dispersion performs well (Figure 65). Nevertheless,
volatility and correlation are themselves correlated, and tend to spike together in a sharp selloff. Losses from short
correlation through variance dispersion can occasionally be very large, especially since the trade becomes short volatility
following adverse moves in correlation. Similarly, the p/l from favourable moves in correlation is diminished as the trade
becomes long volatility under these conditions. This ‘volatilityscaling effect’ helps to account for the premium of variance
dispersion levels compared to correlation swaps and the relatively modest performance of variance dispersion trades in the
low volatility environment since 2003.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
56
Figure 64 : Implied correlation tends to trade at a premium to
realised
Euro Stoxx 50 6m correlation
0.30
0.40
0.50
0.60
0.70
0.80
0.90
Oct01 Oct02 Oct03 Oct04 Oct05 Oct06
6m implied correlation
6m realised correlation
Source : JPMorgan
Figure 65 : rolling p/l of variance dispersion reflects the spread
between implied and realised correlation, but is diminished by low
volatility
vega payoff correlation points
6%
4%
2%
0%
2%
4%
6%
8%
10%
Oct00 Oct01 Oct02 Oct03 Oct04 Oct05 Oct06
0.30
0.20
0.10
0.00
0.10
0.20
0.30
0.40
0.50
0.60
6m correlationweighted
dispersion p/l
6m i mplied minus subsequent
realised correlation
6%
4%
2%
0%
2%
4%
6%
8%
10%
Oct00 Oct01 Oct02 Oct03 Oct04 Oct05 Oct06
0.30
0.20
0.10
0.00
0.10
0.20
0.30
0.40
0.50
0.60
6m correlationweighted
dispersion p/l
6m i mplied minus subsequent
realised correlation
Source: JPMorgan
Note that variance swaps make possible trading equity correlation in this form. Managing a correlation trade through delta
hedged vanilla options on the index and all its constituents would be onerous, whereas unhedged vanillas (short index
options, long options on the members) would not give a true correlation exposure. However such pseudodispersion trades
using naked calls or straddles can give interesting and often attractive p/l profiles.
Indeed the demand for variance swaps as a means of trading correlation has been suggested as a reason for the relatively
good liquidity of variance swaps on equity underlyings in comparison to other asset classes such as bonds and FX where
vanilla options can be even more liquid than for equities.
Equity correlation can also be traded more directly through correlation swaps. For technical reasons, mostly to do with the
observed strong positive correlation between volatility and correlation, correlation swaps levels tend to trade somewhat
below the levels of correlation associated with variance dispersion trades. This tends to mean that for sellers of correlation,
variance dispersion almost always gives more attractive entry levels, at least optically.
See Correlation Vehicles, May 2005 for further details.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
57
3.8: Forward variance and volatility spikes
One of the strengths of variance swaps is the ease of pricing and constructing forwards (see p17). This ability to reexpress
the spot variance term structure into a strip of forward starting variances raises natural analogies with fixed income where
the swap curve can be decomposed into strips of forwards or futures.
The first analogy is to use forward variance to trade the future volatility of an underlying, in a similar way that
investors use Euribor, EuroDollar or Fed Funds interest rate futures to trade the future direction of interest rates. For
example, investors expecting a more prolonged hiking cycle from ECB will sell Euribor futures at far out maturities
expecting the yield curve to steepen (Figure 66). Likewise investors expecting a pick up in volatility that is greater than that
currently priced in, can buy forward starting variance swaps to express their view (Figure 67).
Figure 66 : Forward interest curve compared to history.
3month Euribor
1%
2%
3%
4%
5%
6%
1996 1998 2000 2002 2004 2006 2008 2010
3month Euribor (ex DEM)
3month implied Euribor f rom f utures
Source : JPMorgan
Figure 67 : Forward variance compared to history
3month volatility
0%
10%
20%
30%
40%
50%
60%
1996 1998 2000 2002 2004 2006 2008 2010
3month realised volatility
3month implied f orward variance
Source : JPMorgan
For a forward volatility position, p/l before the forward date will be entirely driven by changes in expectations of
volatility, as captured by movements in the implied volatility term structure. It is only once the forward date is
reached (and the shorterdated leg of the forward has expired) that there is direct exposure to realised volatility. In the same
way, p/l on a Euribor contract is driven by changes in expectations of the level of future 3month rates, and it is only on
expiry of the futures contract that p/l is finally settled against prevailing 3month spot rates.
Forward variance can be an attractive vehicle to be long volatility without accruing realised volatility losses that can occur
with a long spot variance position (see volatility risk premium as discussed in Sections 2.6 and 3.3). Given that investors
have a choice to trade forward volatility across a different maturities and forward dates, as well as being able to use spot
variance swaps, we use a variety of metrics to help us decide where to concentrate exposure. The following discussion
concentrates on index variance swaps.
One important point to note is the difference in behaviour between interest rates and realised volatility. Short term interest
rates are dictated by central bank action and typically move in a relatively smooth trend, hence forward interest can
reasonably accurately capture the future direction of such a trend. In contrast, realised equity volatility can be driven more
by short term spikes, with only a fairly slowly underlying “base” level change. This makes using forward variance to trade
the future path of volatility more unpredictable, as a correct call on the underlying trend can be ruined through an
unfortunate choice of timing. This problem can be mitigated by using longer term variance swaps, which will help to
smooth over the effect of any short term spikes.
Below we introduce the fixed income concepts of slide and carry as applied to volatility. In addition, the performance of a
forward volatility position crucially depends on the dynamics of the volatility term structure, which we characterise using
principal component analysis (PCA). Having done this, we will focus on the ability of long forward variance swaps to
hedge against volatility spikes.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
58
Volatility carry
Carry in fixed income is taken to be the cost of funding a spot position and reflects the difference between short and long
term interest rates. For example, an investor wanting to take a view that long term interest rates will rise will need to either
short sell a bond or short a long dated swap. In order to fund this position, they will have to pay the difference between the
long and short rate. Since yield curves tend to be upward sloping this is usually a positive spread, so the investor will have
to pay the “carry” to be short. Likewise a long investor will receive the carry from holding long maturity positions.
In variance swaps the holder of a long variance position will have to pay the spread of implied over realised every day they
hold the swap. Since this spread is typically positive (see Section 2.6), being long variance often means a negative carry
position, at least at the index level. Likewise, being short index variance swaps is usually viewed as a positive carry trade.
Note that for both fixed income and variance swaps, forward starting instruments have no exposure to carry until the
forward starting date is reached.
Volatility slide
Slide in both fixed income and variance swaps occurs because once a trade is implemented the maturity is fixed, so that
relative maturity diminishes over time. Slide measures the impact of this diminished relative maturity assuming that the
shape of the term structure remains unchanged over the period. For an upward sloping variance term structure – typical of
low volatility environments – a holder of a long variance swap will lose out through time as the variance swap ‘slides’ down
the term structure curve. Both spot and forward starting variance swaps have exposure to slide.
Principal component analysis
The behaviour of changes in the implied variance curve can be analysed using principal component analysis (PCA). PCA
decomposes moves in the volatility curve into its major components, or factors. Figure 68 shows the two most important
components of weekly changes in the Euro Stoxx volatility curve, using 6 years of data. The first, and most important,
factor illustrates how the dominant curve move is to increase (or decrease) much more at the shortend of the curve.
This illustrates how short dated implied variance tends to move about twice as much as longer dated maturities during a
shift up (or down) in the term structure, reflecting the tendency for increases in volatility to flatten the curve and for
decreases in volatility to steepen the curve (consistent with the roottime adjustment described above, see p42). The second
component of curve moves is more of a twist component, with volatility decreasing (or increasing) at the very shortend
of the curve and increasing (or decreasing) further out.
Figure 68 : Principal component decomposition of Euro Stoxx
volatility curve since 2000
Change in implied variance
2%
1%
0%
1%
2%
3%
0 1 2 3 4 5
First factor  " Shift "
Second factor  " Twist "
Maturity, Yrs
Source : JPMorgan
Figure 69 : Principal component decomposition of Euro Stoxx volatility
curve since 2000 … the first component has flattened through time
Change in implied variance
0%
1%
2%
3%
4%
0 1 2 3 4 5
Nov04  Nov06
Nov02  Nov04
Nov00  Nov02
Maturity, Yrs
Source : JPMorgan
If we also look at how the first component has changed through time, Figure 69 shows that primary curve moves have
become more parallel over the last 2 years. This is illustrated by considering the movements in the variance curve
following the volatility spikes in April 2005, October 2005 and May/June 2006 (Figure 73). These moves have often been
relatively parallel, at least up to the 2year maturity, consistent with the observed flattening of the first principal component
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
59
changes. At the same time however, secondary twistlike moves have become relatively more important, as shown in
Figure 71. This means that curve twists have become more independent and less well correlated to the direction of
volatility.
Figure 70: Movements in variance term structure predicted by the
principal components analysis model
10%
12%
14%
16%
18%
20%
22%
0 3 6 9 12 15 18 21 24
1M Vol spikes by 5%
1M Vol remains unchanged
1M Vol drops by 3%
Source : JPMorgan
Figure 71 : Euro Stoxx principal component analysis: weight of second
(twist) component has increased over the last 5 years
weight of 2
nd
component, %
0%
10%
20%
2005 2004 2003 2002 2001 2000
Source : JPMorgan
PCA analysis is useful in calculating expected gains or losses for different parts of the implied variance curve following a
general increase (or decrease) in volatility. In conjunction with carry and slide analysis for forward variance they can help
identify the best placed part of the curve to be long (or short) volatility.
Volatility Spikes
Below we use carry, slide and PCA analysis to help decide which long variance position (spot or forward, long or short
maturity) is likely perform well as a hedge against a volatility spike. Factors to consider include:
• The likely magnitude of a volatility spike and its effect across the variance term structure;
• The likely timing of a volatility spike;
• The cost of holding the position until the spike occurs;
• The possibility of term structure flattening, which could adversely affect the marktomarket of forward starting
variance.
Forecasting when a spike in volatility might occur is well nigh impossible. However, by decomposing a long history of
realised volatility into periodic components using Fourier transforms, one of higher weighted components of the
transform suggests that there is a bias for a spike to occur about every six months. If the shift up in volatility happens
immediately, the strategy that would benefit most is likely to be the spot shortdated variance swap, as the increase in
implied variance will likely be greatest at the short end of the curve.
As a final point it is worth considering, if a spike does occur, whether or not to unwind a long volatility position. If an
investor holding forward variance into a spike believes that fundamentals do not point to a raised base level of realised
volatility they would most likely look to unwind, whereas if they believed that the spike heralded the beginning of a higher
volatility regime they may look to profit by continuing to hold their long variance position. Also if a spike has not occurred
before the forward start, investors may consider rolling their position to a new forward starting variance position, so as to
avoid a realised volatility exposure.
Whilst it is important to consider the likely time of arrival of any spike and the effect such a spike is likely to have on the
variance term structure, the behaviour of volatility, both implied and realised, after a spike is also an important
consideration. Our analysis of spikes (in realised volatility) on major worldwide indices since the 1950’s (See European
Equity Derivatives Weekly, 30 May 2006), suggests the following results:
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
60
• There is not a strong pattern of reduction in volatility in the first month following a spike.
• In the second month after the spike, there is a clear bias for volatility to decrease, on average by about 4.5 vegas.
• For those spikes which subside in this manner (a reduction in volatility within 2 months) the reduction in volatility
is almost the same size as the increase seen in its spike. That is, shortlived spikes tend to subside to prespike
levels within 2 months (Figure 72)
• For those spikes which have not subsided within 2 months, it is likely that there has been a significant further
increase in volatility, and although these kind of spikes tend to eventually subside, they take longer and typically
subside to a significantly higher base level than the previous spike (on average 5.5 vegas higher after 6 months).
• The short lived spikes tend to outnumber the longerlived spikes by about 3 to 1.
Figure 72: Volatility tends to reset higher following longerlasting
spikes
1M realised volatility compared to prespike level
5%
0%
5%
10%
15%
20%
0 1 2 3 4 5 6
realised v olatility follow ing
spikes
... w here v ol decreases in
the follow ing month
... w here v ol increases in
the follow ing month
months after spike
Source: JPMorgan: Data for DJIA since 1900
Figure 73: Changes to the Euro Stoxx 50 variance term structure
during the volatility spike in May/June 2006
Implied variance
12%
14%
16%
18%
20%
22%
24%
26%
0 3 6 9 12 15 18 21 24
11May 06
22May 06
12Jun06
maturity (months)
Source: JPMorgan
Given that we can make an assumption on the likely timing of a spike and have a reasonable idea, from PCA, on how it will
affect the curve, we can estimate the likely p/l from holding different maturity (spot or forward starting) variance swaps.
We can then gauge which variance swap offers the best cost/benefit profile given the prevailing variance term structure.
Figure 74: Variance swap term structure as of 20 October 2006
Implied variance
12%
14%
16%
18%
20%
22%
24%
26%
0 2 4 6 8 10
Maturity
Source: JPMorgan
Figure 75: Expected cost of carry and slide over the next 3months
assuming realised volatility of 12%
1.5%
1.0%
0.5%
0.0%
3M 6M 9M 12M 15M 18M 21M 24M 27M30M 33M 36M
Forward starting maturity
Source: JPMorgan
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
61
Example 15: positioning for a volatility spike through forward starting variance
We consider the best placed forwardstarting 6month variance to position for a move up in volatility, given the prevailing
variance term structure on 20th October 2006 (see Figure 74).
For 6month variance swaps starting at various forward dates, we calculate the slide expected over the next 3months, which
gives our best estimate of the cost of holding that forwardstarting instrument (Figure 75). We find that the 9month
forward starting maturity is the one with the least slide. In fact, it is fairly typical that the 9 or 12 month forward starting
maturity has the least slide as it normally represents the place where the term structure begins to flatten out.
If a shift up in volatility happens immediately, the forward starting term that would benefit most is actually the spot
6month variance swap, as the increase in implied variance will likely be greatest at the short end of the curve. However, if
the spike were to happen in three months time, a spot 6month variance swap would have likely already have accrued a
negative p/l from the usual negative spread of realised to implied volatility (“carry”) and 3month variance would then have
to spike by about 4 vegas just to break even (consistent with 1year variance increasing by about 2 vegas, according to our
PCA analysis of the volatility term structure). In contrast, the 9 and 12month forward starting variance works relatively
well under several different timings of spikes.
Figure 76 and Figure 77 show the expected p/l of 6months variance swaps of various forward starts as a function of the
magnitude of the move in 1year variance. Figure 76 shows the projected p/l for a spike after 3months and Figure 77 for a
spike in 6months. Given any variance move at 1year maturity, the first principal component tells us how to shift the
whole curve and we can calculate p/l of all forward starting swaps. It is worth noting that a 4% move in the 1year maturity
equates to a move of about 7% in the 1month maturity, which is a fairly typical size for a volatility spike.
The shorter maturity forward starting variance does best if the magnitude of the shift is large and happens within
the next few months. If the move up happens later, the low realised volatility before the spike increases the breakeven of
this maturity. The 9month forward starting maturity does well in most scenarios as the low slide helps mitigate the
cost of holding the long volatility position.
Figure 76: p/l if spike in implied volatility were to happen in 3
months, assuming realised volatility of 12% until then
2%
1%
0%
1%
2%
3%
4%
5%
0% 1% 2% 3% 4%
3M start
6M start
9M Start
12M start
Change in 1year variance strike
Source: JPMorgan
Figure 77: p/l if spike in implied volatility were to happen in 4
months, assuming realised volatility of 12% until then
4%
3%
2%
1%
0%
1%
2%
3%
4%
0% 1% 2% 3% 4%
3M start
6M start
9M Start
12M start
Change in 1year variance strike
Source: JPMorgan
Thus for investors expecting a rise in volatility even within the next few months, the forward starting maturities can be a
better place to position for this than the spot maturities. Moreover, the forward starting maturities can provide some
cushion against the precise timing of any spike, which by their very nature are difficult to predict. Note though that entering
and exiting a forwardstarting variance swap can be more costly than running a spot variance swap to maturity, because the
forward has to be traded twice and because it has a (slightly) wider bidoffer spreads.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
62
Example 16: Performance of forward starting variance swap over volatility spikes
To illustrate the potential advantages of using forward starting variance swaps, we consider two specific examples.
As a way of minimising negative carry, we suggested using forward starting variance swaps for a long volatility view in
both January 2006 (European Equity Derivatives Weekly, 6th January 2006) and in April 2006 (European Equity
Derivatives Weekly, 18th April 2006), both of which would have benefited from the May/June volatility spike.
In the later case (April), the spike occurred soon after implementation and hence spot variance performed well; it had not
accrued much negative realised carry prior to the spike (Figure 78).
However in the January example, the spot variance performed relatively poorly, since by the time that the volatility spike
occurred, previous low realised volatility had increased the required breakeven level, and also the residual exposure to
implied volatility had fallen. In contrast the forward starting variance swaps did much better, participating more fully in the
spike in implied volatility. Moreover the long maturity 12month forward did almost as well as the 6month forward, which
had the added advantage of being an almost spot instrument at the time of spike (Figure 79).
Figure 78: The p/l of various long 6month term variance strategies
starting in May
Vega p/l 6month implied variance
2%
0%
2%
4%
6%
8%
10%
12%
18Apr 02May 16May 30May 13Jun
14%
16%
18%
20%
22%
24%
26%
6month spot p/l
6month, 12month forward p/l
6month implied v ariance (RHS)
Source: JPMorgan
Figure 79 The p/l of various long 6month term variance strategies
starting in January
Vega p/l 6month implied variance
4%
2%
0%
2%
4%
6%
8%
10%
06Jan 06Feb 06Mar 06Apr 06May 06Jun
14%
16%
18%
20%
22%
24%
26%
6month spot p/l
6month, 6month forward p/l
6month, 12month forward p/l
6month implied variance (RHS)
Source: JPMorgan
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
63
3.9: Trading the variance swap term structure
Whilst we have shown that variance swaps are useful for taking a direct view on the future value of implied and realised
variance, they can also be used to trade the shape of the variance term structure curve in analogy with the use of spreads and
butterflies in fixed income to trade the shape of the yield curve. One of the main uses of these types of trades is to express a
directional view, but at a more advantageous level or with reduced carry. Again this is best described using examples.
The first example attempts to take advantage of a flattening of volatility term structures (Figure 80). In this case, in
midMay 2006, the Euro Stoxx 50 1year2year implied variance spread had come close to its twoyear lows (Figure 81),
which in turn left 1year variance 1year forward close to its lows as seen in early 2005 (Figure 82). This meant that the
market was implying a similar outlook for forward realised volatility in May 2006 as it had at the beginning of 2005. This
appeared to be somewhat contradictory given the macro economic concerns prevailing at the time.
Investors, believing that these concerns may act to increase volatility, but wishing to avoid a direct long exposure to realised
variance, could have used the recent flattening to enter into 1year, 1year forward variance swaps. For example, the June
07/June08 forward variance was trading around 18.6%, compared to spot 1year variance of around 17%.
Whilst this forward volatility was low by historical standards, it was still well above the recent realised volatility of the Euro
Stoxx 50 and would suffer a negative slide of 1.6% over the next year, assuming an unchanged term structure. Hence a
long 1year variance 1year forward position was only likely to be profitable if implied volatility spikes or realised volatility
rises substantially over the next year.
Figure 80 : Euro Stoxx 50 curve had flattened up to 15th May
Variance swap level
10%
12%
14%
16%
18%
20%
22%
0 1 2 3 4 5
Current
2 months ago
Source : JPMorgan
Figure 81 : … and particularly in the 12 year part of the curve
Volatility Volatility
0%
2%
4%
6%
Mar04 Sep04 Mar05 Sep05 Mar06
0%
1%
2%
3%
5Y1Y (LHS)
2Y1Y (RHS)
Source : JPMorgan
Figure 82 : Euro Stoxx 50: 1 year 1 year forward variance had come
down relative to spot variance
volatility
10%
15%
20%
25%
30%
Mar04 Sep04 Mar05 Sep05 Mar06
f orward spot
Source : JPMorgan
Figure 83 : … especially accounting for the level of volatility
2y – 1y variance spread
y = 0.2001x + 0.0507
R
2
= 0.8639
6%
4%
2%
0%
2%
4%
10% 20% 30% 40% 50%
1y spot variance
Source : JPMorgan, data since Aug 2000
15 May 2006
15 Mar 2006
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
64
Investors unwilling to take on an outright long volatility exposure, could still take advantage of the relatively low forward
volatility strike by recognising that the spread between 1year and 2year variance was low compared to the level of 1year
variance (Figure 83). By combining a variance calendar (long 2year, short 1year) with some additional long 1year
variance (weighted as per the regression beta in Figure 83), investors could be long the curve whilst minimising outright
implied volatility exposure.
For example, the beta of 0.2 in Figure 83 suggested a trade ratio of long 1 unit of 2year variance and short 0.8 units of the
1year variance. This would be equivalent to trading the spread between forward and spot variance, hedged for volatility
directionality. Note that this position would have positive carry providing the market realises less than the 1year strike
(around 16.4%). Also note the trade ratio will need adjustment through time as the position ages.
The second example aims to take advantage of a Euro Stoxx 50 term structure that appears to be overly convex, (late
October 2006) due to the steepness of the volatility term structure between the 1 and 9 month maturity variance swaps.
This steepness can be observed by comparing the Euro Stoxx 50 term structure with that of the FTSE on the same date
(Figure 84), where a bulge is clearly visible in the Euro Stoxx 50 term structure that is not seen in the equivalent area of the
FTSE.
Whilst this could have reflected anticipation on the part of investors of an increase in volatility over the following months,
this slope appeared steep compared to the overall level of implied volatility (Figure 85). As the level of implied volatility
falls, the spread between the near term forwards and those further out usually widens. However as the May2006 volatility
spike receded this spread of forwards stayed narrow, reflecting an expectation on the part of investors for a nearterm move
up in implied volatility.
Figure 84: Euro Stoxx term structure bulged compared to FTSE …
Euro Stoxx 50 Implied variance FTSE implied variance
14%
15%
16%
17%
18%
19%
0 0.5 1 1.5 2
12%
13%
14%
15%
16%
17%
Euro Stox x FTSE
maturity , y ears
Source: JPMorgan
Figure 85: … meaning that near maturity Euro Stoxx forwards look
high compared to further out maturity forwards.
6month variance 9month forward – 3month forward 6month implied variance
4%
3%
2%
1%
0%
1%
2%
3%
Oct02 Oct03 Oct04 Oct05 Oct06
10%
15%
20%
25%
30%
35%
40%
45%
Spread of f orwards (lhs)
Six month implied variance (inverted, rhs)
Source: JPMorgan
Nonetheless, by assuming that the relationship between this spread and spot implied variance would hold, it can be seen that
a spread which is long 6month variance 9 months forward and short 6month variance 3 months forward, is equivalent (at
least according to a regression) to selling a spot 6month variance but at higher than prevailing levels.
Another way to think about a spread of forward variances is as a butterfly of spot variance swaps. In this case the long 1
unit of variance notional of 6month variance 9 months forward is constructed from long 2½ units of 15month spot
variance and short 1½ units of 9month variance. Similarly the short 1 unit of variance notional of 6month variance 3
months forward is constructed from short 1½ units of 9month spot variance and long ½ units of 3month variance (Section
1.7). In total this equates to a variance butterfly trade of long 2½ units of 15month spot variance, short 3 units of 9
month spot variance and long ½ a unit of the 3month spot variance. This example shows how butterflies can sometimes be
used to trade the underlying asset (i.e. variance) at more attractive levels than spot, an often used technique in fixed income.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
65
3.10: Skew and convexity trades
Since variance swaps are priced off the entire skew curve, they take on a significant exposure to skew and convexity.
This is particularly apparent when calculating the marktomarket: if skew and/or convexity instantaneously increase and
nothing else changes a long variance swap will profit. Variance swaps are long skew and convexity, at least if overall levels
of volatility (ATM volatility) do not decrease. Note that in theory exposure to skew is neutral, but typical shapes of skew
surfaces mean that increasing the skew increases the convexity, and really it is this which affects the variance swap price,
see p82 for further details.
Straddles vs. variance swaps
One possibility for owning convexity is to trade variance swaps against deltahedged straddles. Whilst variance swaps
are in effect long volatility and convexity, the exposure to volatility (both implied and realised) can be hedged out by
trading a short straddle against a long variance swap position, leaving a residual long convexity exposure. However, this
hedge is only effective providing that the underlying remains close to the straddle strike, since the straddle’s exposure to
volatility, both in the form of gamma and vega, will be reduced as the underlying trends away from the strike. In order to
maintain as pure as possible an exposure to convexity, the straddle will need to be restruck as the underlying moves.
Variance swaps plus naked short puts
One way of trading realised skew is to combine a put with a variance swap. For example consider an investor wanting to
express a long volatility position via variance swaps. The p/l at maturity is simply given by:
K
K
N l p
Vega
2
) (
/
2 2
−
× =
σ
However, we know from looking at past history that volatility tends to be directional, in particular increasing – often
substantially – in a sell off, and decreasing or maintaining its current level as the underlying rallies. This helps to explain
the commonly observed shape of the index skew curve, with relatively steep put skews representing the expectation of
increased volatility on the downside, and flatter call skews signalling the likelihood of more benign volatility on the upside.
To a first approximation this type of skew curve can look somewhat like the payoff from a long put, suggesting that it
should be possible to partially offset the observed negative correlation of volatility and market direction by trading
short naked puts against long variance swap positions.
Consider an investor who wishes to be long volatility, believing that realised volatility will be above that encapsulated by
the skew. To this end, he would be willing to be long volatility at a higher than prevailing variance strike if the market sells
off, in return for being long volatility at a lower strike if the market rallies. This exposure can be approximated by
combining a long variance swap with a short put.
Suppose we sell a put of strike S
P
for absolute premium P for N
Put
notional of underlying. The combined p/l of the short
put and long variance swap at time T is:
K
S K
N S S MAX P
S
N
K
K
N
T
Vega T P
Put
Vega
2
) ) (
~
(
)) 0 , ( (
2
) (
2 2
0
2 2
−
= − − +
− σ σ
where the adjusted strike is now dependent on the underlying’s price at maturity S
T
and is written as:
)) 0 , ( (
2
) (
~
0
2
T P
Vega
Put
T
S S MAX P
S N
KN
K S K − − − =
We can set the notional size of the put such that on the upside our breakeven is equal to the lowest variance realised if the
market has rallied.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
66
Example 17: Long variance swap plus short put
A Euro Stoxx 50 6month variance swap is trading at 16, the current market level is 4,077 and the 4,050 (99%) put is 175.
Since 1996, the lowest 6month realised volatility in a rally has been about 9%.
For €100K of vega exposure we risk losing €100K x ( 9
2
– 16
2
)/ (2 x 16) = –€547K. Thus to offset this loss we need to
raise the same amount through put premium. We do so by selling €547K x 4077 / 175 = €12.741M notional of the 4050
puts.
If the market were to fall, the losses on the put would act to increase the breakeven volatility for the long variance swap.
For example, if the market falls to 3,600 (11.7%) the equivalent breakeven realised would be 23, calculated from the
equation above:
Breakeven = sqrt{ 16
2
– 2 x 16 x [ 547 – 12,741 x ( 4,050 – 3,600 ) / 4,077 ] / 100 }
Conversely, if the market rallies, the premium retained from the put would act to lower the variance swap breakeven
For example, if the market rallies to 4400 (+7.9%) the breakeven volatility would be 9:
Breakeven = sqrt( 16
2
– 2 x 16 x 547/100 ), as per construction
Table 12 shows breakeven realised volatility levels for different market levels, for two different put overlays. The first
overlay targets an upside breakeven volatility of 9%, as per the calculations above. The second involves selling fewer puts,
which lowers the downside volatility breakeven at the expense of an increased upside breakeven – in this case 12%.
Figure 86 : Long variance and short put can better reflect the
tendency of volatility to rise is a selloff.
Breakeven / realised volatility
0%
10%
20%
30%
40%
50%
40% 20% 0% 20% 40%
v ariance @ 16 + 99% put
v ariance @ 16
realised since Jan1996
Last point
relative move in the underlying
Source: JPMorgan
Table 12: Volatility breakevens for long variance, short put
Market Level Variance + 4,050 Put + 4,050 Put
Target upside breakeven 9.0% 12.0%
3,000 16% 33.6% 28.6%
3,200 16% 30.5% 26.2%
3,400 16% 27.0% 23.7%
3,600 16% 23.0% 20.8%
3,800 16% 18.2% 17.4%
4,000 16% 11.4% 13.3%
4,200 16% 9% 12%
4,400 16% 9% 12%
4,600 16% 9% 12%
4,800 16% 9% 12%
5,000 16% 9% 12%
Source: JPMorgan
This concept of shorting puts against a long variance swap position, effectively neutralises some of the long downside puts
which are embedded within a variance swap (see Part 4). Taken to its extreme, combinations of short options can be used to
remove exposure to realised volatility completely if the underlying moves above or below a certain level. This observation
naturally leads to the concept of third generation volatility products such a conditional variance swaps (see “Conditional
Variance Swaps”, April 2006). Since conditional variance swaps are effectively priced from only part of the volatility
skew, they can be used to gain more direct skew exposures. For example the popular ‘long up95% conditional variance,
short 105% downconditional variance’ trade locks in a premium dependent on the level of skew, providing the underlying
remains within 5% either side of current spot.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
67
3.11: Cross asset class trades: trading equity volatility against credit
Investors may wish to consider trading variance swaps against credit default swaps (CDS) to exploit any discrepancies
between the equity and credit markets. Both variance swaps and CDS give some measure of the risk associated with a
particular company, and should be to some extent correlated. This can be seen when the relationship is aggregated to the
index level. Here credit and CDS indices such as Maggie and iTraxx show a strong relationship to Euro Stoxx 50 implied
variance in Europe (Figure 87) and in US the CDX HG and HY indices is well correlated to S&P variance (Figure 88).
Figure 87: Implied variance is related to credit at the index level in
Europe…
Maggie industrials spread Euro Stoxx 3m implied variance
0
10
20
30
40
50
60
Jan 01 Nov 01 Sep 02 Jul 03 May 04 Mar 05 Jan 06 Nov 06
0%
10%
20%
30%
40%
50%
60%
Credit Spreads (LHS)
Implied v ariance (RHS)
Source: JPMorgan
Figure 88: … and the same is true in the US.
CDX 5Y HG spread SPX 3m implied variance
0
20
40
60
80
100
120
140
Jan 01 Nov 01 Sep 02 Jul 03 May 04 Mar 05 Jan 06 Nov 06
0%
5%
10%
15%
20%
25%
30%
35%
40%
Credit Spreads (LHS)
Implied v ariance (RHS)
Source: JPMorgan
While credit and equity have a very strong relationship over the long run, there have been several instances where short
term opportunities have arisen in one of the asset classes, and where the other asset could have been used as a hedge against
a wider macro economic change in regime.
In May 2005, the credit markets soldoff following the downgrade of GM bonds to noninvestment grade and correlation
worries in structured credit. This credit selloff was particularly noticeable in the US, although was not reflected in a
commensurate increase in S&P equity implied volatility (Figure 88). Credit investors could have sold CDS protection,
taking advantage of the dislocation in prices. However, supposing that credit markets were actually heralding some macro
change in the economic environment, then it would be likely that equity volatility would in that case also rise. Buying
equity volatility against a long credit position could have provided a hedge to possible macro deterioration.
In May 2006, almost the opposite occurred, with equity volatility spiking, mainly in Europe, whilst credit markets remained
largely unmoved (Figure 87). Thus a short equity volatility/long CDS position could have taken advantage of the price
dislocation, whilst providing a certain degree of protection for a more macro change in regime.
Somewhat counterintuitively (5Y) CDS levels are more correlated with shortdated variance than with longerdated
variance, likely due to benchmark effects, with the most liquid variance swaps trading at shorter maturities, and longer
dated variance more susceptible to structured product flows.
See Synthetic Short CDS with Positive Carry, October 2006 for more details on debt/equity trades.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
68
Single stock variance swaps and debt / equity
Whilst indices can provide interesting short term opportunities to trade credit against equity variance, they do suffer from
basis risk arising from different index members. This can be particularly important in credit indices which may be driven
by idiosyncratic default events by single companies. Probably the majority of debt/equity trading is focussed on single
name capital structure analysis.
This analysis normally takes the form of attempting to model the relationship of a company’s share price to the price of its
credit and equity options, for example using reduced form or capital structure models. A typical trade to take advantage of
debt versus equity mispricing is a long/short combination of equity puts traded against CDS. One of the main deficiencies
of this approach, particularly for reducedform models based on regressions of share price versus CDS is that a company’s
share price is nonstationary, compared to CDS which is more mean reverting.
For example, consider the behaviour of Fiat CDS as a function of its share price (Figure 89). In the pre2004 period, the
CDS tended to widen when the share price fell through €15. However more recently it has taken a much lower share price
(below €10) to increase CDS. This is because leverage reduction has reduced the price at which the credit market assumes
that the company will become distressed. A simple regression model would assume that CDS behaved like a €15 put option
and a combined trade using this strike put against CDS would not correctly model the capital structure.
In contrast, because implied variance factors in some of this reduced risk, the relation between Fiat implied variance and
credit is more stationary (Figure 90). Hence, variance swaps can be useful instruments as vehicles for expressing a view on
the capital structure of a company. In practice, some combination of variance and the underlying equity may be used.
Figure 89: Fiat CDS has changed relationship with stock price …
5Y CDS spread
0
200
400
600
800
1000
1200
1400
0 5 10 15 20 25 30
Stock Price
Source: JPMorgan
Figure 90: … whereas Fiat CDS is more linearly related to variance
5Y CDS spread
y = 3567.9x  770.06
R
2
= 0.652
0
200
400
600
800
1000
1200
1400
20% 30% 40% 50% 60%
1Y implied variance
Source: JPMorgan
Although debt/equity trading is probably the major form of crossasset class trading, the advent of commodity variance
swaps introduces the possibility of further opportunities in crossasset arena. For example one could envisage trading
mining company variance against a basket of metals variance or oil company variance against crude oil variance.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 3: Uses of Variance Swaps
69
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
70
71
4. Replication and Hedging
• 4.1 Deltahedging and dollargamma
• 4.2 Theta – the cost of gamma
• 4.3 Options pathdependency
• 4.4 From options to variance swaps
• 4.5 Variance swap replication in one page
• 4.6 Sensitivity to skew and convexity
• 4.7 Variance swap Greeks
• 4.8 Setting up the replicating portfolio
• 4.9 Replicating and hedging in practice
• 4.10 Effects of variance swap hedging
• 4.11 Why not volatility swaps?
P
a
r
t
4
:
R
e
p
l
i
c
a
t
i
o
n
a
n
d
H
e
d
g
i
n
g
Part 4: Replication and Hedging
Overview
In Part 4 we consider the more technical issues surrounding variance swaps and particularly how they are priced and
replicated. From the perspective of marketmakers or others attempting to hedge variance swaps in the options market, the
replication argument is crucial, showing how (deltahedging) a portfolio of vanilla options can be used to reliably capture
realised volatility.
From the point of view of an investor trading a variance swap the hedging mechanism may seem unimportant. The investor
simply buys or sells variance by entering into the contract and receives or pays out an amount at expiry depending only on
the difference between the agreed strike and the level of variance realised. Even if the investor chooses to close out the
variance swap before expiry, the marktomarket is easily calculated (based on realised variance to date and prevailing
variance swap prices in the market) without reference to any other market variables.
Nevertheless, a knowledge of the mechanics of variance swap replication is instructive, helping to clarify how realised
volatility can be accurately captured, and where the prices of these instruments come from. In terms of pricing, variance
swaps can be observed to trade at levels of ATM volatility “plus a bit”. The “bit” depends on the steepness of the volatility
skew (and convexity) and the time to maturity, with steeper skews (higher convexity) and longer maturities both leading to
higher variance strikes. The replication argument shows unambiguously how the volatilities across the skew
contribute to the final variance swap price.
Perhaps the most important point to come out of the replication argument (for a nonreplicating investor) is that if a
marketmaker hedges a variance swap, this can, in theory, act against the investor’s interests: investor sells variance,
marketmaker sells options to hedge their long variance position and the deltahedging of these short options on the close
can act to increase the closeclose volatility of the underlying. This is in contrast to an investor who sells options to a
marketmaker who then delta hedges: the marketmaker is long options, and to replicate they will buy low / sell high at any
time during the trading day and hence act to reduce volatility. In this case the marketmaker will act in favour of the
investor’s interests. See detailed discussion in Section 4.10.
We begin the section with a review of deltahedged options, and particularly how p/l is generated by ‘trading the gamma’.
This exposure to gamma is paid for by the timedecay as measured by the theta. We define dollar gamma, which allows
deltahedging p/l to be calculated directly from the relative moves in the underlying. The main drawback of deltahedged
options as vehicles for trading volatility is that the dollargamma, which characterises the exposure to volatility, changes
over time and with the level of the underlying. This means that the exposure to volatility is path dependent and unless the
volatility is constant and the deltahedging instantaneous, then it is possible that an option could be purchased and delta
hedged at a volatility below that subsequently realised, but the p/l from this deltahedging program ends up negative.
To remove this pathdependency it is natural to look for a contract whose dollargamma is constant, both with the
passage of time and with changes in the underlying. We demonstrate that such a contract can be created with a static
portfolio of options across the continuum of strikes (weighted as 1/strikesquared). This provides an exposure to realised
volatility (squared) independent of the path of the underlying. A variance swap consists of this portfolio, bundled up with
the necessary delta hedging.
In fact, the variance swap replication argument is valid under much wider conditions than the relatively restrictive Black
Scholes assumptions, in particular it assumes very little about the underlying volatility process. After giving an informal
demonstration of the constant dollargamma of the replicating portfolio, we show that we can use this portfolio to capture
realised variance in a much more general setting. The variance swap replication argument can also be extended to apply to
the socalled 3rd generation instruments such as gammaswaps and conditional variance swaps, helping to understanding
how these products can be hedged and to explain their relative attractiveness from a marketmaker’s perspective. In turn,
ease of replication promotes liquidity and the attractiveness of such products to investors.
We go on to give a concrete example of how to construct a replicating portfolio and discuss various practical issues
associated with the hedging and replication of variance swaps.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
72
4.1: Delta hedging and dollar gamma
In this section we outline how vanilla options can be used to trade volatility. Options are exposed to a wide range of
factors (performance of underlying, timetomaturity, volatility, rates, dividends etc). The firstorder exposure to moves in
the underlying can be hedged out by the familiar deltahedging process described below. This leaves the exposure to
volatility, paid for in timedecay as the most important sensitivity. However, as we shall see, this is not a pure volatility
exposure, but is path dependent, varying over time and as the underlying moves towards and away from the option strike.
Suppose we hold a (call) option. Then to make the position instantaneously deltaneutral (insensitive to the direction of the
underlying) we can sell an amount of the underlying equivalent to the delta of the option. Delta measures the sensitivity of
the option to moves in the underlying. For an ATM option delta is about 0.5. For other options the delta is more dependent
on the level of volatility, with deltas of inthemoney calls tending towards 1 as the option becomes more in the money; and
doing so more quickly at lower levels of volatility. Similarly OTM call deltas tend to zero, more quickly for low volatility
(Figure 91). Note by convention deltas are often multiplied by 100, e.g. a delta of 0.30 is quoted as 30.
By frequently readjusting this deltahedge (in theory continuously, in practice usually daily) the sensitivity to direction of
the underlying can be hedged out over the lifetime of the option. Note that this is a dynamic hedge which must
continuously be monitored and rebalanced. P/l will come from the accumulated action of continuously rebalancing the
holding of the underlying, in order to keep the entire portfolio deltaneutral over time. This is referred to as the gamma p/l,
and is paid for in the option premium which is marked to market as lost theta.
Figure 91 : Deltas of 1month maturity ATM call options
Delta
0
0.2
0.4
0.6
0.8
1
75 80 85 90 95 100 105 110 115 120 125
10% v olatility
20% v olatility
40% v olatility
Underlying
Source : JPMorgan
Figure 92 : The convexity of the option
p/l
2
0
2
4
6
8
92 94 96 98 100 102 104 106 108
option v alue
deltaequiv alent amount of underly ing
pay off at ex piry
Positive convexity means the option outperforms
the deltaequivalent position in the underlying for
both up moves and down moves
Underlying
Source : JPMorgan
How is this gamma p/l actually made? In order to understand this, we must consider the p/l made over a single delta
hedging step, typically a day. Essentially the gamma (rate of change of the delta) measures the convexity of the option.
This convexity (curvature) always works in favour of long options positions, meaning that although a replicating hedge is
accurate for small moves in the underlying, for larger moves the long option will outperform the replicating hedge in both
directions (Figure 92). For a deltahedged option, the gamma p/l will be the outperformance of the option over the
replicating hedge. Gamma p/l is largest for shortmaturity, ATM options, where the chance of the option ending inthe
money can change rapidly, even for relatively small moves.
As spot moves the delta changes. The gamma measures the change in delta per unit of the underlying, and a long option is
long gamma. The more the delta changes, the more the replicating deltahedge will underperform the long options position
(Figure 92 again). How much will it underperform? The actual amount depends on the difference between the initial and
final delta of the option, which is gamma x dS, where dS is the change in the underlying. Assuming the gamma remains
constant, the option gains or loses value at the average delta (average delta = initial delta + ½ gamma x dS) whilst the
replicating hedge gains or loses value at the initial delta. Since the difference between the two is ½ gamma x dS, the
difference in performance between the option and the replicating hedge is ½ gamma x dS times the move in the underlying.
Overall option outperformance is therefore ½ gamma x dS
2
(Figure 93).
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
73
That is, the gamma p/l from a move in the underlying is proportional to the gamma of the option and the square of
the move. Thus, the p/l from gamma hedging will be large when gamma is high, and moves in the underlying are large.
For instance the gamma p/l from a 2% move will be four times that of a 1% move.
Figure 93 : Gamma p/l from deltahedged options
option v alue
initial delta
new delta
average delta
S
0
S
1
∆
0
dS
∆
1
(= ∆
0
+Γ
0
dS)
(∆
0
+∆
1
)/2 = ∆
0
+Γ
0
dS/2
p
0
p
1
= p
0
+ (
∆
0
dS+
Γ
0
dS
2
/2)
p/l =
Γ
0
dS
2
/2
Source : JPMorgan
Box 12: Deriving the gamma p/l
1. Spot is initially S
0
, option is worth p
0
, delta and
gamma are initially ∆
0
and Γ
0
respectively.
2. Spot increases, by dS, to S
1
3. Delta increases, to ∆
1
=
∆
0
+ Γ
0
dS
4. Average delta is therefore ∆
Avg
=
∆
0
+ Γ
0
dS/2
5. New premium p
1
= p
0
+ ∆
Avg
dS
6. p
1
= p
0
+ ∆
0
dS + Γ
0
dS
2
/2 (by 4 and 5)
7. p/l on long option is ∆
0
dS + Γ
0
dS
2
/2 (by 6)
8. p/l on short futures is ∆
0
dS
9. total p/l is Γ
0
dS
2
/2 (by 7 and 8)
Box 13: Calculating the gamma
The delta of a call option is given by: ) (
1
d N = ∆
where
( ) ( )
T
T r K S
d
σ
σ 2 ln
2
1
+ +
= and ( ) dt e x N
x
t
∫
∞ −
−
=
2
2
2
1
π
Since gamma is the rate of change of delta with respect to the underlying:
S ∂
∆ ∂
= Γ . differentiating ) (
1
d N w.r.t. S gives
T S
d N
σ
) (
1
′
= Γ where ( )
2
2
2
1
x
e x N
−
= ′
π
Since the delta of a put is given by 1 ) (
1
− = ∆ d N , differentiating has the same effect. Therefore the gamma of a put is
the same as the gamma of call.
Note that as spot S, moves away from the strike K, N′(d
1
)
will become small and gamma will decrease very rapidly. Then,
since N′(d
1
) is independent of the monetary units of the underlying, gamma will be inversely proportional to the value of
the underlying.
Gamma is a useful concept, but since it measures change in delta per unit of underlying, it is dependent on the absolute level
on the underlying. For example the gamma of an option on a stock worth €10 will be double the gamma of the equivalent
option on a stock worth €20 (with the same relative strike, volatility etc). Moving to the concept of the dollargamma is
more useful, since it allows us to directly calculate the gamma p/l for a given percentage underlying move, making it much
easier to compare gamma exposures across different underlyings.
To compute the dollar gamma and hence the actual cash p/l made from the deltahedging process we first define the dollar
delta, which is simply the cash equivalent exposure of the underlying. For example a position which is long $100,000
notional of call options, each with a delta of 30, will be long $30,000 of dollar delta. The dollar gamma is then defined to
be the change in the dollar delta for a 1% move in the underlying. This dollar gamma is therefore equal to the normal
gamma multiplied by the square of the value of the underlying and divided by 100 (Box 14).
S: underlying
K: strike
T: maturity
σ: implied volatility (at the strike)
r: interest rates
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
74
Box 14: Dollar delta and dollar gamma
Dollar delta: $∆ = cash value of the underlying delta hedge (for 1 share of the underlying $∆ = ∆ x S)
Dollar gamma: $Γ = change in $∆ for a 1% move in the underlying S.
It can easily be shown that $Γ = ΓS
2
/100 (S=>S+S/100; ∆=>∆+ΓS/100; $∆=>$∆+ΓS
2
/100)
Once the dollar gamma has been computed, the gamma p/l from a single deltahedging step is easy to calculate: for a
return R, the gamma p/l is equal to 50 $Γ x R
2
(Box 15). For example if we hold a position which is long $100,000 of
dollargamma and the underlying moves by 3%, the p/l will be 50 x $100,000 x 0.03
2
= $4,500.
Box 15: From gamma to dollar gamma
For a change in spot S => S + dS, the gamma p/l will be Γ dS
2
/2
Since $Γ = ΓS
2
/100, the p/l can be written as (100 $Γ)/ S
2
x dS
2
/2 = 50 $Γ (dS/S)
2
Gamma p/l = 50 $Γ R
2
where R is the return dS/S in percentage terms
Figure 94 : Dollar gamma is highest close to the option strike, and
falls away rapidly away from the strike, especially if volatility is low
Dollar gamma
0
1
2
3
4
5
6
7
75 80 85 90 95 100 105 110 115 120 125
ATM option
(20% v ol)
ATM option
(40% v ol)
90strike option
(20% v ol)
underlying
Source : JPMorgan
Figure 95 : Gamma p/l for a 100strike option with spot at different
levels. The p/l is quadratic in the size of the move.
p/l is $s
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
80,000
90,000
5% 4% 3% 2% 1% 0% 1% 2% 3% 4% 5%
spot@100
spot@95
spot@90
spot@85
size of daily move in the underlying
Source : JPMorgan
Example 18: Dollar Gamma
Suppose an underlying is trading at $100. We buy 1,000 contracts (100 shares each) of ATM call options with an implied
volatility 20%. Delta is 0.51. Therefore the equivalent cash holding in the underlying is 10M x 0.51= $5.1M
The gamma is 0.07 which means that the delta will change by approximately 0.07 for a $1 move in the underlying. E.g. if
spot moves to $99, delta will be approximately 0.44; if spot moves to $101, delta will be approximately 0.58.
The dollar gamma is given by multiplying gamma by the spot squared and dividing by 100. In this case it will be 100
2
x
0.07/100 = $7 per share. Since we hold options on 100,000 shares this equates to a dollar gamma of $700,000. This leads
to a p/l of $3,500 for a 1% move in the underlying, $14,000 for a 2% move and $87,500 for a 5% move (Figure 95).
If the stock drops by 5%, then with spot now at 95, the new dollar gamma (for the next time step) will be only $450,00
(Figure 94) and a subsequent 1% move will now lead to a p/l of only $2,250 (Figure 95).
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
75
4.2: Theta  the cost of gamma
Of course, this positive convexity from the gamma must be paid for, and it is paid through the time decay or theta. Theta
can be thought of as the amortized (nonlinear) cost of the option, spread over its lifetime. In order to make money on the
deltahedging strategy over a single deltahedging step (typically a day), it is necessary to make more on the gamma (from a
large daily move) than is lost on the time decay.
Box 16: Gamma vs. theta – calculating the deltahedging p/l
If delta hedging an option, the p/l in one time step is given by:   t R S δ σ
2 2 2
2
1
− Γ or   t R δ σ
2 2
$ 50 − Γ
where R is the return on the underlying, σ represents the implied volatility at which the option was purchased, and δt is the
length of the time step.
This comes from the relationship between theta and gamma: θ = − ½Γ S
2
σ
2
at the heart of the BlackScholes analysis.
(See e.g. Hull for further details.)
If deltahedging daily, the strategy will profit if the daily realised volatility, in the form of the square of the return, R, is
greater than the ‘daily implied volatility’ given by the option implied volatility scaled by the squared root of the timestep.
The above equation can be easily extended to give the p/l from deltahedging over the lifetime of the options, where the
summation is taken over all the deltahedging steps.
   
∑ ∑
− Γ = − Γ =
i
i
i
i i i
t R t R S l p δ σ δ σ
2 2 2 2 2
$ 50
2
1
/
Note that the only reason the exposure to volatility is nonconstant is the changing gamma. See Carr and Madan for a
thorough dissection of the p/l from deltahedging a vanilla option, with possibly nonconstant volatility.
Example 19: Daily breakevens
Suppose an option is bought at 25% implied volatility and deltahedged daily. Then the daily breakeven return can be
calculated by deannualising the implied volatility (dividing by the squareroot of 252) to get a value of 1.57%.
If the underlying moves by more than 1.57% in either direction the long option position will profit. If it moves by less than
this breakeven it will lose. The magnitude of the gain or loss is determined by the dollar gamma, with large dollar gammas
acting to magnify both gains and losses.
If deltahedging weekly, the weekly breakeven will be 25%/sqrt(52) = 3.5%. Whilst monthly deltahedging would give a
breakeven of 7.2%.
Vega and other sensitivities
In reality the implied volatility of the option will change throughout the lifetime of the option. This will affect the option p/l
directly: if volatility goes up, the option becomes more valuable; but also indirectly since the hedging quantities dictated by
the delta will also change. Of course, if holding an option to maturity (and deltahedging at the original implied volatility)
the vega p/l over the lifetime of the option will be zero and the total p/l will the be the gamma vs. theta p/l described above.
However, in practice since the changes in implied volatility alter the hedging amounts, the actual effects can be more
difficult to quantify.
Option prices are also sensitive to changes in rates (measured by the rho) and to changes in dividends (sometimes call mu).
However, for ease of exposition we assume zero interest rates and an underlying with zero dividend yield, unless stated
otherwise.
i
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
76
4.3: Options pathdependency: can volatility be captured by deltahedging?
Suppose a marketmaker buys and deltahedges a vanilla option. Assume that whatever volatility is realised is constant and
the option is deltahedged over infinitesimally small time intervals. Then the marketmaker will profit if and only if realised
volatility exceeds the level of volatility at which the option was purchased. However:
1. The magnitude of the p/l will depend not only on the difference between implied and realised volatility, but where
that volatility is realised, in relation to the option strike. If the underlying trades near the strike, especially close to
expiry (high gamma) the absolute value (either positive or negative) of the p/l will be larger.
2. If volatility is not constant, where and when the volatility is realised is crucial. The differences between implied
and realised volatility will count more when the underlying is close to the strike, especially close to expiry.
Indeed, for nonconstant volatility, it is perfectly possible to buy (and deltahedge) an option at an implied volatility
below that subsequently realised, and still lose from the deltahedging. For example consider the performance of delta
hedging 1year Euro Stoxx 50 options from 1st November 2001 to expiry 1year later on 1st November 2002. The index
was initially at 3500 (with ATM implied volatility at 28.5%) and up until May 2002 remained in the range 35003800,
realising around 20% volatility. After May, the index fell rapidly to around the 2500 level, realising high (around 50%)
volatility on the way. Over the whole year, realised volatility was 36%.
Consider first the performance of the 2500strike option (Figure 96). This was purchased at 33.4%, (above the 28.5% ATM
volatility due to the put skew). Initially the option made a loss due to the low realised volatility, but this was kept small by
the low gamma exposure so far away from the strike. However, as the index sold off toward the 2500 strike, increasing the
gamma, volatility picked up dramatically and the option ended up making a profit of 120 index points (ip), or about 18
vegas.
In contrast the 4000strike option (Figure 97) initially had much higher gamma than the 2500strike option and
consequently lost more due to the lower realised volatility. When the volatility did occur, this happened well away from the
option strike, and the low gamma meant that the deltahedging failed to capitalise on this high volatility with the delta
hedged options ending up making a loss of 60 ip (about 5 vegas). This is despite the fact that the option was purchased at an
implied volatility of 26%, 10% below that subsequently delivered.
Figure 96 : The 2500strike option profited due to the high volatility
and gamma close to expiry…
Index level option p/ (ip)l
2000
2500
3000
3500
4000
Nov01 Feb02 May02 Aug02 Nov02
50
0
50
100
150
spot
strike
long option p/l
Source : JPMorgan
Figure 97 :…whereas the 4000strike option made a loss, despite
being purchased at implied volatility 10% below that later delivered
Index level option p/ (ip)l
2000
2500
3000
3500
4000
Nov01 Feb02 May02 Aug02 Nov02
100
80
60
40
20
0
20
spot
strike
long option p/l
Source : JPMorgan
When deltahedging an option, the p/l is a (daily) accrual of the difference between the realised volatility (in the form of the
return squared) and the implied volatility scaled by the time step. However, the magnitude of the contribution of this (daily)
accrual is weighted by the current dollar gamma of the option, which as it depends on the path of the underlying, is
inherently unpredictable (see Box 16).
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
77
The nonconstancy of exposure to realised – implied volatility (squared) comes from the dollar gamma changing over the
lifetime of the option. Thus the difference between realised and implied volatility will count more when (dollar) gamma is
high. An atthemoney option will have high exposure to the spread between implied and realised volatility whilst the
underlying remains close to its initial value (Figure 98), whilst an OTM option will take on no significant exposure to
volatility until the underlying trades at around the option strike (Figure 100).
Figure 98 : Simulation of p/l for hedging an ATM option: magnitude of
returns is highest if underlying remains close to the money.
p/l
2
1
0
1
2
60 80 100 120 140
underlying at maturity
Simulation: option struck at 100, implied volatility=20%, maturity = 100 days, flat skew
Source : JPMorgan
Figure 99 : … returns are correlated with volatility but not perfectly,
exposure is locally linear for small changes in volatility.
p/l
2
1
0
1
2
14% 16% 18% 20% 22% 24%
realised volatility
Simulation: option struck at 100, implied volatility=20%, maturity = 100 days, flat skew
Source : JPMorgan
Figure 100 : Simulation of p/l for hedging an OTM (80strike) option:
magnitude of p/l is highest when underlying is close to the strike
p/l
1.0
0.5
0.0
0.5
1.0
60 80 100 120 140
underlying at maturity
Simulation: option struck at 80, implied volatility=20%, maturity = 100 days, flat skew
Source : JPMorgan
Figure 101 : … however, due to the mostly low gamma, the exposure
to volatility is poor
p/l
1.0
0.5
0.0
0.5
1.0
14% 16% 18% 20% 22% 24%
p
o
r
t
f
o
l
i
o
2
p
/
realised volatility
Simulation: option struck at 80, implied volatility=20%, maturity = 100 days, flat skew
Source : JPMorgan
To summarise:
• The p/l of a deltahedged option does (generally) depend on the spread between the (square of the) implied
volatility at which the option was purchased and the subsequent (squared) realised volatility
• This dependence is relatively strong if the underlying trades close to the strike of the option
• But the sensitivity to realised volatility tails off rapidly as the underlying moves away from the option strike
• This sensitivity is accounted for by the (dollar) gamma. If there was such an instrument where the dollar
gamma remained constant both over time and for changes in the underlying, then the exposure to volatility
would be constant.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
78
4.4: From options to variance swaps
In order to create a portfolio of options which has constant exposure to volatility (actually variance) we need to ensure it has
a dollar gamma which is constant for moves both in the underlying and over time. The p/l from deltahedging this portfolio
would then be just a constant multiplied by the squared returns summed over the hedging steps, otherwise known as realised
variance. The cost of this portfolio represents the price of exposure to realised variance. When considering how to
construct such at exposure, there are three possible approaches to take:
1. Use a single vanilla option, but buy or sell additional amounts of this option to keep the dollar gamma constant
over time. This has the advantage of using only a single option strike, but the option itself needs to be dynamically
traded, and the position could end up with enormous amounts of the option as the underlying moves away from the
strike and the gamma of a single option decreases; suggesting…
2. Restrike the option to maintain a constant gamma. Start with an ATM option and on each rehedging step,
either sell or hold the previous option, and buy an amount of the new ATM to achieve the constant gamma
exposure. This is better than (1) but still requires (potentially costly) dynamic trading of options; finally …
3. Construct a portfolio of options in such a way that the dollar gamma remains constant over both moves in
the underlying and with the passage of time. This has the advantage of not requiring the dynamic trading of
options, but the initial portfolio turns out to require a strip of options across a continuum of strikes.
It is this third approach that is used to construct variance swaps, at least in theory – in practice it is more like a combination
of (2) and (3) above (see Section 4.9). This method has the advantage of using only a static options portfolio – although the
options themselves must still be dynamically deltahedged with the underlying. Perhaps more importantly this approach is
to some extent independent of the volatility process driving the underlying. The first two approaches would require
continually calculating the gamma over the course of the trade, and this gamma will be highly (and dangerously) dependent
on the assumed implied volatility (and volatility process) at that time.
To achieve a constant dollar gamma across strikes what kind of portfolio is needed? One important observation is that
(peak) dollar gamma increases linearly with strike (Figure 102). Naively, it may be thought that weighting the options in
the portfolio (across all strikes) by the inverse of the strike will achieve a constant dollar gamma. It does have the property
that each option in the portfolio has an equal peak dollar gamma (Figure 103). However, the dollargammas of the higher
strike options ‘spread out’ more, and the effect of summing these 1/Kweighted options across all strikes still leads to a
dollargamma exposure which still increases with the underlying (Figure 105). In fact, in can be shown that this increase is
linear, and therefore weighting each option by the inverse of the strikesquared will achieve a portfolio with constant dollar
gamma (Figure 104, Figure 105).
Figure 102 : Peak dollar gamma of an option increases linearly with
the underlying
Dollar gamma
0
1
2
3
4
5
6
7
8
0 20 40 60 80 100 120 140 160 180 200
20 40
60 80
100 120
140 160
180 200
underlying
Source : JPMorgan
Figure 103 : Peak dollar gamma of options divided by strike is
constant
Dollar gamma
0.00
0.01
0.02
0.03
0.04
0.05
0 20 40 60 80 100 120 140 160 180 200
20 40 60 80
100 120 140 160
180 200
underlying
Source : JPMorgan
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
79
Figure 104 : Weighting options by the inverse of the strikesquared
gives a portfolio with aggregate constant dollargamma
Dollar gamma
0.0000
0.0002
0.0004
0.0006
0.0008
0.0010
0 20 40 60 80 100 120 140 160 180 200
20 40 60
80 100 120
140 160 180
200
underlying
Source : JPMorgan
Figure 105 : Weighting options as the inverse of the strike squared
gives constant dollargamma
Dollar gamma
0.0
0.5
1.0
1.5
2.0
2.5
3.0
0 50 100 150 200
portfolio 1: constant w eighting
portfolio 2: w eighted as 1/K
portfolio 3: w eighted as 1/K^2
underlying
Source : JPMorgan
One way of explaining why a portfolio of (Europeanstyle) options weighted as the inverse strike squared gives a constant
exposure to volatility is as follows. Recall (Box 16) that the p/l at each hedging interval is given by:
 
∑
− Γ =
i
i i i
t R S l p δ σ
2 2 2
2
1
/ . Only the
2
i i
S Γ term prevents this being a direct exposure to variance.
Thus if we find a portfolio with gamma proportional to 1/S
then the exposure to implied minus realised variance will be
constant. Since the gamma of a portfolio represents the second derivative of the portfolio’s value with respect to the
underlying, this means that the second derivative of the portfolio’s value must be proportional to 1/S
2
.
As ( )
S
S
dS
d 1
ln
−
= − and ( )
2 2
2
1
ln
S
S
dS
d
= − , the negative natural log of the underlying represents such a payoff.
In fact (by integrating 1/S
2
twice) any function of the underlying S of the form c bS S a + + − ln will have a constant
dollar gamma (of a). The action of deltahedging this combined portfolio will recover the realised volatility of the
underlying. Note that the bS + c terms represent a static positions in bonds (c) and forwards (b) with no volatility exposure.
Unfortunately, log contracts are not traded in the market. But we can replicate such a contract with vanilla options using an
infinite sum of calls and puts across the continuum of strikes, each weighted by the inverse square of strike. Integrating the
value of this portfolio at expiry, demonstrates that the nonlinear part of the payoff is indeed a negative log contract:
( ) ( ) ( )
∫ ∫ ∫
−
=
−
+
−
∞
+ +
T
S
S
T
S
T
S
T
dK
K
K S
dK
K
K S
dK
K
S K
0 0
0
2 2
0
2
( ) ( ) 1 ln
1
ln
0
0
− + − − = S S
S
S
T T
Note (SK)
+
is a shorthand for Max(SK,0) – e.g. the payout from a call option on S with strike K. A put is (KS)
+
.
To summarise:
1. A portfolio of calls and puts, weighted as 1/strikesquared, has constant dollargamma;
2. Deltahedging this portfolio provides constant exposure to the difference between implied and realised variance
regardless of where the volatility is delivered;
3. Hence the p/l from deltahedging this portfolio is proportional to difference between realised and implied variance.
A variance swap can therefore be created by replicating a log contract with options which are then deltahedged. Although
this argument is easy and intuitive, it does depend on certain strong (BlackScholes) assumptions on the underlying. The
power of the more general replication argument outlined below lies in the fact that it applies under much less restrictive
conditions.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
80
4.5: Variance swap replication in one page
In this section we make few assumptions: namely continuous trading is possible, complete markets (for options of all
strikes) exist, no transaction costs, and the underlying (futures contract) follows a diffusion process (with no jumps).
Let F
t
for t in the interval (0,T) be the price for a future on the underlying maturing at some time after T.
Let ( )
t
F f be the function ( )
− +


.

\

= 1 ln
2
0
0
F
F
F
F
T
F f
t
t
t
Therefore ( )
− = ′
t
t
F F T
F f
1 1 2
0
and ( )
2
2
t
t
TF
F f = ′ ′ Note that both ( )
t
F f and ( )
t
F f ′ disappear at
0
F F
t
=
Using Ito’s lemma: ( ) ( ) ( ) ( )
t
T
t T
T
t t
t
dF F f F f F f dt F f
F
∫ ∫
′ − − = ′ ′
0
0
0
2
2
2
σ for any smooth function f
Hence substituting f as above gives:
∫ ∫
− −
− +


.

\

=
T
t
t
t
t
T
t
dF
F F T F
F
F
F
T
dt
T
0
0 0
0
0
2
1 1 2
1 ln
2 1
σ
The LHS of the above equation represents the average realised variance of the future over the period (0, T). The final term
represents a continuously rebalanced futures position, which is easily replicated and costless to initiate. The first term on the
RHS represents a static position in a contract paying f(F
t
) at expiry. This can be replicated with options:
This replicating portfolio can be found directly by using the fact that every smooth (twice differentiable) payoff of futures f,
can be replicated by a portfolio of cash, the underlying and options as follows (see Carr and Madan for details):
( ) ( ) ( ) ( ) ( )   ( )( ) ( )( )
∫ ∫
∞
+ + + +
− ′ ′ + − ′ ′ + − − − ′ + =
κ
κ
κ κ κ κ dK K F K f dK F K K f F F f f F f
T T T T T
0
for any κ.
Using the function f above, and κ = F
0
gives:
( ) ( )
∫ ∫
∞
+ +
−
+
−
=
− +


.

\

0
0
2
0
2
0
0
1 ln
F
T
F
T T
T
dK
K
K F
dK
K
F K
F
F
F
F
Thus (at expiry)
( ) ( )
∫ ∫ ∫ ∫
− −
−
+
−
=
+ +
T
t
t
F
T
F
T
T
t
dF
F F T
dK
K
K F
T
dK
K
F K
T
dt
T
0
0
0
2
0
2
0
2
1 1 2 2 2 1 0 0
σ
Therefore holding a portfolio of calls and puts, weighted according to the inverse of the strike squared, and simultaneously
holding 1/F
t
 1/F
0
of futures at time t will pay out average realised volatility squared at expiry. (Note, in a BlackScholes
environment, that this continuously rebalanced futures holding encapsulates the deltahedging of the options.) Therefore
the forward cost of this portfolio is simply:
( ) ( )
+ =
∫ ∫
∞
0
0
2
0
0
2
0 2
2
F
F
rT
VAR
dK
K
K C
dK
K
K P
T
e
K
The strikesquared, K
VAR
2
, can be thought of as the (futurevalued) inverse strikesquared weighted sum of the time values
of the options portfolio
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
81
4.6: Sensitivity to skew and convexity
Skew is commonly thought of as an important component of variance swap prices, with put skews seen as having a much
greater impact on prices than call skews. Although, in practice, this is a useful framework for thinking about how variance
swap prices behave, it is not theoretically correct.
For example, suppose an underlying has 3month ATM volatility of 20%, a linear put skew of 5% (per 10 volatility points),
with put volatilities capped at 35% and all OTM call volatilities flat at the level of the ATM volatility (Figure 106). The
theoretical 3month variance strike can then be calculated to be 23.05 (pricing OTM options 1 point apart between 1 and
300). Now consider a situation where the skew is a mirror image of that in Figure 106. ATM volatility is 20%, and all
OTM put volatilities are flat at 20%, but call volatilities increase linearly by 5% per 10 points as they become more OTM,
capped at 35% (Figure 107). In this case the theoretical variance swap price is virtually identical at 23.15.
Figure 106: High put skew increases the variance strike above ATM
volatility…
Volatility / strike level
0%
5%
10%
15%
20%
25%
30%
35%
40%
50 60 70 80 90 100 110 120 130 140 150
skew
v ariance strike
N'(d2)
relative strike
Source : JPMorgan
Figure 107 : …but increasing the call skew has almost exactly the
same effect.
Volatility / strike level
0%
5%
10%
15%
20%
25%
30%
35%
40%
50 60 70 80 90 100 110 120 130 140 150
skew
v ariance strike
N'(d2)
relative strike
Source : JPMorgan
Why does this symmetry exist? Since the 1/strikesquared replicating portfolio has a much higher weighting of puts, it
would be natural to assume that their associated implied volatilities should have a proportionally greater effect on the
variance swap strike. Also, in practice, variance swaps can effectively be priced and hedged with ATM volatility plus a
contribution from the skew.
To answer the first point, the exposure to the skew curve is symmetrical. That is, the contribution to the variance swap
price of the volatility of an OTM call is exactly the same as from an OTM put with same (riskneutral) probability, N(d2), of
ending in the money. The OTM puts have a greater weighting in the replicating portfolio, simply because their dollar
gamma is lower and so they must be scaled up in order to provide constant dollar gamma across the range of strikes.
For the second point, in reality, equity skews look much more like Figure 106 than Figure 107, at least for short dated index
volatility. This explains why Derman’s approximation assuming a linear skew is most successful in these conditions (See
Section 2.4). Directionality of volatility, at least in the case of indices, makes volatility skews highly asymmetric, with
OTM calls trading much closer to ATM volatility than the corresponding OTM puts.
So what exactly determines the contribution of volatilities across the skew curve to the variance swap price? Clearly a very
outofthe money option should make a relatively small contribution to the variance swap price. (This must be true in order
to replicate variance swaps in practice, otherwise the value of the replicating portfolio will be very sensitive to the exact
choice of replicating portfolio.) Alternatively, one can think of the chance of needing the constant dollar gamma in the area
represented by the very outofthemoney option as being very small. It then seems sensible that a variance swap represents
a kind of weighted average of volatilities across the skew curve, with the closertothemoney volatilities higher weighted.
In fact, this is exactly the case, with the average being taken over the variances rather than the volatilities, and the weighting
function simply being the riskneutral probability density function, N′(d2), that the corresponding OTM option ends up in
themoney.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
82
If we define the variable z to be the standard BlackScholes parameter d2:
( ) ( )
T
T r K S
d z
σ
σ 2 ln
2
2
− +
= = then;
It can shown that: ( ) ( )dz z z N K
BS VAR
2 2
σ
∫
∞
∞ −
′ =
The N′(d2) term is the probability density function for the underlying at expiry, T. That is, the cumulative distribution
N(d2), gives the (riskneutral) probability that the underlying will be trading above z at time T. Thus the parameter z,
simply represents the ‘moneyness’ of the corresponding OTM option. This means that the variance swap price is a
weighted sum of squared option implied volatilities weighted by the probability that the (OTM) option will end in
themoney. See Gatheral 2006 for a detailed of the proof of this.
Furthermore, if the skew curve is quadratic in the variable z (the moneyness of the option) of the form:
( )
2 2
0
2
z z z
BS
β α σ σ + + =
then substituting and integrating gives T T K
VAR
β σ + =
2
0
, i.e. in ‘d2space’ the variance swap price is not affected by
the linear component of the skew α, but only on the base level of volatility σ
0
, and the convexity parameter β.
This explains the phenomena seen in Figure 106 and Figure 107: very different (linear) skews, but similar convexities give
almost identical variance swap strikes. The reason that there is any difference at all in variance swap strikes, is because in
d2space, the convexity in Figure 107 is slightly greater, since the maximum volatility is at a strike of 130, which is slightly
less outofthe money than the 70 strike where the maximum volatility is achieved in Figure 106. Indeed, if we keep the
‘linear’ skew at a slope of 5% or 5% but do not apply the cap until it reaches 50%, which it will do at strikes of 40 and 160
respectively, the version with the upward sloping call skew (similar to Figure 107) will now price about 80bps above the
version with the downward sloping put skews (similar to Figure 106) since a 160 strike is significantly more inthemoney
that the 40 strike.
This also helps to explain why the convexity has a greater effect on longer maturity variance. As maturity increases, the
probability of far outofthemoney (in relative strike terms) options ending in the money increases. Therefore the relative
weight of e.g. the 160strike call will increase with maturity. At shorter maturities perhaps only the 80120 portion of the
skew surface significantly affects the variance swap price, and this part of the skew is relatively linear (at least for indices).
At 5 years, strikes out to say 50 180 become relevant to the pricing and the convexity of these strikes can be much higher.
In practice, skew is most often thought about in terms of relative strikes, rather than moneyness – but the point remains: the
contribution of a point on the skew curve to the variance swap price depends on the (riskneutral) probability N(d2) of the
associated OTM option ending inthemoney. Therefore ATM volatility will provide the greatest contribution to variance
swap prices – particularly for short maturities. Both high put skews, and high call skews (where OTM calls have higher
volatilities than ATM) will increase variance swap prices. In fact there is some kind of feedback effect here because the
contribution of each volatility is determined by the probability of an option being inthemoney, but, for example high put
skews will increase the probability of OTM puts ending in the money (greater volatility on the downside makes OTM puts
easier to reach) so both the volatility and the weighting factor will increase.
To conclude, from the equation: ( ) ( )dz z z N K
BS VAR
2 2
σ
∫
∞
∞ −
′ = we can make a number of specific observations:
1. Given a flat skew, variance will price at the same level as ATM volatility.
2. Positive convexity will always act to increase variance swap strikes.
3. With a negatively convex skew – OTM volatilities are (on average) less than ATM volatility – it is theoretically
possible that variance could price below ATM volatility.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
83
4.7: Variance swap Greeks
In this section we consider the Greeks for variance swaps, giving information about the sensitivity of variance swaps to
various market variables. We work directly by differentiating the marktomarket value of the variance swap contract. An
alternative approach would be to calculate the replicating portfolio and sum the Greeks of the contributing options. Note for
simplicity of exposition we assume a world with zero interest rates; alternatively in the presence of nonzero rates, the
following parameters represent the forward Greeks and would have to be discounted by the appropriate discount factor. In
analogy with the Greeks for vanilla options, we return to considering realised volatility and variance swap strikes in
percentage terms rather than the standard wholenumber units.
The value of a variance swap, per unit veganotional, at time t is given by:
 
2
0
2
,
0
2
1
K
K
P
t Expected t
− = σ
where K
0
is the initial (fixed) variance swap strike and σ
Expected, t
is the expected realised volatility, at time t, between trade
inception and maturity. This expected volatility is then calculated in the same way as the marktomarket p/l as a
combination accrued realised variance to date and future implied variance. Note also that intraday there is a term
representing the square of the move which will act to give the variance swap delta on an intraday basis:
2
,
2
1
2
1 , 0
2
,
ln
252 1
T t
t
t
t t Expected
K
T
t T
S
S
T T
t −
+


.

\

+
−
=
−
−
σ σ
where σ
0,t1
is the realised variance accrued to day t1, and K
t,T
is the strike of a variance swap on day t expiring at T. S
t1
is the value of the underlying at the close on day t1, and S
t
is the value of the underlying at the valuation time on day t.
The Greeks of the variance swap can then be calculated by differentiating P
t
.
Gamma:
Gamma comes only from the exposure to realised volatility on each day:
= Γ
=
∂
∂
2
0
2
2
1 252
t t
S T K S
P
Since the dollar gamma is achieved by scaling the gamma by the spot squared, this gives a constant dollar gamma as
expected (ignoring higher order errors).
Theta:
Calculating the theta of the variance swap gives:
= θ
0
2
,
2TK
K
t
P
T t
−
=
∂
∂
In particular, if the variance strike doesn’t change, theta remains constant. Note, that since T is measured in days this value
represents a daily theta. Annualising gives a value for theta of:
0
2
,
2
252
TK
K
T t
−
= θ which can then be shown to satisfy the formula
2 2
2
1
σ θ S Γ − = taking σ
2
as the implied variance from
the variance swap strike (see Box 16, p76).
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
84
Vega:
We can calculate exposure to volatility in terms of (i) sensitivity to changes in ATM volatility, (ii) variance strike or (iii)
implied variance (strike squared).
(i) To compute the vega in terms of sensitivity to ATM volatility, we must make some assumptions about how the variance
swap strike relates to ATM volatility. If we assume the Derman approximation: ( ) ( )
2 2 2
,
3 1 skew t T K
ATM T t
− + = σ then:
( ) ( ) ( ) ( ) ( )
2
0
2 2
0
3 1 3 1
2
1
skew t T
T
t T
K
skew t T
T
t T
K
P
ATM
ATM
ATM ATM
− +
−
= − +
∂
∂ −
=
∂
∂ σ
σ
σ σ
Note also that this approximation allows the calculation of sensitivity to the skew or the skew squared:
( )
2
2
0
2
2
3
ATM
T
t T
K skew
P
σ
−
=
∂
∂
and
( )
skew
T
t T
K skew
P
ATM
2
2
0
3
σ
−
=
∂
∂
i.e. increasing skew will increase the value of the variance swap, and do so by more, if there is more time remaining until
expiry. (Compare Figure 21)
Computing sensitivities to the strike (or strike squared) is more straightforward and needs no such assumptions about the
skew surface:
(ii )
0
,
,
K
K
T
t T
K
P
T t
T t
−
=
∂
∂
and (iii)
−
=
∂
∂
T
t T
K K
P
T t 0
2
,
2
1
.
These all tell us that the exposure to implied variance (or volatility) decreases with time as the accrued realised volatility is
locked in to the p/l.
Delta:
Firstly, assuming that the variance strike K has no sensitivity to the underlying, the variance swap can be seen to take on
delta only intraday:
= ∆
t t
t
t t
t
t
S S
S
T K S S
S
K T
t T
S
P 1
ln
252 1
ln 2
252
1 0 1 0


.

\

=


.

\
 −
=
∂
∂
− −
This represents the replication of the log contract which will have to be done at the end of the day to capture that day’s
realised variance.
If the variance strike itself has a dependency on the underlying (implied variance is directional) then the variance swap
acquires other sources of delta, in addition to the intraday delta discussed above.
For example, using Derman’s approximation (see p30) ( ) ( )
2 2 2
,
3 1 skew t T K
ATM T t
− + = σ where σ
ATM
is the Black
Scholes ATM implied volatility (or atthemoney forward volatility if rates or dividends are nonzero).
Calculating delta (only on the close to avoid the intraday delta discussed above) gives:
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
85
= ∆ ( )
2
,
0
2
1
T t
t t
K
S T
t T
K S
P
∂
∂ −
=
∂
∂
Then since ( ) ( )
2 2 2
,
3 1 skew t T K
ATM T t
− + = σ we have ( ) ( )
2
2
,
3 1 2 skew t T
S S
K
ATM
ATM
T t
− +
∂
∂
=
∂
∂
σ
σ
Which implies:
S
K
S
K
ATM
ATM
T t T t
∂
∂
=
∂
∂
σ
σ
2
,
2
,
2
then skew
K
S
K
ATM
T t T t
× =
∂
∂
σ
2
,
2
,
2
since skew =
S
ATM
∂
∂σ
is skew is linear.
Therefore delta = skew
K
T
t T
K S
P
ATM
T t
×
−
=
∂
∂
σ
2
,
0
2
2
1
.
Thus with a ‘normally shaped’ negatively sloping skew, the delta of the variance swap will be negative, at least if the skew
curve is sticky with strike. This fits with the intuition that (implied) volatility will tend to go down as the underlying rallies
– which is exactly what a (negative) linear skew represents.
Other sensitivities to consider include:
Jumps:
The replication argument assumes a continuous diffusion process without jumps. In reality jumps (discontinuities) can
happen, and the volatility associated with them is not guaranteed to be captured by the replication process. However, the
error associated with this is of order jump (return) cubed and is therefore small enough in practice to be ignored. Although,
in the case of a singlestock where an M&A event is anticipated, the probability of a large jump increases and this may be
reflected in an increase in variance swap prices.
Stochastic Interest Rates:
Our replication argument is constructed using forwards. In reality the variance swap pays out on the volatility of the spot
underlying. If rates (and dividends) are deterministic, this should make no difference to the variance swap strike. If rates
are uncertain, this will act to change the relationship between spot and forward volatility, altering the variance swap strike.
Except for very longdated variance the effect on variance swap prices should be small, and in practice can offset the effect
of jumps.
Convexity:
As explained above, it is really convexity and not skew which acts to increase the variance swap strike above ATM
volatility. Clearly then, rising convexity will increase variance swap strikes as the ‘average’ implied volatility used in
Gatheral’s formula goes up.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
86
4.8: Setting up a replicating portfolio
In the preceding sections we showed that a variance swap could be statically replicated by a portfolio of (outofthemoney)
options, weighted according to 1/strikesquared. How do we weight this portfolio to achieve a specific variancenotional or
veganotional exposure? For example if we wish to replicate a variance swap for €100,000 of vega notional, how many of
the calls and puts do we need in the replicating portfolio? We know the relative proportions of the puts and calls but what
is the absolute amount to hold?
Let Π be forward of the price of the portfolio:
P =
( ) ( )
+
∫ ∫
∞
0
0
2
0
0
2
0
F
F
rT
dK
K
K C
dK
K
K P
e
The above portfolio (together with the dynamic futures position) will pay
2
2
R
T
σ at expiry, where dt
T
T
t R
∫
=
0
2 2
1
σ σ .
Scaling this portfolio by 2/T (which will cost 2Π/T) will produce a payout of realised volatility squared, but in percentage
terms. To put this into the wholenumber units consistent with standard quoted variance swap strikes (e.g. 20% volatility
expressed as 20) we must multiply by 100
2
. The (squareroot of) the cost of this portfolio, the variance strike represents the
fair cost of future realised variance.
Therefore the fair price for 2 ×100
2
/T lots of the portfolio P, paying out realised volatility (squared) in whole number
amounts is given by:
T
K
VAR
Π × ×
=
2
2
100 2
Therefore, to be long N
VAR
of variance notional we must buy
T
N
VAR
× ×
2
100 2
lots of the portfolio P,
equivalent to
2
2
100 2
K T
N
VAR
×
× ×
of each OTM Kstrike option.
This cost of this portfolio will therefore be
2
VAR VAR
K N × . The value of the portfolio and its associated delta hedging at
expiry will be
2
σ ×
VAR
N , giving an overall p/l of ( )
2 2
VAR VAR
K N − × σ as required.
At inception if the cutoff between puts and calls is taken to be the current forward price of the underlying, no net delta
hedge is needed. However, during the lifetime of the variance swap, the delta hedge will need to be adjusted such that,
against the long portfolio of options, there is a constant monetary (dollar) value of a long forward position but short a
constant number of shares of a forward position. Thus at the end of each day’s close the notional value of the delta hedge is
adjusted so that it equals


.

\
 − × ×
0
0
2
100 2
F
F F
T
N
t VAR
where F
0
was the original forward price at time 0, and F
t
is the current forward price.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
87
To achieve a specific vega notional, N
VAR
=N
Vega
/2K
VAR
we must trade
VAR
Vega
K T
N
×
×
2
100
lots of the portfolio P.
Since
T
K
VAR
Π
=
2
100 the required amount of the portfolio is
T
N
Vega
Π
×
2
100
or
T K
N
Vega
Π
×
2
100
2
of each (OTM) option.
In practice it is not possible to trade a continuum of option strikes, so we now examine how to (approximately) construct a
variance swap from a finite, tradable strip of OTM options. For simplicity, we assume the strikes are equally spaced.
Suppose we have available liquid OTM options of strikes K
0
…K
r
, K
r+1
…K
n
with all options up to K
r
puts, and all
subsequent options calls. Suppose the forward F
0
such that K
r
< F
0
≤ K
r+1
and suppose that the options are equally spaced
so that K
r+1
K
r+1
= ∆
K
for all i. The replicating variance swap portfolio is then given by
( ) ( )
∑ ∑
+ = =
∆
+
∆ n
r i
i
i K
r
i
i
i K
K
K C
K
K P
1 2 0 2
approximating the true replicating portfolio:
( ) ( )
∫ ∫
∞
+
0
0
2
0
0
2
0
F
F
dK
K
K C
dK
K
K P
To get a variance notional exposure we will need
2
2
100 2
K T
N
Var K
×
× ∆ × ×
of each option (contract).
There is a slight complication here with contract sizes. If one option contract represents e.g. 10 shares (or €10 per index
point) then we need only 1/10 as many option contracts as we would otherwise. Therefore we must divide the above
amount by the contract size, to compute the actual number of options required.
This kind of approximation slightly overvalues the true theoretical variance swap strike due to convexity issues. Derman
(Derman, 1999) discusses this effect and also presents a robust method for using whatever available strikes are to replicate a
variance swap.
Example 20: Constructing a variance swap
Suppose we are replicating a 6month maturity Euro Stoxx 50 variance swap (variance notional of €2,500). Strikes are 200
ip apart and the forward is 3868. Each option represents €10 per index point. The replicating portfolio will consist of puts
at strikes 3800, 3600, 3400…1200 and calls at 4000, 4200, 4400…6000. How much of each option contract is needed?
No of contracts =
size Contract K T
N
Variance K
_
100 2
2
2
× ×
× ∆ × ×
=
10 5 . 0
500 , 2 200 100 2
2
2
× ×
× × ×
K
=
2
9
10 2
K
×
For example, the number of 3600 puts required is 3 . 154
3600
10 2 10 2
2
9
2
9
=
×
=
×
K
The prices, and numbers of options required for each strike are given in Table 13 below:
Suppose the market rallies by 1% over the following day. In order to maintain a “flatdelta” for the overall portfolio, we will
need to sell on the close the following notional size:
% 1
100 2
2
T
N
VAR
× ×
= % 1
5 . 0
500 , 2 100 2
2
× ×
= €1M notional
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
88
Table 13: Replicating portfolio for a 6month Euro Stoxx variance swap, variance notional €2,500.
Option strike Option
type
Spot
premium (€)
Fwd. premium x
bucket size/strike
2
# Option contracts Price of options (€)
1200 5.90E12 8.36E16 1389 
1400 5.90E09 6.14E13 1020 
1600 1.20E06 9.56E11 781 0
1800 7.10E05 4.47E09 617 0
2000 1.80E03 9.18E08 500 9
2200 0.024 1.01E06 413 100
2400 0.195 6.90E06 347 675
2600 PUTS 1.054 3.18E05 296 3,118
2800 4.165 1.08E04 255 10,624
3000 12.777 2.90E04 222 28,393
3200 26.645 5.31E04 195 52,041
3400 47.968 8.46E04 173 82,989
3600 83.143 1.31E03 154 128,308
3800 140.932 1.99E03 139 195,197
4000 103.483 1.32E03 125 129,354
4200 40.164 4.64E04 113 45,538
4400 11.899 1.25E04 103 12,292
4600 2.853 2.75E05 95 2,696
4800 CALLS 0.69 6.11E06 87 599
5000 0.145 1.18E06 80 116
5200 0.027 2.04E07 74 20
5400 4.40E03 3.08E08 69 3
5600 6.60E04 4.29E09 64 0
5800 9.00E05 5.46E10 59 0
6000 1.10E05 6.23E11 56 0
Price of one portfolio Π
0.00706
Price of replicating
portfolio for €2,500
of variance notional 692,074
x 40,000 (2 x 100
2
/ T)
282.31
Square rooted gives….
16.80
4.9: Replication and hedging in practice
As demonstrated above, a variance swap can be statically hedged with a portfolio of outofthemoney (Europeanstyle)
options, weighted according to the inverse squares of their strikes. This makes it easy, in theory, to calculate the fair value
of a variance swap, assuming option prices are available across the entire range of strikes. In practice, traded strikes are not
continuous, although for major liquid indices they are closely spaced (0.4% notional apart for the S&P, 1% for the FTSE,
1.4% for the Euro Stoxx). A more serious limitation is the lack of liquidity in OTM strikes, especially for puts, as these
provide a relatively large component of the variance swap price in the presence of steep put skews. S&P options are listed
down to a strike of 600 (c45% of current spot), FTSE to 3525 (c60% of spot) and Euro Stoxx down to 600 (c15% of spot),
although in reality, liquidity does not even reach this far.
The problem with the lack of OTM puts can be seen from following through the practical example of setting up the
replicating portfolio above. The long futures position is used to create a payout which is equivalent to a long log contract
plus realised variance. In opposition the long options/short forward position is used to create a short log contract and pay
the fixed strike. Supposing the market falls significantly, the deltahedge will be long the log contract (a will hence lose),
while the options should counteract this by being short the log contract. However, if not enough downside puts were used,
the options portfolio will not fully reflect the short logcontract exposure needed and hence the overall hedge will lose
money. This lack of liquidity at the wings has led to the development of conditional variance swaps which can remove
exposure to volatility once the underlying moves into areas where vanilla options are illiquid.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
89
In practice, marketmakers will not attempt to hedge with the entire strip of options but typically will use only two or three
– including one close to the money and one or more OTM, but liquid, puts. Alternatively they could approach the
replicating by hedging the vega with an OTM put whose implied volatility coincides with the variance swap strike – close
to the money for 1month maturity, 95% for 36 months, 90% for 1year, 85% for 23 years etc. In this case they would
also look to buy back the wings/convexity separately. Other proxies, such as CDS or EDS could also be used instead of the
more deeply outofthemoney puts.
One problem with this kind of approach is that the partial hedge is no longer static, and must be dynamically managed. For
example if the market sells off towards the strike, the market maker will have to trade further OTM puts to ensure that their
exposure to volatility, in the form of dollargamma, remains constant. This makes the actual variance swap replication
more akin to a combination of alternatives 2 and 3 listed on p79. Here the constant dollar gamma would be maintained by a
combination of holding a portfolio which has roughly constant dollar gamma if the underlying does not move too much, and
rehedging by trading more options if the underlying does move significantly.
In practice, pricing models will often only price contributing options for the variance swap hedge between 1 and 99 delta.
For example, on the Euro Stoxx (6month maturity) this would only price puts down to 2700 (ref 3822) or around 70% of
spot. The difference from this truncated and the full replication pricing is typically in the region of 525bps.
Another limitation comes from the discrete nature of adjusting the delta hedge on the close, which introduces possible errors
due to large daily moves. However these moves are actually of order the cube of the move (as in jumps for the continuous
case, see section 4.7) and hence are negligible for all but very large moves. Also as previously noted if interest rate changes
are related to changes in the spot underlying, this can also have an impact on the ability to accurate replicate the logcontract
and realised volatility
This and other differences between the theoretical and midmarket price of a variance swap contribute to the socalled
variance swap basis, see Section 4.7. This basis tends to vary across maturities, being larger for longer maturities. At
shorter dates a value of around 0.5 vegas would be typical.
4.10: Effects of variance swap hedging
Marketmakers who trade variance swaps may hedge their positions by replicating the opposite variance swap position
through the replicating options portfolio. This replicating portfolio then needs to be delta hedged. The effects of delta
hedging this portfolio are different to that from normal deltahedged options for two principal reasons:
1. The actions of deltahedging the options could potentially act to the disadvantage of the counterparty’s position.
2. Since variance swap contracts typically measure closeclose realised volatility, the options must be deltahedged
on the close only to capture this.
We investigate both of these properties below.
Firstly, consider a marketmaker who has sold a call option to a counterparty. The marketmaker will be short the option
(and delta) and will therefore buy back the delta in the underlying. If the underlying rallies, the shortdelta exposure from
the short option will increase, and the marketmaker will therefore have to buy more of the underlying in the market.
Similarly, if the underlying sells off, the marketmaker will have to sell the underlying. In both cases – if the position being
hedged is big enough – the action of deltahedging will have the effect of increasing volatility in the underlying: magnifying
both upmoves and downmoves. The argument for a put option is analogous. Similarly, if the marketmaker is long
options and deltahedging, the hedging will act to suppress volatility in the underlying, potentially to the advantage of the
counterparty who is short the option.
The situation with variance swaps is different. Suppose that the market is such that hedgefunds, or other market
participants, have generally sold index variance swaps to marketmakers (as was the case before the May 2006 correction).
Note that no exchange of options has taken place here – the parties have just taken opposite sides in a contract for
difference. Suppose that the volatility sellers do not hedge their variance swaps (they have sold the variance swaps
specifically for the direct volatility exposure they offer). But assume that marketmakers hedge their short volatility
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
90
exposure. A marketmaker who is long the variance swap can offset the risk by shorting the replicating portfolio of options,
and deltahedging. They will therefore be short gamma in the options market (Figure 108).
As described above, a deltahedger who is short options will act to increase volatility in the underlying – buying as it rallies,
and selling as it sellsoff. However, the action of these market makers hedging their short options will not necessarily act to
increase volatility in the underlying, as the counterparties they have sold options to may be counteracting this effect by
themselves hedging their long volatility positions. However, the important difference between these two groups of hedgers
is that the variance swap marketmakers who are short options, must hedge only on the close to capture the closeclose
realised variance specified in the variance swap contract. In contrast, the hedgers who are long the options, will generally
be free to choose when to deltahedge, as they attempt to capture the true volatility of the underlying process.
Figure 108: Flows in the market as a result of a marketmaker buying a variance swap and replicating it in the market
Therefore, the overall effect of hedging these variance swaps need not have the effect of increasing overall market
volatility, although it may if the long options positions are not being hedged (e.g. they are sold on to endinvestors).
However, the important point is that the hedging of long variance swap positions may act to increase closetoclose
volatility, with option hedges on the close having the potential effect of magnifying daily moves. In practice, this only
becomes an issue if hedgers (variance swap marketmakers) are the same way round and daily moves are large enough.
In fact the expectation of these deltahedging flows could itself act to further amplify moves into the close through possible
‘feedback’ effects. For instance, on big down days, the market could anticipate a selloff into the close as a result of long
variance swap hedging and take short positions into the close to profit from this anticipated effect. This then drives the
underlying market down further and could increases the amount that the variance swap hedgers have to sell on the close.
These feedback effects were much talked about during the volatile market period in May 2006, when the hedging into the
close of long variance swap positions, and anticipation thereof, was blamed for the amplification of some already large daily
moves. In fact, in an efficient market, such effects should be propagated back through the trading day, and may be offset by
intraday deltahedging of opposing options.
Fixed Payment
(implied variance = strike
2
)
Variance
swap
seller
Realised variance Maintain constant $value
short position in
underlying (buy high / sell
low on the close)
Sell portfolio of
options
Market maker
(variance
swap buyer)
Stock Market
Intraday delta hedging
(buy low / sell high)
Options Market
Option premia
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
91
4.11: Why not volatility swaps?
A common complaint about variance swaps is that they payoff based on realised variance (volatility squared) and not
simply realised volatility. Why are there not volatility swaps, products which payout linearly on the difference
between implied and realised volatility? If these are more easily understandable why are they not regularly traded?
Whilst volatility can be seen as more of an intuitive measure (being a standard deviation it is measured in the same units as
the underlying), variance is in some sense more fundamental – especially because it is additive. For example, the variance
of the sum of two independent distributions is just the sum of each of their variances (e.g. see discussion on variance swap
marktomarket, p15). This linearity of variance makes calculating variance swap marktomarkets and forwards
particularly simple.
As pointed out above, the exposure of deltahedged options to volatility, after accounting for the gamma, is actually an
exposure to the difference between implied and realised volatility squared. In this sense, a variance swap mirrors a kind
of ideal deltahedged option whose gamma remains constant. Furthermore, variance swaps are relatively easy to
replicate. Once the replicating portfolio of options has been put in place, only deltahedging is required. No further buying
or selling of options is necessary.
All this explains why variance swaps are attractive instruments to trade, but still does not explain why volatility swaps are
not also frequently traded. The main theoretical difficulty with volatility swaps is that they cannot be statically replicated
through options. A replicating portfolio must dynamically trade options, and make relatively strong assumptions about the
underlying volatility process – in particular about the volatility of volatility. This makes any replication process model
dependent, and therefore much more prone to errors than the theoretically robust variance swap replication. See Derman,
1999 for further details about construction volatility swaps.
In fact the convention of quoting variance swap notionals in vega, rather than variance, amounts can be seen as an
attempt to treat variance swaps like volatility swaps. The vega notional represents the average p/l for a 1 vega change in
volatility, but with the convexity meaning that longs will profit by more if volatility increases, and lose by less if volatility
decreases. Thus for small changes in volatility, where the effect of the variance swap convexity is relatively limited,
variance swaps (measured in vega notional) locally approximate volatility swaps. Volatility swaps can then be thought of
as variance swaps without the convexity. The discount of the volatility swaps to variance swaps should therefore reflect the
value of this convexity, which in turn is determined by the volatility of volatility (see Section 2.8). Seen in this light, the
ability to calculate a fair price for volatility swaps, requires not only a price for the volatility (variance), but also a price for
the volatility of volatility – i.e. the means of valuing options on volatility.
Since it is variance (volatility squared) which arises naturally from deltahedging options, in terms of volatility products,
variance should be thought of as the true underlying. In this framework, volatility swaps are naturally thought of as
derivatives of variance – paying the squareroot of the variance swap contract.
In fact we could dynamically trade a long variance swap (buying more as volatility decreases and selling as volatility
increases) to hedge out this convexity bias. Similar to deltahedging an option the p/l made from the resulting buy low –
sell high strategy (for variance) will lead to a p/l based on the volatility of volatility: the larger this volatilityofvolatility,
the bigger the discount of the dynamically replicated volatility swap to the variance swap. However, besides issues arising
from the transaction costs of dynamically trading variance swaps, estimating this volatilityofvolatility contribution from
the dynamic hedging of the variance swap is problematic and modeldependent, making the volatility swap contracts
difficult to price.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 4: Replication and Hedging
92
93
5. Future Developments
• Variance swap indices
• Variance swap fixings
• Benchmark volatility return index
• Exchangetraded variance swaps
• 3rd generation products
• Derivatives on variance
• Variance swaps on other asset classes
P
a
r
t
5
:
F
u
t
u
r
e
D
e
v
e
l
o
p
m
e
n
t
s
Part 5: Future developments
With the increasing liquidity and breadth of the variance swap market, apart from the obvious benefits of reducing
transaction costs and the ability to trade variance on more stock and index underlyings, other developments are underway
which will in turn help both the liquidity and transparency of the variance swap market.
We highlight a few examples of recent developments, such as:
• Variance swap indices
• Variance swap fixings
• Benchmark volatility return indices
• Exchange traded variance swaps
• 3rd generation products
• Derivatives on variance
• Variance swaps on other asset classes.
Variance swap indices
The volatility indices associated with the major indices (VIX for S&P500, VXN for the Nasdaq, VSTOXX for Euro Stoxx
50, VDAX for DAX and VSMI for the SMI) are now calculated according to a variance swap style calculation. They
represent the theoretical level of a rolling 1month (30 calendar day) maturity variance swap, based on traded option prices.
In fact, theoretical variance swap levels are first calculated for listed option maturities, and then the 30day index level is
interpolated. As such this gives the indices a real interpretation: each index represents the riskneutral expected
variance of the underlying over the next month. This also makes the indices directly comparable.
Table 14: Main variance swap based volatility index details
Underlying Reuters Bloomberg Datastream Calculation details
VIX S&P 500 .VIX VIX Index CBOEVIX www.cboe.com
VXN Nasdaq 100 .VXN VXN Index CBOEVXN www.cboe.com
VDAXNEW Dax 30 .V1X V1X Index VDAXNEW www.deutscheboerse.com
VSTOXX Euro STOXX 50 .V2TX V2X Index VSTOXXI www.stoxx.com
VSMI SMI .VSMI V3X Index VSMIIDX www.swx.com
Source: JPMorgan
All these volatility indices are well correlated with traded variance swap levels on their respective indices. However, the
level of the volatility index is typically somewhat above the actual traded levels for variance swaps of the same maturity.
There are two reasons for this:
• A convex skew should lead to midpoint discrete strike variance swap approximation (as in the VSTOXX index
calculation) overestimating the level of implied variance.
• Given the typical bid/offer on equity options, the width of the arbitragecorridor of trading a variance swap against
a portfolio of options can give rise to a fairly large margin of error. The increasing demand to sell index variance
and correlation over the last few years, as equity volatility has fallen, helped to cheapen index variance to its
theoretical value, but not to the extent at which an arbitrage has been possible.
This ‘basis’ between the volatility index and actual traded (mid) variance swap levels is usually about 0.51 vegas.
However this can vary much more widely during volatility spikes when supply and demand issues become important.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 5: Future Developments
94
These volatility indices, due to their visibility and wide availability, are often referred to by those outside the equity
derivatives space as a ‘market price of risk’ index. While this is to some extent true, the shortdated nature of these variance
swaps means the principal driver of the volatility index level is recent realised volatility (see Section 2.5). In reality, longer
dated (e.g. 1year) variance, spreads of implied to realised variance, skew levels or even ratios of put to call openinterest
would perhaps be a better proxy for the level of riskaversion present in the market.
Variance swap fixings
One of the biggest deficiencies of using the portfolio replication method to calculate a variance swap level is its lack of
tradability. As noted above there exists a basis between the actual variance swap level and theoretical levels calculated
from the portfolio of options. This basis exists probably due either to differences in marketmakers’ volatility skews or in
the supply and demand from end users.
A way to capture a more tradable level for variance swaps is to construct a brokerdealer poll, in a similar way to that used
in interest rates markets. For example, each day at 11.00am an interest rate fixing for different maturity deposits is made
using a trimmed mean average of several banks’ contributed rates. Most notably, the 3 and 6month fixings are then often
used as reference rates which form the basis of many longer dated instruments.
It could be envisaged that the 3month variance swap could itself become a similar benchmark for volatility. Moreover,
longer dated variance swaps could themselves be easily constructed from a strip of forward variance swaps. With
structured products increasingly incorporating payoffs derived from variance swaps, having an independent method of
valuing these instruments at realistic tradable levels will become more desirable to provide transparent marktomarket
valuations.
Benchmark volatility return index
With the availability of tradable and transparent variance swap levels for different maturities, transparent benchmark total
return volatility indices can be constructed using a similar methodology to that described in Section 3.3. Though there will
still need to be some development of the strategies involved (i.e. long/short, spot/forward, maturity etc), it will aid the
development of volatility funds in countries where an appropriate benchmark is necessary. For example, see the variance
component of JPMorgan’s Yield Alpha product, as described in JPMorgan’s Yield Alpha Strategy Guide.
Exchangetraded variance
Following the change in calculation method of the VIX to the variance swap calculation, the CBOE launched futures
contracts on the VIX index in March 2004. This was followed in September 2005 by the launch of futures on the European
VSTOXX, VDAX and VSMI indices.
Note that these futures contracts are forward volatility products, paying out on the value of the relevant volatility index at
expiry, a value which is itself an implied variance. Thus the futures contract p/l is based on the change in expected future
volatility, not on realised volatility or the spread of realised to implied volatility.
These futures do not expire on the normal index (futures) expiry dates, but 30 calendar days beforehand. For example a
Nov06 VXTOXX future expires on 15th November, 30 days before the 15th December options expiry. Expiry on 15th
November is chosen because on that date, the Dec06 listed options have exactly 30 calendar days remaining maturity and
the VSTOXX calculation does not need to interpolate from any other maturities.
Note that there is a convexity bias inherent in marking to market these futures, since they are margined linearly in the level
of forward starting volatility (See VDAXNEW, VSTOXX and VSMI Futures, JPMorgan September 2005).
In April 2005, options on the VIX index were launched. These represented the first available exchange traded options on
variance. As for the futures, these expire 30 days before an index expiry and are listed to expire 30 days before the
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 5: Future Developments
95
corresponding quarterly options expiry dates for the underlying. E.g. in November 2006, the listed VIX options are for
Nov06, Dec 06, Feb07, May07 and Aug07 expiries.
More recently, beginning September 2006, Euronext announced the ability to list variance swaps on a limited number of
indices through its BClear mechanism. This can be useful for end users who wish to take a volatility view with variance, but
can not trade OTC products directly.
3rd generation volatility products
Generalised variance swaps or “3rd generation products” include gamma swaps, corridor swaps and conditional variance
swaps. Gamma swaps are very similar to variance swaps but maintain a dollar gamma linear with spot, rather than a
constant dollar gamma, leading to lower a skew exposure and a slightly lower strike compared to standard variance swaps.
The original corridor variance swaps only accrue variance within a prespecified range, meaning a long would make a
maximal loss when the underlying fails to trade within the range.
Conditional variance swaps have been the most popular (and liquid) of these products, allowing investors to take exposure
to realised variance contingent upon the underlying instrument trading within a prespecified range. Outside this range no
p/l accrues. Conditional variance swaps can be useful for hedging complex volatility exposures, taking a view on the
volatility levels encapsulated in the skew or buying/selling variance at more attractive levels given a view on the
underlying.
As with variance swaps, the sign of the p/l for a conditional variance swap is controlled by the difference between the
volatility realised (in the range) over the lifetime of the swap and the preagreed fixed strike level. In the case of the
conditional, p/l is accrued only when the underlying is within a prespecified range. In addition, the magnitude of the
final p/l is scaled by the proportion of time the underlying has spent in the prespecified range. If the underlying trades
entirely outside the range, the p/l will be zero, so the maximal loss for a long will be when the underlying trades within the
range but with very low volatility.
Whilst investors are free to specify the range associated with a conditional variance swap, the two principal types are up
and down conditional variance swaps (upvariance and downvariance). Upvariance accrues realised volatility only
when the underlying is above a prespecified level (i.e. no upper barrier), while downvariance is accrued only when the
underlying is below the specified barrier (i.e. no lower barrier). In the presence of a positive putskew, downvariance will
normally price above upvariance for close to ATM barrier levels.
Figure 109 : Payout from a conditional variance swap @ 20 strike
p/l, different lines for % days in the range
20%
10%
0%
10%
20%
30%
0% 10% 20% 30% 40%
0%
25%
50%
75%
100%
volatility realised in the range
Source : JPMorgan
Figure 110 : Dec07 conditional variance swaps levels as of Apr 06
Strike
10%
15%
20%
25%
30%
80% 90% 100% 110% 120%
conditional dow n v ar
conditional upv ar
v ariance
Up/down barrier
Source : JPMorgan, data as of Apr 2006
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 5: Future Developments
96
Conditional variance swaps can be useful for expressing views on volatility contingent on market level. For example
investors seeking crash protection may purchase conditional downvariance, which only becomes activated in the event of a
market selloff. That is, if the market stays above the downbarrier, p/l is zero. Conversely investors who believe volatility
will be realised on the upside can buy conditional upvariance swaps, which are usually cheaper than a standard variance
swap for which a significant amount of the premium is used to fund the downside skew exposure.
Conditional variance swaps have typically been traded on index underlyings. More recently, singlestock conditionals
have gained liquidity, allowing investors to trade the volatility of a stock, contingent on stock price.
See Conditional Variance Swaps, April 2006 for further details.
Derivatives on variance
Obviously with the increased liquidity and breadth of the variance swap market, the development of a liquid market in
derivatives based on variance becomes more plausible. This is because the ability to use variance swaps themselves to
deltahedge variance derivatives becomes easier as liquidity increases and transaction costs decrease. For example, options
on variance now trade in the market as investors look to tailor their volatility exposures.
However, as we have highlighted throughout the note, volatility is a process very different from the underlying instrument
itself. For example volatility tends to be a meanreverting process characterised by fairly long periods of subdued volatility
with intermittent spikes of high volatility. Hence, using the simple deltahedging replication option pricing schema that is
basis of BlackScholes becomes an impractical model to use in this case.
The key principle behind BlackScholes, is that exposure to the underlying can be removed through deltahedging, leaving
behind the sole residual risk of exposure to the volatility of the underlying. If a non BlackScholes approach is used
another risk factor, such as the marketpriceofrisk, is introduced in addition to implied volatility.
Nonetheless, this problem of modelling the underlying process has failed to halt the development of derivatives markets
where the BlackScholes approach is also considered invalid, for example: interest rate options, credit options, out
performance options etc. Derivatives on variance look to be no different.
Variance swaps on other asset classes
While variance swap liquidity has increased massively on equity underlyings over the last few years, variance swaps have
yet to really take hold in other asset classes. Nonetheless, variance swaps do trade on assets such as commodities, FX and
interest rates.
One issue for variance swaps on other asset classes is the dependence of the payoff on movements in the price of a spot
underlying. While this is appropriate for equities, it can be troublesome for commodities which tend to be based upon a
rolling forward price. In fact, commodity variance swaps are usually referenced to a futures contract of the same maturity
as the swap.
Moreover in fixed income volatility based on yields is often more useful. This would add additional complexity in
constructing a replicating portfolio, which would have to take into account the convexity of bond prices to yields.
Furthermore, bonds have a defined maturity, unlike equities, which leads to problems defining a meaningful payoff for
longer dated derivative products.
Nonetheless, we expect variance swap markets to continue to develop on other assets, particularly commodities and FX,
driven by both investor demand to trade pure volatility, and at least in some cases the desire for investment banks to recycle
exotic and structured product risk.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
Part 5: Future Developments
97
References
General References
More Than You Ever Wanted to Know About Volatility Swaps, Demeterfi, Derman, Kamal & Zou, 1999
Options, Futures and Other Derivatives, John C. Hull, Fifth Edition, 2003
The Volatility Surface, Jim Gatheral, 2006
Towards a theory of Volatility Trading, Peter Carr & Dilip Madan  2002
JPMorgan Publications
Calling All Overwriters, August 2002
Conditional Variance Swaps, April 2006
Correlation Vehicles, May 2005
European Equity Derivatives Weekly Outlook , various publications
Introducing JPMorgan’s CrossSection Volatility Report, November 2006
Just What You Need to Know About Variance Swaps, February 2005
Synthetic Short CDS with Positive Carry, October 2006
VDAXNEW, VSTOXX and VSMI Futures, September 2005
Yield Alpha Strategy Guide, July 2006
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
References
98
Risks of common option strategies
Please also refer to option risk disclosure documents
Put Sale. Investors who sell put options will own the underlying stock if the stock price falls below the strike price of the
put option. Investors, therefore, will be exposed to any decline in the stock price below the strike potentially to zero, and
they will not participate in any stock appreciation if the option expires unexercised.
Call Sale. Investors who sell uncovered call options have exposure on the upside that is theoretically unlimited.
Call Overwrite or Buywrite. Investors who sell call options against a long position in the underlying stock give up any
appreciation in the stock price above the strike price of the call option, and they remain exposed to the downside of the
underlying stock in the return for the receipt of the option premium.
Booster. In a selloff, the maximum realised downside potential of a doubleup booster is the net premium paid. In a rally,
option losses are potentially unlimited as the investor is net short a call. When overlayed onto a long stock position, upside
losses are capped (as for a covered call), but downside losses are not.
Collar. Locks in the amount that can be realized at maturity to a range defined by the put and call strike. If the collar is not
costless, investors risk losing 100% of the premium paid. Since investors are selling a call option, they give up any stock
appreciation above the strike price of the call option.
Call Purchase. Options are a decaying asset, and investors risk losing 100% of the premium paid if the stock is below the
strike price of the call option.
Put Purchase. Options are a decaying asset, and investors risk losing 100% of the premium paid if the stock is above the
strike price of the put option.
Straddle or Strangle. The seller of a straddle or strangle is exposed to stock increases above the call strike and stock price
declines below the put strike. Since exposure on the upside is theoretically unlimited, investors who also own the stock
would have limited losses should the stock rally. Covered writers are exposed to declines in the long stock position as well
as any additional shares put to them should the stock decline below the strike price of the put option. Having sold a covered
call option, the investor gives up all appreciation in the stock above the strike price of the call option.
Put Spread. The buyer of a put spread risks losing 100% of the premium paid. The buyer of higher ratio put spread has
unlimited downside below the lower strike (down to zero), dependent on the number of lower struck puts sold. The
maximum gain is limited to the spread between the two put strikes, when the underlying is at the lower strike. Investors
who own the underlying stock will have downside protection between the higher strike put and the lower strike put.
However, should the stock price fall below the strike price of the lower strike put, investors regain exposure to the
underlying stock, and this exposure is multiplied by the number of puts sold.
Call Spread. The buyer risks losing 100% of the premium paid. The gain is limited to the spread between the two strike
prices. The seller of a call spread risks losing an amount equal to the spread between the two call strikes less the net
premium received. By selling a covered call spread, the investor remains exposed to the downside of the stock and gives up
the spread between the two call strikes should the stock rally.
Butterfly Spread. A butterfly spread consists of two spreads established simultaneously. One a bull spread and the other a
bear spread. The resulting position is neutral, that is, the investor will profit if the underlying is stable. Butterfly spreads
are established at a net debit. The maximum profit will occur at the middle strike price, the maximum loss is the net debit.
Pricing Is Illustrative Only:
Prices quoted in the above trade ideas are our estimate of current market levels, and are not indicative trading levels
Risks to Strategies:
Not all option strategies are suitable for investors; certain strategies may expose investors to significant potential losses.
We have summarized the risks of selected derivative strategies. For additional risk information, please call your sales
representative for a copy of "Characteristics and Risks of Standardized Options". We advise investors to consult their tax
advisors and legal counsel about the tax implications of these strategies.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
99
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
100
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
101
Analyst Certification:
The research analyst(s) denoted by an “AC” on the cover of this report certifies (or, where multiple research analysts are primarily
responsible for this report, the research analyst denoted by an “AC” on the cover or within the document individually certifies, with
respect to each security or issuer that the research analyst covers in this research) that: (1) all of the views expressed in this report
accurately reflect his or her personal views about any and all of the subject securities or issuers; and (2) no part of any of the research
analyst’s compensation was, is, or will be directly or indirectly related to the specific recommendations or views expressed by the
research analyst(s) in this report.
Important Disclosures
Explanation of Equity Research Ratings and Analyst(s) Coverage Universe:
JPMorgan uses the following rating system: Overweight [Over the next six to twelve months, we expect this stock will outperform the
average total return of the stocks in the analyst’s (or the analyst’s team’s) coverage universe.] Neutral [Over the next six to twelve
months, we expect this stock will perform in line with the average total return of the stocks in the analyst’s (or the analyst’s team’s)
coverage universe.] Underweight [Over the next six to twelve months, we expect this stock will underperform the average total return of
the stocks in the analyst’s (or the analyst’s team’s) coverage universe.] The analyst or analyst’s team’s coverage universe is the sector
and/or country shown on the cover of each publication. See below for the specific stocks in the certifying analyst(s) coverage universe.
JPMorgan Equity Research Ratings Distribution, as of September 29, 2006
Overweight
(buy)
Neutral
(hold)
Underweight
(sell)
JPM Global Equity Research Coverage 42% 41% 17%
IB clients* 44% 46% 35%
JPMSI Equity Research Coverage 38% 48% 15%
IB clients* 62% 55% 47%
*Percentage of investment banking clients in each rating category.
For purposes only of NASD/NYSE ratings distribution rules, our Overweight rating falls into a buy rating category; our Neutral rating falls into a hold
rating category; and our Underweight rating falls into a sell rating category.
Valuation and Risks: Please see the most recent JPMorgan research report for an analysis of valuation methodology and risks on any
securities recommended herein. Research is available at http://www.morganmarkets.com , or you can contact the analyst named on the
front of this note or your JPMorgan representative.
Analysts’ Compensation: The equity research analysts responsible for the preparation of this report receive compensation based upon
various factors, including the quality and accuracy of research, client feedback, competitive factors, and overall firm revenues, which
include revenues from, among other business units, Institutional Equities and Investment Banking.
Other Disclosures
Options related research: If the information contained herein regards options related research, such information is available only to persons who
have received the proper option risk disclosure documents. For a copy of the Option Clearing Corporation’s Characteristics and Risks of
Standardized Options, please contact your JPMorgan Representative or visit the OCC’s website at
http://www.optionsclearing.com/publications/risks/riskstoc.pdf.
Legal Entities Disclosures
U.S.: JPMSI is a member of NYSE, NASD and SIPC. J.P. Morgan Futures Inc. is a member of the NFA. J.P. Morgan Chase Bank, N.A. is a
member of FDIC and is authorized and regulated in the UK by the Financial Services Authority. U.K.: J.P. Morgan Securities Ltd. (JPMSL) is a
member of the London Stock Exchange and is authorised and regulated by the Financial Services Authority. South Africa: J.P. Morgan Equities
Limited is a member of the Johannesburg Securities Exchange and is regulated by the FSB. Hong Kong: J.P. Morgan Securities (Asia Pacific)
Limited (CE number AAJ321) is regulated by the Hong Kong Monetary Authority and the Securities and Futures Commission in Hong Kong.
Korea: J.P. Morgan Securities (Far East) Ltd, Seoul branch, is regulated by the Korea Financial Supervisory Service. Australia: J.P. Morgan
Australia Limited (ABN 52 002 888 011/AFS Licence No: 238188, regulated by ASIC) and J.P. Morgan Securities Australia Limited (ABN 61
003 245 234/AFS Licence No: 238066, a Market Participant with the ASX) (JPMSAL) are licensed securities dealers. Taiwan: J.P.Morgan
Securities (Taiwan) Limited is a participant of the Taiwan Stock Exchange (companytype) and regulated by the Taiwan Securities and Futures
Commission. India: J.P. Morgan India Private Limited is a member of the National Stock Exchange of India Limited and The Stock Exchange,
Mumbai and is regulated by the Securities and Exchange Board of India. Thailand: JPMorgan Securities (Thailand) Limited is a member of the
Stock Exchange of Thailand and is regulated by the Ministry of Finance and the Securities and Exchange Commission. Indonesia: PT J.P.
Morgan Securities Indonesia is a member of the Jakarta Stock Exchange and Surabaya Stock Exchange and is regulated by the BAPEPAM.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
102
Philippines: This report is distributed in the Philippines by J.P. Morgan Securities Philippines, Inc. Brazil: Banco J.P. Morgan S.A. is regulated
by the Comissao de Valores Mobiliarios (CVM) and by the Central Bank of Brazil. Japan: This material is distributed in Japan by JPMorgan
Securities Japan Co., Ltd., which is regulated by the Japan Financial Services Agency (FSA). Singapore: This material is issued and distributed
in Singapore by J.P. Morgan Securities Singapore Private Limited (JPMSS) [mica (p) 069/09/2006 and Co. Reg. No.: 199405335R] which is a
member of the Singapore Exchange Securities Trading Limited and is regulated by the Monetary Authority of Singapore (MAS) and/or JPMorgan
Chase Bank, N.A., Singapore branch (JPMCB Singapore) which is regulated by the MAS. Malaysia: This material is issued and distributed in
Malaysia by JPMorgan Securities (Malaysia) Sdn Bhd (18146x) (formerly known as J.P. Morgan Malaysia Sdn Bhd) which is a Participating
Organization of Bursa Malaysia Securities Bhd and is licensed as a dealer by the Securities Commission in Malaysia
Country and Region Specific Disclosures
U.K. and European Economic Area (EEA): Issued and approved for distribution in the U.K. and the EEA by JPMSL. Investment research
issued by JPMSL has been prepared in accordance with JPMSL’s Policies for Managing Conflicts of Interest in Connection with Investment
Research which can be found at http://www.jpmorgan.com/pdfdoc/research/ConflictManagementPolicy.pdf. This report has been issued in the
U.K. only to persons of a kind described in Article 19 (5), 38, 47 and 49 of the Financial Services and Markets Act 2000 (Financial Promotion)
Order 2001 (all such persons being referred to as "relevant persons"). This document must not be acted on or relied on by persons who are not
relevant persons. Any investment or investment activity to which this document relates is only available to relevant persons and will be engaged
in only with relevant persons. In other EEA countries, the report has been issued to persons regarded as professional investors (or equivalent) in
their home jurisdiction Germany: This material is distributed in Germany by J.P. Morgan Securities Ltd. Frankfurt Branch and JPMorgan Chase
Bank, N.A., Frankfurt Branch who are regulated by the Bundesanstalt für Finanzdienstleistungsaufsicht. Australia: This material is issued and
distributed by JPMSAL in Australia to “wholesale clients” only. JPMSAL does not issue or distribute this material to “retail clients.” The
recipient of this material must not distribute it to any third party or outside Australia without the prior written consent of JPMSAL. For the
purposes of this paragraph the terms “wholesale client” and “retail client” have the meanings given to them in section 761G of the Corporations
Act 2001. Hong Kong: The 1% ownership disclosure as of the previous month end satisfies the requirements under Paragraph 16.5(a) of the
Hong Kong Code of Conduct for persons licensed by or registered with the Securities and Futures Commission. (For research published within
the first ten days of the month, the disclosure may be based on the month end data from two months’ prior.) J.P. Morgan Broking (Hong Kong)
Limited is the liquidity provider for derivative warrants issued by J.P. Morgan International Derivatives Ltd and listed on The Stock Exchange of
Hong Kong Limited. An updated list can be found on HKEx website: http://www.hkex.com.hk/prod/dw/Lp.htm. Korea: This report may have
been edited or contributed to from time to time by affiliates of J.P. Morgan Securities (Far East) Ltd, Seoul branch. Singapore: JPMSI and/or its
affiliates may have a holding in any of the securities discussed in this report; for securities where the holding is 1% or greater, the specific holding
is disclosed in the Legal Disclosures section above. India: For private circulation only not for sale. New Zealand: This material is issued and
distributed by JPMSAL in New Zealand only to persons whose principal business is the investment of money or who, in the course of and for the
purposes of their business, habitually invest money. JPMSAL does not issue or distribute this material to members of "the public" as determined
in accordance with section 3 of the Securities Act 1978. The recipient of this material must not distribute it to any third party or outside New
Zealand without the prior written consent of JPMSAL.
General: Additional information is available upon request. Information has been obtained from sources believed to be reliable but JPMorgan
Chase & Co. or its affiliates and/or subsidiaries (collectively JPMorgan) do not warrant its completeness or accuracy except with respect to any
disclosures relative to JPMSI and/or its affiliates and the analyst’s involvement with the issuer that is the subject of the research. All pricing is as
of the close of market for the securities discussed, unless otherwise stated. Opinions and estimates constitute our judgment as of the date of this
material and are subject to change without notice. Past performance is not indicative of future results. This material is not intended as an offer or
solicitation for the purchase or sale of any financial instrument. The opinions and recommendations herein do not take into account individual
client circumstances, objectives, or needs and are not intended as recommendations of particular securities, financial instruments or strategies to
particular clients. The recipient of this report must make its own independent decisions regarding any securities or financial instruments
mentioned herein. JPMSI distributes in the U.S. research published by nonU.S. affiliates and accepts responsibility for its contents. Periodic
updates may be provided on companies/industries based on company specific developments or announcements, market conditions or any other
publicly available information. Clients should contact analysts and execute transactions through a JPMorgan subsidiary or affiliate in their home
jurisdiction unless governing law permits otherwise.
Revised September 30, 2006.
Copyright 2006 JPMorgan Chase & Co. All rights reserved. This report or any portion hereof may not be reprinted, sold or
redistributed without the written consent of JPMorgan.
European Equity Derivatives Strategy
17 November 2006
Peter Allen
(4420) 73254114
Stephen Einchcomb
(4420) 73259064
Nicolas Granger
(4420) 73257033
103
14. Exploiting crossmarket momentum, Ruy Ribeiro and
Jan Loeys, February 2006
15. A crossmarket bond carry strategy, Nikolaos
Panigirtzoglou, March 2006
16. Bonds, Bubbles and Black Holes, George Cooper,
March 2006
17. JPMorgan FX Hedging Framework, Rebecca
Patterson and Nandita Singh, March 2006
18. Index Linked Gilts Uncovered, Jorge Garayo and
Francis Diamond, March 2006
19. Trading Credit Curves I, Jonny Goulden, March 2006
20. Trading Credit Curves II, Jonny Goulden, March 2006
21. Yield Rotator, Nikolaos Panigirtzoglou, May 2006
22. Relative Value on Curve vs Butterly Trades, Stefano Di
Domizio, June 2006
23. Hedging Inflation with Real Assets, John Normand,
July 2006
24. Trading Credit Volatility, Saul Doctor and Alex
Sbityokov, August 2006
25. Momentum in Commodities, Ruy Ribeiro, Jan Loeys
and John Normand, September 2006
26. Equity Style Rotation, Ruy Ribeiro, November 2006
27. Euro Fixed Income Momentum Strategy, Gianluca
Salford, George Cooper, November 2006
1. RockBottom Spreads, Peter Rappoport, Oct 2001
2. Understanding and Trading Swap Spreads, Laurent
Fransolet, Marius Langeland, Pavan Wadhwa, Gagan
Singh, Dec 2001
3. New LCPI trading rules: Introducing FX CACI, Larry
Kantor, Mustafa Caglayan, Dec 2001
4. FX Positioning with JPMorgan’s Exchange Rate Model,
Drausio Giacomelli, Canlin Li, Jan 2002
5. Profiting from Market Signals, John Normand, Mar 2002
6. A Framework for Longterm Currency Valuation, Larry
Kantor and Drausio Giacomelli, Apr 2002
7. Using Equities to Trade FX: Introducing LCVI, Larry
Kantor and Mustafa Caglayan, Oct 2002
8. Alternative LCVI Trading Strategies, Mustafa Caglayan,
Jan 2003
9. Which Trade, John Normand, Jan 2004
10. JPMorgan’s FX & Commodity Barometer, John Normand,
Mustafa Caglayan, Daniel Ko, Nikolaos Panigirtzoglou
and Lei Shen, Sep 2004.
11. A Fair Value Model for US Bonds, Credit and Equities,
Nikolaos Panigirtzoglou and Jan Loeys, Jan 2005.
12. JPMorgan Emerging Market CarrytoRiskModel,
Osman Wahid, February 2005
13. Valuing crossmarket yield spreads, Nikolaos
Panigirtzoglou, January 2006
Investment Strategies Series
This series aims to offer new approaches and methods on investing and trading profitably in financial markets.
European Equity Derivatives Research
Variance Swaps
November 17, 2006
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033
European Equity Derivatives Strategy 17 November 2006
Table of Contents
Part 1: Variance Swap Mechanics........................................................................................................................ 8 1.1: Realised volatility.......................................................................................................................................... 8 1.2: The variance swap contract ....................................................................................................................... 11 1.3: Vega notional and variance notional .......................................................................................................... 12 1.4: Variance swap convexity............................................................................................................................ 13 1.5: Variance swap example – accruing realised volatility................................................................................ 14 1.6: Variance swap marktomarket .................................................................................................................. 15 1.7: Forward variance........................................................................................................................................ 17 1.8: Variance swap contract specifications ....................................................................................................... 19 1.9: Example Variance Swap Term Sheet ........................................................................................................ 22 Part 2: The Variance Swap Market ..................................................................................................................... 24 2.1: Market development................................................................................................................................... 24 2.2: Historical prices .......................................................................................................................................... 26 2.3: Variance swaps and option volatilities ....................................................................................................... 28 2.4: Pricing rules of thumb ................................................................................................................................ 30 2.5: What drives variance swap levels? ............................................................................................................ 33 2.6: The volatility risk premium.......................................................................................................................... 34 2.7: Variance swaps as predictor of future volatility.......................................................................................... 36 2.8: Is variance swap convexity fairly priced?................................................................................................... 38 2.9: Variance term structure.............................................................................................................................. 41 Part 3: Uses of Variance Swaps ......................................................................................................................... 44 3.1: Exploiting a volatility view........................................................................................................................... 45 3.2: Specific hedging purposes ......................................................................................................................... 46 3.3: Rolling short variance................................................................................................................................. 47 3.4: Diversification ............................................................................................................................................. 50 3.5: Index variance spreads .............................................................................................................................. 51 3.6: Relative value singlestock volatility........................................................................................................... 54 3.7: Variance dispersion and correlation trading............................................................................................... 56 3.8: Forward variance and volatility spikes ....................................................................................................... 58 3.9: Trading the variance swap term structure.................................................................................................. 64 3.10: Skew and convexity trades ...................................................................................................................... 66 3.11: Cross asset class trades: trading equity volatility against credit.............................................................. 68 Part 4: Replication and Hedging ........................................................................................................................ 72 4.1: Delta hedging and dollar gamma ............................................................................................................... 73 4.2: Theta  the cost of gamma ......................................................................................................................... 76 4.3: Options pathdependency: can volatility be captured by deltahedging? .................................................. 77 4.4: From options to variance swaps ................................................................................................................ 79 4.5: Variance swap replication in one page ...................................................................................................... 81 4.6: Sensitivity to skew and convexity............................................................................................................... 82 4.7: Variance swap Greeks ............................................................................................................................... 84 4.8: Setting up a replicating portfolio................................................................................................................. 87 4.9: Replication and hedging in practice ........................................................................................................... 89 4.10: Effects of variance swap hedging ............................................................................................................ 90 4.11: Why not volatility swaps? ......................................................................................................................... 92 Part 5: Future developments .............................................................................................................................. 94 References ........................................................................................................................................................... 98
2
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033
European Equity Derivatives Strategy 17 November 2006
Introduction
Introduction
Variance swaps are instruments which offer investors straightforward and direct exposure to the volatility of an underlying asset such as a stock or index. They are swap contracts where the parties agree to exchange a preagreed variance level for the actual amount of variance realised over a period. Variance swaps offer investors a means of achieving direct exposure to realised variance without the pathdependency issues associated with deltahedged options. Buying a variance swap is like being long volatility at the strike level; if the market delivers more than implied by the strike of the option, you are in profit, and if the market delivers less, you are in loss. Similarly, selling a variance swap is like being short volatility. However variance swaps are convex in volatility: a long position profits more from an increase in volatility than it loses from a corresponding decrease. For this reason variance swaps normally trade above ATM volatility.
Figure 1 : Variance swap payoffs are convex in volatility
p/l for 100K vega notional, variance strike of 20
Figure 2 : Euro Stoxx 50 implied and realised variance levels
Euro Stoxx variance
2,000,000 1,500,000 1,000,000 500,000 0 500,000 1,000,000 1,500,000 5%
Source : JPMorgan
variance swap p/l
payout linear in volatility
25% 20% 15% 10% 5%
3M Implied Variance
3M Realised Variance 3M Realised Volatility
10%
15%
20%
25%
30%
35%
Oct04
Source : JPMorgan
Apr05
Oct05
Apr06
Oct06
realised volatility
Market development Variance swap contracts were first mentioned in the 1990’s, but like vanilla options only really took off following the development of robust pricing models through replication arguments. The directness of the exposure to volatility and the relative ease of replication through a static portfolio of options make variance swaps attractive instruments for investors and marketmakers alike. The variance swap market has grown steadily in recent years, driven by investor demand to take direct volatility exposure without the cost and complexity of managing and deltahedging a vanilla options position. Although it is possible to achieve variance swap payoffs using a portfolio of options, the variance swap contract offers a convenient package bundled with the necessary deltahedging, offering investors a simple and direct exposure to volatility, without any of the path dependency issues associated with deltahedging an option. Variance swaps were initially developed on index underlyings. In Europe, variance swaps on the Euro Stoxx 50 index are by far the most liquid, but DAX and FTSE are also frequently traded. Variance swaps are also tradable on the more liquid stock underlyings – especially Euro Stoxx 50 constituents, allowing for the construction of variance dispersion trades. Variance swaps are tradable on a range of indices across developed markets and increasingly also on developing markets. Bid/offer spreads have come in significantly over recent years and in Europe they are now typically in the region of 0.5 vegas for indices and 1–2 vegas for singlestocks – although the latter vary according to liquidity factors. We expect spreads to come in further in future as liquidity continues to improve. Spreads for indices in the US and Japan are similar with single stock bid/offers in the 2 – 2.5 vegas range. Spreads are naturally higher in emerging markets although these too are becoming more liquid. With volatilities often much higher and less stable than in developed markets, emerging market variance can offer interesting opportunities.
3
For example.g. Diversification: Volatility can be thought of as an asset class in its own right. Variance swaps are useful instruments in debt/equity trades.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Introduction The most liquid variance swap maturities are generally from 3 months to around 2 years. and as such can act to diversify returns within a portfolio. analogous to the use of spreads and butterflies in fixed income to trade the shape of the yield curve. investors have come to view volatility itself as an asset class. Skew and convexity trades: Variance swaps are long skew and convexity. The VIX. Trading the variance term structure: Variance swaps can be used to trade the shape of the variance term structure. Uses of variance swaps Increasingly. either at the index or single name level. Variance swaps can be traded outright or spread against other products to create a wide variety of exposures. systematically selling volatility has historically been profitable. Rolling short index variance is an attractive systematic volatility strategy from a riskreturn perspective. • • • • • • 4 . and are calculated by the exchanges from listed option prices. even though they are only shortdated measures and are not directly tradable. Such trades can be thought of as either relative value volatility trades or as ‘volatilitybeta’ trades aiming to profit from a spread of volatilities widening as volatility increases. being long variance swaps) can help to hedge an equity portfolio. or crosssectional regression volatility models to find rich/cheap singlestock volatilities Variance dispersion and correlation trading: Trading variance swaps on an index against variance swaps on its constituents provides exposure to equity correlation. However. For example. volatility typically rises in a bear market. Below we list some common uses. Good liquidity is also seen in the front 3 months for shortdated index variance. And here investors can look to exploit the meanreverting nature of volatility to add alpha to portfolios. Index variance spreads: Variance swaps can be used to trade the spread of volatilities between two indices. Euro Stoxx and DAX indices respectively. one that can diversify investment returns or hedge unwelcome investment scenarios. Specific hedging purposes: Variance swaps can be used for macrohedging and also for hedging specific volatility exposures. These volatility indices are widely used as benchmark measures of equity market risk. Trading variance against (deltahedged) vanilla options provides interesting exposures to skew and/or convexity. the path to riches is to buy when cheap and sell when expensive. meaning that they can be efficiently hedged with exchangetraded options of the same maturity. Rolling short variance: Short variance swaps can be used to capture the observed equity index volatility risk premium. such as that resulting from structured products or life assurance policies. Cross asset class trades: Equity Volatility and credit spreads are correlated. VSTOXX and VDAX indices represent the theoretical prices of 1month variance swaps on the S&P500. interpolating to get 1month maturity. The most liquid maturities generally coincide with the quarterly options expiry dates. so holding a long volatility position (e. • • • • • Exploiting a volatility view: Variance swaps are ideal for taking a direct view on the volatility of an underlying without the pathdependency issues of a deltahedged option. and can be a useful way of positioning for volatility spikes. both being measures of corporate risk. as with any asset. with returns akin to selling credit protection. Relative value singlestock volatility: Use volatility pairs. although indices and more liquid stocks have variance swaps trading out to 3 or even 5 years and beyond. Forward variance and volatility spikes: Long forward volatility can avoid potentially negative carry at the cost of slide down the term structure.
This should not be surprising. Depending on market positioning. is to exploit the volatility risk premium by systematically rolling a short variance swap position. convexity etc. the pricing of variance swaps (including rules of thumb) and variance swap drivers. is path dependent. We give a concrete example of such a portfolio and consider some practical issues arising from variance swap hedging. Part 4: Replication and Hedging Part 4 is more technical in nature. such hedging activity may exacerbate large moves into the close. We begin by defining realised volatility and consider some of its commonly observed properties. 5 . Realised volatility and riskaversion are major drivers of variance swap levels. and variance swaps on other asset classes are starting to be traded. It is possible to remove this path dependency by trading a carefully chosen portfolio of options of different strikes. although in an efficient market such flows should be propagated back throughout the trading day. since variance is additive. As yet there is no official variance swap fixing mirroring deposit or swap rate fixings. In particular we discuss the potential impact of variance swap hedging on the underlying market. This leads to the simple valuation of forwardstarting variance swaps. at least at the index level. The main drawback of a deltahedged option as a vehicle for trading volatility is that the dollargamma. Such strategies are now being incorporated into structured products such as JPMorgan’s Yield Alpha product. as both can be interpreted as market estimates of future volatility. Part 3: Uses of Variance Swaps We discuss the uses of variance swaps. bundled up with the necessary deltahedging to give direct exposure to realised volatility. One important use of variance swaps. Part 5: Future Developments We discuss 3rd generation variance swap products and future developments. Part 2: The Variance Swap Market We look at the variance swap market development. have been much talked about. and futures exchanges have listed futures on these indices.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Introduction Outline of the document Part 1: Variance Swap Mechanics We consider the mechanics of variance swaps. with structured product flows impacting the long end of the maturity curve. Variance swap marktomarkets are easily calculated. to relative value volatility trades. driven by the need of replicating option portfolios to be deltahedged on the close (unlike other options). Variance swaps trade slightly above ATM volatility levels. Even though returns from a short variance or volatility position have more downside than upside risk. potentially adding to market volatility. at least better than using previous realised. fuelled by investor demand to trade pure volatility and by the desire of investment banks to recycle their exotic and structured product risks. We expect variance swap markets to continue to gain liquidity. Shortdated variance swaps tend to do a relatively good job of predicting future realised volatility. historically investors have been more than compensated for this risk. which characterises the exposure to realised volatility. to more complex trades using variance to trade correlation. essentially due to the fact that variance swaps are long convexity (or volatility of volatility). including a sample term sheet and specifications. and look at the different measures of notional: vega and variance notionals. Commonly watched volatility indices such as VIX and VSTOXX are in fact calculated as variance swap strikes. skew. Hedging flows. and anticipation thereof. We go on to define a variance swap. Conditional variance swaps have also increased in liquidity. Variance swap strikes are well correlated with standard BlackScholes implied volatility derived from options prices. We outline the mechanism whereby a variance swap can be replicated by a portfolio of options. The volatility risk premium reflects the observed tendency of implied volatility to trade above realised volatility. A variance swap consists of just this portfolio. ranging from simple expressions of a volatility view.
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 6 .
5 1.1.9 Realised volatility The variance swap contract Vega notional and variance notional Variance swap convexity Variance swap example Variance swap marktomarket Forward variance Variance swap contract specifications Example variance swap term sheet .4 1.7 1.3 1.8 1.6 1.2 1. Variance Swap Mechanics • • • • • • • • • Part 1: Variance Swap Mechanics 7 7 1.1 1.
Also noticeable is the effect of the 1987 crash. FX and commodities. 1. and conclude Part 1 by giving an example of a variance swap term sheet. Finally we consider some technicalities associated with variance swap contracts. Figure 3 : Long term history of realised volatility 3month realised volatility (Dow Jones Industrial Average) 70% 60% 50% 40% 30% 20% 10% 0% 1900 Source: JPMorgan 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 The following are some of the commonly observed properties of (equity market) volatility: • • • • Volatility tends to be anticorrelated with the underlying over short time periods (Figure 4) Volatility can increase suddenly in ‘spikes’ Volatility can be observed to experience different regimes (Figure 5) Volatility tends to be mean reverting (within regimes) 8 . but when considered in terms of volatility (as it more commonly is) the payout is convex. A variance swap pays out linearly in terms of the realised variance. Before discussing the mechanics of the variance swap contract we begin by defining the notion of realised volatility. We then define a variance swap and describe some common conventions such as the means of expressing trade notionals. We give a concrete example of a variance swap trade to illustrate how the p/l from realised variance accrues over time. through forward starting variance swaps which offer an interesting alternative to the usual spot variance exposure. The linearity of variance which makes markingtomarket so straightforward also allows us to construct exposure to forward variance. Periods of higher volatility can be observed.g. mostly due to an exceptionally large single day move. through changes both to implied and realised variance. They are swap contracts where the parties agree to exchange a preagreed variance level for the actual amount of variance realised over a period (variance is the square of volatility). fixedincome. although much of the discussion could apply to the volatility of other underlying assets such as credit. and to a lesser extent around 2000 with the buildup and unwind of the dotcom bubble. Variance swaps offer investors a means of achieving direct exposure to realised variance without the pathdependency issues associated with deltahedged options. In this note we are concerned with the volatility of equities and equity indices. Figure 3 shows the history of realised volatility on the Dow Jones Industrial Average over the last 100 years. e. in the early 1930’s as a result of the Great Depression. as well as numerous smaller volatility spikes. such as the treatment of dividends and M&A events. This naturally leads into a discussion of how variance swaps are markedtomarket. and look at some of its characteristics.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics Part 1: Variance Swap Mechanics Variance swaps are instruments which offer investors straightforward and direct exposure to the volatility of an underlying asset such as a stock or index.1: Realised volatility Volatility measures the variability of returns of an underlying asset and in some sense provides a measure of the risk of holding that underlying.
However. Volatility can also be calculated using returns of various frequencies (weekly. the realised volatility should be independent of the frequency of the returns used to calculate it. Furthermore. the difference from the traditional standard deviation is very small. or to trade a spread of volatility between related instruments. This simplifies calculations. Futures and Other Derivatives”. J. rhs) 0% Jan06 Source : JPMorgan 3900 4000 4100 Sep06 Mar06 May06 Jul06 1990 1993 1996 1999 2002 2005 Measuring realised volatility and variance Realised (or historical) volatility can be measured in a variety of ways. Volatility is usually quoted as an annualised volatility and expressed as a percentage. over a long enough timeframe. One issue with using this fixed timeperiod volatility is that the realised volatility calculated in this way can suddenly drop as a large magnitude return falls out of the window of observation. a volatility of 16% equates to a standard deviation of daily moves of around 1%. and in practice. in practice. If the returns are truly independent. Figure 4 : in the short term volatility is anticorrelated with index level Euro Stoxx 50 realised volatility index level (inverted) Figure 5 : 20 years of Euro Stoxx realised volatility: the index has experienced both high and low volatility regimes realised volatility 30% 3400 3500 3600 3700 3800 50% 40% 30% 20% 10% 0% 1987 Source : JPMorgan Euro Stox x 50 3m realised v olatility 20% 10% 1m realised volatility (lhs) index level (inverted.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics This list suggests some of the reasons why investors may wish to trade volatility: as a partial hedge against the underlying – especially for a volatility spike caused by a sudden market selloff. This is almost always the type of volatility which dictates the payoff of a variance swap contract. it is this RMS volatility measure which is used to define the payout of the variance swap contract. See “Options. volatility is usually calculated using daily closetoclose returns. Volatility can also be defined by an RMS (rootmeansquared) measure. One of the simplest measures is the annualised standard deviation of daily (log) returns of an underlying. Other. as a diversifying asset class. daily. for all but the shortestdated volatility (Figure 6. for a potential change in volatility regime. which is like a standard deviation but assuming a zero mean (Box 1). intraday returns etc). to take a macro view e.g. calculated from a data set spanning some fixed period of time – for example 1month historical volatility is calculated as the annualised standard deviation of daily returns occurring in one month. assuming returns are independent. Hull. usually taken to be 252. measures of realised volatility exist. The average return is the drift. Note that since the squareroot of 252 is about 16. more sophisticated. The annualised volatility represents the expected annual standard deviation in distribution of the underlying. simply multiply by the squareroot of the number of trading days in the year. which should be near zero on a daily basis. for further details. then. 9 . To move from the standard deviation of daily log returns to annualised volatility. This type of measure has the advantage of being very easy to understand and calculate and it is this which we used throughout this note. which seek to alleviate this problem and provide a potentially better means of forecasting future volatility – for example by giving more weight to recent returns (exponentially weighted measures) and/or attempting to model the meanreverting and regime changing nature of volatility (GARCH models). Figure 7).
000005 0. Longer dated volatility represents a smoother ‘average’ volatility but will include older data possibly outside of the current regime – e.0234394 15. Typical stocks have (annualised) volatilities of around 1535%. especially since it is measured in the same units as the underlying. but will convey more uptodate information.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics Box 1: Definition of realised (RMS) volatility Table 1: Calculating realised volatility (RMS) 13Oct05 14Oct05 17Oct05 18Oct05 19Oct05 20Oct05 21Oct05 24Oct05 25Oct05 26Oct05 27Oct05 Index level 3331.0 3310.8 3334.3% Volatility. Maturities considered usually range from a few days to a few years.9% Log return squared 0. whereas volatility is not.000141 0.3 3241. with equity indices generally somewhat lower.000002 0. Clearly shortdated realised volatility will be much ‘noisier’ than longer dated.4% 1.2% 0.000043 0.000093 0. 5year Euro Stoxx volatility currently remains some 9% above 1year volatility due to the inclusion of data from the high volatility regime in 20022003. is defined by: 252 T S i σ = ∑ ln T i =1 S i −1 2 2 Average the squared log returns where Si is the stock price on day i.6 3284. although the exposure is complicated by path dependency issues.000018 0.2% 0. Most importantly variance is additive. 10 .2% 0. Variance swaps payout purely based on realised variance.8 3271.5% 0. It is realised variance which the deltahedging of options seeks to capture.000030 0. and T is the number of days Variance of log returns Annualised variance Annualised volatility (RMS) Source : JPMorgan Annualise (x 252) Take square root Volatility conveys information about the usual size of moves in the underlying in relation to its absolute price. variance (volatility squared) is in some sense more fundamental. Figure 6 : The difference between RMS volatility and standard realised volatility can be noticeable for shortdated volatility… 5day realised volatility (Euro Stoxx 50) Figure 7 : …but the difference is negligible for longerdated volatility 6month realised volatility (Euro Stoxx 50) 40% 35% 30% 25% 20% 15% 10% 5% 0% Jul05 Source : JPMorgan 17% 16% 15% 14% 13% 12% 11% 10% Jul05 Source : JPMorgan Standard v olatility RMS (zeromean) v olatility Standard v olatility RMS (zeromean) v olatility Oct05 Jan06 Apr06 Jul06 Oct05 Jan06 Apr06 Jul06 Although it is volatility which is the more intuitive concept.1 3292.7% 0.5 3304. σ.000028 0.7% 1.000013 0.1 Log return 0.6 3356.4 3349.g.4% 1.5% 0.8 3279.000372 0.000278 0.
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics 1. Similarly. and will be delivered the difference between the realised variance over the next year and the current level of implied variance. but pay out based on the difference between the level of variance implied by the strike (in fact the strike squared) and the subsequent realised variance. The strike is set according to prevailing market levels so that the swap initially has zero value.0225 (quoted as 225). If the subsequent realised volatility is above the level set by the strike. representing a p/l of €2.500 x (400 – 225) . volatility is scaled by a factor of 100.500 per point difference between realised and implied variance. An investor wishes to gain exposure to the volatility of an underlying asset (e. This means the p/l of a variance swap is nonlinear (convex) with volatility. and if realised volatility is below. for example a strike of 20 represents a volatility of 20%. multiplied by the variance notional. The short will profit by the same amount. not variance. the buyer will be in loss. the investor will make a loss because realised variance is below the level bought. Example 1: Variance swap p/l N Variance is the variance notional. Suppose the trade size is €2. Variance swaps offer investors a means of achieving direct exposure to realised variance without the pathdependency issues associated with deltahedging options. Overall loss to the long = €437.500 = €2. not to be confused with the strike of an option.500 variance notional. 11 . and where K is the variance swap strike. They are swap contracts where the parties agree to exchange a preagreed variance level for the actual amount of variance realised over a specified period (Figure 8). Note that variance swap strikes are quoted in terms of volatility. A buyer of a variance swap is therefore long volatility. The strike of a variance swap. Figure 8: P/L of a variance swap Buyer pays variance swap strike K2 Buyer of variance swap Seller pays realised variance at expiry σ2 Seller of variance swap By convention. Euro Stoxx 50) over the next year. the buyer of a variance swap will be in profit. although of course it is linear in terms of variance. a seller of a variance swap is short volatility and profits if the level of variance sold (the variance swap strike) exceeds that realised.g. Box 2: Variance swap p/l The p/l for a (long) variance swap is given by: p / l = N Variance × σ 2 − K 2 ( ) σ2 is realised variance. The investor buys a 1year variance swap. If the variance swap strike is 20 (implied variance is 400) and the subsequent variance realised over the course of the year is 15%2 = 0.2: The variance swap contract Variance swaps are instruments which offer investors straightforward and direct exposure to the volatility of an underlying asset such as a stock or index. represents the level of volatility bought of sold and is set at trade inception.
the variance swap payout. a gain of €562. The p/l of a (long) variance swap can be calculated by multiplying the vega notional by the difference between the realised variance and the strike squared. The average exposure to the 5% move in realised volatility is therefore 5 vegas. the long will receive €562. trade size is typically expressed in vega notional. However.000 or 5 times the vega notional. giving the same calculation as that used in Example 1.000 / (2 x 20) = €2. If the index realises 25% volatility over the next year.500 = €100. expressed in vega notional.500 = €100.3: Vega notional and variance notional The notional for a variance swap can be expressed either as a variance notional or a vega notional. the long will pay €437.e. Box 3: Vega and variance notionals p/l σ 2 − K 2 = N Vega × 2K where = NVariance × σ 2 − K 2 ( ) NVega is the veganotional. Note that the variance notional is €100. and most market participants are used to thinking in terms of volatility.625 times the vega notional). 12 .500.625 vegas (i. or 5 times the vega notional. That is. the p/l is close to the difference between implied and realised volatility multiplied by the vega notional.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics 1.000. NVega = NVariance × 2K Example 2: Vega notional Suppose a 1year variance swap is stuck at 20 with a vega notional of €100. as expected. Similarly a loss of €437. This ensures that when realised volatility is close to the variance strike. and NVariance is the variance notional. 5. and then dividing by twice the strike. In the example above.365 vegas. The p/l of a variance swap is often expressed in terms of vega notional. and thus gives a more economically meaningful idea of the variance swap’s exposure to volatility.500 represents a loss of 4. The variance notional is in some sense the ‘true’ notional – representing the p/l per point difference between the strike squared (implied variance) and the subsequent realised variance. The vega notional is equal to the variance notional multiplied by twice the variance strike. is locally linear around the strike. Therefore the average exposure for a realised volatility being 5% away from the strike is €500.000 x (152–202) / (2 x 20).000 x (252 – 202) / (2 x 20). since variance is a somewhat unintuitive quantity.500 is expressed as a profit of 5. However if the index only realises 15%. Vega notional represents the average profit or loss for a 1% (1 vega) change in volatility.
it turns out that the maximum vega loss on a short 2. whereas the loss from a short variance swap is not (unless capped). Example 3: Maximum loss of a variance swap Suppose a 1year variance swap is stuck at 20 with a vega notional of €100. is 2. Ultimately this convexity is the reason that variance swaps strikes trade above atthemoney volatility. strike of 20 Figure 10 : … but for large swings in volatility. as expected.g. For a long variance swap the maximum loss will occur when the realised volatility is zero.000 8.000 x 10 = .000 x (502 – 202) / (2 x 20) = 2.€1M This is equal to a loss of ½ times the 20 strike or 10 vegas. In fact a long variance swap position will always profit more from an increase in volatility than it will lose for a corresponding decrease in volatility (as in Example 2).5x the variance strike (see Section 1. Figure 10). strike of 20 1.000.000 1. In the limiting case.000. 13 . unless the variance swap payout is capped. the variance swap payout is similar to a linear “volatility swap”… p/l for 100K vega notional.000. Figure 9 : When volatility remains close to the variance swap strike. The vega notional represents only the average p/l for a 1% change in volatility. the loss from a long variance swap is limited. This difference between the magnitude of the gain and the loss increases with the change in volatility. The maximum loss will occur when realised volatility is at least 50. In this case.8 for a detailed discussion of the value of the variance swap’s convex payoff. The standard cap. with volatility of volatility zero.000 v ariance sw ap p/l hy pothetical "v olatility sw ap" p/l 10.000. but in this case the convexity of the payout works in favour of the long with the result that the maximum vega loss on a long variance swap is half the strike (Example 3). In this case p/l = €100.000 10% Source : JPMorgan 12.000. The higher the variability of volatility.5 x strike. This is the convexity of the variance swap.000 4.625 times the strike (see Example 3).000.500. For the short. if applied.000 0 500.000 500.625 x strike x vega notional. The price paid for this convexity should depend on the expected variability of the (realised) volatility.000.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics 1.000 2.000. For a 2.000 x (02 – 202) / (2 x 20) = €100.000 6. Because of the convexity.8 for details of caps).000 v ariance sw ap p/l hy pothetical "v olatility sw ap" p/l 15% 20% 25% 30% 0% Source : JPMorgan 10% 20% 30% 40% 50% 60% 70% realised volatility realised volatility In common with long options positions.000 0 2. a variance swap will always outperform a contract linear in volatility of the same strike (e.000 1.2M = €100. In the case of options. In the case of variance swaps the maximum loss occurs when realised volatility is zero.5x capped variance swap is 2. the larger the likely discrepancy between the variance strike and the realised volatility. losses are potentially unlimited.000. See Section 2.4: Variance swap convexity Although variance swap payoffs are linear with variance they are convex with volatility. and hence the more beneficial the convexity becomes (Figure 9. the convexity of the variance swap payout becomes important p/l for 100K vega notional.000.5x capped short variance swap the volatility on which the variance swap pays out will be capped at 50 = 2. the loss from the long is limited to the premium paid. the p/l of the variance swap will always be zero (volatility is constant) and hence the variance swap strike and hypothetical volatility swap strike must coincide. Figure 9). This gives a maximum loss of €5.
554 12. we would expect a p/l from the short position of €220.9% 0.5 for €100.4% 16.0% 0.777 65.5 3304.8 3271.000000 0.000278 0.622 30.1% 15.7% 30.3 3241.194 38.2% log return squared 0.5 Day 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Date 13Oct05 14Oct05 17Oct05 18Oct05 19Oct05 20Oct05 21Oct05 24Oct05 25Oct05 26Oct05 27Oct05 28Oct05 31Oct05 01Nov05 02Nov05 03Nov05 04Nov05 07Nov05 08Nov05 09Nov05 10Nov05 Index Level 3331.4% 12.2% 0.5% 0.000004 0.000028 0.7% 1.2% 0.234 29.6% 13.0 3310. Table 2: Daily p/l and cumulative realised volatility from a 20day short variance swap struck at 16. The variance swap was sold on 14 October 2005.000151 0.6 3361.714 Note how most days were low volatility and produced modest positive returns for the short variance position.457 94.2 times the vega notional).5% 15.7% 0.714.4% 15.9% 15. Since the vega notional gives the expected p/l for a 1 point (1%) move in volatility.210 40.9% 8.5% 2.000018 0.023 91.000013 0.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics 1.377 36.000510 0. Figure 11 : Index performance over the lifetime of the variance swap Euro Stoxx level Figure 12 : Accrued volatility vs.2% lower than the level of implied sold. Over the 20day period realised volatility was 14.5% 15.880 39.962 69.6% 8.312 39.2% 0. due to the convexity of the payout working in favour of the long.1 3292.000002 daily volatility (annualised) 8. However.304 34.6% 6.3 3362.0% 14.3%. a few days with relatively large moves caused significant daily losses (Table 2.2% 16.4% 10.824 40.000043 0.5% 3. expiring on the 10th November 2005.1% 2.959 45.040 40. with a strike of 16.5 3320.165 69.000003 0.778 158.000005 0.9% 19.615 39.6% 13.6% 0.962 39.8 3279. 2.4 3349.6% 17.157 126. and against the short.000141 0. Figure 12).3% 0.4% 1.1 3312.0 3320.714 accrued realised volatility (annualised) 8.2% 1.367 166.8% 16.000030 0.5% daily p/l (€) 29.3% accrued p/l (€) 29.000002 0.357 97.307 206.271 5.0 return 0.000006 0.6% 2.000005 0.403 100.8% 2.2% 0.469 104.8 3361.181 57. This pattern is common: short variance will profit modestly most of the time.407 206.000 (=2.5% 6.6% 3.401 121.5% 0.3% 14.2% 0.6% 18.6% 14. the actual p/l is slightly less: €206.7 3355. but will lose heavily on large moves.858 16.7% 3.4% 26.2% 0.1 3246.000372 0.8 3361. daily returns Accrued realised volatility absolute daily return (rhs) 20% realised v olatility to date (lhs) 3400 3350 3300 3250 3200 daily return 3% index level 15% 10% 5% 0% 13Oct 20Oct 27Oct 03Nov 10Nov 2% 1% 0% 13Oct 20Oct 27Oct 03Nov 10Nov Source : JPMorgan 14 Source : JPMorgan .839 85.664 22.9% 3.000 of vega notional (equivalent to €3.494 24. The variance swap maturity was 20 (business) days.1% 12.935 16.4% 35. However.5% 5.8 3334.6 3284. using prevailing market prices.0% 3.400 153.000005 0.030 variance notional).318 41.5: Variance swap example – accruing realised volatility In this example we show the evolution of the p/l of a short Euro Stoxx 50 variance swap through the accrual of realised variance.4% 1.070 40.5% 2.2% 0.6 3355.000002 0.6 3356.757 55.2% 0.0% 13.
but the volatility realised over the following 9 months is 25%. If this is currently trading at 25.400. we wish to realise this p/l now (after 3months). and the implied variance over the remaining time to maturity. If the volatility realised over the first 3 months is 15%.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics 1.500.000 (variance notional of €2. This is simply the prevailing strike of a 9month variance swap. suppose again that realised volatility was 15% over the first 3 months. 9month LIBOR is 4%. All that is needed to compute the marktomarket of a variance swap is: • • The realised variance since the start of the swap. then.500 Now.500).9 volatility). In order to value the variance swap after 3 months we need to know both the (accrued) realised volatility to date (15%) and the fair value of the expected variance between now and maturity.9 – 202) = €312. the discount factor applied would be 1/(1+¾ x 0. and The implied variance (variance strike) from the present time until expiry Since the variance swap is usually settled at maturity. the expected variance at maturity is simply the timeweighted sum of the variance realised over the time elapsed. since variance is additive. an especially important consideration for long dated variance swaps. after 3months.6: Variance swap marktomarket Although the example on page 14 illustrates how the variance swap accrues realised volatility over time. Example 4: Variance swap marktomarket Suppose a 1year variance swap is stuck at 20 with a vega notional of €100.500. this change in value of future expected variance can be the main driver of p/l. It is therefore necessary to apply an appropriate interest rate discount factor. a discount factor between the present time and expiry is also required. the variance realised over the year is: Variance = [ ¼ x 152 ] + [ ¾ x 252 ] = 525 2 (≈22. Indeed for variance swaps valued relatively soon after inception. At expiry the p/l would be €2. valuation is a function of realised variance before t.04) = 0. Box 4: Variance swap marktomarket p/l = Nvar x ( [σt2–Kold2] Variance swap inception x %time elapsed + [Knew 2 – Kold 2] x %time remaining ) Variance swap expiry (1Y) Knew (=25) At time t (3m). Although the fair value at maturity (now 9 months in the future) is €312. At an intermediate point in the lifetime of a variance swap. and implied variance from t to T Kold (=20) σt (=15%) 0 t (3m) T (1Y) 15 . If. Hence the marktomarket p/l after 3 months would be equal to about €303. if an investor wishes to close out a variance swap position before maturity they must also consider any change in value of the exposure to volatility over the remainder of the variance swap term.500 x (22.97. then the same calculation as above gives a fair value at maturity for the 1year variance swap of €312. Marking to market of variance swaps is easy: variance is additive.
T − K 02. This is exactly equivalent to the marktomarket equation given above and presents a useful framework for thinking about forward variance.t represent the strike for a variance swap at time s maturity at time t.T + K t .i −1 2 is the daily annualised realised volatility equal to 252 ln Each daily p/l is given by a daily variance exposure equal to the total variance notional divided by the total number of days between the start and end dates of the swap.T − K 02. The remaining variance notional size will be equal to (Tt)/T times the original notional.t be the realised volatility between time s and time t.t − K 02.97 as explained in Example 4 gives a marktomarket p/l of €303.T − K 02. Let σs. whilst the second term represents a fixed stream of daily accruals up to expiry at time T. This is done in order to cancel out the remaining exposure to the daily realised volatility component.t + (T − t )K t2. Pi −1 where σ i .i −1 − K 02. Then the p/l (in terms of variance notional) at time t. proportional to the time remaining to maturity on the swap.T T T ( ) ( ) = Example 5: Variance swap marktomarket continued 1 2 tσ 0.T T [ ] Returning to Example 4 we can calculate the marktomarket p/l (at maturity) directly from the formula above as follows: p/l = €2. the subject of the next section. Calculating a marktomarket or closing out an existing position requires entering into a new but opposing variance swap position in the remaining variance notional. is given by: Variance (forward) p/l at time t (per unit variance notional): = t 2 T −t 2 σ 0.400 Box 6: Variance swap marktomarket through daily accruals An alternative approach to calculating marktomarket is to consider a variance swap as a sequence of daily variance swaps. It is important to notice that the total remaining exposure to variance decreases linearly with time.T ) T t +1 T The first term represents the realised variance accrued by the variance swap up to time t. of a variance swap maturing at time T.500 = €2. On each day the p/l accrued by a long variance swap is given by: Daily accrual on day i (per unit of variance notional) = x ( [152 – 202] x ¼ x 125 + [252 – 202] x ¾) x ( 175 x ¼ + 225 x ¾ ) 1 2 (σ i . Thus p/l of the trade at time t will be equal to: T t 1 2 (σ 0.T ) . T Pi and K 02.500 Applying the interest rate discount factor of 0.500 = €2.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics Box 5: Calculating the variance swap marktomarket Let Ks.500 = €312. 16 .t − K 02.T ) + ∑ ( K t2.T is the initial strike. not the initial notional.
and expiring 9 months later? Since variance is additive. thus enabling us to compute this expected forward variance. T t 1 t T N N 2 2 2 2 Combining the long and short variance positions gives an overall p/l of: ∑ T ( K t − K T ) + ∑ T (σ i . This equates to the fair strike of the forwardstarting variance swap. using the additivity of variance as for the marktomarket calculation: 202 = [¼ × 152] + [¾ × F2 ] Solving for F.i −1 is realised daily volatility T 1 t +1 T t Consider the p/l for a short variance swap. To illustrate this we being by considering the p/l of a long variance swap. which because of the additivity of variance can be spilt into two periods (1 t and t+1 T) as follows T NT 2 N 2 2 (σ i .2T = K T − t T t 1 t ∑ ( K t2 − K T2 ) = K T2 − T − t ( K t2 − K T2 ) = T − t K T2 − T − t K t2 T −t 1 The overall p/l becomes ∑ T − t NT (σ i2. ending at time T of notional size: T −t NT . Then. F. such that 2 Ft . the variance expected over the year is the time weighted sum of the variance expected in the first three months. but 3month variance is struck at 15.i −1 − K T ) T 1 t +1 T ∑ t The first p/l term represents a fixed quantity which can be subtracted from the fixed strike in the second p/l term. What is the fair strike for a variance swap starting in 3 months time. Suppose we know the strikes for a shortmaturity variance swap expiring at time t. a combination of long and short variance swaps can be used to cancel out exposure to realised volatility over a particular time period. the longer maturity variance swap is simply the timeweighted sum of the short maturity variance and the expected variance over the forward period.i −1 − K T ) where NT is the variance notional and σ i .2T ) T T −t t +1 T This just represents the p/l of a variance swap starting a time t.i −1 − K T ) + ∑ T (σ i2.i −1 − Ft . We then define a forward variance to be equal to this combined strike. gives a strike for the forward variance of 21. This shows how a forward variance swap can be constructed. Box 7: Calculating forward variance through daily accruals Using the approach that a variance swap can be thought of as a strip of daily variance swaps (see p16). T 17 . and a longer maturity variance swap expiring at time T. maturity t.4. Example 6: Calculating forward variance Suppose that one year variance is struck at 20. and that expected in the remaining nine months. We want to find the expected realised variance. Since variance is additive. This is equal to: T t t NT 2 −∑ (σ i .7: Forward variance The reasoning used to compute the variance swap marktomarket in the preceding section can also be used to calculate the expected variance over a forward starting window. of variance notional N T (with delayed payment).i −1 − K t2 ) which will offset the realised variance from the second term in the p/l from the long above.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics 1. This second value is the forward strike F. giving exposure to realised volatility starting after a future date. between time t and time T.
2T = Example 7: Forward variance construction As in Example 5. This can have a practical impact of increasing the bid/offer of forward starting variance swaps. and Ft.115 variance notional of the 12month variance swap. and long 2 x 20 x €3.8 and 3.T = KT − K t2 T −t T −t T −t T −t Solving for F gives a value for the forward variance: Ft .000 of the 12month variance.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics The calculations in Box 7 show that a long position in forward starting variance (of unit variance) can be constructed from: • • Long T/(Tt) variance notional of spot variance maturity T.T (as in the marktomarket calculation in Box 4) T T T t T t 2 K T2 − K t2 or Ft .T is the strike of the forward starting variance swap (with exposure to variance between t and T). See Sections 3. The 9month variance swap. this equates to €100. 3month forward variance will be struck at 21. then: K T2 = t 2 T −t 2 Kt + Ft . forward starting variance is increasingly quoted as a single level rather than a weighted spread with the consequence that bid/offers are converging to similar levels as seen for spot variance. KT is the strike of the longermaturity variance swap (maturity T). Suppose we wish to trade €100.4) = €2.000 of forward variance swap notional. However. and 12/9 x €2. and a 1year variance swap is struck at 20. suppose a 3month variance swap is struck at 15. and Short t/(Tt) variance notional of spot variance maturity t.4.000 vega notional of the 3month variance swap. is that more variance is need on the longer leg (which tends to be less liquid) than the shorter leg and that the total notional of the two legs will be greater than the notional of the forward. Forward variance swaps are useful for taking a direct view on the future value of implied variance and/or the future shape of the variance term structure curve in the same way that Eurodollar and Euribor futures are used to take view on the future level of interest rates.000 / (2 x 21. 18 . Furthermore the fixed income concepts of slide and carry are useful concepts to apply to both spot and forward starting variance swaps. This in turn equates to short 2 x 15 x €778 = €24. Hence we require 3/9 x €2.336 = €778 variance notional of the 3month variance swap (with payment delayed). The ability to reexpress the spot variance term structure into a strip of forward starting variances raises natural analogies with fixed income – in particular the swap curve and a strip of forwards/futures.115 = €125. As in the case of a spot variance swap.9 for examples of the uses of forward variance swaps. but with payment delayed until maturity T.336 = 3. One of the things worth noting on forward starting variance. as the variance swap market evolves and matures. Box 8: Calculating forward starting variance Suppose Kt is the strike of the shortmaturity variance swap (of maturity t).336 variance notional.
the return used in calculating the realised volatility for the variance swap payout will be 94/951 = 1. see p10) must also be multiplied by 100. This means that the return on the exdividend date is calculated after adjusting for the dividend. For example. Dividend adjustments Variance swaps on single names are typically adjusted for dividends. This means that changes to the underlying due to dividend payments do not count towards the realised volatility calculation for the variance swap. down 2. 18 January 2006 the Tokyo Stock Exchange (TSE) closed early following revelations of the investigation into Livedoor. Thursday. Margins and collateral Variance swaps are usually margined in a similar manner to options. Index reconstitution risk Variance swaps on indices are defined to pay out on the returns of the index and not on the weighted returns of the basket of current constituents. Such a move will act to increase realised volatility more than by the separate 5% and 6% moves. Conversely. 19 . this will count as a single. This can be important for stocks with high dividend yields which pay entire annual dividend in a single payment – e. the day was declared disrupted. if a stock is worth €100 on the day before the dividend. The next day. pays a dividend of €5 and closes at €94 on its exdividend day. The occurrence of a disrupted day could potentially work for or against a long variance position. was considered a normal trading day. potentially with different volatility characteristics than when it was originally traded. the cap on the weight of HSBC. However. Disrupted days The realised variance used in a variance swap payout is calculated from closing (or official observed) prices on observation dates over a specified period. At the point when trading was suspended on the TSE the Nikkei was at 15341. compounded 1day loss of 10. and the variance swap may end with an exposure to a very different set of stocks. with an initial amount to be posted as collateral (3 times the vega amount in the example termsheet below). For this reason the calculated value of the realised variance (using an RMS calculation. Settlement is calculated at maturity and cashflows exchanged shortly afterwards (T+2 in the example below).7% will be counted (see Example 8). its closing price will not be used in the calculation of the realised volatility and only the single combined return of 0. Italian stocks such as Telecom Italia and ENEL (Figure 13).3% on the last level on the disrupted Wednesday.9% on the previous day’s close of 15806. both special and regular.7% counted towards the calculation of realised volatility. A long variance swap position would therefore have lost out due to the Wednesday being declared a disrupted day. and the provisional closing price was not accepted for variance swap calculations.9.7%. Observation dates are defined as all those scheduled trading days which are not disrupted. and the Nikkei recovered most of the previous day’s losses. Suppose a 5% loss is followed by a 6% gain over 2 consecutive trading days. closing at 15696. a low volatility stock. in the Hang Seng. if a 5% loss on a day declared disrupted is followed by a 6% loss the next day which is nondisrupted.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics 1.g. January 19th. Strikes are typically expressed as 100 times the annualised volatility represented – for example 25 will represent a breakeven realised volatility of 25%. only the change from the close on the Tuesday to the close on the Thursday.8: Variance swap contract specifications Variance swap conditions are set out on term sheets such as the one shown in Section 1. and will work in the favour of the long. Example 8: Disrupted days On Wednesday. For example. Further margin calls will be made during the course of the trade as necessary. If the first day is declared as disrupted.05% (not 94/1001 = 6%). a return of 0. and the forthcoming addition of a number of potentially higher volatility Chinese HShares may act to increase the overall volatility of the index. This means that index variance swaps (especially longdated ones) will be exposed to the risk of index reconstitution. For the purposes of the variance swap calculation. up 2.
5 times the strike. Singleday moves needed to cause a variance swap cap to be hit are large and increase with maturity.g. 20 . especially on singlestocks (and sector indices). potentially giving an artificial boost to the realised volatility. Example 9: Offsetting capped variance swaps Suppose that in an investor buys a 6month variance swap with a strike of 20%. there are cases of indices containing high dividend paying stocks some of which pay on the same day – e. frees up capital and avoids any technical issues with changes in cap level (see Example 9). In theory this second alternative is simpler as it removes all collateral obligations. are usually sold with caps. he will have sold a ‘50%/75% call spread’ on volatility. in the course of trying to close out his position. Closing out variance swaps Investors wanting to realise marktomarket p/l by closing out a variance swap before expiry generally have two choices. it is often due to a single large move – e. When caps are hit. For lower strikes the required moves are also lower. potentially with a different counterparty. the residual volatility exposure is presumably unwanted.625 times the strike.5 times the strike of the swap capping realised volatility above this level. With the strike now at 30%. see Box 9).5 x 30% = 75% Then if the investors sells this 30strike variance swap in an attempt to close out his position the difference in caps will mean he takes on a short volatility exposure if the subsequent realised volatility is above 50% (although capped at 75%) In effect. These are often set at 2. or possibly from a dramatic selloff on an index. the SPMIB and the SXKE (the Eurozone Telecoms index) – where index moves as a result of dividend payments can be significant.5K })2 − K 2 × 2K N Vega × Note that this sets the maximum loss for the short to (2. In these cases the extra volatility resulting from dividend payments will count in the calculation for the variance swap payout. 1month later. due to an M&A event or major earning surprise on an individual name. They can either enter into an opposing offsetting contract. or they can agree with the original counterparty on a level at which to close out the contract. will hit its cap. This has the standard 2.2. However. where investors are then able to quantify their maximum possible loss (to 2.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics By convention index variance swaps are not usually adjusted for dividends – since the payments are usually more spread out and the impact of a few index points worth of dividend is generally small in comparison to the average daily move. Variance swap caps are useful for short variance positions.5x cap meaning the exposure to realised volatility will be capped at 50%. For example a 3month variance swap struck at 20 and realising 20% on all days except for one day which has a oneoff 24% move. Then the p/l (for the long) is given by: p / l = N Vega (Min{σ . the cap on an new variance swap contract will by default be set at 2. Whilst the price he gets for selling the variance swap will reflect this higher cap. However.5K )2 − K 2 2K = N Vega × 2. The investor wants to lock in this profit.625K In practice caps are rarely hit – especially on index underlyings and on longerdated variance swaps.g. the (5month) variance swap is now trading at a strike of 30% leading to a significant marktomarket p/l – the exact amount depending on the level of volatility realised over the month. A similar 1month maturity swap would need only a 14% move to hit the cap. and the investor would be best either trading directly with the original counterparty or negotiating a bespoke contract with another counterparty in order to fully close out his outstanding contract. Variance swap caps Variance swaps. in practice it is possible to negotiate bespoke contracts specifically to close out existing positions in a process known as novation which allows for the different counterparties to close out the contract fully. Box 9: Variance swap caps Suppose a variance swap is traded with a cap set at 2. whereas the required 1day move on a 1year swap would be 46%.
the options.5 Nov05 Feb06 May06 Aug06 Source : JPMorgan . options are unwound immediately after the offer has been declared unconditional. Bloomberg 0% Nov06 Aug05 Sep05 Oct05 Nov05 Dec05 Jan06 Feb06 Mar06 Source : JPMorgan. given the size of the daily move on the day of the bid (+ 25%) the 3month realised volatility from the beginning of October up until the unconditional date on January 3rd. was around 50%. as an underlying being acquired can effectively cease to exist. IDM in Italy unwinds at intrinsic and MEFF in Spain works on a case by case basis. and hence the variance swap. In this case.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics M&A events M&A events present particular issues for variance swaps.0 25% 20% 15% 10% 5% 220 200 180 160 140 120 100 OOM LN Equity 1M realised v olatility 3M realised v olatility 100% offer goes unconditional 7. Although outstanding options were settled at the implied volatility prevailing before the October bid. One further issue with cash bids. For share offers. In the case of cash bids. will usually transfer into a variance swap on the new underlying after the transaction date. and variance swap contracts generally follow these rules.0 6. and the time up until unconditional acceptance is longer. or composed offers consisting of less than 66% cash. the low realised volatility over this period could hurt long variance swaps. div pay ment) Figure 14 : O2 realised volatility collapsed following the bid. But the large 1day move from the bid was large enough to keep realised volatility high Share price (p) 1m realised volatility 8. Figure 13 : The apparent drop in spot and consequent spike in volatility was entirely due to the payment of a single 6. The various options exchanges have rules for how such an event is dealt with. it took until 3 January 2006 for the offer to go unconditional (Figure 14). In Europe.5 dividend paid 80% 60% 40% 20% 0% cash bid 7. Bloomberg 21 .3% dividend. The Euronext exchange unwinds at fair value. Eurex unwinds at intrinsic but is switching to fair value from the beginning of 2007. in cases where the move on the bid is smaller. For example when Telefonica bid for O2 on 31 October 2005. Share price (€) 1m realised volatility ENEL IM Equity 1M realised v olatility (incl. leading to two months of virtually zero accrued volatility on any outstanding variance swaps. However. often with close to zero volatility. the options are not unwound until the offer is declared unconditional. In any case documentation should either specify the exact rules in the case of an M&A event or should refer to the rules of the appropriate exchange. However. they were not unwound until the beginning of January. is that the underlying will usually trade at close to the bid level after the offer has been accepted. well in excess of levels of implied variance trading before the bid. some bourses unwind vanillas at intrinsic and others at the theoretical fair value (generally taking the implied volatility to be the average over the 5 days before the bid). div pay ment) 1M realised v olatility (ex cl. Investors should be aware in which exchange the variance swap is based since M&A rules are exchange specific.
if there is no official price. Settlement Price means the official price or. the Settlement Price per Underlying on the Observation Day immediately preceding such Observation Day. Calculation Agent: Documentation: ISDA 22 . the Settlement Price per Underlying on the Observation Start Date and. Observation Period The period from. ln means the natural logarithm. Pt1 means in respect of: (i) the first Observation Day.000 16. the official level of the Index at the Valuation Time on such Observation Day. Where FRV = P 252 × ∑ ln t P t =1 t −1 × 100 Expected _ N N 2 t means each Observation Day. but excluding. the Observation End Date. the midmarket price per Underlying on the Exchange at the scheduled weekday closing time (without regard to after hours or any other trading outside of the regular trading session hours) on an Observation Day. and (ii) each other Observation Day.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 1: Variance Swap Mechanics 1. the Equity Amount will be calculated and paid in accordance with the following formula: Vega Amount 2 2 2 × Min (FRV . If Equity Amount is negative the Variance Buyer will pay the Variance Seller an amount equal to the absolute value of the Equity Amount. the Observation Start Date to. (ii) each subsequent Observation Day.9: Example variance swap term sheet Sep06 OTC Variance Swap: FTSE Index Variance Instrument: Trade Date: Maturity Date: Variance Buyer: Variance Seller: Underlying: Denominated Currency: Vega Amount: Strike Price: Cap Level: Equity Payment: Variance Swap 10th August 2006 15th September 2006 JPMORGAN ******* FTSE100 INDEX (Bloomberg: UKX Index) Great British Pounds (“GBP”) GBP 100.95 T+2 after the Observation End Date.95 36. JPMorgan Securities Ltd. as of the Trade Date. CapLevel ) − Strike Price 2 × Strike [ ] If Equity Amount is positive the Variance Seller will pay the Variance Buyer the Equity Amount. are expected to be Scheduled Trading Days in the Observation Period. the Official Settlement Price of the Exchangetraded Contract on the Observation End Date. Expected_N means 25. Pt means in respect of: (i) the final Observation Day. Observation Day Each Scheduled Trading Day that is not a Disrupted Day during the Observation Period. Observation Start Date Trade Date. Observation End Date Maturity Date. and including. the number of days which. N means number of Observation Days.
2 2.9 Market development Historical prices Variance swaps and option volatilities Pricing rules of thumb What drives variance swap levels? The volatility risk premium Variance as a predictor of future volatility Is variance swap convexity fairly priced? Variance term structure .8 2.1 2.7 2.3 2. The Variance Swap Market • • • • • • • • • Part 2: The Variance Swap Market 23 2.5 2.4 2.2.6 2.
9.6 1 –.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market Part 2: The Variance Swap Market Part 2 begins with an overview of the development and current state of the variance swap market. Topix HSI. RDX. 1999 or “Towards a Theory of Volatility Trading. without any of the path dependency issues associated with deltahedged options. 2. DAX. This leads in to a discussion of the variance swap term structure and curve dynamics. Russell 2000 Nikkei 225. The variance swap market has grown steadily in recent years.g. to a certain extent levels now deviate from the theoretical price obtained from a replicating portfolio of options. 2001). allowing for the construction of variance dispersion trades. Emerging Markets Euro Stoxx 50. the variance swap contract offers a convenient package bundled with the necessary deltahedging.2 3–5 Typical singlestock bid offers. We discuss the convexity premium for variance swaps in terms of the spread of variance swap prices over ATM options implied volatility. Carr and Madan’. Nasdaq.6 0.5 24 5–6 Source: JPMorgan estimates. concentrating particularly on the Euro Stoxx 50 and S&P500 indices. November 2006 24 .4 – 0. Derman. offering investors a simple and direct exposure to volatility. We also consider the drivers of variance swaps prices: principally realised volatility and risk aversion at shorter maturities and structured product flow at the longer end of the curve. vegas (most liquid stocks and maturities) 1–2 1–2 2 – 2. variance swaps on the Euro Stoxx 50 index are by far the most liquid. driven by investor demand to take direct volatility exposure without the cost and complexity of managing and deltahedging a vanilla options position. Japan Asia Ex. Variance swaps are also tradable on the more liquid equity underlyings – especially Euro Stoxx 50 constituents. Variance swaps were initially developed on index underlyings.4 – 1 0. Variance swaps can be thought of as the market’s best estimate of future realised variance and we consider how good a job they do of predicting this. (See e. The directness of the exposure to volatility and the relative ease of replication through a static portfolio of options make variance swaps attractive instruments for investors and marketmakers alike. Although it is possible to achieve the variance swap payoff through a portfolio of options. As liquidity in the variance swap market has increased. Kospi CECE. but like vanilla options only really took off following the development of robust pricing models through replication arguments. Wig20. HSCEI.vegas (most liquid maturities) 0.S. but DAX and FTSE are also frequently traded.4 – 0. CAC S&P500. Table 3: Global variance swap liquidity Major indices on which variance is traded Developed Europe U. We discuss practical replication issues in more detail in Section 4. We then look at where variance swap levels have traded historically. This is partly because the construction of the replicating portfolio includes a relatively large contribution from the deep outofthemoney options and the reduced liquidity in these options gives rise to some discrepancy in the overall pricing of variance swaps. Variance swaps are tradable on a range of indices across developed markets and increasingly also on emerging markets. FTSE. In Europe. Bovespa Typical index bid offers . We leave the detailed theoretical pricing argument until later (see Part 4). Top40.1: Market development Variance swap contracts were first mentioned in the 1990’s. but discuss briefly the relationship between variance levels and traded option volatilities and consider some pricing rules of thumb.
These volatility indices are widely used as benchmark measures of equity market risk. S&P500) are increasingly quoted with a single bid/offer spread. See p94 for further details. including prices for forwards and spreads. meaning that they can be efficiently hedged with exchangetraded options of the same maturity. Traders in Europe and Japan estimate that approximately 1030% of the vega traded in the market is done through variance swaps. See Section 3. commodities and credit – albeit with less liquidity. 25 . although indices and more liquid stocks have variance swaps trading out to 3 or even 5 years and beyond. Such products may be leveraged and/or offer capital guarantees.000 vega notional for singlestock underlyings. In theory there is no reason why variance swaps should not gain liquidity in these asset classes. with a variety of structured products offering access to these indices. As dedicated ‘volatilityarbitrage’ funds have grown. In addition. or only slightly higher than. These developments make forward variance and index spreads much more attractive for investors who may previously have found the transaction costs prohibitive. and conditional variance prices.3 and JPMorgan’s Yield Alpha Strategy Guide for further details. We expect spreads to come in further in future as liquidity continues to improve. Forward variance and index variance spreads (e. without the onerous human costs (and transaction costs) of deltahedging a portfolio of options.000 of vega for index underlyings and € 50. the marketmaking community can offset variance swap exposure with the advantage of economy of scale. encouraged in part by the attractive returns available from shorting variance in 20032004 as equity volatility fell dramatically. interpolating to get 1month maturity. Variance swap liquidity has so far been concentrated on the equity markets. estimating total traded volumes is problematic. Of course. More recently. the investor base has widened further with the advent of structured products which embed variance swap returns. depending on investor demand and providing that the underlying option markets are liquid enough. and are calculated by the exchanges from listed option prices.g. The advent of the variance swap market has allowed the investor community to trade variance directly. often comparable to. VSTOXX and VDAX indices represent the theoretical prices of 1month variance swaps on the S&P500.000 – 200. Euro Stoxx 50 vs. For example. JPMorgan’s recently launched Yield Alpha and Variance Alpha indices incorporate variance swap exposure.S. Maturities generally coincide with the quarterly options expiry dates. As with any OTC market. With volatilities often much higher and less stable than in developed markets. interest rates.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market Bid/offer spreads have come in significantly over recent years and in Europe they are now typically in the region of 0. in part fuelled by the need for investment banks to recycle volatility and correlation exposures that are embedded in structured products. meaning the second component of the spread can effectively be traded at midmarket levels. selling stock correlation using variance swaps (socalled ‘variance dispersion’ trades) has also been a popular trade in the hedge fund community. with recent estimates for daily traded volumes on indices in the region of $35 million vega (source: Risk Magazine). emerging market variance can offer interesting opportunities.5 vegas range. even though they are only shortdated measures and are not directly tradable. Typical sizes of variance swap trades are around € 100.5 vegas for indices and 1 – 2 vegas for singlestocks – although the latter especially vary according to liquidity factors. with a slightly lower proportion in the U. The most liquid variance swap maturities are generally from 3 months to around 2 years. making them attractive to a broader range of institutions including private banks and sophisticated private investors. Good liquidity is also seen in the front 3 months for shortdated index variance. Euro Stoxx and DAX indices respectively. Spreads for indices in the US and Japan are similar with single stock bid/offers in the 2 – 2. the spread on a single leg. JPMorgan publishes its variance swap runs. on Bloomberg JPVS.or 10fold increase on volumes three years ago. However variance swaps do trade on nonequity underlyings – for example FX. variance swaps have allowed them to directly express their views. The VIX. Spreads are naturally higher in emerging markets although these too are becoming much more liquid. This represents an estimated 5.
with 3year implied variance remaining in the 1535 range.2: Historical prices Figure 15 shows Euro Stoxx 50 implied variance levels between 2000 and 2006 for maturities of 1month. The VIX gives a proxy for rolling 1month maturity variance swap levels on the S&P 500 (see p94). Figure 15 : Euro Stoxx implied variance levels since 2000 Implied variance Figure 16 : VIX (shortdated S&P500 variance) since 1990 VIX level realised volatility 60% 50% 40% 30% 20% 10% Sep00 Source : JPMorgan 50 1month implied v ariance 6month implied v ariance 50% 40% 30% 20% VIX Index SPX 3m realised v olatility 40 30 20 10 0 3y ear implied v ariance 10% 0% Sep01 Sep02 Sep03 Sep04 Sep05 Sep06 May90 May93 May96 May99 May02 May05 Source : JPMorgan Singlestock variance swaps generally trade at higher levels than index variance. Longer maturity variance swap levels vary less. and in particular reacting less to spikes. 6months and 3years. See Table 4 below for some examples of index and singlestock variance prices. Note also that variance swaps tend to trade somewhat above levels of realised volatility – in part reflecting the volatility risk premium. Below we discuss the two principal reasons why variance swaps tend to trade above realised volatility. with implied variance priced in the 2030 range. 26 . for example ahead of an M&A event or following a profit warning or unexpected earnings announcements. is perhaps less likely to cause the same level of elevated volatility over the next few years. interrupted by a large spike in September 2001. and within this period different volatility regimes are clearly discernible: • • • A ‘medium’ volatility regime up until mid 2002. Single stock variance can increase substantially in times of high company uncertainty. Figure 16 shows VIX levels since 1990. A ‘low’ volatility regime since mid 2004 with shortdated variance below 20 (excepting the May 2006 spike). A sudden unexpected event likely to dramatically increase shortterm volatility dramatically. with investor risk aversion and hedging programmes in the market supporting demand for index options leading to a persistent overpricing of (index) variance. Variance swap prices can be observed to follow high and low regimes in a similar manner to realised variance (Figure 5). Again variance swap levels can be seen to be well correlated with realised volatility both in the short term (including spikes) and in the long run. due to the fact that index volatility will be less that the average of its constituents because of diversification effects. with 1month variance experiencing highs above 60 and lows of almost 10 over the six year period.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market 2. following the underlying pattern of volatility regimes. Shortdated variance swap prices tend to be more variable than those for longer maturities. A ‘high’ volatility regime from mid 2002 – mid 2003 with shortdated variance priced consistently above 40.
and then variance swap levels trading above ATM options implied volatility due to the additional premium required to pay for the convexity. for example selling index variance swaps or traditional index covered call strategies.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market In general. and to some extent value. The value of this convex payout depends on how much future realised volatility is likely to deviate from the variance swap strike. Variance swaps are convex in volatility and the variance swap buyer should fairly pay for this convexity. the volatility risk premium can be viewed as a potential source of alpha. We begin by considering the relationship between ATM options prices and variance swap levels for various underlyings. the contributions of these two factors – convexity and volatility risk premium – presented above. at least for S&P500 variance. for two distinct reasons. with the discussion centring around attempts to quantify. that ‘volatility of volatility’ also tends to be overpriced: perhaps of ‘volatility of volatility risk premium’? 27 . We go on to explicitly consider the drivers of variance swap prices: principally realised volatility/variance and the convexity and riskpremium discussed above as the main reasons for variance swaps to trade above these levels. Thus a simple framework has (ATM) vanilla option implied volatilities trading above prevailing levels of realised volatility due to the volatility risk premium. Of these two factors. Nevertheless. The difference comes down to value of the convex payout which is represented by the convexity of the volatility skew. which is particularly noticeable at the index level. 2. 1. and is manifest in the premium of implied ATM volatility to realised volatility. Indeed. with large potential moves in volatility increasing the value of the variance swap. Note that this kind of increase in risk aversion also increases skew convexity. the second applies both to variance swaps and vanilla options.6 we explicitly discuss and attempt to measure the volatility risk premium. the convexity premium is worth perhaps 12 vegas with the volatility risk premium around 23 vegas. The volatility risk premium mentioned above. we find like volatility. and the reason why they trade above ATM implied volatility. exploited by strategies which systematically short volatility. This is especially true at short maturities where market information priced into the variance swap market. In Section 2. The convexity premium should thus be linked to the ‘volatility of volatility’ and by comparing the spread of variance over ATM implied volatility with realised volatility of volatility we find that this is the case. but not represented in realised volatility can give variance swaps some real forward looking attributes. index variance swap strikes tend to trade above the level of prevailing realised volatility. Risk aversion (driving the risk premium) can perhaps be measured in the market as the excess demand for put options or its natural effect – increasing put skews. whereas a short position can be likened to selling insurance (similar to selling credit protection) with a steady income punctuated with occasional large drawdowns. Typically. leading to a higher implied volatility of volatility and more value being assigned to the convex variance swap payout. demonstrating that they are well correlated (both across underlyings and over time). a simple regression demonstrates that variance swaps tend to give a better estimate of future volatility (after adjusting for the bias) than that given by prevailing levels of realised. In Section 2.8 we turn our attention to convexity. This leads into a discussion of rules of thumb for pricing variance swaps based on vanilla options prices – using assumptions about the skew to produce an estimate for the variance swap price which in theory is priced using the entire skew curve. The following sections consider the drivers of variance swap prices. whereas the first applies only to variance swaps. This risk premium can partly be explained by the asymmetry of returns of volatility (and variance) products: longs occasionally make large gains but mostly suffer small losses. at least for shortdated index variance. the other cause of variance swaps trading above realised volatility. meaning that equity variance swaps trade above implied ATM volatility. representing the spread of ATM implied volatility over realised volatility. Although convexity and the risk premium cause variance swaps to trade above prevailing levels of realised volatility.
This should not be surprising.9% Variance swap strike 15.8% Dec07 ATM implied volatility 16. Note also that at 6month maturity the index variance appears somewhat rich in comparison to its ATM volatility. 27%).9% Variance swap strike 14.1% 21.1% Variance swap strike 19.08 30.8% 14.3% 15.6% 12.7% Variance swap strike 20. there is variation in the spread.0% 27.3% 18. the greater upside convexity of the singlestocks becomes more important and the index variance appears to be priced in line with its ATM volatility in comparison to the single stocks.3% 17.4% 16.3% 28. Ahold doesn’t pay a dividend) the skew surface of Ahold is relatively flat.0% 22. Figure 17 : 6m variance swaps trade above 6m ATM implied volatility 6M implied variance Figure 18 : . Table 4 shows ATM volatility and variance swap levels for a number of European stocks and indices. independent of future market level.7% 32.8% 19. However.2% 14.5% 14.44 8.7% 15. as the effect of the particular shape of the skew becomes more important.9% 23. Whilst this is partly due to the different forwards (ING has a dividend yield of 3.1% 15. whereas that of ING is more convex.7% 12.8% 14.6% Mar07 ATM implied volatility 14.2% 12. Note that these variance swap levels are typically about 12 vegas above ATM volatility. ATM option volatility reflects the market estimate of future volatility realised around the current level.0% 16.247.0% 28.8% 15.5% 16.4% 32. Table 4: ATM volatility and variance levels (as of COB 24 Oct 2006) Ref Dec06 ATM implied volatility 12.5% Dec09 ATM implied volatility 18.1% 20.1% 16.2% 34..5 8.7% 28.4% 19.9% 11. whereas variance swaps represent the market estimate of variance.6% 14.6% 18. Whilst all variance swaps price above their corresponding ATM volatility.8% 14. since both can be interpreted as market estimates of future volatility.5 424.6%.4% 29.7% 27.4% 24. in part due to the differing skews.9% 16.1% 35.182.0 6.2% 21.662.1% 33.5% 32.3% 20.3% 16.1% 16.5% 15. This is caused by the index having higher put skews than the singlestocks.2% 37. This spread of variance above ATM volatility becomes larger and more unpredictable at longer maturities.2% 26.7% 21.3% 25.2% 24.0% Dec11 ATM implied volatility 19.2% 34.2% Euro Stoxx 50 FTSE SMI DAX Nestle BP Axa Ahold Rio Tinto Source : JPMorgan 4. at 3year maturities.6 6.2% Variance swap strike 17.5% 27.50 For example note that at 3year maturity (Figure 18) the variance strikes of Ahold and ING are similar (around 27) even though their ATM implied volatilities differ by more (24% vs.5% 22.5% 34. We have used indicative midmarket levels as of 24th October 2006.9% 24. Figure 17 and Figure 18 show the relationship between ATM volatility and variance swap strikes for the members of the Euro Stoxx 50.0 6.3% 23.3% 11.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market 2.8% 17.3% 13.3% 34.4% 12.9% 27.1% 38.4% 18.0% 28.3: Variance swaps and option volatilities Variance swap strikes are well correlated with standard BlackScholes implied volatility derived from options prices. As explained below this means that variance swaps tend to trade somewhat above levels of ATM volatility.1% 17.2% 29. However.9% 17.2% 26. due to much steeper put skews.014.8% 18..9% 23.9% 13.9% 13.26 28.6% 17.0% 18.3% 21.4% 17.3% 27. and longer maturities trade at a higher premium 3Y Implied variance 30% 28% 26% 24% 22% 20% 18% 16% 14% 14% Euro Stox x Members Euro Stox x Index CGE FP AH NA 32% 28% Euro Stox x Members Euro Stox x Index CGE FP ALBK ID AH NA INGA NA line y = x BN FP 24% 20% TEF SQ IBE SQ BN FP line y = x ENEL IM 16% 18% 20% 22% 24% 26% 28% 16% 16% 18% 20% 22% 24% 26% 28% 30% 6M ATM implied volatility 3Y ATM implied volatility Source : JPMorgan Source : JPMorgan 28 .8% 13.
• Ultimately both of these explanations come down to the fact that a long variance swap is also long volatility of volatility. In practice. 29 . Variance swap prices arise from a kind of weighted average of volatilities across the skew surface. average volatilities will usually be above ATM volatility. and put skews are more significant than call skews. See p82 for details. due to the convex nature of the payoff discussed previously. • Variance swaps trade above ATM volatility because you pay extra for the convexity of the variance swap: the gain from an increase in volatility is more than the corresponding loss from a decrease in volatility (Figure 9. See the next section for more details. which is particularly useful for estimating variance swaps where a limited number of option volatilities are available. The theoretical price of the variance swap is calculated from the prices of a replicating portfolio of options. which in effect price in this volatility of volatility (see Part 4 for details). This convex payoff is paid for through a replicating portfolio of options. Why is this? There are two ways of looking at this which both turn out to be the same explanation. across the whole range of strikes. This assumption lies behind Derman’s approximation to the variance swap level. and convexity (in terms of the overall shape of the skew surface) become more important factors for pricing at longer dates. the contribution of the skew component means that variance swap strikes tend to trade at similar levels to OTM puts – usually around 95100 strike for maturities of around 13 months (Figure 19). with the weighting relating to the probability of the corresponding option expiring in the money. variance swaps invariably trade above the level of ATM implied volatility of the corresponding maturity. the price can be thought of as a function of ATM volatility level and the slope of skew. Figure 10). The skew. Figure 19 : The VSTOXX (theoretical interpolated mid variance price) generally trades just below the EuroStoxx 1M 95% put volatility Implied volatility/variance 28% Euro Stox x 1M ATM implied v olatility Figure 20 : 1Y EuroStoxx variance swaps trade closer to 90strike put volatility Implied volatility/variance Euro Stox x 1Y ATM implied v olatility 24% 20% 16% 12% 8% Jul04 Source : JPMorgan VSTOXX index Euro Stox x 1M 95strike implied v olatility 24% 20% 16% 12% Euro Stox x 1Y Variance sw ap strike (estimated) Euro Stox x 1Y 90strike implied v olatility Jan05 Jul05 Jan06 Jul06 Jul04 Source : JPMorgan Jan05 Jul05 Jan06 Jul06 Variance swap pricing is explored in much greater detail in the Part 4. making the variance swap more expensive. where the probability of reaching more OTM strikes increases (Figure 18). as noted above. This makes explicit the precise impact of skew and convexity on variance levels. p82). This enables the variance swap price to be expressed as a weighted average of option volatilities across the skew surface.g. This does not come free. In the presence of skew and skew convexity. where we outline how the theoretical variance swap price is arrived at by considering a replication argument using a static portfolio of vanilla options. so the variance swap strike can be thought of as a weighted average of vanilla option volatilities (see e.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market Although they move in close step with changes in ATM volatility. and 9095 strikes for maturities of about a year (Figure 20). For short dates where the skew is relatively linear around ATM volatility.
This presupposes a linear (put) skew. We discuss their effectiveness and applicability and give examples of their use. However. but can underestimate the variance swap mid. this approximation tends to work best for shortdated index variance (up to about 1year). Flat skew In the hypothetical case where the skew surface is flat meaning that all strikes trade at identical implied volatilities – the variance swap level will be the (constant) implied volatility level. As maturity increases and the OTM strikes have a greater effect on the variance swap price. Box 10: Derman’s approximation K VAR ≈ σ ATMF 1 + 3T × skew 2 where KVAR is the strike of the variance swap σATMF is the implied volatility of the forward strike T is the maturity in years skew is generally taken to be the slope of 90/100 skew In practice. loglinear and quadratic skews curves.4: Pricing rules of thumb Whilst it is necessary to have prices available for the entire strip of (OTM) options in order to calculate the true theoretical price of a variance swap. but the inability of the approximation to account for the skew convexity can make it less accurate.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market 2. In this section we introduce some commonly used approximations obtained by assuming linear. this is not a particularly useful observation because a flat skew implies zero volatility of volatility (i. Derman.e. Indicative Euro Stoxx 6month variance swap prices 50% 40% 30% 20% 10% 0% Source : JPMorgan 1M Variance Sw ap 3M Variance Sw ap 1Y Variance Sw ap 2Y Variance Sw ap 24% 22% 20% 18% 16% 14% 12% v arsw ap estimate v arsw ap bid v arsw ap ask 2% 4% 6% 8% 10% 90/100 skew Jul05 Source : JPMorgan Oct05 Jan06 Apr06 Jul06 30 . Where the estimate is accurate it tends to be closer to the traded bid than the offer (Figure 22). the slope of the skew and the maturity of the swap. even at shorter dates. Derman’s approximation (ATM volatility = 20%) Figure 22: Derman’s approximation has been relatively accurate for index variance. 1999). effectively assuming the call skew is flat. at least close to the money. constant volatility) in which case volatility cannot change and the p/l of a variance swap could only be zero. Similarly. Linear skew If the skew is assumed to be linear. where the convexity can be much more significant. and calculates the variance swap strike as a function of three variables: ATM (forward) volatility. Figure 21: Variance swap prices are more sensitive to skew at longer maturities Variance swap price. the contribution of the skew becomes more important (Figure 21). at least for strikes relatively close to the money. where put skews are often relatively linear and call skews relatively flat. for singlestocks. then Derman’s approximation can be used (See “More than you ever wanted to know about volatility swaps”. reasonable approximations for variance swap prices can be made under certain assumptions about the skew. the approximation can be less successful.
0% 0.7% 21.3% 17.4% 0. Maturity.1% 2.2% 29. T=0.4% 3.9% 13.3% 27.3% 25.9% 17.24) = 23.9% ATMF volatility 12. the approximation is less successful.1% 20.2% 26. which is 2.2% 12.8% 33.38.7% 15.9% 13.0% 27.2% 1.3% 14.8% 27.22%)/10% = 0.0% 28. Data from 23 October 2006.3% Variance estimate (Derman) 15.0% SX5E FTSE SMI DAX NESN VX BP/ LN CS FP AH NA RIO LN SX5E FTSE SMI DAX NESN VX BP/ LN CS FP AH NA RIO LN Source : JPMorgan.5% 16.5% 23.3% 28.9% 1. skew and variance ATM volatility 18.3% 18.9% 33.2% ATMF volatility 16.8% 15.0% 18.4% 1.7% 20.3% 15.7% 15. skew and variance ATM volatility 12.8% 13.5strike volatility) is 21%.6% Variance estimate (Derman) 19. Source : JPMorgan.9% 11.1% ATMF volatility 18.5.1% 1.2% 21.3% 13.0% 16.2% 26.1% 18.2% 24.3% 20.5% 14.3% 20.8% 14.5. Data from 23 October 2006.8% 14.1% 15. Table 5: Dec06 (2month) volatility.3% 32.9% Variance estimate (Derman) 13. variance swap levels and their approximations for a number of European stocks and indices as of 23 October 2006. with the 6month forward price at 102.2% 2.1% 1.2% 11.9% 1.2% 14.5% 2.7% 24.6% 26.1% 17.5strike: 21% Slope of the skew = (26% .9% 16.4% 12.24 Therefore KVAR = 21 x sqrt(1.8% 17.0% 28.1% 17.0% 21.3% 21.8% 11.3% 27.8% 14.5% Table 8: Dec09 (c.6% 18.6% 3.5% 15.9% 5.7% 27.7% 3.4% 19.5% 18.9% 24.2% Actual Variance Swap 19. skew and variance ATM volatility 14. Table 7: Dec07 (c.9% Variance estimate (Derman) 17.2% 37.2% 26.6% Skew (90/100) 1.0% 13.6% 1.9% ATMF volatility 14.0% 12.3% 15.8% Skew (90/100) 3.2% 34.3% 16.4% 19. 3year) volatility.1% 5.5% 13.7% 0. skew and variance ATM volatility 16.8% 5.1% 1.7% 15.2% 23.9% Skew (90/100) 2. Derman’s approximation is relatively accurate for indices in the shorter maturities.5% 13.8% 19.8% 0.1% 1.7% 32.4% 12.6% Table 6: Mar07 volatility (5month). 1year) volatility. 6month volatility surface: 90strike: 26% 100strike: 22% 102.8% SX5E FTSE SMI DAX NESN VX BP/ LN CS FP AH NA RIO LN SX5E FTSE SMI DAX NESN VX BP/ LN CS FP AH NA RIO LN Source : JPMorgan. Data from 23 October 2006.6% 14.2% 21.8% 13. 100strike volatility is 22% and the ATM forward volatility (102.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market Example 10: Derman’s approximation An index is trading at 100.0% 26.3% 23.4% 0.2% 14.1% 33.9% 26.6% 16. For singlestocks and longerdated index estimates.8% 1.9% 3.9% 27.7% 1.4 vegas above the ATM forward volatility Table 5Table 8 show volatilities.5% 33.1% 2.2% 34.4% 17.5% 34.2% 33. Data from 23 October 2006.9% 32.3% 34.6% 2.7% 12.7% 15.6% 12.5% 13. 31 . The 90strike volatility is 26%.8% 13.2% 18.4.4% 32.5% 14.7% 16.1% 16.1% Actual Variance Swap 14.9% 27.0% 25.5% 16.3% 1.3% 3.1% Actual Variance Swap 15.8% 17.3% 14.8% 16.4% 12.4% 14.7% 12.1% 16.3% 11.7% 26.9% 23. 1 + 3T x skew2 = 1.5% 0.2% 25.6% 21.4% Actual Variance Swap 17.6% 17.8% Skew (90/100) 5. Source : JPMorgan.5% 27.7% 32. Note that consistent with the discussion above.4% 22.1% 35.3% 1.1% 15.9% 23.1% 15.
100strike volatility is 22% and the ATM forward volatility (102. or higher order polynomial to the skew surface parameterised in terms of d2. i. α. in ‘d2space’ the linear component of the skew. 90strike volatility is 26%.5strike volatility) is 21%. JPMorgan. Gatheral.9) Then the loglinear approximation gives KVAR = 23. 2006 for further details 32 .5 ln(0.5strike: 21% β= − 4% ≈ 0. The loglinear model described below is one possibility.38 . with 6month forward at 102. This formula characterises the skew curve in terms of the BlackScholes d2 parameter. which measures the ‘moneyness’ of the associated OTM option. February 2005 for further details Example 11: Loglinear skew An underlying is trading at 100. it is then possible to directly calculate a theoretical variance swap price from this parameterisation. 6month volatility surface: 90strike: 26% 100strike: 22% 102.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market Loglinear skew In reality the volatility skew is not linear across all option strikes – and more accurate variance swap approximations can be used. T = 0. has no effect on the variance swap price: only on the base level of volatility σ0. and the convexity β affect the swap strike See p82 and “The Volatility Surface”.55%. in this case slightly higher than the value given by Derman’s approximation in Example 10 Gatheral’s formula A further approximation to the variance swap price can be obtained from Gatheral’s formula which expresses the variance swap strike as an integral (or summation) of the implied volatilities across the entire range of strikes.e. if the skew curve is quadratic in the variable z = d2 : σ theoretical variance swap strike is K VAR = σ T + β T 2 0 2 (z ) = σ 02 + αz + βz 2 then it can be shown that the . Implied volatilities of traded options can be used to parameterise the skew curve in this form. Box 11: Loglinear skew Suppose that the skew curve is loglinear of the form: σ (K ) = σ ATMF − β ln(K F ) . For example. by converting the strikes into their d2 equivalents. This leads to potentially powerful methods of variance swap approximation: by fitting a quadratic.5. where F is the forward price and K is the strike and β represents the slope of the skew Then: 2 3 K VAR ≈ σ ATMF + βσ ATMF + β2 4 (12σ 2 ATMF 4 T + 5σ ATMF T 2 ) See “Just what you need to know about variance swaps”.
80 0. with volatility tending to increase if the market sells off. realised volatility is not the only driver. as well as volatilities. Since variance swap prices in effect carry a significant skew component.83. we do observe that they tend to trade above realised volatility.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market 2. which is the subject of the next section.00 Rsquared of implied v ariance v s 1M realised v olatility beta of implied v ariance v s 1M realised v olatility 10% 20% 30% 40% 50% 60% 70% 0 1 2 3 4 variance swap maturity 5 trailing 1M realised volatility Source : JPMorgan (Euro Stoxx 50 data since August 2000) Source : JPMorgan (Euro Stoxx 50 data since August 2000) Clearly riskaversion is also an important driver.91 (R2 = 0. Thus the drivers of variance swap levels (at least theoretical levels) are essentially the same drivers as for options volatilities and associated skews.057 R = 0. The skew (in fact more correctly the skew convexity) directly influences variance swap levels. Indeed under certain conditions the effects of variance swap hedging can potentially influence the underlying market. usually within the arbitrage corridor defined by bid/offer spreads. variance swaps included. However. demand for protection in the form of put options will clearly drive up put skews. Whatever the drivers of variance swap levels. is historical realised volatility (e. the remainder is the socalled volatility risk premium. 33 . The correlation between strikes of 1month variance swaps and the volatility realised over the previous month is 0. Ultimately volatility itself is directional.20 0. at longer dates structured product flow and the longerterm macroeconomic outlook also become important considerations. In this sense. it should be added that variance swap flows can have a feedback effect into the rest of the market. variance swaps can be driven in the short term by the (inverse of) market level (Figure 25.00 0.5: What drives variance swap levels? Since variance swap levels move in close step with options implied volatility. dictating supply and demand factors for volatility products. the correlation (and beta) between variance swap prices and realised volatility decreases (Figure 24). with its effect becoming more pronounced at longer maturities (Figure 21).g. most notably at the index level. evidenced by the fact that shortdated variance is superior to realised volatility as a predictor of future volatility (see p37). See p90 for a more detailed explanation. as well as by the shape of the skew curve.829 2 1. With variance swaps currently accounting for perhaps 1030% of the traded vega demand.848x + 0. In common with options volatilities. Finally.40 0. at least over short time frames. they are in large part driven by the same factors which drive these volatilities. see Figure 16). as maturity increases. Figure 26). Even at short maturities. one of the most important drivers of variance swap levels. Figure 23). at least at shorter maturities.60 0. supply and demand issues particular to the variance swap market can cause levels to differ slightly from their theoretical values. Figure 23: Shortdated implied variance is driven by shortdated realised volatility 1M Implied/variance Figure 24 : Shortdated variance is most sensitive to realised volatility Correlation / beta 70% 60% 50% 40% 30% 20% 10% 0% 0% y = 0. However. This is most clearly seen during market shocks when volatility tends to spike as the underlying sells off. variance swap hedging by market makers can influence implied volatilities and particularly skews. Whilst this variance premium is partly due to the convexity of the variance swap payout. In addition. and thus have a feedthrough into variance swap prices.
the potential rewards can be high if timed correctly. In this sense long volatility is like buying insurance: paying a relatively small premium for a potentially large payout if things go wrong.Sep 2006 Dec05 Mar06 Jun06 Sep06 2500 3000 3500 Euro Stoxx 50 4000 2. Conversely selling volatility can be likened to selling insurance (or credit protection): potentially very 34 .. with the relationship changing as volatility switches regimes Euro Stoxx 6m implied variance 24% 22% 20% 18% 16% 14% 12% Sep05 Source : JPMorgan Euro Stox x 6m implied v ariance (lhs) Euro Stox x index lev el. Conversely. profitable. since returns from short volatility positions are limited to the premium received in the case of options.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market Figure 25 : Over short time periods.Apr 2003 May 2003 . of the premium on most occasions. often in the form of protection. Thus although expected returns from long volatility positions are.Apr 2006 Apr 2006 . at least on index underlyings. Implied variance index level (inverted) Figure 26 : .. This bias is referred to as the volatility risk premium. returns from long volatility positions can be potentially unlimited. or all. In this section we investigate why this bias exists. On the other hand. on average. are prepared to pay slightly higher than the theoretical fair value for exposure to long volatility.. or half the strike multiplied by the veganotional in the case of short variance swaps..Dec 2004 Jan 2005 . Figure 27 : Average returns from long variance are negative… Variance swap p/l (vegas) Figure 28 : …but the tail is heavily skewed to the upside Frequency of long variance swap p/l 40% 30% 20% 10% 0% 10% 20% Aug00 Source : JPMorgan long Euro Stox x 1month v ariance sw ap p/l long Euro Stox x 1y ear v ariance sw ap p/l 30% 25% 20% 15% 10% 5% 0% 20% 16% 12% 12% 16% 20% 24% 28% 32% 36% 0% 4% 8% 4% 8% Aug01 Aug02 Aug03 Aug04 Aug05 Aug06 Source : JPMorgan (p/l from 1m Euro Stoxx variance since Aug 2000) Investors in the real world are not riskneutral and. what it represents and how it changes over time. are biased to make a loss (Figure 27). with only part of the spread attributable to the convexity of the variance swap payout. However variance swap levels tend to trade consistently above comparable realised volatility (Figure 23).6: The volatility risk premium The variance swap strike represents the market’s fair price for exposure to realised variance over a specified period of time. in general. negative. variance swap levels are inversely correlated with market level . but expecting to forfeit some. on average. short index volatility positions are. This is in some sense understandable. Backtesting shows that long volatility positions. inv erted (rhs) 3000 3200 3400 3600 3800 4000 50% 45% 40% 35% 30% 25% 20% 15% 10% 2000 Source : JPMorgan Jan 2002 .
From the point of view of a short volatility position. Given these caveats. but in the long run likely to make lucrative returns. These volatility indices represent rolling 1month mid implied variance levels on their respective indices but tend to be slightly higher (around 0.5 – 2 vegas) than real variance swap prices. part of the risk premium is attributable to the convexity of the variance swap payout. real variance swap data is limited. Finally. For example. Secondly. We can attempt to quantify the volatility risk premium by computing the spread of implied variance over contemporaneous realised.g. and from 5. “Calling All Overwriters” JPMorgan.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market risky. The above discussion illustrates that the returns from long variance positions. not €1M. though indicative variance swap levels back in the 1990’s can be inferred from levels of the corresponding volatility indices (VIX for the S&P. around 12% on the Euro Stoxx). We have estimated the convexity premium for 1month variance swaps at somewhere around 1% (1 vega) under normal market conditions (see p 38). we characterise the risk premium as the 1year rolling average of the spread between (estimated) 1month variance swap strikes and 1month realised volatility. have long upside tails (Figure 28). but can be exposed to very large losses if volatility spikes up significantly (see Section 3.5M Example 12 above shows why sellers often fail to drive variance swap prices below certain thresholds (e. Short positions usually make modest gains.4% over 2 years on the Euro Stoxx (Figure 29).5M profit from the long is €7. when volatility had subsided from its highs but riskaversion remained elevated. This strategy can work best as a rolling shortmaturity strategy. That is. although on average negative. Values seem to be relatively stable over time with the S&P risk premium varying between 2% and 9% since 1990. due to the favourable effect of convexity for the long. providing enough capital is available to absorb the occasional inevitable losses. Whilst we make no attempt here to remove this from this characterisation of the risk premium. or to selling credit protection. The recent peak in risk premiums for both indices occurred in 2004.g.8% to 2. the observed premium of implied over realised volatility is often referred to as the carry of the short position and can be a good source for generating alpha. an average spread of implied to realised volatility needs to be taken over a suitable time period to smooth out some of the noise.g. see Example 3) profit from the long is €1. especially in the form of strategies continuously selling variance through variance swaps (see p47). Note that these measures include a convexity premium compensating for the negative convexity of a short variance swap as mentioned above. This is complicated by a number of factors. with an average value of 5%. Firstly.5M (this is the maximum loss. even when shortdated realised volatility drops to very low levels. with the occasional (large) losses from being short volatility diversified away over time. which usually generates positive carry but has a small chance probability of large losses. This is somewhat analogous to the vanilla options equivalents: straddle overlays or covered call overwrite strategies (See e. Euro Stoxx since 1999 is similar with an average riskpremium of about 4%. August 2002). Realised volatility Realised volatility Realised volatility 0%: 20%: 40%: loss for the long is €0. Example 12: The insurancelike nature of long volatility positions A variance swap is struck at 10% for €100K of vega notional. Figure 28). since volatility can spike from time to time. the yintercept on Figure 23 suggests an absolute floor of 6% for implied variance in the hypothetical case that realised volatility drops to zero. but can occasionally make large profits. VSTOXX for the Euro Stoxx – see page 94 for more details). returns are not normally distributed: long positions tend to lose a little and often.8% on the S&P (Figure 30). 35 .3). any backtest of variance swap payoffs will implicitly take this into account (e. Since then the risk premium has fallen from about 7% to 2.
1month implied variance will likely be priced well above 10%. We find similar results (implied variance a good predictor. with a slight edge over trailing realised) using VIX. • • 36 .Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market Figure 29 : Variance riskpremium on the Euro Stoxx 50 Volatility/ points Figure 30 : Variance riskpremium on the S&P500 Volatility points 60% 50% 40% 30% 20% 10% 0% 10% 20% 1M realised v olatility 1M implied realised v ariance v ariance risk premium 40% 30% 20% 10% 0% 10% Jan90 Source : JPMorgan 1M realised v olatility 1M implied realised v ariance v ariance risk premium Jan99 Jan00 Jan01 Jan02 Jan03 Jan04 Jan05 Jan06 Source : JPMorgan Jan93 Jan96 Jan99 Jan02 Jan05 2.61 between 1month realised volatility and that realised volatility 1month later (Figure 31). This compares with an R2 of 0. In this sense implied variance can be truly forward looking. Risk aversion in the market: If investors are nervous and willing to pay extra for protection. at least at shortdates. Thus an increase in implied variance could precede an increase in realised volatility if the market sells off.71 (Figure 32). using data back to the inception of the relevant volatility indices. Knowledge of forthcoming events: volatilities (especially shortdated) will tend to be marked up ahead of earnings seasons or other events likely to increase volatility. This is complicated somewhat by supply and demand factors. Nevertheless. Euro Stoxx and DAX respectively. they are perhaps also prepared to liquidate their positions quickly in a downturn and increase market volatility. This suggests that implied variance better predicts changes in future realised volatility in comparison with previous realised volatility. VSTOXX and VDAX levels as predictors for future realised volatility on the S&P. Using data since August 2000. especially at longer maturities where prices can be driven by structured product flow. the R2 between 1month Euro Stoxx variance swap strikes and subsequent 1month realised volatility is 0. Given historical data. At shorter maturities riskaversion and protection buying can play a role in biasing the variance swap strike to be above a ‘fair’ expectation of future realised variance as explained in the section above. If volatility over the last year has been say 30%. Given that implied variance appears to have better predictive power than prevailing realised volatility. although results using variance itself are similar. variance is itself a biased estimator. a simple model of future realised volatility as 0. as it generally overestimates realised volatility. Note that here we are correlating the square roots of implied and realised variance. what extra information could the variance swap and option markets be pricing in? • Mean reversion of volatility: the implied variance can take account of the current medium term “average” of volatility and the fact that volatility tends to be meanreverting within a regime.7: Variance swaps as predictor of future volatility In theory the variance swap strike represents the market prediction of realised variance over the term of the swap.9x variance strike provides the best fit with the data (Figure 32). not least because units are more tractable. but over the last month 10%. whereas past realised volatility is not. As discussed in the previous section. implied variance tends to do a relatively good job of forecasting future realised variance.
10 0.50 0.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market Figure 31 : 1month realised volatility is relatively well correlated with realised volatility over the next month… subsequent realised 1month volatility Figure 32 : … but the correlation of 1month implied variance with subsequent realised variance is somewhat stronger subsequent realised 1month volatility 70% 60% 50% 40% 30% 20% 10% 0% 0% 10% 20% 30% 40% y = 0. variance swap prices really do seem to be adding additional information to that encapsulated in backward looking volatility measures.20 0.71 0.7087 2 1month realised volatility Source : JPMorgan (Euro Stoxx 50 data since Aug 2000) Instead of trying to predict future 1month realised volatility using volatility realised over the past month. Further in the future is harder to predict.9037x . subsequent realised v olatility Previous 1M realised variance R2=0. Although shortdated implied variance is a relatively good predictor of subsequent realised volatility. That is. Even choosing the weightings of these by performing an insample regression still does not give as good a correlation with future realised as using the implied variance.7843x + 0.6157 50% 60% 70% 2 70% 60% 50% 40% 30% 20% 10% 0% 0% 10% 20% 30% 40% 50% 60% 70% 1month implied variance Source : JPMorgan (Euro Stoxx 50 data since Aug 2000) y = 0. Figure 33 : Correlation between implied variance and past and future realised variance Figure 34 : Correlation of implied variance with subsequent realised of similar tenor decreases with maturity.0022 R = 0. subsequent realised v olatility correlation of realised v olatility v s.0. rsquared vs.0452 R = 0.00 0 3 correlation of implied v ariance v s.g.83 R2=0. Moreover.61 Future 1M realised variance 6 9 maturity (months) 12 Source : JPMorgan (Euro Stoxx 50 data since Aug 2000) Source : JPMorgan (Euro Stoxx 50 data since Aug 2000) 37 . the same is not true for longer maturities.70 0. 1month.80 0. 5day. future realised volatility Current 1M implied variance R2=0.60 0. 3month and 1year).40 0. we could use a combination of realised volatilities of different maturities (e. any edge of implied over realised volatility as a predictor for future realised effectively disappears for maturities longer than 3months (Figure 34).30 0.
The convex payout means that a variance swap must be worth more than a hypothetical ‘volatility swap’ struck at the same level.5% (Figure 35).7087 0% 10% 20% 30% 40% 50% 60% 70% 2 5% 10% 15% 20% 0% 1month implied variance Source : JPMorgan (Euro Stoxx 50 data since Aug 2000) change in volatility from strike level As demonstrated above. this standard error is around 6% (6 vegas).9037x .g. as represented by a convex skew. Equivalently.0.9x) the variance swap strike. struck at ATM implied volatility. This 6 vegas represents an ‘expected’ range of values for future volatility. How much is the 6% volatility of volatility worth? It depends on the strike of the variance swap. but for strikes in the region of 1520% the convexity is worth an extra 11. We can attempt to quantify the price paid for the convexity by computing the difference between the variance swap strike and the ATM volatility. Taking data from the last 6years. which can be thought of in the form of the extra value of the convex variance swap payout. shortdated variance swaps have some value as a predictor of future volatility. the standard deviation of the difference between implied variance and future realised variance is 6%. we can calculate the standard error of the future volatility as estimated by the variance swap strike (Figure 36). In this section we attempt to quantify the value of the convex variance swap payout in order to account for the observed premium of variance over ATM volatility. Figure 9). We can then compare the (linear) payoff from the volatility swap with the (convex) payoff of the variance swap to estimate the value of convexity. say a vega or two either way (e. and the variance swap should price this in by trading significantly above ATM volatility. This is essentially due to convexity. 38 . This tells us the average RMS error in the 1month variance strike as a predictor of realised volatility over the coming month. convexity can be thought of as the extra volatility paid for the OTM options needed to replicate the variance swap. Figure 35 : Measuring the value for variance swap convexity Extra vega p/l due to convexity (var swap – vol swap p/l) Figure 36 : 1month implied variance has some predictive power for future realised variance subsequent realised 1month volatility 14% 12% 10% 8% 6% 4% 2% 0% 0% Source : JPMorgan 70% strike = 15% strike = 20% strike = 25% strike = 30% 60% 50% 40% 30% 20% 10% y = 0. which corresponds fairly closely to what is observed in practice for shortdated Euro Stoxx variance.g. If volatility is unlikely to change much. As well as looking at the regression R2. implicitly assuming that we could trade a “volatility swap” linear in volatility. that is. then the convexity is likely to be valuable (e. In contrast if the likely range of volatility is high. How much more the variance swap is worth depends on how much volatility is likely to change (volatility of volatility) and hence how much effect variance swap convexity will have.0022 R = 0. then the convexity of the variance swap will be worth little and the variance swap should trade only just above ATM volatility. thus determining the magnitude of the convexity of the variance swap payoff.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market 2. Figure 10).8: Is variance swap convexity fairly priced? As previously discussed (see p29) variance swap strikes usually trade above corresponding atthemoney implied volatilities. The idea is that this ‘average error’ is giving information about the historical variability of the future realised volatility around the estimate provided by (0.
Euro Stoxx implied volatility of volatility has been more or less in line with the actual variability of volatility (variance appears fairly priced with respect to ATM volatility). the changes roughly reflecting ambient levels of realised volatility (Figure 39). Although. the difference between Euro Stoxx ATM volatility and variance swap strikes suggests an average implied ‘volatility of volatility’ of about 7% . we can back out a value for this ‘volatility of volatility’. we see that it has a minimum of just below 2% and a maximum around 10%. skew/convexity is too high) Figure 37 : Variance swaps trade at a premium to ATM volatility Implied volatility / variance Figure 38 : The spread between variance and ATM volatility reflects the implied volatility of volatility Spread / implied volatility of volatility 70% 60% 50% 40% 30% 20% 10% Aug00 Aug01 Aug02 Aug03 Variance sw ap strike ATM implied v olatility 30% 25% 20% 15% 10% 5% 0% Variance strike . but in the US. on a rolling 1year basis. like volatility. Notice also that over the last few years. but slightly lower range for the S&P (Figure 40). S&P implied volatility of volatility has consistently overestimated the actual variability of volatility (variance is overpriced with respect to ATM volatility. and the only difference between the price of this instrument and a standard variance swap of the same maturity is driven by the expected volatility of volatility.fairly close to the 6% average error of the variance swap in predicting volatility. it changes over time.ATM implied v olatility implied v olatility of v olatility Aug04 Aug05 Aug06 Aug00 Aug01 Aug02 Aug03 Aug04 Aug05 Aug06 Source : JPMorgan (Euro Stoxx 50 data since Aug 2000) Source : JPMorgan (Euro Stoxx 50 data since Aug 2000) Figure 39 : Euro Stoxx 50: implied and realised volatility of volatility Implied/realised volatility of volatility Figure 40 : S&P 500: implied and realised volatility of volatility Implied/realised volatility of volatility 20% 15% 10% 5% 0% Aug00 Source : JPMorgan SX5E Implied v olatility of v olatility (from Varw sap v s ATM spread) SX5E realised v olatility of v olatility (std error of 12m regression) 14% 12% 10% 8% 6% 4% 2% 0% S&P Implied v olatility of v olatility (from Varw sap v s ATM spread) S&P realised v olatility of v olatility (std error of 12m regression) Aug01 Aug02 Aug03 Aug04 Aug05 Aug06 Aug00 Source : JPMorgan Aug01 Aug02 Aug03 Aug04 Aug05 Aug06 39 . By assuming that a hypothetical ‘volatility swap’ would be struck at ATM volatility. By estimating this standard error. Over the last six years. We observe a similar pattern.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market We can also calculate an ‘implied’ variability of volatility by considering the spread between short dated ATM volatility and variance strikes. this figure of 6% volatility of volatility for the Euro Stoxx is a long term average (Aug 2000 – Sep 2006).
Figure 43 : A capped variance swap is equivalent to a covered call on variance p/l (vegas) Table 9: P/L from a long variance swap @ 20 strike Realised volatility Realised minus implied volatility Uncapped variance swap p/l (vegas) Capped variance swap p/l (vegas) 100% 80% 60% 40% 20% 0% 20% 0% capped variance swap p/l (cap at 2. if we believe that the volatility of volatility is correlated with volatility (Figure 41. a variance swap with a cap will always be worth less than an uncapped variance swap of the same strike. Therefore capped variance swaps must trade with strikes slightly below their uncapped equivalents – the difference. However in the case of variance swaps they are also of use for pricing variance swap caps. annualised) 60% 50% 40% 30% 20% 10% 0% 0% y = 1. Figure 42).69x .5% 30.0% 150.5x strike) uncapped variance swap p/l (strike = 20) 20% 40% 60% 80% 0 10% 20% 30% 40% 50% 60% 80% 100% Source : JPMorgan 20% 10% 0% 10% 20% 30% 40% 60% 80% 10. with the variability of the volatility realised over each of the 12 months Stddev of monthly realised volatilities over the year 70% 60% 50% 40% 30% 20% 10% 0% Jan87 Jan90 Jan93 Jan96 Jan99 Jan02 Jan05 Source : JPMorgan (Euro Stoxx 50 data since 1987) stdev of 1month realised v olaitlity (last 12 months. representing the current value of the call on variance.5% 52.0% 12.5% realised volatility Source : JPMorgan 40 . This makes sense.5% 52. assuming that the volatility of volatility is not too large. A standard cap of 2. This is not surprising: when volatility is low.5% 0. This means that the value of the cap should be relatively small compared to the variance swap strike and should not have a major effect on pricing. stuck at the cap level (Figure 43). when volatility is high it tends to swing about more wildly.0% 7.0% 240.0% 10. it is generally stable (give or take the odd spike).5% 52. That is.0% 7.07 R = 0.5% 0. in theory.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market Note that there is a strong correlation between volatility and the spread between variance swap levels and ATM volatility.79 2 1987 2003 1998 2001 2002 1991 1990 1989 1992 1999 1988 2006 1997 2000 1996 2004 1995 1994 2005 1993 10% 20% 30% 1year realised volatility 40% Source : JPMorgan (Euro Stoxx 50 data since 1987) Estimates for the expected volatility of volatility are used for pricing options on variance.0% 52. Figure 41 : Volatility is correlated with volatility of volatility Volatility / volatility of volatility 1month realised v olatility Figure 42 : Correlation of 1year realised volatility.5x current implied variance strike is relatively far outofthemoney. for a long position. large (absolute) changes in volatility are more likely when volatility is high that when volatility is low.0% 52.5% 80. However.0% 12.0.5% 30. A long capped variance swap can be thought of as a standard variance swap plus a short call on variance.
following increases in volatility. Variance swap term structures (in common with ATM volatility term structures) are usually upwards sloping (Figure 44) and have a tendency to flatten. the chance of a change in volatility regime of the course of a 5year trade could be significant. Therefore.g. whereas even if volatility has been low for 23 years. this is driven by factors not necessarily correlated with current realised variance (structured product flows etc. longerdated volatility is also driven by structured product flows and tends to be more susceptible to supply/demand dynamics. 41 . As it happened. As observed above. years 5 4% Jul04 Source : JPMorgan Jan05 Jul05 Jan06 Jul06 An upward sloping curve in a lowtomedium volatility environment can also be thought of pricing in the chance of a change of volatility regime sometime in the future. especially for longer maturity trades. longer dated variance swaps take on significant exposure to changes in implied variance before expiry. Volatility regimes represent a shift in the medium/longterm average of realised volatility and are generally considered to last a number of years. so this particular trade may well have lost even if it had been held to maturity. Whilst a shortdated variance swap is principally exposed to realised variance. The variance swap curve can also be thought of representing the meanreverting nature of volatility. Term structures for variance are generally steeper than the equivalent ATM volatility curves due to the increasing effect of the skew at longer maturities (Figure 44). their principal exposure will initially be to 5year implied variance. in 1998 LTCM experienced crippling marktomarket losses on short positions on longdated volatility. The shape and potential movement of the implied variance curve is important in determining the most promising parts of the curve to buy or sell variance. including positioning for volatility exposure through forward variance.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market 2.) and the p/l can be unpredictable. or even invert at the short end. if an investor has just bought or sold a 5year variance swap. whereas the long end (5years+) tends to be more anchored to some longterm estimate of average volatility. Another important point to consider. However. is the effect of the marktomarket p/l. These observations help to explain why investors can be reluctant to sell longdated variance at seemingly high levels (e. The short end is most sensitive to prevailing levels of realised volatility (Figure 45). 19982003 was probably the worst 5year period in the last halfcentury to be short realised volatility. Over a short time period the chance of a switch to a higher volatility regime may be negligible. 25% for Euro Stoxx 5years) when the relevant indices have almost never realised these levels over any 5 year period (Figure 46). sometimes correlated to economic cycles.9: Variance term structure The variety of drivers of short and longdated variance can lead to interesting dynamics of the variance swap term structure curve. For example. Figure 44 : Euro Stoxx term structure COB 23 October 2006 Implied volatility/variance Figure 45 : Shortdated volatility is driven by shortdated realised Implied volatility/variance 22% 20% 18% 16% 14% 12% 0 Source : JPMorgan 28% 24% 20% 16% ATM i mplied volatility Implied variance VSTOXX Index Euro Stox x 1M realised v olatility 12% 8% 1 2 3 4 Maturity.
1Y 1Y .5%.1M 5% 07Sep06 22May 06 01Jan06 0% Sep04 M ar05 Sep05 M ar06 Sep06 5 5% Variance swap maturity Source : JPMorgan (Euro Stoxx 50 data) 42 . and movements in the curve. Direct trades on the predicted evolution of the variance maturity curve are also possible (Section 3. 3month variance moves up by 2% and 4year variance moves up by 0. and steepening it if volatility decreases. These kind of movements will have the effect of flattening a normal upwardssloping term structure as volatility increases. See Section 3. the loss due the slide from moving down a normal termstructure is tradedoff against the savings in carry due to the elimination of exposure to realised volatility. For example.8 for details of using forward variance to position for an increase in volatility. whereas longer data variance will initially be more exposed to changes in implied variance (vega).9). In general shortdated variance will have the greatest exposure to realised volatility (gamma). Here.4σ sqrt(T) a roottime move in volatility ensures that prices for (ATM) vanilla options will change in the same proportion across the maturity curve. Analysis of term structure movements is very important especially for long forward volatility trades. Since the relative price for an ATM option is approximately 0.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 2: The Variance Swap Market Figure 46 : Realised index volatility has rarely exceed 25% over 5 years 5year realised volatility 30% 25% 20% 15% 10% 5% Dow Jones 0% Jan50 Source : JPMorgan N ikkei S&P500 DAX FTSE Euro Stox x 50 Jan55 Jan60 Jan65 Jan70 Jan75 Jan80 Jan85 Jan90 Jan95 Jan00 Jan05 Investors looking to buy (or sell) volatility through variance swaps will clearly need to consider the best placed part of the term structure curve to target. Figure 47 : Variance term structure moves more at the short end… Implied variance Figure 48 : …as demonstrated by the May 2006 volatility spike when the short end inverted and the longer end flattened Variance spread 24% 22% 20% 18% 16% 14% 12% 0 1 2 3 4 Maturity (years) Source : JPMorgan (Euro Stoxx 50 data) 10% 5Y . Typical movements of term structures can be explained in part by the “roottime” rule sometimes used in the market for characterising curve changes. and changes will be most visible at the shorter end of the curve. That is for a ‘normal’ move in volatility the change in implied variance at a given point on the curve will be proportional to the inverse square root of the maturity. if 1year variance increases by 1%.
2 3.5 3.8 3.3.10 3.6 3.9 3. Uses of Variance Swaps • • • • • • • • • • • Part 3: Uses of Variance Swaps 3.1 3.4 3.3 3.11 Exploiting a volatility view Specific hedging purposes Rolling short variance Diversification Index variance spreads Relative value singlestock volatility Variance dispersion and correlation trading Forward variance and volatility spikes Trading the variance swap term structure Skew and convexity trades Cross asset class trades .7 3.
Diversification: Volatility can be thought of as an asset class in its own right. By combining variance swaps with other (volatility) instruments trading on the same underlying.g.g. Index variance spreads: Variance swaps can be used to trade the spread of volatilities between two indices.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps Part 3: Uses of Variance Swaps Overview In Part 3 we consider uses of variance swaps. hedging specific exposures. Skew and convexity trades: Variance swaps are long skew and convexity. a variance swap is just a contract providing direct exposure to (realised) volatility. Trading variance against (deltahedged) vanilla options provides interesting exposures to skew and/or convexity. Rolling short index variance is an attractive systematic volatility strategy from a riskreturn perspective. Rolling short variance: Short variance swaps can be used to capture the observed equity index volatility risk premium. and exploiting market inefficiencies. Forward variance and volatility spikes: Long forward volatility can avoid potentially negative carry at the cost of slide down the term structure. diversifying returns. outlined briefly in the list below. Relative value singlestock volatility: Use volatility pairs. variance swaps can be used together with credit default swaps (CDS) to trade equity volatility against credit. But why is it useful to have the ability to trade such an instrument? Variance swaps can be used for trading macro volatility views. such as that resulting from structured products or life assurance policies. Taken at face value. or crosssectional regression volatility models to find rich/cheap singlestock volatilities Variance dispersion and correlation trading: Trading variance swaps on an index against variance swaps on its constituents provides exposure to equity correlation. Such trades can be thought of as either relative value volatility trades or as ‘volatilitybeta’ trades aiming to profit from a spread of volatilities widening as volatility increases. it is possible to take exposures to other factors such as skew and convexity. Trading the variance term structure: Variance swaps can be used to trade the shape of the variance term structure. moves in the volatility term structure or correlation. either at the index or single name level. • • • • • Exploiting a volatility view: Variance swaps are ideal for taking a direct view on the volatility of an underlying without the pathdependency issues of a deltahedged option. Variance swaps are useful instruments in debt/equity trades. forward variance) or on different underlyings (variance spread trades and dispersion trades) it is possible to use variance swaps to create exposure to factors other than realised volatility – e. both being measures of corporate risk. By trading combinations of variance swaps either of different maturities (e. deltahedged options. or conditional variance swaps. such as straddles. • • • • • • 44 . Cross asset class trades: Equity Volatility and credit spreads are correlated. future implied volatility. and as such can act to diversify returns within a portfolio. analogous to the use of spreads and butterflies in fixed income to trade the shape of the yield curve. Specific hedging purposes: Variance swaps can be used for macrohedging and also for hedging specific volatility exposures. Finally. and can be a useful way of positioning for volatility spikes.
Buying variance can prove profitable if the correction is relatively sudden (volatile) and occurs any time within the maturity of the variance swap. If this part of the business was sold at a fair level. if the underlying trends up before the event and then afterwards sellsoff suddenly back towards the straddle strike. The nature of variance swap payouts means that large singleday stock moves will generate large p/l for long variance positions. Hence. See European Equity Derivatives Weekly Outlook. but otherwise known timing.g. Variance swaps have none of these issues and provide a pure exposure to volatility. say within the next year. but the result of the trial is in the balance. which may gradually trend up. either outright or traded against vanilla options such as straddles. and nonpathdependent. However. Variance swaps can also be used by investors to express their view on the likelihood of an M&A event. Put options are a possibility to use for hedging. the likely long term volatility expected should fall and selling long dated variance may be profitable. This could be a catalyst for large moves in the stock price in either direction depending on the outcome. whilst it may be far from obvious how to play such an event using the underlying alone. This can be particularly valuable if an event timeline for a company is known. Straddles are not perfect since a low volatility market often displays trending behaviour. At the index level investors may use variance swaps to express a macro view – for example to buy volatility ahead of possible recession. However the investor must be careful to maintain the required volatility exposure and must also consider whether an exposure to realised volatility is desirable. rising market environment since 2004. the spot price of the company may remain relatively unchanged. where the usually negative carry from holding a long variance position can be compared to the put premium. 45 . even in comparison to vanilla options. but the volatility characteristics of the company may alter significantly.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps 3. but will need to be restruck in order to roll up the protection as the market rallies. especially if an investor feels that the options market has not correctly priced in the risk associated with such an event.1: Exploiting a volatility view The most obvious use for variance swaps is to express a direct view on the volatility of the underlying. On singlename underlyings. Similarly.8) to position for a sudden increase in volatility. suppose an investor feels there will be a sharp downwards correction at some stage in the future. even deltahedging the straddles will prove suboptimal as the market moves away from the strikes and the (favourable) exposure to volatility is reduced through the decreased gamma. in a bull market. Depending on what is priced in. of an event. which is difficult to capitalise on using the underlying alone. but the likely outcome of the event is unknown. but where and when this volatility is realised. they may fail to capitalise e. way of capitalising on this low volatility view. Whilst straddles could be attractive. For example. To deal with these issues. For example. Indeed backtesting shows that shortvariance swap strategies have outperformed short straddle strategies (both deltahedged and not) in the low volatility. Note that the lack of path dependency can make this a more attractive alternative than buying/selling the underlying (or options thereon) since the p/l depends on the outcome over the entire lifetime of the trade and not just where the underlying ends up at maturity. and the investor’s view. Deltahedged options can be used for this purpose but have two principal drawbacks: • • The deltahedging requires active management (and therefore incurs human and transaction costs). variance swaps could be used to take a view on the uncertainty of a company. suppose an investor expects an industrial conglomerate to sell a more volatile subsidiary. suppose a pharmaceutical company is due to announce the results of a drug trial. long or short variance swap positions may be attractive. selling variance is an efficient. or sell volatility into a temporary spike. 6 November 2006 for more details. In this instance variance swaps can be used to implement a view on the uncertain outcome. The volatility exposure is pathdependent: the p/l of a deltahedged option depends not only on the amount of volatility realised by the underlying. it is sometimes preferable to use forward starting variance swaps (Section 3. investors who expect a quiet market. may want to find ways of boosting their alpha based on this view. The investor does not want to sell his position in the underlying as the market may rally further. For example. Another use of variance swaps is to take advantage of an expected structural change in a company.
But if the market is benign. Forward variance swaps are not directly exposed to realised volatility so do not suffer negative carry in a benign market environment. Forward variance swaps can offer a useful alternative to holding spot variance. whilst capital adequacy requirements (Solvency II) require portfolios to be reserved against changes in market level. They do however suffer slide as the position ages (assuming term structure is positively sloped).2: Specific hedging purposes Variance swaps can be useful for hedging purposes. realised volatility losses will accrue if the market is relatively benign. although longerdated maturities accrue realised volatility losses at a relatively slow rate. In practice. Figure 25) at least over short time horizons. hedged or soldon into the market. this will not hedge marktomarket volatility risk or shifts in implied volatility. 46 . and variance swaps provide a very useful way for desks to manage their volatility exposures.g. Such hedging flows. Sometimes variance swaps do offer more attractive riskreturn characteristics for hedging. can be more efficient. effectively analysing the relative attractiveness of long variance and long put positions. or anticipation thereof. interest rates etc. particularly using combinations of vanilla options to provide protection within a targeted range – e. depending on both hedging need and relative value. Alternatives such as forward starting variance and conditional variance can also be explored as ways of targeting hedge exposure.10 for a discussion of forward starting variance swaps and Conditional Variance Swaps. the variance protection will be effective and only a small notional amount will be required. variable annuities. particularly at longer dates. as will a long put position. withprofits funds). See Section 2. returns for which they attempt to generate by holding equities. which include a delta exposure. a variance swap will likely suffer negative carry as realised volatility is below the level bought. without any large moves. Changes to accounting standards are encouraging marktomarket valuation of these embedded liabilities. they can provide a hedge against an equity portfolio. The aggregate risks of their positions needs to be managed. putladders. The p/l on forward variance is determined by changes in the implied volatility term structure. Similarly. Firstly it is not clear how much variance should be used to hedge a fixed quantity of the underlying. There are problems however with this approach of attempting to hedge an underlying using variance swaps. These companies have essentially sold equity put options to their policy holders and hence have taken on a short volatility exposure. life assurance companies offer many products which offer some form of guaranteed benefits (e. If the selloff is rapid. can have a large effect on the volatility surface. Although assurers may hedge embedded optionality by deltahedging. structured products desks provide investors with a wide variety of equitybased investments which incorporate volatility (and correlation) exposures. collars etc. ranging from simple capital guaranteed products to more complex multiasset payoffs. and to the extent that implied volatility reacts to market downturns. However.10 we analyse the valuation and payoffs of long variance swaps combined with short put positions. if the underlying slowly drifts down. putspreads. other types of protection. which is often what insurance companies need to hedge. For example. In Section 3. April 2006 for a discussion of conditional variance swaps.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps 3.8 for a discussion of spot and forward variance swaps including carry and slide analysis. insurers are looking to use variance swaps (either spot or forward) to hedge volatility exposure. volatility. Increasingly. variance swaps will likely profit if markets selloff significantly and especially if this happens suddenly. See Section 3. which will itself change as the market moves. likewise a long put will lose its premium. Since volatility is directional (see Section 2. the long variance swap may fail to provide protection altogether.5. For any long variance swap position. Forward variance can also be useful for hedging against volatility spikes. Variance swaps can also be useful for hedging out specific volatility exposures. Note also that longdated variance swaps offer particular exposure to changes in implied volatility. Being long a variance swap can be compared to owning a put option: a variance swap will likely profit in a market downturn.g. increasing the demand for longdated variance – and potentially steepening the volatility term structure.
Results from backtesting are very encouraging (Figure 49). In practice 1month maturity variance swaps appear to be a good maturity to target – shorter dated options are available but are liable to incur relatively higher transaction costs. From a riskreturn perspective. we have only sold capped variance swaps (capped at 2. 5 vegas) once a year. to limit the risk of the strategy. compared with only two months with gains greater than 10% – indicating the negative tails of the return distribution (see e.g. In addition 1month trades correspond to VIX/VDAX/VSTOXX levels allowing for greater visibility and ease of backtesting. we have modelled the selling price for capped variance swaps at 1.2%) with that of the positive returns (only 2. Figure 49 : Performance of rolling short variance on the S&P Vega points 40% 30% 20% 10% S&P 1month v ariance bid subsequent S&P 1month realised v olatility short 1month S&P v ariance p/l 0% 10% 20% M ar90 Source : JPMorgan M ar92 M ar94 M ar96 M ar98 M ar00 M ar02 M ar04 M ar06 47 . with 161 of the 196 months yielding positive returns (Table 10).5%). but with a large negative tail from volatility shocks. 2.g. a level we believe is somewhat conservative.g.5x the strike) and assumed that these trade 50 bps below uncapped variance swaps. losses are relatively infrequent but can be large and unpredictable.6. Nevertheless.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps 3. Finally. Figure 28). Then to get a realistic variance swap bid level we have deducted a further 50bps.5–1 vega) we have estimated variance swap mid levels at 75 bps below the VIX. it may be better to regularly take in a relatively small premium (e.75% below prevailing VIX levels. Short variance presents a good example of such a carry trade.3 vegas per month. By implementing a longterm rolling strategy we attempt to diversify away these occasional large losses over time. To summarise.3 vegas) every month than to take in a larger premium (e. the maximum gain (14. Average returns are 2. Returns from this strategy are on average positive.6).8%) is much less than the maximum loss (22. As variance swap prices were not available for the whole period.3: Rolling short variance Historically one of the most successful volatility strategies has been the systematic selling of shortdated index variance. The various forms of the Yield Alpha product aim to take advantage of different types of carry trade across a spectrum of asset classes. Since the VIX tends to trade somewhat above real variance swap prices (typically 0. meaning the shorterterm trades earn less carry. JPMorgan). This is also seen when comparing the standard deviation of negative returns (5. we have estimated the variance swap strike from VIX levels. even though term structures are generally upward sloping. Although positive returns dominate. by taking advantage of the observed equity volatility risk premium (Section 2. Recall that the VIX represents the theoretical 1month variance swap price interpolated from exchangetraded S&P options (see p94). We have backtested the strategy of selling 1month S&P variance on the first day of every month since March 1990. This strategy takes advantage of the volatility risk premium described in Section 2.6%) and there were four months with losses greater than 10%. A diversification argument also argues for using shortdated variance swaps. by choosing a basket of strategies with attractive ‘carrytorisk’ characteristics. The backtest presented here comprises the short variance component of JPMorgan’s “Yield Alpha” product (see “Yield Alpha Strategy Guide”.
In contrast if realised volatility over the next month was only 10%.5% of the new capital amount.3% (uncompounded).6% 60% 80% 1990 1993 1996 1999 2003 2006 Source : JPMorgan.5 x 7. can be risked (with the 2. In fact the greatest level of realised compared to the level of implied bought at was in July 2002 when the realised volatility of 42% was about 1. At the end of each month when the variance swap expires. e. the index level has reached 741 (Figure 50). Given the 2. the capital will increase or decrease depending on the p/l of the short variance swap.75.g.51875 index points. Example 13: Calculation of the ‘Variance Alpha’ index Suppose the index is at 100.5% of the available capital (e. results from Mar 1990 – Jul 2006 Similar results are obtained from this strategy using the Euro Stoxx 50 or DAX indices.5 = 3.502 ) / ( 2 x 20 ) = 26. and the new vega notional equal to 0. Incidentally.5 index points (ip).25ip. Each month we sell a 1month variance swap with vega notional equal to 0. Note that if we assume we can actually trade at prevailing VIX levels.g.5% of the current index level (Example 13).8.7x the 25% level of variance sold. By July 2006. Given this pattern of returns we can consider creating an index based on the performance of shorting variance swaps.3% 15.5 x (202102)/(2x20) = 0. with the VSTOXX and VDAX as the basis for variance swap prices.3% gives an information ratio of 1.5x cap was never even close to being hit. $500K). over the course of the backtest.10% Largest return 195 100% 4.5x cap. Using slightly longer maturity variance swaps.5 x ( 202 . starting in 1990 with a value of 100 and each month entering into a short 1month variance swap with vega notional equal to 0. 48 . the maximum loss cannot be greater than the capital held due to the cap. the p/l would be: 0. which when compared to the annualised standard deviation of returns of 15.4% 5. and a new short variance swap position for the next month is initiated using 0.5% 2 14.78 741 Negative 34 17.3% 27. We assume that a fixed amount of capital.4% 6 Positive 161 82. 16 years since inception. results from Mar 1990 – Jul 2006 Source : JPMorgan. the average monthly return rises to 4% (48% annualised) and the information ratio to about 3. the subsequent veganotional will be 0.75ip.5x cap allowing the maximum losses to be known). The veganotional will be 0.2% 4 22. and the 1month variance swap strike is 20.6% 2. works well although the timing of the index rolldates then becomes more important. say $100M. This means that as long as the variance swap strike does not exceed 76. the maximum loss will occur if realised volatility is 50% or more In this case the monthly loss will be capped at 0. and if this loss occurred. Figure 50 : Index performance from rolling short variance strategy Index performance Table 10: Performance statistics for rolling short variance strategy 1000 800 600 400 200 0 20% 0% 20% 40% index performance monthly short v ariance p/l (rhs) Average Return (monthly) Average Return (annualised) Annualised Stdev Riskreturn Index performance (1990 = 100) All Number of months Percentage Standard deviations Returns In excess of +/. the new index level would be 103. We consider such an index. 3months. the 2.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps The annualised return is 27.1.8% 2.3% 1.36875 index points.
Rolling short 1month naked straddle strategies perform well. but also suffer in trending markets. short variance swaps appear to deliver the most consistent alpha across different volatility environments. However. and avoiding the negatively convex payoff which can hurt short variance positions in times of high volatility. 6 November 2006). Deltahedged straddles do a better job than unhedged straddles of capturing the premium of implied to realised volatility. these types of strategy work well as overlay strategies aimed at boosting alpha and diversifying returns.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps It is an interesting exercise to compare the performance of rolling short volatility strategies using variance swaps with similar strategies using short straddles. 20042006). 49 .g. Whilst all three strategies perform well. and most effectively capture the premium of implied to realised volatility. profiting from the somewhat rangebound behaviour of equity indices. as otherwise positive returns from selling volatility above the level subsequently realised are eroded as the market trends away from the straddle strike. With returns from short volatility trades somewhat uncorrelated with the underlying. their p/l is extremely path dependent and they can suffer in times of lowvolatility trending markets (e. and the gamma exposure of the straddle is reduced. both deltahedged or not (see European Equity Derivatives Weekly Outlook.
it can be used within a portfolio to diversify returns (Figure 53.6 Required portfolio return In Figure 53 we show the risk/return for each individual strategy – equity (long Euro Stoxx).4: Diversification One of the notable features of the returns of a rolling short volatility index is its similarity to returns of a bond index (Figure 51. Moreover since there is relatively low correlation between the p/l from short variance and the underlying equity market. in the case of credit portfolios. the implied realised volatility premium can be thought of as the payment required for providing “equityinsurance” capital. including returns from short variance swaps helps to push out the efficient frontier. Asset allocation. short variance swaps generally replace bonds within an efficiently allocated portfolio (Figure 54). Figure 51 : Index performance from rolling short variance strategy 20% 1000 800 600 400 200 0 1990 Source : JPMorgan Figure 52 : Bond index performance – JPMorgan USD traded bonds 400 5% 350 300 250 200 150 100 50 1990 Source : JPMorgan 0% 20% 40% index performance monthly short v ariance p/l (rhs) 0% 5% 10% bond index performance 60% 80% 15% 20% 2006 monthly p/l (rhs) 1993 1996 1999 2003 2006 1993 1996 1999 2003 Hence. Figure 54). defaults. In bond indices. the regular p/l resulting from coupon payments and accrued interest is punctuated by capital losses caused by rising yields. due to somewhat bondlike nature of short variance swap returns.1 Source : JPMorgan Short equity variance Short equity variance Bond Bond Equity 5 10 15 20 25 Portfolio risk 6. In rolling short variance. regular periods of positive p/l resulting from the spread between implied and realised are punctuated by large losses arising from spikes in volatility. Figure 53 : Efficient frontier of stock plus bond moves further out by adding short variance… Portfolio return Figure 54 : … and variance generally replaces bonds in asset allocation along the efficient frontier. Because variance returns are not fully correlated to any combination of bonds and equities. It is this “equityinsurance” capital that can be considered an alternative asset class that can be invested in and the payment received is captured through a rolling short variance swap strategy. volatility (or more correctly the implied – realised spread) can be thought of as an asset class in its own right. Figure 52). short Euro Stoxx 50 variance – overlaid with an efficient frontier for a twoasset bond/equity mix (grey line) and an efficient frontier for all three asset classes (blue line).8 7.0 6. or.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps 3. 50 .7 8. white dashed line is optimal allocation split for bond + equity only 10 100% 9 8 7 6 5 4 0 Source : JPMorgan Equity 80% 60% 40% 20% 0% 5. In the same way as interest is the payment required for providing financial capital. bonds (long Euro Governments). Moreover when used in conjunction with bonds and equities.
but where one index tends to be more sensitive to increases in volatility than the other. The volatility pairs regression analysis discussed in the following section can also be used to screen for these kind of opportunities. 51 . such as protection buying. to see if an opportunity exists. though this may also be a consequence of the trend of increasing correlation between indices globally over the last few years (Figure 56). Some index variances have a tendency to appear either structurally rich or cheap as a result of local activity.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps 3. However. 6M Average pairwise correlation 0. ‘Volatilitybeta’ trades which are long variance on a higher beta index and short variance on a wellcorrelated but lower beta index. Volatilitybeta trades The second type of spread trading involves taking opposing volatility positions in two wellcorrelated indices. many companies of the companies in the index have now ‘matured’ and the Nasdaq behaves much more like other developed market indices – realising around 1020% volatility in 20052006. Such trades. For example the Nasdaq was a very high volatility index in 20002001 realising well over 50% volatility. JPMorgan.7 0. Opportunities of the first type can be evaluated by looking at current spreads and comparing these to historic spreads of realised (and implied) variance. where the variance of one index appears to be trading at an unwarranted premium or discount to another.dev Euro Stoxx 50 6M realised volatility (lhs) 25% crosssectional stddev of 6M index realised volatilites (rhs) 20% 0. This is highlighted by the contraction of crosssectional deviation between different index realised volatilities whenever overall market volatility becomes more subdued (Figure 55).8 60% 50% 40% 30% 20% 10% 0% 1986 Source: JPMorgan 6M realised volatility std. which will probably exist for both indices.5: Index variance spreads One use of variance swaps is to trade the spread of volatility between two indices. in the current period of low volatility (20042006).1 1991 1993 1995 Top 4 indices Developed indices Developed + EM indices 1990 1994 1998 2002 1997 1999 2001 2003 2005 Source: JPMorgan Moreover. whilst maintaining some overall exposure to increased volatility through the higher sensitivity of the buy side index. Figure 55: Spreads between different indices have compressed as implied volatility has fallen… 6M realised volatility Figure 56: …while there has been a general trend of increasing correlation between indices IndexIndex. One of the characteristics of a low market volatility regime is the compression of volatility spreads across stocks and indices.4 15% 10% 5% 0% 2006 0. Note though that some indices have local ‘regimechanges’ as their composition or behaviour of the members changes. We have seen previously that there exists a premium of implied to realised volatility. are often referred to as volatilitybeta trades. Thus a spread trade may to some degree cancel this extra cost.6 0. 2. August 2004”. this deviation of index volatility is now close to its alltime lows. as previously stated. See also “Volatility Pairs. at least in similar volatility environments. These type of trades fall into two main categories: 1.5 0. with the short variance leg hedging out some or all of the otherwise negative carry if volatility remains at prevailing levels.3 0.2 0. particularly as volatility increases. or other supply and demand dynamics. These trades aim to be net long volatility. following the unwind of the dotcom bubble. Relative value volatility trades.
data since 1991 y = 1.6% of the long DAX variance trade. However. 6month DAX and Euro Stoxx implied variance were trading at 18. coupled with the likelihood that index volatilities will probably diverge in a rising volatility market.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps The high correlation between indices. reflecting the usually negative carry of a long volatility position. Suppose it had been possible to own DAX variance at the 18. Figure 57: DAX tends to realise more volatility than the Euro Stoxx … 6M realised volatility Figure 58: … and spread tends to expand as volatility increases 6M DAX realised volatility 60% 50% 40% 30% 20% 10% 0% SX5E DAX SX5E . then the average profit would have been 3. which has been about 1. To compare these two alternatives fairly. Also note that the returns for the spread trade are more normally distributed than those of the long DAX trade (Figure 60). with the historical realised volatility over recent years.0142 R2 = 0. leading to a (mid) spread of 1. and outside of this period the average p/l would have been 3. However. We can quantify this effect by looking at the beta of a regression of DAX on Euro Stoxx 6month realised volatility. 52 . but with a standard deviation of returns of only 2%. This is a consequence of the volatility beta effect.6% and 17. This is indeed the case. we look at the behaviour of DAX volatility compared to that of Euro Stoxx 50. As an example of this type of trade. with the average spread of realised volatility at around 2% over the last fifteen years (Figure 57). it makes sense to compare a long variance strategy on the DAX with a spread trade which is long DAX / short Euro Stoxx 50 variance.9721 10% 1991 1993 1995 1997 1999 2001 2003 2005 Source: JPMorgan Given the directionality shown in Figure 58. and the spread between them.0493x + 0.2% respectively. it should be expected to realise less volatility than the DAX on a purely correlation effect alone.25%. much of this difference is due to the higher assumed bid offer costs for executing the spread trade.6 vegas) with a standard deviation of 13%.7%. Example 14: DAX vs.DAX 60% 50% 40% 30% 20% 10% 0% 0% 10% 20% 30% 40% 50% 60% 6M Euro Stoxx 50 realised volatility Source: JPMorgan. Given that the Euro Stoxx 50 is a panregion index with a greater number of members and a more balanced sector composition than the DAX. Another important feature of the DAX / Euro Stoxx volatility spread is that it tends to widen as volatility increases.05 over the last fifteen years (Figure 58). gives opportunity to investors to position themselves more efficiently for an increase in volatility using volatility spread trades rather than outright long volatility positions. the majority of profits would have been made in the high volatility period between 1997 and 2003. This would have lead to an average p/l of about 0.9% level (mid + 30bps) throughout the last 15 years. this leads to a p/l of 1. Euro Stoxx volatility beta trade On 25 September 2006.25 over the last three years and 1. which sees DAX realised volatility increase (and then decrease) by more than Euro Stoxx realised volatility. Suppose instead we consider the p/l of being long the DAX/Euro Stoxx variance spread. In this case. we scale the p/l of the spread trade by its standard deviation of returns. We can backtest this directly by comparing the prevailing levels of implied variance of the two indices.6% (3. less than the 3.4% of DAX over Euro Stoxx.8%.
DAX 35% 30% 25% 20% 15% 10% 5% 0% 0% 5% 5% DAX p/l SX5E . but additionally using a correlated index with lower beta to mitigate the carry of such positions.5X) Figure 60: … but the spread trade does have a more normally distributed p/l Population of p/l scenarios 40% 60% 50% 40% 30% 20% 10% SX5E DAX SX5E . 50% 53 0% 1991 1993 1995 1997 1999 2001 2003 2005 10% . 5. to find index pairs where the volatilities are well correlated and which display the volatility beta effect described above. and scaling the spread p/l to give a similar level of risk. We can asses the attractiveness of these pairs by backtesting the variance spread trade against going long variance on the higher beta index.9 vegas and the spread traded p/l. One point worth making in this analysis of volatilitybeta trades is that it does not attempt to directly measure the relative value between index volatilities. indices with a low correlation between their realised volatilities are not considered as a direct volatilitybeta trade. In general we can look for these kinds of spread trades across a wide range of indices.DAX p/l (SD Scaled) 10% 15% 20% 25% 30% 35% 40% 45% 15% 10% 6M variance p/l Source: JPMorgan.6% for the same level of risk.9 vegas in comparison with the long DAX trade (transaction cost 30bps).Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps Figure 59: Long DAX variance p/l is similar to the spread trade against Euro Stoxx … 6M DAX variance p/l 6M DAXEuro Stoxx 50 variance p/l (6. assuming prevailing variance levels as of 25 Sep 2006 Source: JPMorgan. Without these transaction costs the average long DAX p/l would have been 3. in some sense the indices were too correlated in this case. What we are attempting to achieve is to get a long volatility exposure. Hence. However.5 times also necessitated effectively scaling the 60bps transaction costs up to 3. assuming prevailing variance levels as of 25 Sep 2006 The above example shows that in this case the spread trade had no real benefit compared to just holding long DAX variance. Scaling up the spread trade by 6.
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps 3. Volatility pairs analysis In this case.6: Relative value singlestock volatility Variance swaps can also be used to trade spreads between singlestock volatilities – either as a way of partially hedging the variance of a singlename against wider market/sector moves. Thus the trick is to look for pairs with a high zscore in their implied volatility ratio.20 1. typically from the same sector with similar volatility characteristics.00 Aug05 Oct05 Dec05 Feb06 Apr06 Jun06 20% Jun05 Source: JPMorgan JPMorgan produces a daily Volatility Pairs Report considering all possible pairs within 11 broadbased sectors and ranks them according to the difference between the implied and realised zscores. and no trading opportunity exists. June 9 2006 for further details. Of course sometimes a divergence in implied variances of two stocks whose variances were previously highly correlated may simply reflect real changes in the underlying market. which is not reflected in a corresponding high zscore for their realised volatility ratio (Figure 61. then pairs with zscores of large magnitude. We have backtested the results of trading volatility on the top 3 ranked pairs in sector.00 2.00 3. If the realised volatility ratio has recently diverged from its longterm average in the same way as the ratio of implied variances. August 2004 and “Identifying and exploiting dislocations in singlestock volatility”. consider the case of two similar companies with similar volatility characteristics where one company issued an unexpected profit warning raising concerns about its future viability. Clearly in this case.00 0. could potentially represent opportunities for volatility pairs trades.10 35% 30% 25% Alcatel 3M realised v olatility Cap Gem 3M realised v olatility ratio of realised v olatilities (rhs) 4..90 Aug05 Oct05 Dec05 Feb06 Apr06 Jun06 15% 10% Jun05 Source: JPMorgan 1. Figure 61: Alcatel implied volatility increased in relation to Cap Gem. or in order to directly trade relative value singlestock volatility. with encouraging results. For example. expressing a large deviation from their long term average. Two possible methods for assessing relative value single stock variances are temporal regression and crosssectional regression. the ratio of implied volatilities would not necessarily be expected to meanrevert and selling the ratio in the hope that it would close up would not necessarily be a sensible trading strategy. Assuming these ratios of implied variance have meanreverting tendencies. A zscore can be computed to express how far the ratio is from its long term average. temporal regressions work by looking at the evolution of the ratio (or spread) of the implied variances between two stocks. then the change in the implied variance ratio may be justified. See Volatility Pairs. This could lead to a large change in ratio of implied variances as the variance of one company spikes but the other remains relatively stable. 1Y Implied volatility implied volatility ratio Figure 62:…but this is not reflected in the ratio of realised volatilities 3M realised volatility realised volatility ratio 35% Alcatel 1Y implied v olatility Cap Gem 1Y implied v olatility ratio of implied v olatilities (rhs) 1. To some extent we can screen out these type of scenarios by also considering the zscore arising from the ratios of realised volatilities.00 0.. This applies particularly to well correlated stocks. Note that variance swaps are particularly suited to this type of trade due to the direct nature of the exposure to volatility they offer.30 1.00 30% 25% 20% 1. 54 . Figure 62).
but is simply trying to model all current stock implied volatilities as a multilinear function of a number of various other stock properties.4 1.6 1.0 1. 12 June 1006. or even using forward starting variance. The implied volatilities of Centrica.2 1. those with cheap volatility below it 3m implied volatility Table 11: The top 5 richest and cheapest European volatilities according to the crosssectional model (9 November 2006) 3m realised volatility 16.3 1. See European Equity Derivatives Weekly.g. and should correlate with implied volatility) Figure 63 and Table 11 show the results of the crosssectional regression model for 112 major European names as of 9 November 2006.9% 25. Investors who believe that the spread between the implied volatilities may meanrevert in the short term.7% 25. Figure 63: Crosssectional model: stocks with rich volatility appear above line. 55 .6% 25.7 NESN VX 16% 20% 24% 28% 32% crosssectional model implied volatility Source: JPMorgan This method can also be used to select stocks with cheap implied volatility/variance to trade against index variance in a bespoke dispersion trade. November 2006 for further details.8% 20. Scottish Power and Rio Tinto appear rich.6% 21. European Equity Derivatives Weekly. See e. with around 6 vegas average p/l (prebid/offer) since 2003 from trading the spread between the long and short variance baskets. We have backtested this method by trading selected stock variances from a universe of major European names .9% 15. UBS.9% 25.4 1. Ciba.2% 23. having apparently cheap volatilities.3% model 3m implied volatility 18.3% 20. 21 November 2005 and “Introducing JPMorgan’s CrossSectional Volatility Report”. The idea here is to compare the current implied volatilities (or variances) of a universe of stocks against a number of other factors which may have some bearing on their volatilities – for example their historic volatilities and credit spreads. may be better trading longdated variance with the intention of closing out their positions early.2% 25.4% 3m ATM implied volatility 24.8% 22.9% 29. Nestle. this regression is not trying to model the historical relationship of stock volatilities.buying variance on the 10 stocks with the cheapest looking variance according to this metric and selling variance on the 10 richest looking names. BG Group.0% 30.4% 21.5% 34. the market (high beta should translate into high volatility) Dividend yield (high yielding should perhaps be less volatile) 3month return (poor performers may tend to have higher volatility) CDS spread (reflects a company’s credit risk and leverage.0% 29.8% 32% 28% 24% 20% 16% 12% 12% Source: JPMorgan EAD FP BG/ LN SPW LN RIO LN 1 2 3 4 5 … 108 109 110 111 112 CNA LN MAN GY IBE SQ CIBN VX UBSN VX Bloomberg Ticker CNA LN EAD FP BG/ LN SPW LN RIO LN NESN VX IBE SQ CIBN VX MAN GY UBSN VX zscore 2.4% 11.2% 33.8% 17.5% 33. EADS.7% 26.3% 21.5% 30. should trade this spread through relatively short dated variance swaps with the intention of holding them to expiry.9% 24.9% 19. all with actual implied volatilities well above those suggested by the regression model.2 1. The following variables are used in our regression model: • • • • • Historic volatility (should usually be the main factor determining a stock’s implied volatility) Stock beta vs.2 1. Iberdrola and MAN appear below the regression line. On both the temporal and crosssectional regressions investors should consider whether they wish to be exposed to the relative realised or implied variance of the stocks.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps Crosssectional regression An alternative means of screening for rich/cheap singlestock variance is to use a crosssectional regression methodology. Results are good.8% 23. Investors believing that the current spread between the implied volatilities of a pair of stocks represents an unreasonable view of the difference between their future realised volatilities. It is important to note that unlike the temporal regression method explained above.6 1.7% 12.
due to the diversifying effects of the index. and tend to spike together in a sharp selloff. Index variance should always be less than the average variance of the constituents. 56 . However. the index and singlestock legs of the trade must be weighted with more index veganotional than singlestock vega notional – the exact amounts depending on the level of implied correlation. As in the case of realised correlation. the size of this discount effectively being related to the correlation between the constituents of the index. Like volatility. for a well diversified index. ρ= (∑ w σ ) − ∑ w σ 2 i i i i 2 i σ I2 − ∑i wi2σ i2 2 i or approximately (for a well diversified index): ρ≈ (∑ w σ ) i i i σ I2 2 The advantage of this approach is that the average implied correlation. Nevertheless. backtesting shows that the premium of index implied over realised variance tends to be greater than the corresponding premium for singlestock variance. the p/l from favourable moves in correlation is diminished as the trade becomes long volatility under these conditions. whilst fund managers seek to sell options on singlestocks on which they have a view in order to raise premium. This naturally leads to the concept of trading correlation. the resulting position being short correlation. Since these are realised quantities. especially since the trade becomes short volatility following adverse moves in correlation. Realised index volatility can be thought of as arising from two factors: (a) the volatilities of the constituents.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps 3. This can be partly explained by demand for protection at the index level. Another way of expressing this ‘average correlation’ is to assume that the correlation is the same for all distinct pairs of stocks and then rewrite the above equation to solve for this average correlation. making it another attractive candidate for carrystyle trades: rolling short correlation through variance dispersion performs well (Figure 65). This implied correlation can be traded by selling index variance and buying singlestock variance. and averaged to give an overall correlation figure for the index. Realised index variance is described by the equation: σ I2 = ∑ i ωi2σ i2 + 2 ∑ i< j ωiω jσ iσ j ρ ij where th ωi and σ i are respectively the weight and volatility of the ith stock in the index and ρ ij th is the pairwise correlation of the i and j stocks. Losses from short correlation through variance dispersion can occasionally be very large. This ‘volatilityscaling effect’ helps to account for the premium of variance dispersion levels compared to correlation swaps and the relatively modest performance of variance dispersion trades in the low volatility environment since 2003. correlation usually trades at a premium to realised levels (Figure 64). this turns out to be approximately equal to the square of the ratio of index volatility to average singlestock volatility. volatility and correlation are themselves correlated. which we cannot directly measure. together with (b) the correlation between them. That is. at least initially. Such a trade is known as a variance dispersion trade. they can be observed in the market and the pairwise correlations directly calculated. Similarly. Greater correlation means less diversification and hence higher index volatility. selling variance on an index and buying variance on its members has generally been a profitable strategy. can be backed out in exactly the same way from the implied variance of the index and its constituents. To hedge out the exposure to volatility.7: Variance dispersion and correlation trading An extension to trading volatility pairs is to trade the volatility of an index against a the volatility of a basket of stocks.
but is diminished by low volatility vega payoff correlation points 6m correlationweighted dispersion p/l 6m i mplied minus subsequent realised correla tion 0.10 0. long options on the members) would not give a true correlation exposure. correlation swaps levels tend to trade somewhat below the levels of correlation associated with variance dispersion trades. at least optically.30 0.20 0.60 0.90 0.40 0.10 0.00 0. See Correlation Vehicles.30 Oct02 Oct03 Oct04 Oct05 Oct06 Note that variance swaps make possible trading equity correlation in this form. For technical reasons.30 Oct01 Source : JPMorgan 6m implied correlation 6m realised correlation 10% 8% 6% 4% 2% 0% 2% 4% 6% Oct00 Oct01 Source: JPMorgan Oct02 Oct03 Oct04 Oct05 Oct06 0.40 0. Managing a correlation trade through deltahedged vanilla options on the index and all its constituents would be onerous.50 0. May 2005 for further details.60 0.20 0. 57 .Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps Figure 64 : Implied correlation tends to trade at a premium to realised Euro Stoxx 50 6m correlation Figure 65 : rolling p/l of variance dispersion reflects the spread between implied and realised correlation. Equity correlation can also be traded more directly through correlation swaps.50 0. mostly to do with the observed strong positive correlation between volatility and correlation. whereas unhedged vanillas (short index options. This tends to mean that for sellers of correlation. Indeed the demand for variance swaps as a means of trading correlation has been suggested as a reason for the relatively good liquidity of variance swaps on equity underlyings in comparison to other asset classes such as bonds and FX where vanilla options can be even more liquid than for equities.80 0.70 0. However such pseudodispersion trades using naked calls or straddles can give interesting and often attractive p/l profiles. variance dispersion almost always gives more attractive entry levels.
can buy forward starting variance swaps to express their view (Figure 67). In contrast.8: Forward variance and volatility spikes One of the strengths of variance swaps is the ease of pricing and constructing forwards (see p17). hence forward interest can reasonably accurately capture the future direction of such a trend. Forward variance can be an attractive vehicle to be long volatility without accruing realised volatility losses that can occur with a long spot variance position (see volatility risk premium as discussed in Sections 2. The following discussion concentrates on index variance swaps. Having done this. Figure 66 : Forward interest curve compared to history. This problem can be mitigated by using longer term variance swaps. One important point to note is the difference in behaviour between interest rates and realised volatility. p/l before the forward date will be entirely driven by changes in expectations of volatility. Likewise investors expecting a pick up in volatility that is greater than that currently priced in. The first analogy is to use forward variance to trade the future volatility of an underlying. This makes using forward variance to trade the future path of volatility more unpredictable. realised equity volatility can be driven more by short term spikes. This ability to reexpress the spot variance term structure into a strip of forward starting variances raises natural analogies with fixed income where the swap curve can be decomposed into strips of forwards or futures.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps 3. EuroDollar or Fed Funds interest rate futures to trade the future direction of interest rates. Below we introduce the fixed income concepts of slide and carry as applied to volatility. we will focus on the ability of long forward variance swaps to hedge against volatility spikes. It is only once the forward date is reached (and the shorterdated leg of the forward has expired) that there is direct exposure to realised volatility.DEM) 3month implied Euribor from futures 60% 50% 40% 30% 20% 3month realised volatility 3month implied forw ard variance 5% 4% 3% 2% 1% 1996 10% 1998 2000 2002 2004 2006 2008 2010 0% 1996 1998 2000 2002 2004 2006 2008 2010 Source : JPMorgan Source : JPMorgan For a forward volatility position. and it is only on expiry of the futures contract that p/l is finally settled against prevailing 3month spot rates. with only a fairly slowly underlying “base” level change.3). in a similar way that investors use Euribor. the performance of a forward volatility position crucially depends on the dynamics of the volatility term structure. 58 . 3month Euribor 6% Figure 67 : Forward variance compared to history 3month volatility 3month Euribor (ex. Given that investors have a choice to trade forward volatility across a different maturities and forward dates. investors expecting a more prolonged hiking cycle from ECB will sell Euribor futures at far out maturities expecting the yield curve to steepen (Figure 66). p/l on a Euribor contract is driven by changes in expectations of the level of future 3month rates. as captured by movements in the implied volatility term structure. which we characterise using principal component analysis (PCA). In addition. as well as being able to use spot variance swaps. we use a variety of metrics to help us decide where to concentrate exposure. Short term interest rates are dictated by central bank action and typically move in a relatively smooth trend.6 and 3. as a correct call on the underlying trend can be ruined through an unfortunate choice of timing. In the same way. which will help to smooth over the effect of any short term spikes. For example.
Principal component analysis The behaviour of changes in the implied variance curve can be analysed using principal component analysis (PCA).Nov04 Nov00 .Nov06 Nov02 . Figure 68 : Principal component decomposition of Euro Stoxx volatility curve since 2000 Change in implied variance Figure 69 : Principal component decomposition of Euro Stoxx volatility curve since 2000 … the first component has flattened through time Change in implied variance 3% 2% 1% 0% 1% 2% 0 Source : JPMorgan First factor . forward starting instruments have no exposure to carry until the forward starting date is reached. see p42). October 2005 and May/June 2006 (Figure 73). For example. consistent with the observed flattening of the first principal component 59 . This is illustrated by considering the movements in the variance curve following the volatility spikes in April 2005. factor illustrates how the dominant curve move is to increase (or decrease) much more at the shortend of the curve. Likewise a long investor will receive the carry from holding long maturity positions. with volatility decreasing (or increasing) at the very shortend of the curve and increasing (or decreasing) further out. so that relative maturity diminishes over time. or factors. In variance swaps the holder of a long variance position will have to pay the spread of implied over realised every day they hold the swap. The first. Figure 69 shows that primary curve moves have become more parallel over the last 2 years." Twist " 4% 3% 2% 1% 0% 0 Source : JPMorgan Nov04 . reflecting the tendency for increases in volatility to flatten the curve and for decreases in volatility to steepen the curve (consistent with the roottime adjustment described above.Nov02 1 2 3 4 5 Maturity. being long variance often means a negative carry position. using 6 years of data. Note that for both fixed income and variance swaps. In order to fund this position. Figure 68 shows the two most important components of weekly changes in the Euro Stoxx volatility curve.6)." Shift " Second factor . Volatility slide Slide in both fixed income and variance swaps occurs because once a trade is implemented the maturity is fixed. Likewise. Since yield curves tend to be upward sloping this is usually a positive spread. PCA decomposes moves in the volatility curve into its major components. For an upward sloping variance term structure – typical of low volatility environments – a holder of a long variance swap will lose out through time as the variance swap ‘slides’ down the term structure curve.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps Volatility carry Carry in fixed income is taken to be the cost of funding a spot position and reflects the difference between short and long term interest rates. Yrs 1 2 3 4 5 Maturity. Since this spread is typically positive (see Section 2. at least up to the 2year maturity. These moves have often been relatively parallel. and most important. Both spot and forward starting variance swaps have exposure to slide. so the investor will have to pay the “carry” to be short. Slide measures the impact of this diminished relative maturity assuming that the shape of the term structure remains unchanged over the period. The second component of curve moves is more of a twist component. they will have to pay the difference between the long and short rate. Yrs If we also look at how the first component has changed through time. being short index variance swaps is usually viewed as a positive carry trade. This illustrates how short dated implied variance tends to move about twice as much as longer dated maturities during a shift up (or down) in the term structure. an investor wanting to take a view that long term interest rates will rise will need to either short sell a bond or short a long dated swap. at least at the index level.
At the same time however. suggests the following results: 60 . Forecasting when a spike in volatility might occur is well nigh impossible. the strategy that would benefit most is likely to be the spot shortdated variance swap. Factors to consider include: • • • • The likely magnitude of a volatility spike and its effect across the variance term structure. investors may consider rolling their position to a new forward starting variance position. slide and PCA analysis to help decide which long variance position (spot or forward. 30 May 2006). Our analysis of spikes (in realised volatility) on major worldwide indices since the 1950’s (See European Equity Derivatives Weekly. Figure 70: Movements in variance term structure predicted by the principal components analysis model 22% 20% 18% 16% 14% 12% 10% 0 Source : JPMorgan Figure 71 : Euro Stoxx principal component analysis: weight of second (twist) component has increased over the last 5 years weight of 2nd component. Also if a spike has not occurred before the forward start. if a spike does occur. If the shift up in volatility happens immediately. Whilst it is important to consider the likely time of arrival of any spike and the effect such a spike is likely to have on the variance term structure. This means that curve twists have become more independent and less well correlated to the direction of volatility. However. one of higher weighted components of the transform suggests that there is a bias for a spike to occur about every six months. whereas if they believed that the spike heralded the beginning of a higher volatility regime they may look to profit by continuing to hold their long variance position. whether or not to unwind a long volatility position. Volatility Spikes Below we use carry. after a spike is also an important consideration. so as to avoid a realised volatility exposure. secondary twistlike moves have become relatively more important. The possibility of term structure flattening. % 20% 1M Vol spikes by 5% 1M Vol remains unchanged 1M Vol drops by 3% 10% 3 6 9 12 15 18 21 24 0% 2000 Source : JPMorgan 2001 2002 2003 2004 2005 PCA analysis is useful in calculating expected gains or losses for different parts of the implied variance curve following a general increase (or decrease) in volatility. As a final point it is worth considering.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps changes. as the increase in implied variance will likely be greatest at the short end of the curve. If an investor holding forward variance into a spike believes that fundamentals do not point to a raised base level of realised volatility they would most likely look to unwind. The cost of holding the position until the spike occurs. long or shortmaturity) is likely perform well as a hedge against a volatility spike. as shown in Figure 71. which could adversely affect the marktomarket of forward starting variance. In conjunction with carry and slide analysis for forward variance they can help identify the best placed part of the curve to be long (or short) volatility. the behaviour of volatility. The likely timing of a volatility spike. by decomposing a long history of realised volatility into periodic components using Fourier transforms. both implied and realised.
0% 24% 22% 20% 18% 16% 14% 12% 0 Source: JPMorgan 0.5% 2 4 6 8 10 Maturity 3M 6M 9M 12M 15M 18M 21M 24M 27M 30M 33M 36M Forward starting maturity Source: JPMorgan 61 . That is.5 vegas. on how it will affect the curve.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps • • • There is not a strong pattern of reduction in volatility in the first month following a spike. there is a clear bias for volatility to decrease. Figure 74: Variance swap term structure as of 20 October 2006 Implied variance 26% Figure 75: Expected cost of carry and slide over the next 3months assuming realised volatility of 12% 0. w here v ol increases in the follow ing month 26% 24% 22% 20% 18% 16% 14% 11May 06 22May 06 12Jun06 4 5 months after spike 6 12% 0 Source: JPMorgan 3 6 9 12 15 18 21 24 Source: JPMorgan: Data for DJIA since 1900 maturity (months) Given that we can make an assumption on the likely timing of a spike and have a reasonable idea.5% 1. shortlived spikes tend to subside to prespike levels within 2 months (Figure 72) For those spikes which have not subsided within 2 months. on average by about 4. The short lived spikes tend to outnumber the longerlived spikes by about 3 to 1.. We can then gauge which variance swap offers the best cost/benefit profile given the prevailing variance term structure.0% 1. we can estimate the likely p/l from holding different maturity (spot or forward starting) variance swaps.. For those spikes which subside in this manner (a reduction in volatility within 2 months) the reduction in volatility is almost the same size as the increase seen in its spike. they take longer and typically subside to a significantly higher base level than the previous spike (on average 5.. and although these kind of spikes tend to eventually subside.. from PCA. it is likely that there has been a significant further increase in volatility.5 vegas higher after 6 months). w here v ol decreases in the follow ing month . Figure 73: Changes to the Euro Stoxx 50 variance term structure during the volatility spike in May/June 2006 Implied variance • • Figure 72: Volatility tends to reset higher following longerlasting spikes 1M realised volatility compared to prespike level 20% 15% 10% 5% 0% 5% 0 1 2 3 realised v olatility follow ing spikes . In the second month after the spike.
the forward starting maturities can provide some cushion against the precise timing of any spike. because the forward has to be traded twice and because it has a (slightly) wider bidoffer spreads. as the increase in implied variance will likely be greatest at the short end of the curve.month forward starting maturity has the least slide as it normally represents the place where the term structure begins to flatten out. it is fairly typical that the 9. assuming realised volatility of 12% until then 5% 4% 3% 2% 1% 0% 1% 2% 0% Source: JPMorgan Figure 77: p/l if spike in implied volatility were to happen in 4months. The 9month forward starting maturity does well in most scenarios as the low slide helps mitigate the cost of holding the long volatility position.or 12. Figure 76 and Figure 77 show the expected p/l of 6months variance swaps of various forward starts as a function of the magnitude of the move in 1year variance. Note though that entering and exiting a forwardstarting variance swap can be more costly than running a spot variance swap to maturity. Given any variance move at 1year maturity. It is worth noting that a 4% move in the 1year maturity equates to a move of about 7% in the 1month maturity. 62 . which is a fairly typical size for a volatility spike. the forward starting term that would benefit most is actually the spot 6month variance swap. However. If the move up happens later. In contrast. the first principal component tells us how to shift the whole curve and we can calculate p/l of all forward starting swaps. we calculate the slide expected over the next 3months.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps Example 15: positioning for a volatility spike through forward starting variance We consider the best placed forwardstarting 6month variance to position for a move up in volatility. If a shift up in volatility happens immediately. We find that the 9month forward starting maturity is the one with the least slide.and 12month forward starting variance works relatively well under several different timings of spikes. the 9. the low realised volatility before the spike increases the breakeven of this maturity. the forward starting maturities can be a better place to position for this than the spot maturities. assuming realised volatility of 12% until then 4% 3% 2% 1% 0% 1% 2% 3% 4% 3M start 6M start 9M Start 12M start 3M start 6M start 9M Start 12M start 1% 2% 3% 4% 0% Source: JPMorgan 1% 2% 3% 4% Change in 1year variance strike Change in 1year variance strike Thus for investors expecting a rise in volatility even within the next few months. a spot 6month variance swap would have likely already have accrued a negative p/l from the usual negative spread of realised to implied volatility (“carry”) and 3month variance would then have to spike by about 4 vegas just to break even (consistent with 1year variance increasing by about 2 vegas. if the spike were to happen in three months time. which gives our best estimate of the cost of holding that forwardstarting instrument (Figure 75). In fact. according to our PCA analysis of the volatility term structure). The shorter maturity forward starting variance does best if the magnitude of the shift is large and happens within the next few months. which by their very nature are difficult to predict. Figure 76: p/l if spike in implied volatility were to happen in 3months. given the prevailing variance term structure on 20th October 2006 (see Figure 74). Moreover. For 6month variance swaps starting at various forward dates. Figure 76 shows the projected p/l for a spike after 3months and Figure 77 for a spike in 6months.
Figure 78: The p/l of various long 6month term variance strategies starting in May Vega p/l 12% Figure 79 The p/l of various long 6month term variance strategies starting in January Vega p/l 10% 8% 6% 4% 20% 2% 0% 2% 18% 16% 14% 6month spot p/l 6month. However in the January example. the spot variance performed relatively poorly. As a way of minimising negative carry. which had the added advantage of being an almost spot instrument at the time of spike (Figure 79). both of which would have benefited from the May/June volatility spike. participating more fully in the spike in implied volatility. In contrast the forward starting variance swaps did much better.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps Example 16: Performance of forward starting variance swap over volatility spikes To illustrate the potential advantages of using forward starting variance swaps. 6th January 2006) and in April 2006 (European Equity Derivatives Weekly. the spike occurred soon after implementation and hence spot variance performed well. and also the residual exposure to implied volatility had fallen. previous low realised volatility had increased the required breakeven level. Moreover the long maturity 12month forward did almost as well as the 6month forward. 6month forward p/l 6month. it had not accrued much negative realised carry prior to the spike (Figure 78). we suggested using forward starting variance swaps for a long volatility view in both January 2006 (European Equity Derivatives Weekly. since by the time that the volatility spike occurred. 12month forw ard p/l 6month implied v ariance (RHS) 6month implied variance 26% 6month implied variance 26% 24% 22% 10% 8% 6% 4% 2% 0% 2% 18Apr Source: JPMorgan 24% 22% 20% 18% 16% 14% 02May 16May 30May 13Jun 4% 06Jan Source: JPMorgan 06Feb 06Mar 06Apr 06May 06Jun 63 . 12month forward p/l 6month implied variance (RHS) 6month spot p/l 6month. we consider two specific examples. 18th April 2006). In the later case (April).
This meant that the market was implying a similar outlook for forward realised volatility in May 2006 as it had at the beginning of 2005. believing that these concerns may act to increase volatility. but at a more advantageous level or with reduced carry. the Euro Stoxx 50 1year2year implied variance spread had come close to its twoyear lows (Figure 81). compared to spot 1year variance of around 17%.6% over the next year. This appeared to be somewhat contradictory given the macro economic concerns prevailing at the time. Whilst this forward volatility was low by historical standards. Hence a long 1year variance 1year forward position was only likely to be profitable if implied volatility spikes or realised volatility rises substantially over the next year. One of the main uses of these types of trades is to express a directional view.9: Trading the variance swap term structure Whilst we have shown that variance swaps are useful for taking a direct view on the future value of implied and realised variance.2001x + 0. assuming an unchanged term structure. 1year forward variance swaps. Again this is best described using examples. which in turn left 1year variance 1year forward close to its lows as seen in early 2005 (Figure 82).8639 1y spot variance 20% 2% 4% 6% Sep04 Mar05 Sep05 Mar06 15% 10% Mar04 Source : JPMorgan 10% 20% 30% 40% 50% Source : JPMorgan. could have used the recent flattening to enter into 1year.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps 3. 64 . In this case. it was still well above the recent realised volatility of the Euro Stoxx 50 and would suffer a negative slide of 1. the June07/June08 forward variance was trading around 18. but wishing to avoid a direct long exposure to realised variance. Figure 80 : Euro Stoxx 50 curve had flattened up to 15th May Variance swap level 22% 20% 18% 4% 6% Figure 81 : … and particularly in the 12 year part of the curve Volatility Volatility 5Y1Y (LHS) 2Y1Y (RHS) 3% 2% 16% 14% 12% 10% 0 Source : JPMorgan 15 May Current 2006 215 Mar 2006 months ago 1 2 3 4 5 2% 1% 0% Mar04 Source : JPMorgan 0% Sep04 Mar05 Sep05 Mar06 Figure 82 : Euro Stoxx 50: 1 year 1 year forward variance had come down relative to spot variance volatility 30% forw ard 25% spot Figure 83 : … especially accounting for the level of volatility 2y – 1y variance spread 4% 2% 0% y = 0. data since Aug 2000 Investors. For example. The first example attempts to take advantage of a flattening of volatility term structures (Figure 80).6%. they can also be used to trade the shape of the variance term structure curve in analogy with the use of spreads and butterflies in fixed income to trade the shape of the yield curve. in midMay 2006.0507 R2 = 0.
y ears 1. In total this equates to a variance butterfly trade of long 2½ units of 15month spot variance. This example shows how butterflies can sometimes be used to trade the underlying asset (i. rhs) Oct03 Oct04 Oct05 45% Oct06 Nonetheless.e. short 3 units of 9month spot variance and long ½ a unit of the 3month spot variance. 6month variance 9month forward – 3month forward 3% 2% 6month implied variance 10% 15% 20% 25% 30% 35% 19% 18% 17% 16% 15% 14% 0 Source: JPMorgan 17% 16% 15% 14% Euro Stox x FTSE 13% 12% 0. Similarly the short 1 unit of variance notional of 6month variance 3 months forward is constructed from short 1½ units of 9month spot variance and long ½ units of 3month variance (Section 1. This steepness can be observed by comparing the Euro Stoxx 50 term structure with that of the FTSE on the same date (Figure 84). This would be equivalent to trading the spread between forward and spot variance. the beta of 0. Also note the trade ratio will need adjustment through time as the position ages.and 9. by assuming that the relationship between this spread and spot implied variance would hold. In this case the long 1 unit of variance notional of 6month variance 9 months forward is constructed from long 2½ units of 15month spot variance and short 1½ units of 9month variance. 65 .month maturity variance swaps. short 1year) with some additional long 1year variance (weighted as per the regression beta in Figure 83).7). where a bulge is clearly visible in the Euro Stoxx 50 term structure that is not seen in the equivalent area of the FTSE.5 2 1% 0% 1% 2% 3% 4% Oct02 Source: JPMorgan Spread of forw ards (lhs) 40% Six month implied variance (inverted. Another way to think about a spread of forward variances is as a butterfly of spot variance swaps. hedged for volatility directionality. Whilst this could have reflected anticipation on the part of investors of an increase in volatility over the following months. could still take advantage of the relatively low forward volatility strike by recognising that the spread between 1year and 2year variance was low compared to the level of 1year variance (Figure 83). The second example aims to take advantage of a Euro Stoxx 50 term structure that appears to be overly convex. investors could be long the curve whilst minimising outright implied volatility exposure. Note that this position would have positive carry providing the market realises less than the 1year strike (around 16. an often used technique in fixed income.2 in Figure 83 suggested a trade ratio of long 1 unit of 2year variance and short 0. By combining a variance calendar (long 2year. As the level of implied volatility falls.5 1 maturity . However as the May2006 volatility spike receded this spread of forwards stayed narrow. variance) at more attractive levels than spot.4%). this slope appeared steep compared to the overall level of implied volatility (Figure 85). the spread between the near term forwards and those further out usually widens. reflecting an expectation on the part of investors for a nearterm move up in implied volatility. For example. Figure 84: Euro Stoxx term structure bulged compared to FTSE … Euro Stoxx 50 Implied variance FTSE implied variance Figure 85: … meaning that near maturity Euro Stoxx forwards look high compared to further out maturity forwards.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps Investors unwilling to take on an outright long volatility exposure.8 units of the 1year variance. (late October 2006) due to the steepness of the volatility term structure between the 1. it can be seen that a spread which is long 6month variance 9 months forward and short 6month variance 3 months forward. is equivalent (at least according to a regression) to selling a spot 6month variance but at higher than prevailing levels.
10: Skew and convexity trades Since variance swaps are priced off the entire skew curve. the exposure to volatility (both implied and realised) can be hedged out by trading a short straddle against a long variance swap position. Consider an investor who wishes to be long volatility. This helps to explain the commonly observed shape of the index skew curve.0)) = N Vega 2K 2K S0 where the adjusted strike is now dependent on the underlying’s price at maturity ST and is written as: 2 KN Put ~ ( P − MAX ( S P − S T . and decreasing or maintaining its current level as the underlying rallies. we know from looking at past history that volatility tends to be directional. he would be willing to be long volatility at a higher than prevailing variance strike if the market sells off. the straddle will need to be restruck as the underlying moves. Variance swaps are long skew and convexity. leaving a residual long convexity exposure. This exposure can be approximated by combining a long variance swap with a short put. However. The combined p/l of the short put and long variance swap at time T is: N Vega ~ (σ 2 − K ( S T ) 2 ) (σ 2 − K 2 ) N Put + ( P − MAX ( S P − S T . they take on a significant exposure to skew and convexity. at least if overall levels of volatility (ATM volatility) do not decrease. both in the form of gamma and vega. Variance swaps plus naked short puts One way of trading realised skew is to combine a put with a variance swap. To this end. but typical shapes of skew surfaces mean that increasing the skew increases the convexity.0)) K (ST ) = K 2 − N Vega S 0 We can set the notional size of the put such that on the upside our breakeven is equal to the lowest variance realised if the market has rallied. in particular increasing – often substantially – in a sell off. and really it is this which affects the variance swap price. Whilst variance swaps are in effect long volatility and convexity. In order to maintain as pure as possible an exposure to convexity. since the straddle’s exposure to volatility.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps 3. 66 . The p/l at maturity is simply given by: p / l = N Vega × (σ 2 − K 2 ) 2K However. Straddles vs. will be reduced as the underlying trends away from the strike. variance swaps One possibility for owning convexity is to trade variance swaps against deltahedged straddles. For example consider an investor wanting to express a long volatility position via variance swaps. see p82 for further details. Suppose we sell a put of strike SP for absolute premium P for NPut notional of underlying. with relatively steep put skews representing the expectation of increased volatility on the downside. and flatter call skews signalling the likelihood of more benign volatility on the upside. in return for being long volatility at a lower strike if the market rallies. This is particularly apparent when calculating the marktomarket: if skew and/or convexity instantaneously increase and nothing else changes a long variance swap will profit. believing that realised volatility will be above that encapsulated by the skew. To a first approximation this type of skew curve can look somewhat like the payoff from a long put. suggesting that it should be possible to partially offset the observed negative correlation of volatility and market direction by trading short naked puts against long variance swap positions. this hedge is only effective providing that the underlying remains close to the straddle strike. Note that in theory exposure to skew is neutral.
April 2006). Figure 86 : Long variance and short put can better reflect the tendency of volatility to rise is a selloff.9%) the breakeven volatility would be 9: Breakeven = sqrt( 162 – 2 x 16 x 547/100 ). For €100K of vega exposure we risk losing €100K x ( 92 – 162 )/ (2 x 16) = –€547K.600 4.050 (99%) put is 175. which lowers the downside volatility breakeven at the expense of an increased upside breakeven – in this case 12%.5% 27. as per the calculations above. Taken to its extreme.000 4.400 3.0% 33.741M notional of the 4050 puts. if the market falls to 3.8% 17. for two different put overlays.050 – 3. Breakeven / realised volatility Table 12: Volatility breakevens for long variance. calculated from the equation above: Breakeven = sqrt{ 162 – 2 x 16 x [ 547 – 12.600 (11. For example the popular ‘long up95% conditional variance.2% 23. Since 1996. Since conditional variance swaps are effectively priced from only part of the volatility skew.600 3.077 and the 4. This observation naturally leads to the concept of third generation volatility products such a conditional variance swaps (see “Conditional Variance Swaps”. effectively neutralises some of the long downside puts which are embedded within a variance swap (see Part 4). if the market rallies to 4400 (+7.7% 20.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps Example 17: Long variance swap plus short put A Euro Stoxx 50 6month variance swap is trading at 16. If the market were to fall. the premium retained from the put would act to lower the variance swap breakeven For example.6% 26.050 Put 9. providing the underlying remains within 5% either side of current spot. We do so by selling €547K x 4077 / 175 = €12.800 5.077 ] / 100 } Conversely.0% 18. the lowest 6month realised volatility in a rally has been about 9%.6% 30.600 ) / 4.200 4.0% 23. The first overlay targets an upside breakeven volatility of 9%.000 16% 16% 16% 16% 16% 16% 16% 16% 16% 16% 16% 20% 0% 20% 40% Source: JPMorgan relative move in the underlying This concept of shorting puts against a long variance swap position.200 3.800 4. The second involves selling fewer puts. if the market rallies. short put Market Level Variance + 4. they can be used to gain more direct skew exposures.2% 11.4% 9% 9% 9% 9% 9% + 4.400 4.000 3. the losses on the put would act to increase the breakeven volatility for the long variance swap. 67 .4% 13.0% 28. combinations of short options can be used to remove exposure to realised volatility completely if the underlying moves above or below a certain level.7%) the equivalent breakeven realised would be 23.3% 12% 12% 12% 12% 12% 50% 40% 30% 20% 10% 0% 40% Source: JPMorgan v ariance @ 16 + 99% put v ariance @ 16 realised since Jan1996 Last point Target upside breakeven 3.050 Put 12. Thus to offset this loss we need to raise the same amount through put premium.741 x ( 4. short 105% downconditional variance’ trade locks in a premium dependent on the level of skew. the current market level is 4. as per construction Table 12 shows breakeven realised volatility levels for different market levels. For example.
This can be seen when the relationship is aggregated to the index level. This credit selloff was particularly noticeable in the US. Both variance swaps and CDS give some measure of the risk associated with a particular company. October 2006 for more details on debt/equity trades. mainly in Europe. the credit markets soldoff following the downgrade of GM bonds to noninvestment grade and correlation worries in structured credit. However. Credit investors could have sold CDS protection. although was not reflected in a commensurate increase in S&P equity implied volatility (Figure 88).Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 3: Uses of Variance Swaps 3. whilst providing a certain degree of protection for a more macro change in regime. See Synthetic Short CDS with Positive Carry. then it would be likely that equity volatility would in that case also rise. Buying equity volatility against a long credit position could have provided a hedge to possible macro deterioration. with the most liquid variance swaps trading at shorter maturities. 68 . Somewhat counterintuitively (5Y) CDS levels are more correlated with shortdated variance than with longerdated variance. Thus a short equity volatility/long CDS position could have taken advantage of the price dislocation. In May 2005. almost the opposite occurred. Figure 87: Implied variance is related to credit at the index level in Europe… Maggie industrials spread 60 Figure 88: … and the same is true in the US. with equity volatility spiking. In May 2006. and should be to some extent correlated. supposing that credit markets were actually heralding some macro change in the economic environment.11: Cross asset class trades: trading equity volatility against credit Investors may wish to consider trading variance swaps against credit default swaps (CDS) to exploit any discrepancies between the equity and credit markets. and where the other asset could have been used as a hedge against a wider macro economic change in regime. Here credit and CDS indices such as Maggie and iTraxx show a strong relationship to Euro Stoxx 50 implied variance in Europe (Figure 87) and in US the CDX HG and HY indices is well correlated to S&P variance (Figure 88). CDX 5Y HG spread 140 SPX 3m implied variance Euro Stoxx 3m implied variance 60% 40% 35% 30% 25% 20% 15% 10% 5% 0% 50 40 30 20 10 0 Credit Spreads (LHS) Implied v ariance (RHS) 50% 40% 30% 20% 10% 0% 120 100 80 60 40 20 0 Credit Spreads (LHS) Implied v ariance (RHS) Jan 01 Nov 01 Sep 02 Jul 03 May 04 Mar 05 Jan 06 Nov 06 Source: JPMorgan Jan 01 Nov 01 Sep 02 Jul 03 May 04 Mar 05 Jan 06 Nov 06 Source: JPMorgan While credit and equity have a very strong relationship over the long run. taking advantage of the dislocation in prices. and longerdated variance more susceptible to structured product flows. whilst credit markets remained largely unmoved (Figure 87). likely due to benchmark effects. there have been several instances where short term opportunities have arisen in one of the asset classes.
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033
European Equity Derivatives Strategy 17 November 2006
Part 3: Uses of Variance Swaps
Single stock variance swaps and debt / equity Whilst indices can provide interesting short term opportunities to trade credit against equity variance, they do suffer from basis risk arising from different index members. This can be particularly important in credit indices which may be driven by idiosyncratic default events by single companies. Probably the majority of debt/equity trading is focussed on single name capital structure analysis. This analysis normally takes the form of attempting to model the relationship of a company’s share price to the price of its credit and equity options, for example using reduced form or capital structure models. A typical trade to take advantage of debt versus equity mispricing is a long/short combination of equity puts traded against CDS. One of the main deficiencies of this approach, particularly for reducedform models based on regressions of share price versus CDS is that a company’s share price is nonstationary, compared to CDS which is more mean reverting. For example, consider the behaviour of Fiat CDS as a function of its share price (Figure 89). In the pre2004 period, the CDS tended to widen when the share price fell through €15. However more recently it has taken a much lower share price (below €10) to increase CDS. This is because leverage reduction has reduced the price at which the credit market assumes that the company will become distressed. A simple regression model would assume that CDS behaved like a €15 put option and a combined trade using this strike put against CDS would not correctly model the capital structure. In contrast, because implied variance factors in some of this reduced risk, the relation between Fiat implied variance and credit is more stationary (Figure 90). Hence, variance swaps can be useful instruments as vehicles for expressing a view on the capital structure of a company. In practice, some combination of variance and the underlying equity may be used.
Figure 89: Fiat CDS has changed relationship with stock price …
5Y CDS spread 1400
Figure 90: … whereas Fiat CDS is more linearly related to variance
5Y CDS spread 1400
1200 1000 800 600 400 200 0 0 5 10 15 20 25 30
Stock Price
Source: JPMorgan
1200 1000 800 600 400 200 0 20% y = 3567.9x  770.06 R2 = 0.652 30% 40% 50% 60%
1Y implied variance
Source: JPMorgan
Although debt/equity trading is probably the major form of crossasset class trading, the advent of commodity variance swaps introduces the possibility of further opportunities in crossasset arena. For example one could envisage trading mining company variance against a basket of metals variance or oil company variance against crude oil variance.
69
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033
European Equity Derivatives Strategy 17 November 2006
70
4. Replication and Hedging
• • • • • • • • • • •
Part4: Replication and Hedging
71
4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10 4.11
Deltahedging and dollargamma Theta – the cost of gamma Options pathdependency From options to variance swaps Variance swap replication in one page Sensitivity to skew and convexity Variance swap Greeks Setting up the replicating portfolio Replicating and hedging in practice Effects of variance swap hedging Why not volatility swaps?
both with the passage of time and with changes in the underlying. bundled up with the necessary delta hedging. We go on to give a concrete example of how to construct a replicating portfolio and discuss various practical issues associated with the hedging and replication of variance swaps.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging Part 4: Replication and Hedging Overview In Part 4 we consider the more technical issues surrounding variance swaps and particularly how they are priced and replicated. Nevertheless. In turn. This exposure to gamma is paid for by the timedecay as measured by the theta. the replication argument is crucial. changes over time and with the level of the underlying. in theory. act against the investor’s interests: investor sells variance. This means that the exposure to volatility is path dependent and unless the volatility is constant and the deltahedging instantaneous. The main drawback of deltahedged options as vehicles for trading volatility is that the dollargamma. We define dollar gamma. helping to understanding how these products can be hedged and to explain their relative attractiveness from a marketmaker’s perspective. We demonstrate that such a contract can be created with a static portfolio of options across the continuum of strikes (weighted as 1/strikesquared). After giving an informal demonstration of the constant dollargamma of the replicating portfolio. this can. marketmaker sells options to hedge their long variance position and the deltahedging of these short options on the close can act to increase the closeclose volatility of the underlying. Perhaps the most important point to come out of the replication argument (for a nonreplicating investor) is that if a marketmaker hedges a variance swap. the variance swap replication argument is valid under much wider conditions than the relatively restrictive BlackScholes assumptions. helping to clarify how realised volatility can be accurately captured. which characterises the exposure to volatility. and where the prices of these instruments come from. which allows deltahedging p/l to be calculated directly from the relative moves in the underlying. A variance swap consists of this portfolio. the marktomarket is easily calculated (based on realised variance to date and prevailing variance swap prices in the market) without reference to any other market variables. In this case the marketmaker will act in favour of the investor’s interests. The replication argument shows unambiguously how the volatilities across the skew contribute to the final variance swap price. To remove this pathdependency it is natural to look for a contract whose dollargamma is constant. but the p/l from this deltahedging program ends up negative. The “bit” depends on the steepness of the volatility skew (and convexity) and the time to maturity. This provides an exposure to realised volatility (squared) independent of the path of the underlying.10. See detailed discussion in Section 4. In fact. The investor simply buys or sells variance by entering into the contract and receives or pays out an amount at expiry depending only on the difference between the agreed strike and the level of variance realised. then it is possible that an option could be purchased and deltahedged at a volatility below that subsequently realised. In terms of pricing. in particular it assumes very little about the underlying volatility process. We begin the section with a review of deltahedged options. This is in contrast to an investor who sells options to a marketmaker who then delta hedges: the marketmaker is long options. 72 . and to replicate they will buy low / sell high at any time during the trading day and hence act to reduce volatility. From the point of view of an investor trading a variance swap the hedging mechanism may seem unimportant. From the perspective of marketmakers or others attempting to hedge variance swaps in the options market. ease of replication promotes liquidity and the attractiveness of such products to investors. Even if the investor chooses to close out the variance swap before expiry. and particularly how p/l is generated by ‘trading the gamma’. we show that we can use this portfolio to capture realised variance in a much more general setting. with steeper skews (higher convexity) and longer maturities both leading to higher variance strikes. variance swaps can be observed to trade at levels of ATM volatility “plus a bit”. The variance swap replication argument can also be extended to apply to the socalled 3rd generation instruments such as gammaswaps and conditional variance swaps. showing how (deltahedging) a portfolio of vanilla options can be used to reliably capture realised volatility. a knowledge of the mechanics of variance swap replication is instructive.
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033
European Equity Derivatives Strategy 17 November 2006
Part 4: Replication and Hedging
4.1: Delta hedging and dollar gamma
In this section we outline how vanilla options can be used to trade volatility. Options are exposed to a wide range of factors (performance of underlying, timetomaturity, volatility, rates, dividends etc). The firstorder exposure to moves in the underlying can be hedged out by the familiar deltahedging process described below. This leaves the exposure to volatility, paid for in timedecay as the most important sensitivity. However, as we shall see, this is not a pure volatility exposure, but is path dependent, varying over time and as the underlying moves towards and away from the option strike. Suppose we hold a (call) option. Then to make the position instantaneously deltaneutral (insensitive to the direction of the underlying) we can sell an amount of the underlying equivalent to the delta of the option. Delta measures the sensitivity of the option to moves in the underlying. For an ATM option delta is about 0.5. For other options the delta is more dependent on the level of volatility, with deltas of inthemoney calls tending towards 1 as the option becomes more in the money; and doing so more quickly at lower levels of volatility. Similarly OTM call deltas tend to zero, more quickly for low volatility (Figure 91). Note by convention deltas are often multiplied by 100, e.g. a delta of 0.30 is quoted as 30. By frequently readjusting this deltahedge (in theory continuously, in practice usually daily) the sensitivity to direction of the underlying can be hedged out over the lifetime of the option. Note that this is a dynamic hedge which must continuously be monitored and rebalanced. P/l will come from the accumulated action of continuously rebalancing the holding of the underlying, in order to keep the entire portfolio deltaneutral over time. This is referred to as the gamma p/l, and is paid for in the option premium which is marked to market as lost theta.
Figure 91 : Deltas of 1month maturity ATM call options
Delta
Figure 92 : The convexity of the option
p/l
1 0.8 0.6 0.4 0.2 0 75 80 85 90 95 100 105 110 115 120 125
Underlying
Source : JPMorgan
8 6 4
10% v olatility
20% v olatility
40% v olatility
Positive convexity means the option outperforms the deltaequivalent position in the underlying for both up moves and down moves
2 0 2 92 94 96 98
option v alue deltaequiv alent amount of underly ing pay off at ex piry
100
102
104
106
108
Underlying
Source : JPMorgan
How is this gamma p/l actually made? In order to understand this, we must consider the p/l made over a single deltahedging step, typically a day. Essentially the gamma (rate of change of the delta) measures the convexity of the option. This convexity (curvature) always works in favour of long options positions, meaning that although a replicating hedge is accurate for small moves in the underlying, for larger moves the long option will outperform the replicating hedge in both directions (Figure 92). For a deltahedged option, the gamma p/l will be the outperformance of the option over the replicating hedge. Gamma p/l is largest for shortmaturity, ATM options, where the chance of the option ending inthemoney can change rapidly, even for relatively small moves. As spot moves the delta changes. The gamma measures the change in delta per unit of the underlying, and a long option is long gamma. The more the delta changes, the more the replicating deltahedge will underperform the long options position (Figure 92 again). How much will it underperform? The actual amount depends on the difference between the initial and final delta of the option, which is gamma x dS, where dS is the change in the underlying. Assuming the gamma remains constant, the option gains or loses value at the average delta (average delta = initial delta + ½ gamma x dS) whilst the replicating hedge gains or loses value at the initial delta. Since the difference between the two is ½ gamma x dS, the difference in performance between the option and the replicating hedge is ½ gamma x dS times the move in the underlying. Overall option outperformance is therefore ½ gamma x dS2 (Figure 93).
73
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033
European Equity Derivatives Strategy 17 November 2006
Part 4: Replication and Hedging
That is, the gamma p/l from a move in the underlying is proportional to the gamma of the option and the square of the move. Thus, the p/l from gamma hedging will be large when gamma is high, and moves in the underlying are large. For instance the gamma p/l from a 2% move will be four times that of a 1% move.
Figure 93 : Gamma p/l from deltahedged options
option v alue initial delta new delta av erage delta
Box 12: Deriving the gamma p/l
∆1 (= ∆ 0+Γ 0dS) p 1 = p 0 + (∆0 dS+Γ 0 dS2 /2)
1. Spot is initially S0, option is worth p0, delta and gamma are initially ∆0 and Γ0 respectively. 2. Spot increases, by dS, to S1 3. Delta increases, to ∆1 = ∆0 + Γ0dS 4. Average delta is therefore ∆Avg = ∆0 + Γ0dS/2 5. New premium p1 = p0 + ∆Avg dS
(∆0+∆1)/2 = ∆0+Γ 0dS/2 p0 S0
Source : JPMorgan
∆0
p/l = Γ 0 dS /2 dS S1
2
6. p1 = p0 + ∆0dS + Γ0dS2/2 (by 4 and 5) 7. p/l on long option is ∆0dS + Γ0dS2/2 (by 6) 8. p/l on short futures is ∆0dS 9. total p/l is Γ0dS2/2 (by 7 and 8)
Box 13: Calculating the gamma The delta of a call option is given by: ∆ = N ( d1 ) where d 1 = S: underlying K: strike T: maturity σ: implied volatility (at the strike) r: interest rates
ln(S K ) + r + σ 2 2 T
(
)
σ T
and
N (x ) =
1 2π
∫
x
−∞
e
−
t2 2
dt Γ=
Since gamma is the rate of change of delta with respect to the underlying:
∂∆ , differentiating N ( d 1 ) w.r.t. S gives ∂S
Γ=
N ′(d1 ) Sσ T
where
N ′( x ) =
1 2π
e
− x2 2
Since the delta of a put is given by ∆ = N ( d 1 ) − 1 , differentiating has the same effect. Therefore the gamma of a put is the same as the gamma of call. Note that as spot S, moves away from the strike K, N′(d1) will become small and gamma will decrease very rapidly. Then, since N′(d1) is independent of the monetary units of the underlying, gamma will be inversely proportional to the value of the underlying. Gamma is a useful concept, but since it measures change in delta per unit of underlying, it is dependent on the absolute level on the underlying. For example the gamma of an option on a stock worth €10 will be double the gamma of the equivalent option on a stock worth €20 (with the same relative strike, volatility etc). Moving to the concept of the dollargamma is more useful, since it allows us to directly calculate the gamma p/l for a given percentage underlying move, making it much easier to compare gamma exposures across different underlyings. To compute the dollar gamma and hence the actual cash p/l made from the deltahedging process we first define the dollar delta, which is simply the cash equivalent exposure of the underlying. For example a position which is long $100,000 notional of call options, each with a delta of 30, will be long $30,000 of dollar delta. The dollar gamma is then defined to be the change in the dollar delta for a 1% move in the underlying. This dollar gamma is therefore equal to the normal gamma multiplied by the square of the value of the underlying and divided by 100 (Box 14).
74
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033
European Equity Derivatives Strategy 17 November 2006
Part 4: Replication and Hedging
Box 14: Dollar delta and dollar gamma Dollar delta: Dollar gamma: $∆ = cash value of the underlying delta hedge (for 1 share of the underlying $∆ = ∆ x S) $Γ = change in $∆ for a 1% move in the underlying S. (S=>S+S/100; ∆=>∆+ΓS/100; $∆=>$∆+ΓS2/100)
It can easily be shown that $Γ = ΓS2/100
Once the dollar gamma has been computed, the gamma p/l from a single deltahedging step is easy to calculate: for a return R, the gamma p/l is equal to 50 $Γ x R2 (Box 15). For example if we hold a position which is long $100,000 of dollargamma and the underlying moves by 3%, the p/l will be 50 x $100,000 x 0.032 = $4,500. Box 15: From gamma to dollar gamma For a change in spot S => S + dS, the gamma p/l will be Γ dS2/2 Since $Γ = ΓS2/100, the p/l can be written as (100 $Γ)/ S2 x dS2/2 = 50 $Γ (dS/S)2 Gamma p/l = 50 $Γ R2 where R is the return dS/S in percentage terms
Figure 94 : Dollar gamma is highest close to the option strike, and falls away rapidly away from the strike, especially if volatility is low
Dollar gamma
Figure 95 : Gamma p/l for a 100strike option with spot at different levels. The p/l is quadratic in the size of the move.
p/l is $s
7 6 5 4 3 2 1 0 75 80 85 90 95 100 105 110
ATM option (20% v ol) ATM option (40% v ol) 90strike option (20% v ol)
90,000 80,000 70,000 60,000 50,000 40,000 30,000 20,000 10,000 0
spot@100
spot@95
spot@90
spot@85
115
120
125
5% 4% 3% 2% 1% 0%
Source : JPMorgan
1%
2%
3%
4%
5%
underlying
Source : JPMorgan
size of daily move in the underlying
Example 18: Dollar Gamma Suppose an underlying is trading at $100. We buy 1,000 contracts (100 shares each) of ATM call options with an implied volatility 20%. Delta is 0.51. Therefore the equivalent cash holding in the underlying is 10M x 0.51= $5.1M The gamma is 0.07 which means that the delta will change by approximately 0.07 for a $1 move in the underlying. E.g. if spot moves to $99, delta will be approximately 0.44; if spot moves to $101, delta will be approximately 0.58. The dollar gamma is given by multiplying gamma by the spot squared and dividing by 100. In this case it will be 1002 x 0.07/100 = $7 per share. Since we hold options on 100,000 shares this equates to a dollar gamma of $700,000. This leads to a p/l of $3,500 for a 1% move in the underlying, $14,000 for a 2% move and $87,500 for a 5% move (Figure 95). If the stock drops by 5%, then with spot now at 95, the new dollar gamma (for the next time step) will be only $450,00 (Figure 94) and a subsequent 1% move will now lead to a p/l of only $2,250 (Figure 95).
75
in practice since the changes in implied volatility alter the hedging amounts. Of course.2%. In order to make money on the deltahedging strategy over a single deltahedging step (typically a day). spread over its lifetime. the p/l in one time step is given by: 1 2 2 ΓS R − σ 2δt or 50$Γ R 2 − σ 2δt 2 [ ] [ ] where R is the return on the underlying. Hull for further details. Theta can be thought of as the amortized (nonlinear) cost of the option. with large dollar gammas acting to magnify both gains and losses.g. this positive convexity from the gamma must be paid for. the strategy will profit if the daily realised volatility. Vega and other sensitivities In reality the implied volatility of the option will change throughout the lifetime of the option.2: Theta . and δt is the length of the time step. unless stated otherwise. but also indirectly since the hedging quantities dictated by the delta will also change. This will affect the option p/l directly: if volatility goes up. Whilst monthly deltahedging would give a breakeven of 7. the actual effects can be more difficult to quantify.57%.) θ = − ½Γ S2σ2 at the heart of the BlackScholes analysis. with possibly nonconstant volatility. theta – calculating the deltahedging p/l If delta hedging an option. in the form of the square of the return. If deltahedging daily.the cost of gamma Of course. p/l = 1 2 2 ∑i Γi S i2 Ri − σ 2δt = 50∑i $Γ Ri − σ 2δt i 2 [ ] [ ] Note that the only reason the exposure to volatility is nonconstant is the changing gamma.57% in either direction the long option position will profit. However. theta p/l described above. Then the daily breakeven return can be calculated by deannualising the implied volatility (dividing by the squareroot of 252) to get a value of 1. See Carr and Madan for a thorough dissection of the p/l from deltahedging a vanilla option. is greater than the ‘daily implied volatility’ given by the option implied volatility scaled by the squared root of the timestep. If deltahedging weekly. Example 19: Daily breakevens Suppose an option is bought at 25% implied volatility and deltahedged daily. where the summation is taken over all the deltahedging steps. This comes from the relationship between theta and gamma: (See e. for ease of exposition we assume zero interest rates and an underlying with zero dividend yield. Box 16: Gamma vs. the weekly breakeven will be 25%/sqrt(52) = 3.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging 4. If the underlying moves by more than 1. The above equation can be easily extended to give the p/l from deltahedging over the lifetime of the options. if holding an option to maturity (and deltahedging at the original implied volatility) the vega p/l over the lifetime of the option will be zero and the total p/l will the be the gamma vs. the option becomes more valuable. 76 . However. and it is paid through the time decay or theta. it is necessary to make more on the gamma (from a large daily move) than is lost on the time decay. If it moves by less than this breakeven it will lose. σ represents the implied volatility at which the option was purchased. R. The magnitude of the gain or loss is determined by the dollar gamma.5%. Option prices are also sensitive to changes in rates (measured by the rho) and to changes in dividends (sometimes call mu).
as the index sold off toward the 2500 strike. this happened well away from the option strike. where and when the volatility is realised is crucial. Indeed. Initially the option made a loss due to the low realised volatility. However. However. realising high (around 50%) volatility on the way.4%. in relation to the option strike. despite being purchased at implied volatility 10% below that later delivered Index level option p/ (ip)l 4000 3500 3000 2500 2000 Nov01 Source : JPMorgan option p/ (ip)l spot 150 strike long option p/l 20 0 4000 3500 3000 2500 2000 Nov01 Source : JPMorgan 100 50 0 50 Feb02 May02 Aug02 Nov02 spot strike long option p/l 20 40 60 80 100 Feb02 May02 Aug02 Nov02 When deltahedging an option. If the underlying trades near the strike. If volatility is not constant. In contrast the 4000strike option (Figure 97) initially had much higher gamma than the 2500strike option and consequently lost more due to the lower realised volatility. 2. and still lose from the deltahedging.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging 4. This is despite the fact that the option was purchased at an implied volatility of 26%. it is perfectly possible to buy (and deltahedge) an option at an implied volatility below that subsequently realised. However: 1. especially close to expiry. but this was kept small by the low gamma exposure so far away from the strike. increasing the gamma. When the volatility did occur. For example consider the performance of deltahedging 1year Euro Stoxx 50 options from 1st November 2001 to expiry 1year later on 1st November 2002. but where that volatility is realised. volatility picked up dramatically and the option ended up making a profit of 120 index points (ip). realising around 20% volatility. The magnitude of the p/l will depend not only on the difference between implied and realised volatility. Then the marketmaker will profit if and only if realised volatility exceeds the level of volatility at which the option was purchased. Figure 96 : The 2500strike option profited due to the high volatility and gamma close to expiry… Index level Figure 97 :…whereas the 4000strike option made a loss. 77 . and the low gamma meant that the deltahedging failed to capitalise on this high volatility with the deltahedged options ending up making a loss of 60 ip (about 5 vegas). 10% below that subsequently delivered. Over the whole year. realised volatility was 36%.5%) and up until May 2002 remained in the range 35003800. Assume that whatever volatility is realised is constant and the option is deltahedged over infinitesimally small time intervals.5% ATM volatility due to the put skew). The index was initially at 3500 (with ATM implied volatility at 28. (above the 28.3: Options pathdependency: can volatility be captured by deltahedging? Suppose a marketmaker buys and deltahedges a vanilla option. This was purchased at 33. is inherently unpredictable (see Box 16). the index fell rapidly to around the 2500 level. After May. the p/l is a (daily) accrual of the difference between the realised volatility (in the form of the return squared) and the implied volatility scaled by the time step. the magnitude of the contribution of this (daily) accrual is weighted by the current dollar gamma of the option. Consider first the performance of the 2500strike option (Figure 96). especially close to expiry (high gamma) the absolute value (either positive or negative) of the p/l will be larger. or about 18 vegas. which as it depends on the path of the underlying. The differences between implied and realised volatility will count more when the underlying is close to the strike. for nonconstant volatility.
maturity = 100 days. An atthemoney option will have high exposure to the spread between implied and realised volatility whilst the underlying remains close to its initial value (Figure 98).5 0. then the exposure to volatility would be constant.5 1. flat skew Source : JPMorgan 2 14% 16% 18% 20% 22% 24% realised volatility Simulation: option struck at 100. exposure is locally linear for small changes in volatility. implied volatility=20%.5 1. 78 . If there was such an instrument where the dollar gamma remained constant both over time and for changes in the underlying.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging The nonconstancy of exposure to realised – implied volatility (squared) comes from the dollar gamma changing over the lifetime of the option. Thus the difference between realised and implied volatility will count more when (dollar) gamma is high. implied volatility=20%.0 60 80 100 120 140 underlying at maturity 0. implied volatility=20%. flat skew Source : JPMorgan realised volatility Simulation: option struck at 80. maturity = 100 days. maturity = 100 days. p/l Figure 99 : … returns are correlated with volatility but not perfectly. p/l 2 1 2 1 0 0 1 1 2 60 80 100 120 140 underlying at maturity Simulation: option struck at 100.5 0.0 0. flat skew Source : JPMorgan To summarise: • • • • The p/l of a deltahedged option does (generally) depend on the spread between the (square of the) implied volatility at which the option was purchased and the subsequent (squared) realised volatility This dependence is relatively strong if the underlying trades close to the strike of the option But the sensitivity to realised volatility tails off rapidly as the underlying moves away from the option strike This sensitivity is accounted for by the (dollar) gamma.0 0.0 16% 18% 20% 22% 24% Simulation: option struck at 80. Figure 98 : Simulation of p/l for hedging an ATM option: magnitude of returns is highest if underlying remains close to the money. the exposure to volatility is poor p/l 1. whilst an OTM option will take on no significant exposure to volatility until the underlying trades at around the option strike (Figure 100). due to the mostly low gamma.0 14% portfolio 2 p/ 1. maturity = 100 days. flat skew Source : JPMorgan Figure 100 : Simulation of p/l for hedging an OTM (80strike) option: magnitude of p/l is highest when underlying is close to the strike p/l Figure 101 : … however. implied volatility=20%.0 0.
otherwise known as realised variance. suggesting… Restrike the option to maintain a constant gamma. To achieve a constant dollar gamma across strikes what kind of portfolio is needed? One important observation is that (peak) dollar gamma increases linearly with strike (Figure 102). It does have the property that each option in the portfolio has an equal peak dollar gamma (Figure 103). Naively. and buy an amount of the new ATM to achieve the constant gamma exposure.9). When considering how to construct such at exposure.05 0. at least in theory – in practice it is more like a combination of (2) and (3) above (see Section 4. but the initial portfolio turns out to require a strip of options across a continuum of strikes. This is better than (1) but still requires (potentially costly) dynamic trading of options. The p/l from deltahedging this portfolio would then be just a constant multiplied by the squared returns summed over the hedging steps. and the position could end up with enormous amounts of the option as the underlying moves away from the strike and the gamma of a single option decreases. but buy or sell additional amounts of this option to keep the dollar gamma constant over time. and the effect of summing these 1/Kweighted options across all strikes still leads to a dollargamma exposure which still increases with the underlying (Figure 105). However. the dollargammas of the higher strike options ‘spread out’ more. Perhaps more importantly this approach is to some extent independent of the volatility process driving the underlying.00 20 100 180 40 120 200 60 140 80 160 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200 underlying underlying Source : JPMorgan Source : JPMorgan 79 . either sell or hold the previous option.03 0. The cost of this portfolio represents the price of exposure to realised variance. This has the advantage of using only a single option strike. 3. in can be shown that this increase is linear.4: From options to variance swaps In order to create a portfolio of options which has constant exposure to volatility (actually variance) we need to ensure it has a dollar gamma which is constant for moves both in the underlying and over time. there are three possible approaches to take: 1. Use a single vanilla option.04 0. Start with an ATM option and on each rehedging step. and this gamma will be highly (and dangerously) dependent on the assumed implied volatility (and volatility process) at that time. The first two approaches would require continually calculating the gamma over the course of the trade. and therefore weighting each option by the inverse of the strikesquared will achieve a portfolio with constant dollar gamma (Figure 104. but the option itself needs to be dynamically traded. finally … Construct a portfolio of options in such a way that the dollar gamma remains constant over both moves in the underlying and with the passage of time.01 0. Figure 102 : Peak dollar gamma of an option increases linearly with the underlying Dollar gamma Figure 103 : Peak dollar gamma of options divided by strike is constant Dollar gamma 8 7 6 5 4 3 2 1 0 0 20 20 60 100 140 180 40 80 120 160 200 0. It is this third approach that is used to construct variance swaps. In fact.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging 4. This has the advantage of not requiring the dynamic trading of options. This method has the advantage of using only a static options portfolio – although the options themselves must still be dynamically deltahedged with the underlying. Figure 105).02 0. it may be thought that weighting the options in the portfolio (across all strikes) by the inverse of the strike will achieve a constant dollar gamma. 2.
0008 0. The power of the more general replication argument outlined below lies in the fact that it applies under much less restrictive conditions. 2 [ ] Thus if we find a portfolio with gamma proportional to 1/S then the exposure to implied minus realised variance will be constant. As dS S S dS In fact (by integrating 1/S2 twice) any function of the underlying S of the form − a ln S + bS + c will have a constant dollar gamma (of a). The action of deltahedging this combined portfolio will recover the realised volatility of the underlying. Integrating the value of this portfolio at expiry.0 2. Recall (Box 16) that the p/l at each hedging interval is given by: p/l = 1 ∑i Γi S i2 Ri2 − σ 2δt . the negative natural log of the underlying represents such a payoff. Deltahedging this portfolio provides constant exposure to the difference between implied and realised variance regardless of where the volatility is delivered. Hence the p/l from deltahedging this portfolio is proportional to difference between realised and implied variance. the payout from a call option on S with strike K. this means that the second derivative of the portfolio’s value must be proportional to 1/S2.0002 0.g.5 0. Unfortunately. A portfolio of calls and puts.5 1. each weighted by the inverse square of strike.0 portfolio 1: constant w eighting portfolio 2: w eighted as 1/K portfolio 3: w eighted as 1/K^2 0 Source : JPMorgan 20 40 60 80 100 120 140 160 180 200 underlying 0 Source : JPMorgan 50 100 150 200 underlying One way of explaining why a portfolio of (Europeanstyle) options weighted as the inverse strike squared gives a constant exposure to volatility is as follows. demonstrates that the nonlinear part of the payoff is indeed a negative log contract: ∫ S0 0 (K − S T )+ dK + ∞ (ST K2 ∫ S0 S T (S − K ) − K) 1 T dK = ∫ dK = − ln (S T ) − S T + ln (S 0 ) − 1 2 2 S0 S0 K K + Note (SK)+ is a shorthand for Max(SK.0006 0. But we can replicate such a contract with vanilla options using an infinite sum of calls and puts across the continuum of strikes. A variance swap can therefore be created by replicating a log contract with options which are then deltahedged. has constant dollargamma. Although this argument is easy and intuitive. Note that the bS + c terms represent a static positions in bonds (c) and forwards (b) with no volatility exposure.0) – e.0000 20 80 140 200 40 100 160 60 120 180 3.0010 0. Since the gamma of a portfolio represents the second derivative of the portfolio’s value with respect to the underlying. log contracts are not traded in the market. To summarise: 1. Only the Γi Si2 term prevents this being a direct exposure to variance.0004 0. it does depend on certain strong (BlackScholes) assumptions on the underlying.0 1. A put is (KS)+. weighted as 1/strikesquared.5 2.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging Figure 104 : Weighting options by the inverse of the strikesquared gives a portfolio with aggregate constant dollargamma Dollar gamma Figure 105 : Weighting options as the inverse of the strike squared gives constant dollargamma Dollar gamma 0.0 0. d −1 d2 (− ln S ) = and 2 (− ln S ) = 12 . 80 . 3. 2.
The final term represents a continuously rebalanced futures position. T). weighted according to the inverse of the strike squared. Let f (Ft ) be the function f (Ft ) = 2 F0 ln T Ft Ft + − 1 F 0 Therefore f ′(Ft ) = 2 21 1 Note that both f (Ft ) and f ′(Ft ) disappear at Ft = F0 − and f ′′(Ft ) = T F0 Ft TFt 2 Using Ito’s lemma: ∫ T 0 T Ft 2 f ′′(Ft )σ t2 dt = f (FT ) − f (F0 ) − ∫ f ′(Ft )dFt for any smooth function f 0 2 Hence substituting f as above gives: 1 T ∫ T 0 σ t2 dt = 2 F0 ln T Ft 2 T 1 Ft 1 + − 1 − ∫ − dFt F 0 0 T F0 Ft The LHS of the above equation represents the average realised variance of the future over the period (0. no transaction costs. in a BlackScholes environment.1/F0 of futures at time t will pay out average realised volatility squared at expiry. the underlying and options as follows (see Carr and Madan for details): f (FT ) = f (κ ) + f ′(κ )(FT − κ ) − (κ − FT ) + ∫ f ′′(K )(K − FT ) dK + ∫ f ′′(K )(FT − K ) dK for any κ.T) be the price for a future on the underlying maturing at some time after T. KVAR2. which is easily replicated and costless to initiate. and κ = F0 gives: ln F0 FT + + F0 (K − F ) ∞ (F − K ) FT T T + − 1 = ∫ dK + ∫ dK F F0 0 K2 K2 0 Thus (at expiry) 1 T ∫ T 0 σ t2 dt = + + 2 F0 (K − FT ) 2 F0 (FT − K ) 2 T 1 1 dK + ∫ dK − ∫ − dFt 2 ∫0 K 2 T T 0 T 0 F0 Ft K Therefore holding a portfolio of calls and puts. complete markets (for options of all strikes) exist. This can be replicated with options: This replicating portfolio can be found directly by using the fact that every smooth (twice differentiable) payoff of futures f. that this continuously rebalanced futures holding encapsulates the deltahedging of the options. can be thought of as the (futurevalued) inverse strikesquared weighted sum of the time values of the options portfolio 81 . and simultaneously holding 1/Ft . The first term on the RHS represents a static position in a contract paying f(Ft) at expiry. + + + + 0 [ ] κ ∞ κ Using the function f above. and the underlying (futures contract) follows a diffusion process (with no jumps). (Note. can be replicated by a portfolio of cash.) Therefore the forward cost of this portfolio is simply: 2 K VAR = ∞ C (K ) 2e rT F0 P0 (K ) 0 ∫0 K 2 dK + ∫F0 K 2 dK T The strikesquared. Let Ft for t in the interval (0.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging 4.5: Variance swap replication in one page In this section we make few assumptions: namely continuous trading is possible.
in practice. Now consider a situation where the skew is a mirror image of that in Figure 106. it would be natural to assume that their associated implied volatilities should have a proportionally greater effect on the variance swap strike. For example. So what exactly determines the contribution of volatilities across the skew curve to the variance swap price? Clearly a very outofthe money option should make a relatively small contribution to the variance swap price. In fact. in practice. with put skews seen as having a much greater impact on prices than call skews. a linear put skew of 5% (per 10 volatility points).Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging 4. N(d2). N′(d2). 82 . at least in the case of indices.15. That is. one can think of the chance of needing the constant dollar gamma in the area represented by the very outofthemoney option as being very small. Volatility / strike level 40% 35% 30% 25% 20% 15% 10% 5% 0% skew v ariance strike N'(d2) 40% 35% 30% 25% 20% 15% 10% 5% 0% skew v ariance strike N'(d2) 50 60 70 80 90 100 110 120 130 140 150 relative strike 50 60 70 80 90 100 110 120 130 140 150 relative strike Source : JPMorgan Source : JPMorgan Why does this symmetry exist? Since the 1/strikesquared replicating portfolio has a much higher weighting of puts. suppose an underlying has 3month ATM volatility of 20%. simply because their dollar gamma is lower and so they must be scaled up in order to provide constant dollar gamma across the range of strikes.) Alternatively. Although. otherwise the value of the replicating portfolio will be very sensitive to the exact choice of replicating portfolio. that the corresponding OTM option ends up inthemoney. It then seems sensible that a variance swap represents a kind of weighted average of volatilities across the skew curve. with put volatilities capped at 35% and all OTM call volatilities flat at the level of the ATM volatility (Figure 106). with the average being taken over the variances rather than the volatilities. at least for short dated index volatility. with the closertothemoney volatilities higher weighted. ATM volatility is 20%. this is exactly the case. equity skews look much more like Figure 106 than Figure 107. the contribution to the variance swap price of the volatility of an OTM call is exactly the same as from an OTM put with same (riskneutral) probability. makes volatility skews highly asymmetric. To answer the first point. This explains why Derman’s approximation assuming a linear skew is most successful in these conditions (See Section 2. The theoretical 3month variance strike can then be calculated to be 23. In this case the theoretical variance swap price is virtually identical at 23.4).6: Sensitivity to skew and convexity Skew is commonly thought of as an important component of variance swap prices. and all OTM put volatilities are flat at 20%. Directionality of volatility. it is not theoretically correct. in reality. the exposure to the skew curve is symmetrical. and the weighting function simply being the riskneutral probability density function. this is a useful framework for thinking about how variance swap prices behave. of ending in the money. capped at 35% (Figure 107). Figure 106: High put skew increases the variance strike above ATM volatility… Volatility / strike level Figure 107 : …but increasing the call skew has almost exactly the same effect.05 (pricing OTM options 1 point apart between 1 and 300). variance swaps can effectively be priced and hedged with ATM volatility plus a contribution from the skew. with OTM calls trading much closer to ATM volatility than the corresponding OTM puts. but call volatilities increase linearly by 5% per 10 points as they become more OTM. For the second point. Also. The OTM puts have a greater weighting in the replicating portfolio. (This must be true in order to replicate variance swaps in practice.
and the convexity parameter β. but similar convexities give almost identical variance swap strikes. Positive convexity will always act to increase variance swap strikes. the version with the upward sloping call skew (similar to Figure 107) will now price about 80bps above the version with the downward sloping put skews (similar to Figure 106) since a 160 strike is significantly more inthemoney that the 40 strike. In fact there is some kind of feedback effect here because the contribution of each volatility is determined by the probability of an option being inthemoney. and high call skews (where OTM calls have higher volatilities than ATM) will increase variance swap prices. in ‘d2space’ the variance swap price is not affected by 2 the linear component of the skew α. In practice. Both high put skews. from the equation: K VAR = 2 ∫ ∞ −∞ N ′( z ) BS ( z )dz we can make a number of specific observations: σ2 1. skew is most often thought about in terms of relative strikes. Therefore ATM volatility will provide the greatest contribution to variance swap prices – particularly for short maturities. and this part of the skew is relatively linear (at least for indices). gives the (riskneutral) probability that the underlying will be trading above z at time T. To conclude. the 160strike call will increase with maturity. This also helps to explain why the convexity has a greater effect on longer maturity variance. the cumulative distribution N(d2). The reason that there is any difference at all in variance swap strikes. Indeed. since the maximum volatility is at a strike of 130. rather than moneyness – but the point remains: the contribution of a point on the skew curve to the variance swap price depends on the (riskneutral) probability N(d2) of the associated OTM option ending inthemoney. That is. As maturity increases. T. variance will price at the same level as ATM volatility. the probability of far outofthemoney (in relative strike terms) options ending in the money increases. At 5 years. for example high put skews will increase the probability of OTM puts ending in the money (greater volatility on the downside makes OTM puts easier to reach) so both the volatility and the weighting factor will increase.g. but. See Gatheral 2006 for a detailed of the proof of this. Furthermore.e. With a negatively convex skew – OTM volatilities are (on average) less than ATM volatility – it is theoretically possible that variance could price below ATM volatility. ∫ ∞ −∞ N ′( z ) BS ( z )dz σ2 The N′(d2) term is the probability density function for the underlying at expiry. if the skew curve is quadratic in the variable z (the moneyness of the option) of the form: 2 σ BS ( z ) = σ 02 + αz + βz 2 then substituting and integrating gives K VAR = σ 0 T + β T . the convexity in Figure 107 is slightly greater.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging If we define the variable z to be the standard BlackScholes parameter d2: z = d 2 = It can shown that: K VAR = 2 ln (S K ) + r − σ 2 2 T ( ) σ T then. simply represents the ‘moneyness’ of the corresponding OTM option. 3. strikes out to say 50 180 become relevant to the pricing and the convexity of these strikes can be much higher. This explains the phenomena seen in Figure 106 and Figure 107: very different (linear) skews. Given a flat skew. 83 . but only on the base level of volatility σ0. Thus the parameter z. This means that the variance swap price is a weighted sum of squared option implied volatilities weighted by the probability that the (OTM) option will end inthemoney. At shorter maturities perhaps only the 80120 portion of the skew surface significantly affects the variance swap price. Therefore the relative weight of e. which it will do at strikes of 40 and 160 respectively. is because in d2space. which is slightly less outofthe money than the 70 strike where the maximum volatility is achieved in Figure 106. i. if we keep the ‘linear’ skew at a slope of 5% or 5% but do not apply the cap until it reaches 50%. 2.
The Greeks of the variance swap can then be calculated by differentiating Pt.T T T t −1 2 where σ0.T is the strike of a variance swap on day t expiring at T. the following parameters represent the forward Greeks and would have to be discounted by the appropriate discount factor. In analogy with the Greeks for vanilla options. t is the expected realised volatility. this gives a constant dollar gamma as expected (ignoring higher order errors). that since T is measured in days this value represents a daily theta.T 2TK 0 which can then be shown to satisfy the formula θ = − ΓS 2σ 2 taking σ2 as the implied variance from 1 2 the variance swap strike (see Box 16. 84 . An alternative approach would be to calculate the replicating portfolio and sum the Greeks of the contributing options.t1 is the realised variance accrued to day t1. We work directly by differentiating the marktomarket value of the variance swap contract.T = θ= ∂t 2TK 0 In particular. between trade inception and maturity. giving information about the sensitivity of variance swaps to various market variables. Annualising gives a value for theta of: θ = 252 − K t2. theta remains constant.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging 4. we return to considering realised volatility and variance swap strikes in percentage terms rather than the standard wholenumber units. p76). Note.t t −1 2 252 S t T −t 2 = σ 0. per unit veganotional. at time t is given by: Pt = 1 2 σ Expected . Note for simplicity of exposition we assume a world with zero interest rates. alternatively in the presence of nonzero rates. This expected volatility is then calculated in the same way as the marktomarket p/l as a combination accrued realised variance to date and future implied variance.t −1 + ln S + T K t . Note also that intraday there is a term representing the square of the move which will act to give the variance swap delta on an intraday basis: σ 2 Expected . The value of a variance swap.7: Variance swap Greeks In this section we consider the Greeks for variance swaps. at time t. if the variance strike doesn’t change. St1 is the value of the underlying at the close on day t1.t − K 02 2K 0 [ ] where K0 is the initial (fixed) variance swap strike and σExpected. Theta: Calculating the theta of the variance swap gives: 2 ∂P − K t . and Kt. and St is the value of the underlying at the valuation time on day t. Gamma: Gamma comes only from the exposure to realised volatility on each day: Γ= ∂ 2 P 252 1 = ∂S t2 K 0T S t2 Since the dollar gamma is achieved by scaling the gamma by the spot squared.
If we assume the Derman approximation: K t . Calculating delta (only on the close to avoid the intraday delta discussed above) gives: 85 . and do so by more.e. if there is more time remaining until expiry. the variance swap can be seen to take on delta only intraday: ∆= ∂P T − t 252 S t 1 252 S t 1 = ln = 2 ln T K 0 S t −1 S t K 0T S t −1 S t ∂S t This represents the replication of the log contract which will have to be done at the end of the day to capture that day’s realised variance. we must make some assumptions about how the variance swap strike relates to ATM volatility. using Derman’s approximation (see p30) K t .T = σ ATM 1 + 3(T − t )skew 2 2 ( 2 ) then: ∂P ∂σ ATM = σ 1 T −t ∂ T −t 2 σ ATM 1 + 3(T − t )skew 2 = ATM 1 + 3(T − t )skew 2 2 K 0 T ∂σ ATM K0 T ( ( )) ( ) Note also that this approximation allows the calculation of sensitivity to the skew or the skew squared: ∂P ∂P 3 (T − t ) 2 3 (T − t ) 2 σ ATM and σ ATM skew = = 2 T ∂skew K 0 2K 0 T ∂skew 2 2 i. These all tell us that the exposure to implied variance (or volatility) decreases with time as the accrued realised volatility is locked in to the p/l. For example.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging Vega: We can calculate exposure to volatility in terms of (i) sensitivity to changes in ATM volatility. If the variance strike itself has a dependency on the underlying (implied variance is directional) then the variance swap acquires other sources of delta.T T K0 ∂K t . assuming that the variance strike K has no sensitivity to the underlying.T = σ ATM 1 + 3(T − t )skew 2 2 ( 2 ) where σ ATM is the Black Scholes ATM implied volatility (or atthemoney forward volatility if rates or dividends are nonzero). in addition to the intraday delta discussed above. Delta: Firstly. (i) To compute the vega in terms of sensitivity to ATM volatility.T 2 K 0 T − t T . (Compare Figure 21) Computing sensitivities to the strike (or strike squared) is more straightforward and needs no such assumptions about the skew surface: (ii ) ∂P T − t K t . increasing skew will increase the value of the variance swap.T ∂P 1 and (iii) = = 2 ∂K t . (ii) variance strike or (iii) implied variance (strike squared).
∂S Therefore delta = 2 1 T − t 2 K t . Convexity: As explained above.T ∂K t2. this should make no difference to the variance swap strike. Stochastic Interest Rates: Our replication argument is constructed using forwards.T ∂P = × skew . If rates (and dividends) are deterministic.T ∂σ ATM = Which implies: σ ATM ∂S ∂S then ∂K t2. the probability of a large jump increases and this may be reflected in an increase in variance swap prices.T = 2σ ATM ∂σ ATM 1 + 3(T − t )skew 2 ∂S ( ) 2 K t2. If rates are uncertain. rising convexity will increase variance swap strikes as the ‘average’ implied volatility used in Gatheral’s formula goes up. Although. at least if the skew curve is sticky with strike. Clearly then.T σ ATM × skew since skew = ∂σ ATM is skew is linear. altering the variance swap strike. However. in the case of a singlestock where an M&A event is anticipated. In reality jumps (discontinuities) can happen. In reality the variance swap pays out on the volatility of the spot underlying. this will act to change the relationship between spot and forward volatility.T ∂S = 2 K t2. Except for very longdated variance the effect on variance swap prices should be small. ∂S 2 K 0 T σ ATM Thus with a ‘normally shaped’ negatively sloping skew. and in practice can offset the effect of jumps.T ) = ∂S t 2 K 0 T ∂S t 2 2 Then since K t . the delta of the variance swap will be negative.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging ∆= ∂P 1 T −t ∂ (K t2. This fits with the intuition that (implied) volatility will tend to go down as the underlying rallies – which is exactly what a (negative) linear skew represents. 86 .T = σ ATM 1 + 3(T − t )skew ( 2 ) we have ∂KS ∂ 2 t . it is really convexity and not skew which acts to increase the variance swap strike above ATM volatility. the error associated with this is of order jump (return) cubed and is therefore small enough in practice to be ignored. and the volatility associated with them is not guaranteed to be captured by the replication process. Other sensitivities to consider include: Jumps: The replication argument assumes a continuous diffusion process without jumps.
how many of the calls and puts do we need in the replicating portfolio? We know the relative proportions of the puts and calls but what is the absolute amount to hold? Let Π be forward of the price of the portfolio: P = e rT ∫ F0 0 ∞ C (K ) P0 (K ) dK + ∫ 0 2 dK 2 F0 K K The above portfolio (together with the dynamic futures position) will pay T 2 1 2 σ R at expiry. no net deltahedge is needed. the delta hedge will need to be adjusted such that. against the long portfolio of options. ) At inception if the cutoff between puts and calls is taken to be the current forward price of the underlying. 20% volatility expressed as 20) we must multiply by 1002. where σ R = T 2 ∫ T 0 σ t2 dt . Therefore the fair price for 2 ×1002/T lots of the portfolio P.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging 4. Thus at the end of each day’s close the notional value of the delta hedge is adjusted so that it equals 2 × 100 2 × N VAR T F0 − Ft F 0 where F0 was the original forward price at time 0. during the lifetime of the variance swap.g. the variance strike represents the fair cost of future realised variance. weighted according to 1/strikesquared. T equivalent to 2 × 100 2 × N VAR of each OTM Kstrike option. Scaling this portfolio by 2/T (which will cost 2Π/T) will produce a payout of realised volatility squared. and Ft is the current forward price. However.8: Setting up a replicating portfolio In the preceding sections we showed that a variance swap could be statically replicated by a portfolio of (outofthemoney) options. giving an overall p/l of N VAR × 2 (σ 2 2 − K VAR as required. but in percentage terms. paying out realised volatility (squared) in whole number amounts is given by: 2 K VAR = 2 × 100 2 × Π T 2 × 100 2 × N VAR Therefore. to be long NVAR of variance notional we must buy lots of the portfolio P. The value of the portfolio and its associated delta hedging at expiry will be N VAR × σ . To put this into the wholenumber units consistent with standard quoted variance swap strikes (e. T ×K2 2 This cost of this portfolio will therefore be N VAR × K VAR . How do we weight this portfolio to achieve a specific variancenotional or veganotional exposure? For example if we wish to replicate a variance swap for €100. there is a constant monetary (dollar) value of a long forward position but short a constant number of shares of a forward position. 87 .000 of vega notional. The (squareroot of) the cost of this portfolio.
3400…1200 and calls at 4000. 10 shares (or €10 per index point) then we need only 1/10 as many option contracts as we would otherwise. Each option represents €10 per index point. to compute the actual number of options required. 3600. Therefore we must divide the above amount by the contract size.g. so we now examine how to (approximately) construct a variance swap from a finite. the number of 3600 puts required is 2 × 10 9 2 × 10 9 = = 154. Derman (Derman.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging To achieve a specific vega notional. The replicating portfolio will consist of puts at strikes 3800. and numbers of options required for each strike are given in Table 13 below: Suppose the market rallies by 1% over the following day.500). In order to maintain a “flatdelta” for the overall portfolio. Since K VAR = 100 100 × N Vega 100 × N Vega 2Π the required amount of the portfolio is or of each (OTM) option. T 2Π T K 2 2Π T In practice it is not possible to trade a continuum of option strikes. The replicating variance swap portfolio is then given by ∑ ∆ K P (K i ) ∆ C (K ) n + ∑i = r +1 K 2 i approximating the true replicating portfolio: 2 i =0 Ki Ki r ∫ F0 0 ∞ C (K ) P0 (K ) dK + ∫ 0 2 dK F0 K2 K To get a variance notional exposure we will need 2 × 100 2 × ∆ K × N Var of each option (contract). 4400…6000. Example 20: Constructing a variance swap Suppose we are replicating a 6month maturity Euro Stoxx 50 variance swap (variance notional of €2. T ×K2 There is a slight complication here with contract sizes. 4200. This kind of approximation slightly overvalues the true theoretical variance swap strike due to convexity issues. we will need to sell on the close the following notional size: 2 × 100 2 × N VAR 2 × 100 2 × 2. and all subsequent options calls. Suppose we have available liquid OTM options of strikes K0…Kr. tradable strip of OTM options. NVAR=NVega/2KVAR we must trade 100 2 × N Vega T × K VAR lots of the portfolio P. If one option contract represents e.5 × K 2 × 10 K2 For example. Suppose the forward F0 such that Kr< F0 ≤ Kr+1 and suppose that the options are equally spaced so that Kr+1Kr+1= ∆K for all i.5 T 88 . Kr+1…Kn with all options up to Kr puts.3 3600 2 K2 The prices. Strikes are 200 ip apart and the forward is 3868. 1999) discusses this effect and also presents a robust method for using whatever available strikes are to replicate a variance swap.500 2 × 10 9 No of contracts = = = T × K 2 × Contract _ size 0. we assume the strikes are equally spaced. How much of each option contract is needed? 2 × 100 2 × ∆ K × N Variance 2 × 100 2 × 200 × 2.500 1% = 1% = €1M notional 0. For simplicity.
56E11 4.23E11 0.40E03 6.60E04 9.01E06 6. Supposing the market falls significantly. while the options should counteract this by being short the log contract. especially for puts. traded strikes are not continuous. The problem with the lack of OTM puts can be seen from following through the practical example of setting up the replicating portfolio above.143 140.20E06 7.027 4. the deltahedge will be long the log contract (a will hence lose).08E04 2. although for major liquid indices they are closely spaced (0.46E10 6.777 26.69 0. as these provide a relatively large component of the variance swap price in the presence of steep put skews. In opposition the long options/short forward position is used to create a short log contract and pay the fixed strike.18E06 2. 1.968 83.165 12.195 1.164 11. premium x bucket size/strike2 8.696 599 116 20 3 0 0 0 692.054 4.75E05 6.04E07 3.47E09 9.14E13 9.18E05 1.99E03 1.31 16.10E05 Fwd. In practice.145 0. Option strike 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000 3200 3400 3600 3800 4000 4200 4400 4600 4800 5000 5200 5400 5600 5800 6000 Price of one portfolio Π Option type Spot premium (€) 5.00E05 1.64E04 1. a variance swap can be statically hedged with a portfolio of outofthemoney (Europeanstyle) options. variance notional €2.31E04 8. FTSE to 3525 (c60% of spot) and Euro Stoxx down to 600 (c15% of spot).932 103. to calculate the fair value of a variance swap. However.80E03 0.393 52.31E03 1.90E06 3. weighted according to the inverse squares of their strikes.9: Replication and hedging in practice As demonstrated above.90E12 5. in theory. although in reality.10E05 1.08E08 4.18E08 1. assuming option prices are available across the entire range of strikes. 4.46E04 1.4% notional apart for the S&P.483 40. The long futures position is used to create a payout which is equivalent to a long log contract plus realised variance.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging Table 13: Replicating portfolio for a 6month Euro Stoxx variance swap.118 10.90E04 5.292 2.538 12.899 2.25E04 2.000 (2 x 1002 / T) Square rooted gives…. 1% for the FTSE.354 45.4% for the Euro Stoxx).80 # Option contracts 1389 1020 781 617 500 413 347 296 255 222 195 173 154 139 125 113 103 95 87 80 74 69 64 59 56 Price of replicating portfolio for €2.500 of variance notional Price of options (€) 0 0 9 100 675 3. This lack of liquidity at the wings has led to the development of conditional variance swaps which can remove exposure to volatility once the underlying moves into areas where vanilla options are illiquid. the options portfolio will not fully reflect the short logcontract exposure needed and hence the overall hedge will lose money.32E03 4.197 129.074 PUTS CALLS x 40. S&P options are listed down to a strike of 600 (c45% of current spot).00706 282. A more serious limitation is the lack of liquidity in OTM strikes.041 82.500.11E06 1. liquidity does not even reach this far.36E16 6. This makes it easy.645 47.308 195.29E09 5.853 0.90E09 1. 89 . if not enough downside puts were used.624 28.989 128.024 0.
7) and hence are negligible for all but very large moves. In practice. if the marketmaker is long options and deltahedging.10: Effects of variance swap hedging Marketmakers who trade variance swaps may hedge their positions by replicating the opposite variance swap position through the replicating options portfolio.5 vegas would be typical. but liquid. 85% for 23 years etc. This basis tends to vary across maturities.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging In practice. Similarly.7. see Section 4. If the underlying rallies. Note that no exchange of options has taken place here – the parties have just taken opposite sides in a contract for difference. Another limitation comes from the discrete nature of adjusting the delta hedge on the close. the market maker will have to trade further OTM puts to ensure that their exposure to volatility. Alternatively they could approach the replicating by hedging the vega with an OTM put whose implied volatility coincides with the variance swap strike – close to the money for 1month maturity. the options must be deltahedged on the close only to capture this. We investigate both of these properties below. The actions of deltahedging the options could potentially act to the disadvantage of the counterparty’s position. and the marketmaker will therefore have to buy more of the underlying in the market. and rehedging by trading more options if the underlying does move significantly. But assume that marketmakers hedge their short volatility 90 . This replicating portfolio then needs to be delta hedged. In both cases – if the position being hedged is big enough – the action of deltahedging will have the effect of increasing volatility in the underlying: magnifying both upmoves and downmoves. in the form of dollargamma. this can also have an impact on the ability to accurate replicate the logcontract and realised volatility This and other differences between the theoretical and midmarket price of a variance swap contribute to the socalled variance swap basis. see section 4. For example if the market sells off towards the strike. In this case they would also look to buy back the wings/convexity separately. being larger for longer maturities. the shortdelta exposure from the short option will increase. The difference from this truncated and the full replication pricing is typically in the region of 525bps. The marketmaker will be short the option (and delta) and will therefore buy back the delta in the underlying. At shorter dates a value of around 0. on the Euro Stoxx (6month maturity) this would only price puts down to 2700 (ref 3822) or around 70% of spot. remains constant. Here the constant dollar gamma would be maintained by a combination of holding a portfolio which has roughly constant dollar gamma if the underlying does not move too much. However these moves are actually of order the cube of the move (as in jumps for the continuous case. which introduces possible errors due to large daily moves. Since variance swap contracts typically measure closeclose realised volatility. or other market participants. One problem with this kind of approach is that the partial hedge is no longer static. such as CDS or EDS could also be used instead of the more deeply outofthemoney puts. Other proxies. pricing models will often only price contributing options for the variance swap hedge between 1 and 99 delta. consider a marketmaker who has sold a call option to a counterparty. marketmakers will not attempt to hedge with the entire strip of options but typically will use only two or three – including one close to the money and one or more OTM. 90% for 1year. Suppose that the volatility sellers do not hedge their variance swaps (they have sold the variance swaps specifically for the direct volatility exposure they offer). puts. the hedging will act to suppress volatility in the underlying. The argument for a put option is analogous. 95% for 36 months. potentially to the advantage of the counterparty who is short the option. 4. Similarly. Suppose that the market is such that hedgefunds. if the underlying sells off. Also as previously noted if interest rate changes are related to changes in the spot underlying. The situation with variance swaps is different. For example. and must be dynamically managed. This makes the actual variance swap replication more akin to a combination of alternatives 2 and 3 listed on p79. Firstly. The effects of deltahedging this portfolio are different to that from normal deltahedged options for two principal reasons: 1. the marketmaker will have to sell the underlying. 2. have generally sold index variance swaps to marketmakers (as was the case before the May 2006 correction).
this only becomes an issue if hedgers (variance swap marketmakers) are the same way round and daily moves are large enough. they are sold on to endinvestors). although it may if the long options positions are not being hedged (e. 91 . This then drives the underlying market down further and could increases the amount that the variance swap hedgers have to sell on the close. was blamed for the amplification of some already large daily moves. and deltahedging. the important point is that the hedging of long variance swap positions may act to increase closetoclose volatility. They will therefore be short gamma in the options market (Figure 108). and anticipation thereof. However. the action of these market makers hedging their short options will not necessarily act to increase volatility in the underlying. must hedge only on the close to capture the closeclose realised variance specified in the variance swap contract. In contrast. such effects should be propagated back through the trading day. and may be offset by intraday deltahedging of opposing options. However. These feedback effects were much talked about during the volatile market period in May 2006. the hedgers who are long the options.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging exposure. a deltahedger who is short options will act to increase volatility in the underlying – buying as it rallies. As described above. in an efficient market. as they attempt to capture the true volatility of the underlying process. In fact. will generally be free to choose when to deltahedge. For instance. the important difference between these two groups of hedgers is that the variance swap marketmakers who are short options.g. with option hedges on the close having the potential effect of magnifying daily moves. Figure 108: Flows in the market as a result of a marketmaker buying a variance swap and replicating it in the market Fixed Payment (implied variance = strike2) Variance swap seller Realised variance Market maker (variance swap buyer) Sell portfolio of options Option premia Options Market Intraday delta. In fact the expectation of these deltahedging flows could itself act to further amplify moves into the close through possible ‘feedback’ effects. as the counterparties they have sold options to may be counteracting this effect by themselves hedging their long volatility positions. A marketmaker who is long the variance swap can offset the risk by shorting the replicating portfolio of options. the overall effect of hedging these variance swaps need not have the effect of increasing overall market volatility. on big down days. In practice. when the hedging into the close of long variance swap positions. and selling as it sellsoff.hedging (buy low / sell high) Stock Market Maintain constant $value short position in underlying (buy high / sell low on the close) Therefore. the market could anticipate a selloff into the close as a result of long variance swap hedging and take short positions into the close to profit from this anticipated effect. However.
Once the replicating portfolio of options has been put in place. is actually an exposure to the difference between implied and realised volatility squared. but still does not explain why volatility swaps are not also frequently traded. In this framework. the ability to calculate a fair price for volatility swaps. the bigger the discount of the dynamically replicated volatility swap to the variance swap. Thus for small changes in volatility. Volatility swaps can then be thought of as variance swaps without the convexity. In fact the convention of quoting variance swap notionals in vega. and lose by less if volatility decreases. rather than variance. see discussion on variance swap marktomarket. where the effect of the variance swap convexity is relatively limited. The main theoretical difficulty with volatility swaps is that they cannot be statically replicated through options. For example. in terms of volatility products. requires not only a price for the volatility (variance). variance should be thought of as the true underlying. variance swaps (measured in vega notional) locally approximate volatility swaps.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 4: Replication and Hedging 4. As pointed out above. variance is in some sense more fundamental – especially because it is additive. products which payout linearly on the difference between implied and realised volatility? If these are more easily understandable why are they not regularly traded? Whilst volatility can be seen as more of an intuitive measure (being a standard deviation it is measured in the same units as the underlying). variance swaps are relatively easy to replicate. Similar to deltahedging an option the p/l made from the resulting buy low – sell high strategy (for variance) will lead to a p/l based on the volatility of volatility: the larger this volatilityofvolatility. volatility swaps are naturally thought of as derivatives of variance – paying the squareroot of the variance swap contract. estimating this volatilityofvolatility contribution from the dynamic hedging of the variance swap is problematic and modeldependent. No further buying or selling of options is necessary.11: Why not volatility swaps? A common complaint about variance swaps is that they payoff based on realised variance (volatility squared) and not simply realised volatility. making the volatility swap contracts difficult to price. Seen in this light. which in turn is determined by the volatility of volatility (see Section 2. See Derman. A replicating portfolio must dynamically trade options. 1999 for further details about construction volatility swaps. All this explains why variance swaps are attractive instruments to trade.g. In this sense. However. but with the convexity meaning that longs will profit by more if volatility increases. only deltahedging is required. The discount of the volatility swaps to variance swaps should therefore reflect the value of this convexity. This linearity of variance makes calculating variance swap marktomarkets and forwards particularly simple.8). This makes any replication process model dependent. Why are there not volatility swaps. and make relatively strong assumptions about the underlying volatility process – in particular about the volatility of volatility. p15). The vega notional represents the average p/l for a 1 vega change in volatility.e. besides issues arising from the transaction costs of dynamically trading variance swaps. a variance swap mirrors a kind of ideal deltahedged option whose gamma remains constant. the means of valuing options on volatility. the exposure of deltahedged options to volatility. 92 . the variance of the sum of two independent distributions is just the sum of each of their variances (e. Since it is variance (volatility squared) which arises naturally from deltahedging options. In fact we could dynamically trade a long variance swap (buying more as volatility decreases and selling as volatility increases) to hedge out this convexity bias. but also a price for the volatility of volatility – i. and therefore much more prone to errors than the theoretically robust variance swap replication. after accounting for the gamma. Furthermore. amounts can be seen as an attempt to treat variance swaps like volatility swaps.
5. Future Developments • • • • • • • Part 5: Future Developments 93 Variance swap indices Variance swap fixings Benchmark volatility return index Exchangetraded variance swaps 3rd generation products Derivatives on variance Variance swaps on other asset classes .
Given the typical bid/offer on equity options. This also makes the indices directly comparable. VSTOXX for Euro Stoxx 50.cboe. the width of the arbitragecorridor of trading a variance swap against a portfolio of options can give rise to a fairly large margin of error. Variance swap indices The volatility indices associated with the major indices (VIX for S&P500. VXN for the Nasdaq. The increasing demand to sell index variance and correlation over the last few years. However. VDAX for DAX and VSMI for the SMI) are now calculated according to a variance swap style calculation. theoretical variance swap levels are first calculated for listed option maturities.VXN . This ‘basis’ between the volatility index and actual traded (mid) variance swap levels is usually about 0.com VDAXNEW www.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 5: Future Developments Part 5: Future developments With the increasing liquidity and breadth of the variance swap market. and then the 30day index level is interpolated. as equity volatility has fallen.V2TX .V1X . other developments are underway which will in turn help both the liquidity and transparency of the variance swap market.com All these volatility indices are well correlated with traded variance swap levels on their respective indices.deutscheboerse.stoxx.VIX . but not to the extent at which an arbitrage has been possible. Table 14: Main variance swap based volatility index details VIX VXN VDAXNEW VSTOXX VSMI Source: JPMorgan Underlying S&P 500 Nasdaq 100 Dax 30 Euro STOXX 50 SMI Reuters . apart from the obvious benefits of reducing transaction costs and the ability to trade variance on more stock and index underlyings.com VSMIIDX www.com CBOEVXN www. 94 .swx. based on traded option prices. In fact. However this can vary much more widely during volatility spikes when supply and demand issues become important. As such this gives the indices a real interpretation: each index represents the riskneutral expected variance of the underlying over the next month.cboe.51 vegas. the level of the volatility index is typically somewhat above the actual traded levels for variance swaps of the same maturity. There are two reasons for this: • • A convex skew should lead to midpoint discrete strike variance swap approximation (as in the VSTOXX index calculation) overestimating the level of implied variance. They represent the theoretical level of a rolling 1month (30 calendar day) maturity variance swap.com VSTOXXI www. helped to cheapen index variance to its theoretical value. We highlight a few examples of recent developments.VSMI Bloomberg VIX Index VXN Index V1X Index V2X Index V3X Index Datastream Calculation details CBOEVIX www. such as: • • • • • • • Variance swap indices Variance swap fixings Benchmark volatility return indices Exchange traded variance swaps 3rd generation products Derivatives on variance Variance swaps on other asset classes.
these expire 30 days before an index expiry and are listed to expire 30 days before the 95 . longer dated (e. While this is to some extent true. VDAX and VSMI indices. Benchmark volatility return index With the availability of tradable and transparent variance swap levels for different maturities. the Dec06 listed options have exactly 30 calendar days remaining maturity and the VSTOXX calculation does not need to interpolate from any other maturities. For example.5). longer dated variance swaps could themselves be easily constructed from a strip of forward variance swaps. As noted above there exists a basis between the actual variance swap level and theoretical levels calculated from the portfolio of options. Thus the futures contract p/l is based on the change in expected future volatility. For example a Nov06 VXTOXX future expires on 15th November. In April 2005.g. since they are margined linearly in the level of forward starting volatility (See VDAXNEW. VSTOXX and VSMI Futures.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 5: Future Developments These volatility indices. Variance swap fixings One of the biggest deficiencies of using the portfolio replication method to calculate a variance swap level is its lack of tradability. see the variance component of JPMorgan’s Yield Alpha product. A way to capture a more tradable level for variance swaps is to construct a brokerdealer poll. spot/forward. the shortdated nature of these variance swaps means the principal driver of the volatility index level is recent realised volatility (see Section 2. In reality. each day at 11. but 30 calendar days beforehand. having an independent method of valuing these instruments at realistic tradable levels will become more desirable to provide transparent marktomarket valuations. This basis exists probably due either to differences in marketmakers’ volatility skews or in the supply and demand from end users. For example. This was followed in September 2005 by the launch of futures on the European VSTOXX. skew levels or even ratios of put to call openinterest would perhaps be a better proxy for the level of riskaversion present in the market. as described in JPMorgan’s Yield Alpha Strategy Guide. spreads of implied to realised variance. options on the VIX index were launched. These futures do not expire on the normal index (futures) expiry dates. As for the futures. With structured products increasingly incorporating payoffs derived from variance swaps. Note that there is a convexity bias inherent in marking to market these futures. Though there will still need to be some development of the strategies involved (i. transparent benchmark total return volatility indices can be constructed using a similar methodology to that described in Section 3. paying out on the value of the relevant volatility index at expiry. Exchangetraded variance Following the change in calculation method of the VIX to the variance swap calculation. 1year) variance. it will aid the development of volatility funds in countries where an appropriate benchmark is necessary. maturity etc). It could be envisaged that the 3month variance swap could itself become a similar benchmark for volatility.e. Moreover. long/short.00am an interest rate fixing for different maturity deposits is made using a trimmed mean average of several banks’ contributed rates. are often referred to by those outside the equity derivatives space as a ‘market price of risk’ index. Expiry on 15th November is chosen because on that date. the 3. 30 days before the 15th December options expiry.3. not on realised volatility or the spread of realised to implied volatility. Most notably. the CBOE launched futures contracts on the VIX index in March 2004. Note that these futures contracts are forward volatility products. JPMorgan September 2005). due to their visibility and wide availability. These represented the first available exchange traded options on variance. in a similar way to that used in interest rates markets. a value which is itself an implied variance.and 6month fixings are then often used as reference rates which form the basis of many longer dated instruments.
so the maximal loss for a long will be when the underlying trades within the range but with very low volatility. the listed VIX options are for Nov06.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 5: Future Developments corresponding quarterly options expiry dates for the underlying. Whilst investors are free to specify the range associated with a conditional variance swap. allowing investors to take exposure to realised variance contingent upon the underlying instrument trading within a prespecified range. rather than a constant dollar gamma. As with variance swaps. More recently. downvariance will normally price above upvariance for close to ATM barrier levels. Dec 06. In addition. Outside this range no p/l accrues. In the case of the conditional. 3rd generation volatility products Generalised variance swaps or “3rd generation products” include gamma swaps. Feb07. no lower barrier). Figure 109 : Payout from a conditional variance swap @ 20 strike p/l. Conditional variance swaps have been the most popular (and liquid) of these products. Conditional variance swaps can be useful for hedging complex volatility exposures. different lines for % days in the range 30% 0% Figure 110 : Dec07 conditional variance swaps levels as of Apr 06 Strike 30% 25% 20% 15% 10% conditional dow n v ar conditional upv ar v ariance 20% 10% 0% 10% 20% 0% Source : JPMorgan 25% 50% 75% 100% 80% 10% 20% 30% volatility realised in the range 40% 90% 100% Up/down barrier 110% 120% Source : JPMorgan. no upper barrier). taking a view on the volatility levels encapsulated in the skew or buying/selling variance at more attractive levels given a view on the underlying. but can not trade OTC products directly. The original corridor variance swaps only accrue variance within a prespecified range.g.e. p/l is accrued only when the underlying is within a prespecified range.e. the two principal types are upand down. beginning September 2006. the sign of the p/l for a conditional variance swap is controlled by the difference between the volatility realised (in the range) over the lifetime of the swap and the preagreed fixed strike level. In the presence of a positive putskew. the p/l will be zero. corridor swaps and conditional variance swaps. in November 2006. meaning a long would make a maximal loss when the underlying fails to trade within the range. Gamma swaps are very similar to variance swaps but maintain a dollar gamma linear with spot. Euronext announced the ability to list variance swaps on a limited number of indices through its BClear mechanism. May07 and Aug07 expiries.conditional variance swaps (upvariance and downvariance). Upvariance accrues realised volatility only when the underlying is above a prespecified level (i. leading to lower a skew exposure and a slightly lower strike compared to standard variance swaps. data as of Apr 2006 96 . This can be useful for end users who wish to take a volatility view with variance. the magnitude of the final p/l is scaled by the proportion of time the underlying has spent in the prespecified range. while downvariance is accrued only when the underlying is below the specified barrier (i. If the underlying trades entirely outside the range. E.
as we have highlighted throughout the note. For example volatility tends to be a meanreverting process characterised by fairly long periods of subdued volatility with intermittent spikes of high volatility. p/l is zero. volatility is a process very different from the underlying instrument itself. bonds have a defined maturity. it can be troublesome for commodities which tend to be based upon a rolling forward price. which would have to take into account the convexity of bond prices to yields. the development of a liquid market in derivatives based on variance becomes more plausible. and at least in some cases the desire for investment banks to recycle exotic and structured product risk. This would add additional complexity in constructing a replicating portfolio. unlike equities. One issue for variance swaps on other asset classes is the dependence of the payoff on movements in the price of a spot underlying. For example. In fact. Nonetheless.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Part 5: Future Developments Conditional variance swaps can be useful for expressing views on volatility contingent on market level. Conditional variance swaps have typically been traded on index underlyings. for example: interest rate options. contingent on stock price. variance swaps have yet to really take hold in other asset classes. driven by both investor demand to trade pure volatility. is that exposure to the underlying can be removed through deltahedging. allowing investors to trade the volatility of a stock. outperformance options etc. Furthermore. Nonetheless. April 2006 for further details. variance swaps do trade on assets such as commodities. credit options. is introduced in addition to implied volatility. However. This is because the ability to use variance swaps themselves to deltahedge variance derivatives becomes easier as liquidity increases and transaction costs decrease. 97 . using the simple deltahedging replication option pricing schema that is basis of BlackScholes becomes an impractical model to use in this case. Derivatives on variance look to be no different. commodity variance swaps are usually referenced to a futures contract of the same maturity as the swap. If a non BlackScholes approach is used another risk factor. Moreover in fixed income volatility based on yields is often more useful. Conversely investors who believe volatility will be realised on the upside can buy conditional upvariance swaps. The key principle behind BlackScholes. While this is appropriate for equities. See Conditional Variance Swaps. For example investors seeking crash protection may purchase conditional downvariance. we expect variance swap markets to continue to develop on other assets. particularly commodities and FX. FX and interest rates. such as the marketpriceofrisk. singlestock conditionals have gained liquidity. if the market stays above the downbarrier. options on variance now trade in the market as investors look to tailor their volatility exposures. Variance swaps on other asset classes While variance swap liquidity has increased massively on equity underlyings over the last few years. Nonetheless. leaving behind the sole residual risk of exposure to the volatility of the underlying. this problem of modelling the underlying process has failed to halt the development of derivatives markets where the BlackScholes approach is also considered invalid. which are usually cheaper than a standard variance swap for which a significant amount of the premium is used to fund the downside skew exposure. More recently. Derivatives on variance Obviously with the increased liquidity and breadth of the variance swap market. Hence. which leads to problems defining a meaningful payoff for longer dated derivative products. which only becomes activated in the event of a market selloff. That is.
Peter Carr & Dilip Madan . John C. May 2005 European Equity Derivatives Weekly Outlook . April 2006 Correlation Vehicles. September 2005 Yield Alpha Strategy Guide. July 2006 98 . various publications Introducing JPMorgan’s CrossSection Volatility Report. Hull. VSTOXX and VSMI Futures. October 2006 VDAXNEW. Jim Gatheral.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 References References General References More Than You Ever Wanted to Know About Volatility Swaps. Futures and Other Derivatives. 2003 The Volatility Surface. Demeterfi. February 2005 Synthetic Short CDS with Positive Carry.2002 JPMorgan Publications Calling All Overwriters. 2006 Towards a theory of Volatility Trading. Kamal & Zou. 1999 Options. Derman. August 2002 Conditional Variance Swaps. Fifth Edition. November 2006 Just What You Need to Know About Variance Swaps.
The buyer of higher ratio put spread has unlimited downside below the lower strike (down to zero). By selling a covered call spread. they give up any stock appreciation above the strike price of the call option. Locks in the amount that can be realized at maturity to a range defined by the put and call strike. and investors risk losing 100% of the premium paid if the stock is below the strike price of the call option. Butterfly spreads are established at a net debit. Call Overwrite or Buywrite. One a bull spread and the other a bear spread. Butterfly Spread. and are not indicative trading levels Risks to Strategies: Not all option strategies are suitable for investors. Investors. Call Purchase. investors risk losing 100% of the premium paid. Investors who own the underlying stock will have downside protection between the higher strike put and the lower strike put. Call Spread. A butterfly spread consists of two spreads established simultaneously. We have summarized the risks of selected derivative strategies. please call your sales representative for a copy of "Characteristics and Risks of Standardized Options". the investor will profit if the underlying is stable. The gain is limited to the spread between the two strike prices. When overlayed onto a long stock position. 99 . should the stock price fall below the strike price of the lower strike put. Call Sale. Put Spread. In a selloff. dependent on the number of lower struck puts sold. Put Purchase. We advise investors to consult their tax advisors and legal counsel about the tax implications of these strategies. The seller of a straddle or strangle is exposed to stock increases above the call strike and stock price declines below the put strike. upside losses are capped (as for a covered call). and they will not participate in any stock appreciation if the option expires unexercised. Options are a decaying asset. Investors who sell uncovered call options have exposure on the upside that is theoretically unlimited. The seller of a call spread risks losing an amount equal to the spread between the two call strikes less the net premium received. the investor gives up all appreciation in the stock above the strike price of the call option. when the underlying is at the lower strike. Since investors are selling a call option. In a rally. However. The maximum gain is limited to the spread between the two put strikes. therefore. Pricing Is Illustrative Only: Prices quoted in the above trade ideas are our estimate of current market levels. The resulting position is neutral. option losses are potentially unlimited as the investor is net short a call. Investors who sell call options against a long position in the underlying stock give up any appreciation in the stock price above the strike price of the call option. Having sold a covered call option. and investors risk losing 100% of the premium paid if the stock is above the strike price of the put option. Covered writers are exposed to declines in the long stock position as well as any additional shares put to them should the stock decline below the strike price of the put option. the maximum realised downside potential of a doubleup booster is the net premium paid. will be exposed to any decline in the stock price below the strike potentially to zero. The buyer risks losing 100% of the premium paid. Since exposure on the upside is theoretically unlimited. the maximum loss is the net debit. Options are a decaying asset. that is. For additional risk information. Straddle or Strangle. but downside losses are not.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Risks of common option strategies Please also refer to option risk disclosure documents Put Sale. investors regain exposure to the underlying stock. investors who also own the stock would have limited losses should the stock rally. the investor remains exposed to the downside of the stock and gives up the spread between the two call strikes should the stock rally. Investors who sell put options will own the underlying stock if the stock price falls below the strike price of the put option. The buyer of a put spread risks losing 100% of the premium paid. and they remain exposed to the downside of the underlying stock in the return for the receipt of the option premium. Collar. certain strategies may expose investors to significant potential losses. and this exposure is multiplied by the number of puts sold. The maximum profit will occur at the middle strike price. If the collar is not costless. Booster.
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 100 .
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 101 .
Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Analyst Certification: The research analyst(s) denoted by an “AC” on the cover of this report certifies (or. Morgan Futures Inc. Other Disclosures Options related research: If the information contained herein regards options related research. JPMorgan Equity Research Ratings Distribution. a Market Participant with the ASX) (JPMSAL) are licensed securities dealers.K. is a member of FDIC and is authorized and regulated in the UK by the Financial Services Authority.P. Research is available at http://www. South Africa: J. Morgan Australia Limited (ABN 52 002 888 011/AFS Licence No: 238188.P.P. Hong Kong: J. 102 . Morgan Securities Ltd. we expect this stock will outperform the average total return of the stocks in the analyst’s (or the analyst’s team’s) coverage universe. among other business units.com/publications/risks/riskstoc. is a member of the NFA. 2006 Overweight (buy) 42% 44% 38% 62% Neutral (hold) 41% 46% 48% 55% Underweight (sell) 17% 35% 15% 47% JPM Global Equity Research Coverage IB clients* JPMSI Equity Research Coverage IB clients* *Percentage of investment banking clients in each rating category.P. J. and overall firm revenues.: J. (JPMSL) is a member of the London Stock Exchange and is authorised and regulated by the Financial Services Authority. regulated by ASIC) and J. is.P. Morgan Securities Australia Limited (ABN 61 003 245 234/AFS Licence No: 238066. which include revenues from. Morgan Chase Bank. Seoul branch. Analysts’ Compensation: The equity research analysts responsible for the preparation of this report receive compensation based upon various factors.P. Morgan India Private Limited is a member of the National Stock Exchange of India Limited and The Stock Exchange. Mumbai and is regulated by the Securities and Exchange Board of India. our Overweight rating falls into a buy rating category. Taiwan: J. Institutional Equities and Investment Banking. such information is available only to persons who have received the proper option risk disclosure documents. client feedback. Australia: J. For purposes only of NASD/NYSE ratings distribution rules.] Neutral [Over the next six to twelve months. J.morganmarkets. Korea: J. Legal Entities Disclosures U. U. Thailand: JPMorgan Securities (Thailand) Limited is a member of the Stock Exchange of Thailand and is regulated by the Ministry of Finance and the Securities and Exchange Commission. Morgan Securities Indonesia is a member of the Jakarta Stock Exchange and Surabaya Stock Exchange and is regulated by the BAPEPAM. Indonesia: PT J. as of September 29. and (2) no part of any of the research analyst’s compensation was. is regulated by the Korea Financial Supervisory Service. NASD and SIPC.] The analyst or analyst’s team’s coverage universe is the sector and/or country shown on the cover of each publication. See below for the specific stocks in the certifying analyst(s) coverage universe. Morgan Securities (Asia Pacific) Limited (CE number AAJ321) is regulated by the Hong Kong Monetary Authority and the Securities and Futures Commission in Hong Kong. or will be directly or indirectly related to the specific recommendations or views expressed by the research analyst(s) in this report.optionsclearing. Valuation and Risks: Please see the most recent JPMorgan research report for an analysis of valuation methodology and risks on any securities recommended herein.pdf.P.S. Important Disclosures Explanation of Equity Research Ratings and Analyst(s) Coverage Universe: JPMorgan uses the following rating system: Overweight [Over the next six to twelve months. Morgan Securities (Far East) Ltd. including the quality and accuracy of research. we expect this stock will perform in line with the average total return of the stocks in the analyst’s (or the analyst’s team’s) coverage universe. and our Underweight rating falls into a sell rating category. we expect this stock will underperform the average total return of the stocks in the analyst’s (or the analyst’s team’s) coverage universe.P.com .: JPMSI is a member of NYSE. N. our Neutral rating falls into a hold rating category. the research analyst denoted by an “AC” on the cover or within the document individually certifies.P. India: J.P. where multiple research analysts are primarily responsible for this report. please contact your JPMorgan Representative or visit the OCC’s website at http://www. Morgan Equities Limited is a member of the Johannesburg Securities Exchange and is regulated by the FSB. competitive factors.A.P. or you can contact the analyst named on the front of this note or your JPMorgan representative.] Underweight [Over the next six to twelve months. For a copy of the Option Clearing Corporation’s Characteristics and Risks of Standardized Options. with respect to each security or issuer that the research analyst covers in this research) that: (1) all of the views expressed in this report accurately reflect his or her personal views about any and all of the subject securities or issuers.Morgan Securities (Taiwan) Limited is a participant of the Taiwan Stock Exchange (companytype) and regulated by the Taiwan Securities and Futures Commission.
N. which is regulated by the Japan Financial Services Agency (FSA). Morgan Securities Singapore Private Limited (JPMSS) [mica (p) 069/09/2006 and Co.Peter Allen (4420) 73254114 Stephen Einchcomb (4420) 73259064 Nicolas Granger (4420) 73257033 European Equity Derivatives Strategy 17 November 2006 Philippines: This report is distributed in the Philippines by J. No. The recipient of this material must not distribute it to any third party or outside New Zealand without the prior written consent of JPMSAL. Japan: This material is distributed in Japan by JPMorgan Securities Japan Co. Past performance is not indicative of future results..com. Inc. New Zealand: This material is issued and distributed by JPMSAL in New Zealand only to persons whose principal business is the investment of money or who. Brazil: Banco J.S. Frankfurt Branch who are regulated by the Bundesanstalt für Finanzdienstleistungsaufsicht. or needs and are not intended as recommendations of particular securities. the report has been issued to persons regarded as professional investors (or equivalent) in their home jurisdiction Germany: This material is distributed in Germany by J. is regulated by the Comissao de Valores Mobiliarios (CVM) and by the Central Bank of Brazil. Morgan Malaysia Sdn Bhd) which is a Participating Organization of Bursa Malaysia Securities Bhd and is licensed as a dealer by the Securities Commission in Malaysia Country and Region Specific Disclosures U. affiliates and accepts responsibility for its contents. Australia: This material is issued and distributed by JPMSAL in Australia to “wholesale clients” only. All rights reserved. This report has been issued in the U.A.htm. This material is not intended as an offer or solicitation for the purchase or sale of any financial instrument. This document must not be acted on or relied on by persons who are not relevant persons. (For research published within the first ten days of the month.. 2006. JPMSAL does not issue or distribute this material to “retail clients. All pricing is as of the close of market for the securities discussed.hkex.P. JPMSAL does not issue or distribute this material to members of "the public" as determined in accordance with section 3 of the Securities Act 1978. Singapore: JPMSI and/or its affiliates may have a holding in any of the securities discussed in this report.: 199405335R] which is a member of the Singapore Exchange Securities Trading Limited and is regulated by the Monetary Authority of Singapore (MAS) and/or JPMorgan Chase Bank. Morgan Securities Philippines. Periodic updates may be provided on companies/industries based on company specific developments or announcements. 103 . in the course of and for the purposes of their business.K. for securities where the holding is 1% or greater. In other EEA countries.) J. Ltd. 38. and the EEA by JPMSL. Clients should contact analysts and execute transactions through a JPMorgan subsidiary or affiliate in their home jurisdiction unless governing law permits otherwise. An updated list can be found on HKEx website: http://www. Hong Kong: The 1% ownership disclosure as of the previous month end satisfies the requirements under Paragraph 16. market conditions or any other publicly available information.com/pdfdoc/research/ConflictManagementPolicy. Korea: This report may have been edited or contributed to from time to time by affiliates of J.K.S.P.P. the disclosure may be based on the month end data from two months’ prior. Frankfurt Branch and JPMorgan Chase Bank. only to persons of a kind described in Article 19 (5). This report or any portion hereof may not be reprinted.5(a) of the Hong Kong Code of Conduct for persons licensed by or registered with the Securities and Futures Commission. the specific holding is disclosed in the Legal Disclosures section above. Reg. Seoul branch.P. and European Economic Area (EEA): Issued and approved for distribution in the U. Copyright 2006 JPMorgan Chase & Co. Any investment or investment activity to which this document relates is only available to relevant persons and will be engaged in only with relevant persons. Revised September 30.hk/prod/dw/Lp.” The recipient of this material must not distribute it to any third party or outside Australia without the prior written consent of JPMSAL. Information has been obtained from sources believed to be reliable but JPMorgan Chase & Co. objectives.A. General: Additional information is available upon request. For the purposes of this paragraph the terms “wholesale client” and “retail client” have the meanings given to them in section 761G of the Corporations Act 2001.. Morgan Securities (Far East) Ltd. Morgan S. Malaysia: This material is issued and distributed in Malaysia by JPMorgan Securities (Malaysia) Sdn Bhd (18146x) (formerly known as J. N..pdf. Morgan Broking (Hong Kong) Limited is the liquidity provider for derivative warrants issued by J. Singapore: This material is issued and distributed in Singapore by J. Morgan International Derivatives Ltd and listed on The Stock Exchange of Hong Kong Limited.P. The opinions and recommendations herein do not take into account individual client circumstances.P. India: For private circulation only not for sale.P.K. Morgan Securities Ltd. sold or redistributed without the written consent of JPMorgan. JPMSI distributes in the U. The recipient of this report must make its own independent decisions regarding any securities or financial instruments mentioned herein. or its affiliates and/or subsidiaries (collectively JPMorgan) do not warrant its completeness or accuracy except with respect to any disclosures relative to JPMSI and/or its affiliates and the analyst’s involvement with the issuer that is the subject of the research. unless otherwise stated.jpmorgan. research published by nonU. 47 and 49 of the Financial Services and Markets Act 2000 (Financial Promotion) Order 2001 (all such persons being referred to as "relevant persons"). Investment research issued by JPMSL has been prepared in accordance with JPMSL’s Policies for Managing Conflicts of Interest in Connection with Investment Research which can be found at http://www. Opinions and estimates constitute our judgment as of the date of this material and are subject to change without notice.A. habitually invest money. Singapore branch (JPMCB Singapore) which is regulated by the MAS. financial instruments or strategies to particular clients.P.
Nikolaos Panigirtzoglou. Alternative LCVI Trading Strategies. Gianluca Salford. FX Positioning with JPMorgan’s Exchange Rate Model. Hedging Inflation with Real Assets. November 2006 . Larry Kantor and Drausio Giacomelli. Oct 2002 8. Rebecca Patterson and Nandita Singh. Stefano Di Domizio. New LCPI trading rules: Introducing FX CACI. Momentum in Commodities. Understanding and Trading Swap Spreads. Peter Rappoport. Trading Credit Curves I. March 2006 19. Using Equities to Trade FX: Introducing LCVI. Jorge Garayo and Francis Diamond. May 2006 22. RockBottom Spreads. Ruy Ribeiro. JPMorgan FX Hedging Framework. March 2006 18. July 2006 24. Daniel Ko. Laurent Fransolet. 12. January 2006 14. Canlin Li. Relative Value on Curve vs Butterly Trades. Jan 2002 5. Trading Credit Curves II. Nikolaos Panigirtzoglou and Jan Loeys. George Cooper. March 2006 16. 11. March 2006 21. Index Linked Gilts Uncovered. Equity Style Rotation. Mar 2002 6. John Normand. Mustafa Caglayan. JPMorgan’s FX & Commodity Barometer. Jonny Goulden. March 2006 17. Dec 2001 3. 1. Valuing crossmarket yield spreads. Credit and Equities. Drausio Giacomelli. Ruy Ribeiro. Jan 2003 9. Pavan Wadhwa. Marius Langeland. John Normand. Nikolaos Panigirtzoglou. Which Trade. Jonny Goulden.European Equity Derivatives Research Variance Swaps November 17. Exploiting crossmarket momentum. 2006 Investment Strategies Series This series aims to offer new approaches and methods on investing and trading profitably in financial markets. Trading Credit Volatility. John Normand. Apr 2002 7. Osman Wahid. Mustafa Caglayan. Jan 2004 10. Sep 2004. Dec 2001 4. Nikolaos Panigirtzoglou and Lei Shen. March 2006 20. John Normand. Bubbles and Black Holes. Larry Kantor. A Framework for Longterm Currency Valuation. June 2006 23. Jan Loeys and John Normand. February 2005 13. Profiting from Market Signals. February 2006 15. Mustafa Caglayan. September 2006 26. Saul Doctor and Alex Sbityokov. Nikolaos Panigirtzoglou. A crossmarket bond carry strategy. Larry Kantor and Mustafa Caglayan. A Fair Value Model for US Bonds. George Cooper. JPMorgan Emerging Market CarrytoRiskModel. Yield Rotator. Bonds. August 2006 25. Euro Fixed Income Momentum Strategy. Gagan Singh. Jan 2005. November 2006 27. Oct 2001 2. Ruy Ribeiro and Jan Loeys.
This action might not be possible to undo. Are you sure you want to continue?