You are on page 1of 52

European Corporate Research

19 November 2007

A Framework for Credit-Equity Investing


Introducing the JPM CEV (Constant Elasticity of Variance) Credit-Equity Investment Framework
The relationship between Credit and Equity markets is an important signal for both markets but not simple to capture. To address this, we present a new framework for Credit-Equity investing using JPMorgans new Credit-Equity CEV model. We use equity market inputs in a reduced-form CEV model to imply a credit (CDS) spread from tradable equity instruments. This model gives us: A relative value signal between a companys credit and equity Trade structure using CDS against Stock and Variance Swaps Sensitivities (Greeks) of Credit-Equity trades
European Credit Derivatives Strategy Jonny Goulden
AC

(44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com

European Equity Derivatives Strategy Peter S AllenAC


(44-20) 7325-4114 peter.allen@jpmorgan.com

Stephen EinchcombAC
(44-20) 7325-9064 stephen.einchcomb@jpmorgan.com

Our back-testing which we detail shows that our JPM CEV Model provides a successful valuation signal through the cycle. This note gives an introduction to the theory of Credit-Equity modeling and explains our choice of a reduced-form CEV model A future note will focus on Credit-Equity trade structure and tools for spotting credit-equity valuation signals.
Figure 1: CDS Spreads versus JPM CEV Model Implied CDS Spreads
Spread (bp), for companies (in Europe and US) in our analysis universe 600 Av erage CDS Spread

500 400 300 200 100 0 2000 2001 2002 2003

Av erage CEV-implied CDS Spread

2004

2005

2006

2007

Source: JPMorgan

www.morganmarkets.com

J.P. Morgan Securities Ltd.

See page 50 for analyst certification and important disclosures, including investment banking relationships.
JPMorgan does and seeks to do business with companies covered in its research reports. As a result, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of this report. Investors should consider this report as only a single factor in making their investment decision.

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Table of Contents
1. Executive Summary .............................................................3 2. Why Analyse Credit versus Equity? ...................................5
Motivation for Linking Credit and Equity ...................................................................5 Formally Relating Credit and Equity: The Merton Model...........................................6

3. Debt-Equity Modeling Approaches .....................................9


Structural versus Reduced-Form Models...................................................................10 Why Use a Reduced-Form Model?............................................................................11 The Need for a CEV Model .......................................................................................11 Implying CDS Spreads from Equities Using the JPM CEV Model...........................15 A Worked Example ...................................................................................................16

4. Does Our CEV Model Work? .............................................19


1. Does CEV Spread Deviation from Actual CDS Spread Indicate Mis-Pricing? .....20 2. Are Deviations from our Calibrated CEV-to-CDS Level Good Trading Signals? 23 3. Can we Construct Trades to Capitalise on These Signals? ....................................26 4. Does our Trading Signal Work as a Real Trading Back-test? ...............................29

5. Conclusions ........................................................................36 Appendix I: The Merton Model Option Pricing Formula......37 Appendix II: The CEV Model from Equity Inputs to Credit Spreads ...................................................................................40
1. Market Inputs.........................................................................................................40 2. Solve for and ....................................................................................................41 3. Calculating the Default Probability........................................................................42 4. Calculating the CDS Spread ..................................................................................43 5. Using the JPM CEV-Implied CDS Spreads as an Investment Signal ....................43

Appendix III: Universe of Companies For Back-test Results .................................................................................................44 Bibliography ...........................................................................46
Papers.........................................................................................................................46 Books .........................................................................................................................46

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

1. Executive Summary
In this note we introduce a new framework for identifying value between equities and credit: the JPM CEV Model1. This framework allows us to: Identify companies where the credit spread (CDS (Credit Default Swap) spread) is too high or too low given the pricing in the equities markets. It does this by calculating an implied CDS spread from the equity markets. It also allows us to profit from these mis-pricings by using tradable instruments: a company's stock, equity variance swaps and credit default swaps. It provides both trade structure and sensitivities (Greeks). Finally, it is shown to be a reliable trading signal and we detail how our model would have given a profitable investment strategy historically. The model provided profitable trades 64% of the time (success rate), to give an annualised Information Ratio of 1.96 as a strategy.

The need to analyse credit against equity markets for any company is both theoretical, in that the same underlying company fundamentals should drive the value of the stock price, equity volatility and credit spread, and observed, in that there are often clear relationships between a companys stock price, its stock price (implied) volatility and credit spread. The relationship between equity markets and credit markets is an important signal but it is not simple to model or to capture through trading strategies. This is the reason we need a more robust model the JPM CEV Model to indicate when there are investment opportunities and how to construct trades to profit from these. We believe there is both a maturity of products to trade these credit-equity signals and also a range of investors who are already investing across both markets, or the whole of the 'capital structure' as it is sometimes called. The recent rise in credit spreads and equity volatilities also indicates we are potentially entering a period where more opportunities will present themselves. The way we will introduce this framework is as follows: In Section 2 we discuss the motivation for linking equity and credit markets for a company and give an overview of the classical model for doing this, the Merton Model. We also discuss what we will require from a usable credit / equity framework and why we believe this is still needed in the wake of many other attempts to do so. Section 3 justifies our choice and need for our JPM CEV Model. We begin by outlining different credit-equity models available to us, both structural and reducedform. We chose a reduced-form model as it reduces the burden of forecasting the entire capital structure of a firm and also allows us to trade our inputs of equity (stock) price and implied volatility, against our output of credit spread. We show why a CEV (Constant Elasticity of Variance) model is needed to allow us to capture the probability that the stock price falls to zero, which is our default probability. The calculation steps to go from equity price and equity volatility to credit spread using our JPM CEV Model are shown, along with a worked example. Appendix II gives a much more technical description of the JPM CEV Model.
We would like to thank Jerry Hanweck for his invaluable insight and work in helping with this project.
3
1

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

The final stage of our analysis is to test whether our JPM CEV Model works. Section 4 shows how our approach leads to a historically profitable trading strategy. We start by highlighting how the model gives investment signals and the need for calibration. It is then shown that the relationship between the CEV-implied Spread and actual market CDS Spread is mean-reverting. Using a Z-score signal is therefore likely to work to spot trade opportunities and we also show that we can successfully trade this signal by creating a basket of stock and variance swaps against the CDS spread. Finally we back-test our model to show that the traded signals would have given a profitable investment strategy. The Appendices show some more technical aspects of the work, with Appendix I giving a technical description of the Merton Model and Appendix II detailed the calculations involved in the JPM CEV Model. Appendix III shows the companies we use in our back-test. The field of debt-equity modeling is expansive in both academic literature and practical studies. In the Bibliography we highlight some of the key papers and books on the topic. This note focuses on introducing our framework for credit-equity investing. We will look to follow this with a more practically orientated note showing how we use our trading signals and crucially how we deal with trade construction and the Greeks (sensitivities) of our trades involving stock, variance swaps and CDS.

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

2. Why Analyse Credit versus Equity?


Motivation for Linking Credit and Equity
The basic motivations behind wanting to analyse credit and equities together are both fundamental and observed. Fundamentally, equity valuation and credit risk are both driven in part by a company's financial condition. If a companys financial condition improves, its equity price should rise, all else being equal, and credit spread (compensating for the default risk) should tighten. This means that information on a companys financial health should be reflected in both the equity markets and credit (corporate bond or credit default swap) markets. Put another way, equities and credit should both reflect the fundamentals of the same underlying company and this should allow us to spot valuation anomalies or trading opportunities between these markets. Reflecting this fundamental connection, we can easily observe how equity and credit move together in the markets. This movement is not always a simple linear relationship (equity prices up, credit spreads tighter) as looking at a longer history of the co-movement of equity and credit markets shows in Figure 2. At times (A on graph) equity markets rally when credit markets are broadly stagnant, at times both sell-off together (B on graph) and at times credit markets can rally strongly while equity markets are only marginally improving (C on graph). This shows that there is an observable relationship between credit and equity, but that this is more complex than a straight 1-for-1 linear relationship.
Figure 2: Credit Markets versus Equity Markets
Credit = Maggie Credit Industrials Asset Swap Spread (bp). Equity = E-Stoxx 50 closing level.

Figure 3: British Airways Plc Credit and Equity Co-Movement


Equity price (, left axis), Credit Spread (bp, right axis inverted)

60 5.5 80 100 120 3.5 Equity (left ax is) Credit Spread (right inv erted) 140 160

5500 4500 3500 2500 C 1500 0


Source: JPMorgan

Apr 05 - Apr 07 Mar 03 - Apr 05 A B Feb 00 - Mar 03 Jan 99 - Feb 00

4.5

2.5
75 100 125 150

25

50

Apr-06

Jul-06

Oct-06

Jan-07

Apr-07

Source: JPMorgan, Thomson Financial

We can also observe more simply how at times credit and equity markets move together on a single company. For example, Figure 3 shows how British Airways Plcs credit spreads (on CDS) have closely tracked movements in the share price over a 12 month period. Both the credit and equity markets seem to be reacting to similar underlying factors in the same way. So, there are reasons to look at the relationship between credit and equity from an investment perspective. We now examine more formally a methodology to evaluate credit (debt) and equity on a company.

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Formally Relating Credit and Equity: The Merton Model


The formal relation between credit (or debt as it is often called) and equity is originally described in the Merton Model2. The theory behind the Merton model is that debt and equity can both be thought of as derivative securities on the underlying company's asset value. A simplified company balance sheet, as in Figure 4, shows that a company's asset value is equal to the value of its liabilities (debt) plus the residual value which is its equity. The liabilities of a firm are that which is owed to creditors and these can be paid as long as the asset value is sufficient. The residual asset value of the firm once these are paid out belongs to the shareholders, i.e. the equity value. So the value of both the debt and equity is driven by the underlying asset value.
Figure 4: A Simplified Company Balance Sheet
Asset Value = Liabilities (Debt) + Equity

Figure 5: Merton Framework: Debt and Equity Holders as Options Positions at Maturity
Value of debt, D Pay off Lev el (strike price)

Liabilities (Debt) Asset Value Equity

Debt holder (sold a put) Equity holder (bought a call)

Source: JPMorgan

Company Asset Value, V


Source: JPMorgan

Looking at this relationship in more detail, Merton highlighted that a debt holder and equity holder both have payout profiles similar to holders of options on the underlying company asset value (i.e. derivatives on the company asset value). The payoff profile of debt and equity holders at maturity is illustrated in Figure 5. A debt holder will be paid the value of his debt at maturity as long as the value of the firms assets (V) is sufficient to pay the debt (D). If the firms asset value is lower than the value of the debt, then the equity holder will put the firm to the debt holders and the debt holders will receive the company value (V). The debt holder has therefore sold a put on the companys asset value, with strike D. This also means that the equity holders own the residual asset value of the firm. If the firm value is less than the debt this residual value is zero. If the company asset value is above the value of the debt, this difference represents the value of the equity. The equity holder has therefore bought a call on the companys asset value, with strike D. The payoffs of the equity and credit at maturity of the debt are therefore: Bond value (B) = min(D,V)3

2 After Robert Merton, who originally outlined this relationship in 1974 (see On the Pricing of Corporate Debt: The Risk Structure of Interest Rates, R Merton, Journal of Finance, 1974).

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Equity value (S) = max (V-D, 0) This framework for thinking about a companys debt and equity as options can also help explain the changing (i.e. non-linear) nature of the relationship of credit to equity Figure 6 illustrates this using Deutsche Telecom. Here we can see the short put type nature of a long debt (credit) position. If we use the equity price as a proxy for the asset value for now (well deal with this more rigorously later), we can see that as the asset value rises towards a theoretical strike price (the grey line) the credit spread tightens as well (bond values rise). Once the asset value is at the point where the debt of the company is more secure , further rises in asset value do not lead to spread tightening as the short put that the debt holders have is already well in-themoney.
Figure 6: Deutsche Telecom AG - Changing Relationship of Credit versus Equity
x-axis: Equity price (), y-axis: CDS spread (bp, inverted) The dotted lines reflects the inflection point, which is a function of the strike of the Merton options. This changes over time as well.

0 100 200 300 400 500


Source: JPMorgan. Data 2000 2007.

20

40

60

80

100

The Ingredients of a Credit-Equity Model This framework leads us to think about using an options pricing framework to value the credit and equity of a company consistently. Remember, that both instruments are options on a company's asset value. We can therefore price these options with a standard option-pricing framework4. The key ingredients of this price (along with the risk free rate, r) will be: The asset value of the company (V) The volatility of the asset value of the company (v) The level of debt of the company (D)

Appendix I gives a formal statement of the Merton Model option pricing formula.
To show this in standard put notation: Bond Value (B) = D max(0, D-V). Using no-arbitrage conditions, we can set-up a portfolio of the company debt and S units of the companys stock, which is fully hedged as both securities are derivatives of the underlying asset value. This will give us a partial differential equation for the Debt or Equity price, which can be solved using the boundary conditions of our options set-up (i.e. debt is a short put position, so Bond value (B) = min(D,V)).
4 3

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

This general framework still leaves us plenty of scope for different valuation models and the next section gives an overview of debt-equity models, after which we deal in more detail with our JPM CEV model. Before we move on to this, we will outline the general characteristics that will make a debt-equity framework useful and deal with the question of whether we need another model, given that the road to heaven is littered with discarded debt-equity models. Our Requirements of a Useful Debt-Equity Model Many feasible debt-equity models can fit within the general theoretical framework that links equity with debt. The restricting factors should be the accuracy of the model (obviously) and the usefulness of the model to investor needs. This makes us require the following of our debt-equity framework: It must be theoretically sound It must highlight valuation anomalies between credit and equity markets It should provide profitable trading strategies with some consistency It is not too complex to implement and the inputs should be as simple as is feasible It must give credit/equity trade structure, hedge ratios and Greeks

We outline the model that we believe best achieves these later in the note. We also show how this is a useful investment tool, in our view. The Road to Heaven is Littered with Debt-Equity Models To finish this section, well deal with an obvious question about the need for another debt-equity model. Since Merton first published his model in 1974, there have been many laudable attempts to accurately capture the debt-equity relationship. These have often been of increasing complexity and, to be honest, have often been discarded after a period. What then is the need for us to have another go? Although JPMorgan Research has published several studies of the debt/equity relationship5, we have not, until now, implemented a model-based debt-equity framework. Our motivation for this is simple. First, in our analysis of the credit and equity markets we find ourselves unable to ignore the fact that there must be some relationship between debt-equity no investor or market-practitioner can ignore the information that each market gives about the other. We have seen the theory behind this relationship and all our model seeks to do is consistently value credit and equity. Second, we believe there are now enough investors in the market who can trade across credit and equity (as well as the instruments to make this possible) for there to be use in a framework for trading and investing. Our motivation is therefore honest in attempting to usefully link credit and equities. We are also realistic about this and look to evaluate the results and usefulness of our model with all the rigour (i.e. scepticism) that we usually apply. In short, all we are seeking to do is to extract the information available in the liquid equity and credit markets to gain profitable insights or trading strategies. In that sense, the motivation is obvious. We have seen in this first section the reasons for analysing credit and equity in a common framework. We now turn to the more detailed questions of what models are available to us, to explain our choice of the JPM CEV Model.

See both Using Equities to Price Credit, JPMorgan, Sept 2004 and Using Equity Prices to Trade Corporate Bonds: An Evaluation, P Rappoport, Aug 2002 for empirical analysis of KMV's model. More recently our colleagues in US Corporate Quantitative Research have been publishing useful debt-equity reports, for example, see Introduction to the JPMorgan Cross Asset Class Relative Performance Report, E Beinstein, Aug 2005.

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

3. Debt-Equity Modeling Approaches


We have seen the fundamental rationale for analysing credit and equity with a common valuation approach. But as is often the case, the devil is in the details. The heart of the issue for credit investors is that pricing of credit relies on capturing the probability that a company defaults. In the classical Merton framework, this happens if the company's asset value is below its debt at the time the debt matures, effectively assuming there is only one bond or class of debt outstanding. This makes an assumption that default can only happen at maturity of the debt and that the debt structure is very simple. In practice, default could happen at any time and companies have many liabilities. So, we typically make these assumptions more realistic which would lead to saying that default happens whenever the company's asset value falls below some measure of the outstanding debt. We will therefore need to model how the firms asset value changes through time and will look to calculate the probability that the asset value falls below the debt value of the firm at any time. This can be shown diagrammatically in Figure 7.
Figure 7: Model of the Firm's Asset Value Through Time
x-axis: Time, y-axis: Asset value

Probability distribution of 100 80 60 40 20 0 -20 -40


Source: JPMorgan

Firm asset v alue

firm's asset v alue


(asset value volat ility) M ean asset value (with drif t)

Debt (default barrier) Asset v alue < Debt

What this will require is some mathematical description of how the asset value of the firm evolves through time (i.e. the asset value process). This looks fairly simple, as all we'll need is a measure of the firm's asset value, the volatility of that asset value and the level of debt through time and we can price the credit and equity using standard option pricing (as shown in Appendix I). However, none of these inputs can be easily observed. We cannot observe the asset value of the firm directly, let alone its volatility. Calculating the liabilities of a company through time will also involve a degree of complexity as companies have many forms of borrowing and other liabilities, and change these frequently. The response to these challenges has been broadly in two camps: structural models or reduced-form models. We outline these in the next section6.

The academic literature on these topics is vast and comprehensive. We highlight some of the key pieces in the Bibliography.
9

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Structural versus Reduced-Form Models


We will use the term structural model to refer to any model that attempts to accurately model the company's asset and liability structure in order to calculate the firm's asset value and debt profile through time. This will typically involve a large number of inputs of liabilities (short-term and long-term) and a projection of how these change through time. The model essentially takes the Merton framework and applies it as faithfully as possible to the real world complexity, which makes it necessary to specify the company structure in detail. The starting point for structural models will be an attempt to specify the companys balance sheet: its assets and liabilities through time. We will use the term reduced-form models to refer to any models that somehow simplify the modeling process (hence reduced-form) to price credit with fewer inputs. Typically these will not need detailed inputs of a firms asset and liability structure. In debt-equity modeling, this will mean using simplified inputs about a company's health (e.g. equity price and volatility) to calculate a default probability and hence price credit. Reduced-Form in Credit The term reduced-form covers a very wide range of credit models. For example, the market standard CDS pricing model JPMorgans CDSW Model is referred to as a reduced-form model in that it has simple inputs. In the market standard pricing model, we use market CDS spreads to calculate default probabilities so that we can consistently price credit instruments. How these default probabilities arise (where we get these from in a fundamental sense) is not needed in the model. When we move to anything within the debt-equity framework, we need to model the default probability as fundamentally linked to a companys performance. That will involve taking in some inputs about the company. What makes a model reduced-form is that these inputs are usually much simpler than specifying the entire company structure.

Structural Model Challenges The challenge that any structural model faces is in accurately specifying the point (or barrier) at which the firm defaults, as we have difficulty specifying all of the firm's liabilities and the asset value of the company through time. We briefly look at two well-known models that look to tackle this issue in different ways. i) KMV: Moodys KMV is a practical implementation of a structural model7 (although some would argue whether this really follows the Merton structural approach to its conclusion, see Schnbucher in the Bibliography). This model uses structural information about the company's assets and liabilities to solve the Merton option-pricing formula. However, instead of calculating the probability of the companys asset value falling below the debt value, KMV uses a metric called the distance-to-default (DtD), which gives a standardised probability of the company's asset value hitting the default barrier (it is the number of standard deviations that the company's asset value is away from default given the asset value volatility). This is then empirically compared to other companies with that DtD to see what default probability companies with a given DtD have exhibited in the past. From this, a probability of default is taken (the EDF or Expected Default Frequency in KMV's model) which is then used to derive a CDS spread.
We tread with care when discussing Moodys KMVs model. The last time JPMorgan research wrote a paper evaluating the usefulness of this model it drew a public rebuttal from Moodys. In the placid world of credit research, this was the equivalent of 'pistols at dawn'.
10
7

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

ii) CreditGrades: This structural model (part of the RiskMetrics family) also takes in detailed information of a companys asset and liability structure but has an alternative way of modeling the uncertainty around when default is actually triggered. In order to deal with the difficulty of accurately measuring the structural model inputs and whether a company has defaulted at each point in time, CreditGrades calculates a non-static barrier which the asset value process has to cross for a default to occur. In Figure 7 this means that the straight line representing the default barrier should actually be represented as a probability distribution around that value. This represents the uncertainty around the firms actual value and when it has crossed the default threshold. The model calculates the default probability using both a stochastic (changing with time) asset value process and a stochastic default barrier, which it then uses to calculate credit spreads. These two well-known structural models highlight that most implementations of a structural model involve both accurate inputs about the companys balance sheet, and also significant additional modeling to cope with uncertainty about observing asset values and the point of default. We therefore prefer to implement a reduced-form model as we will now discuss.

Why Use a Reduced-Form Model?


The JPM CEV Model is a reduced-form model. In other words, we will be looking to value credit and equity based on simple market inputs without having to specify the entire firm capital structure. We chose this model as: a) It simplifies the inputs required for the model to market observable levels we prefer to use unambiguous inputs that reflect market pricing. For us, this will mean looking at equity prices, equity volatilities and credit (CDS) spreads. b) It reduces the calculation complexity whilst not solvable on the back of an envelope, our model is solvable without too much calculation complexity which aids its usefulness. c) It highlights investment opportunities with tradeable instruments and gives appropriate trade structure this is vital for us in terms of usefulness of the model. In general, the complexity of inputs needed to accurately specify a structural model is too great for our purposes and the evidence around whether structural models add much over reduced-form models is not clear8.

The Need for a CEV Model


The model we will use to relate credit and equity is the JPM CEV Model CEV stands for Constant Elasticity of Variance. Having explained our preference for a reduced-form model, we now go through the need for this CEV model.

A good study of this is in Reduced Form vs. Structural Models of Credit Risk: A Case Study of Three Models, Arora, Bohn and Zhu at Moodys KMV. They conclude that a reduced form Hull-White model performs as well as a Vasicek-Kealhofer structural model.
11

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Modeling the Default Point We have seen that a key issue with the pure structural Merton model is modeling the point at which at firm's asset value falls below some default barrier, as both of these are unobservable. We have also seen the complexity introduced by trying to model these parameters. Our approach will be different. We rely on the basic balance sheet equation, namely: Assets = Liabilities + Shareholders Equity Assets Liabilities = Shareholders Equity (1A) (1B) or

In a Merton framework, default happens when the firms assets fall to below the liabilities (Assets < Liabilities). At the first point that this happens, the Shareholders Equity should have an intrinsic value of zero, from equation (1B). This can be shown diagrammatically in Figure 8 where we change from Figure 7 to model the evolution of the firms share price (as opposed to Asset Value) through time. In terms of modeling, if the firm defaults, the intrinsic value of the share price will be zero (as opposed to the Asset Value being equal to the Debt).
Figure 8: Model of the Firm's Share Price Through Time
x-axis: Time, y-axis: Share price

Probability distribution of 100 80 60 40 20 0 -20 -40


Source: JPMorgan

Share price

firm's share price


(share price volat ility) M ean share price (wit h drift )

Share Price =0 (default barrier)

Share price = 0

Shifting from modeling the Asset Value versus Debt Value relationship to focusing on the Share Price has the advantage of replacing a non-observable firm asset value and asset value volatility with a well known and observable share price and share price volatility. All we then need to do is calculate the probability of a firm's share price going to zero using standard Black-Scholes option pricing. Unfortunately, the standard process underlying the Black-Scholes model of a stock price does not allow the share price to go to zero. Thats one reason why we use a Constant Elasticity of Variance (CEV) model, which will allow this. We now move to describe this CEV model in more detail9.

9 See Options, Futures and Other Derivatives, J Hull, 5th Edition, Ch 21 for more on CEV models.

12

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Introducing the CEV The standard model of how equity prices change over time is given by Equation 1 below. In plain language it states that the change in a share price (dS) is a function of a drift term (the share price rises over time) and a volatility around that value. Equation 1: Standard Black-Scholes Model Equation 2: CEV Model

Volatility term

Volatility term

dS = Sdt + SdZ
Drift term

dS = Sdt + S dZ
Drift term

Where, S = Share price dS = Change in share price = Expected equity return in % = Share volatility in % dZ = A standard Wiener process (i.e. a random variable term) = Elasticity term, alpha (to be explained later) A brief look at the terms of our standard Black-Scholes model will show why this isn't an appropriate model if we want the possibility (a non-zero probability) that equity can go to zero, i.e. a firm can default. The Black-Scholes model states that the change in a share price has two components: a drift term (share prices rise over time) and a volatility term. represents the share price return in % terms, so S means that the share price drift is proportional to its price. represents the volatility (standard deviation) in % terms, so S again means that the share price changes are proportional to the share price. Technically we say that Equation 1 represents Geometric Brownian Motion, the geometric meaning that percentage moves are close to normally distributed or actual (price) moves are lognormally distributed. This is opposed to Arithmetic Brownian Motion where actual (price) moves are normally distributed. In ordinary language, if the share price changes are normally distributed, the firm share price is equally likely to move up or down 1 (if volatility is static at 1), whether the share price is at 5 or at 50. If percentage moves are normally distributed then share price is equally likely to move up or down 20%, which equates to a 1 move when the share price is at 5 but a 10 move when the share price is at 50. As this is a key point, a practical example will be useful. If price volatility is constant then the price is equally likely to move up or down a given price amount (1 in our previous example) and the share price can feasibly go to zero by moving down enough. However, if we say that percentage volatility is constant (e.g. at 20% in our example) then as the share price falls actual price moves get smaller. For example, when the share price is 5, the share price is likely to move 1 (5 20%) over a period. But as the share price falls to 2, it is only likely to move by 0.4 (2 20%) and at a share price of 1, the likely move is only 0.1 etc. This will mean (for a continuous process) that the share price can never reach zero.

13

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Of course, in the real world a companys share price can go to zero and this is just one of the practical inaccuracies of the Black-Scholes framework we will deal with some others later on. This is the reason we cannot use a standard Black-Scholes model for the equity process. CEV Model Allows Stock to Go to Zero The Constant Elasticity of Variance (CEV) model crucially allows the share price to fall to zero. Furthermore, we assume that if the share price goes to zero the company is in default in other words, the share price falling to zero is an absorbing state. So our CEV model will allow a non-zero probability that the firm defaults. In practice, default for a company may not result in a zero share price or vice versa, but under the CEV model the probability of the share price being exactly zero will be close to the probability of the share price being near to zero. The CEV model also has some other real-world accuracies that Black-Scholes doesn't capture, such as the volatility skew and leverage effect which we highlight further on.
Figure 9: % Volatility Increases as Share Price Falls with CEV Model (alpha = 0.5, sigma = 0.1)
% volatility (y-axis), Share price (x-axis)

Table 1: % Volatility Increases as Share Price Falls with CEV Model (alpha = 0.5, sigma = 0.1)
Share Price S 10 8 6 4 2 1
Source: JPMorgan

12% 10% 8% 6% 4% 2% 0% 0
Source: JPMorgan

Volatility Term S 0.316 0.283 0.245 0.200 0.141 0.100

Black-Scholes % Volatility S / S 3.16% 3.54% 4.08% 5.00% 7.07% 10.00%

10

14

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Figure 10: Schematic of JPM CEV Model


1. Market Inputs Equity price Equity implied option v olatilities

The CEV model uses the volatility term S as opposed to S in the Black-Scholes framework (the term represents the elasticity of variance). This means that BlackSholes volatility, or % volatility, is S (-1), which is proportional to the share price raised to some power. For example, if alpha () = 0.5, and CEV volatility is a constant 0.10, we get the relationship between share price, the volatility term and the CEV volatility term shown in Figure 9 and Table 1. This will allow Black-Scholes % volatility to rise as the share price falls (if <1) and so will give us some probability that the share price falls to zero and there is a default. Note that the lognormal distribution of the Black-Scholes model appears as a special case of the CEV model, namely when =1. Also, Arithmetic Brownian Motion mentioned above is covered by the case =0. The inputs that we will need for our CEV Model are just the share price and a range of options with different strikes to give us the volatilities. These are readily available in the market and crucially also tradable inputs.

2. Solve for and This creates an equity process consistent w ith market inputs

3. Calculate Default Probability Using equity process, calculate probability equity price = 0, i.e. default

4. Calculate CDS Spread Use default probability and recov ery rate to calculate CDS spread

Implying CDS Spreads from Equities Using the JPM CEV Model
The way we imply CDS spreads from equity inputs is shown as a schematic in Figure 10. Appendix II describes the JPM CEV Model more formally. The usefulness of the model is in allowing us to take observable and tradable market inputs using equity price and implied volatilities and imply a CDS spread from them. The CEV is also analytically tractable, meaning that it allows solvable solutions to real world pricing issues. Empirical Backing for CEV Models The alpha factor in the CEV model that allows percentage volatility to rise as stock price falls is not just a convenient technicality that allows us to infer a default probability (and thereby a CDS spread) from equity prices. It can also fit the observed equity options skew (difference between implied volatilities for options struck At-The-Money and those Out-of-The-Money) and the observed rise in percentage volatility as equity prices decline (the leverage effect). Neither of these effects should be observable in the standard Black-Scholes equity model where percentage volatility should be constant for different strikes and for different share prices. A simple look at ING Group Plc shows both a clearly observable leverage effect (% volatility rises as the share price falls in Figure 11) and skew (% implied volatility rises for strikes out-of-the-money, particularly on the downside in Figure 12). These effects are more accurately captured within a CEV model which will allow volatility to rise as share price falls. This shows that the CEV Model has both empirical rationale and theoretical backing.

5. Investment Signal Compare market CDS spread to CEVimplied CDS spread for signal
Source: JPMorgan

15

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Figure 11: ING Group Leverage Effect


x-axis: Share price (), y-axis: 2m Realised volatility (%)

Figure 12: Equity Volatility Skew for ING Group


x-axis: Option strike (%); y-axis: 3m implied volatility 35

140 120 100 80 60 40 20 0 5


Source: JPMorgan

30 25 20 15

15

25

35

45

40

50

60

70

80

90 100 110 120 130 140 150

Source: JPMorgan

A Worked Example
We will use France Telecom to illustrate how the JPM CEV model calculates a credit spread from equity market inputs working through the steps in the schematic in Figure 10 10. Appendix II works through the theoretical stages of the model in detail. In order to derive the probability of equity value falling to zero we first need the share price (25.64 as of 14th November 2007) and option prices for different strikes (the ATM options price and skew). We take 70% to 110% strike options in units of 10% of stock price, for both 6 month and 1 year maturities. Table 2 and Figure 13 show these volatility surface inputs, illustrating the Black-Scholes volatilities for the different strikes. In reality, we take the options prices and use our CEV option pricing model to fit the equity process implied by the share price and options prices. This fitting involves solving for the CEV volatility () and alpha () that best fit the current share price and options skew, using the CEV Model described in Equation 2 previously.

10

We will use data as of 14th November 2007.

16

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Table 2: Options Skew Inputs


Black-Scholes volatility (%) Strike Tenor (years) 6m 1y
Source: JPMorgan

Figure 13: Options skew input


Volatility surface (x-axis: Strike (% of ATM); y-axis: Black-Scholes implied volatility (%); z-axis: Tenor, (years)) 100 26.2% 26.5% 110 25.3% 25.4%

70 33.1% 32.2%

80 30.4% 29.9%

90 28.0% 28.0%

35% 30% 25% 20% 15% 100 90 1 0.5


Source: JPMorgan

110

80 70

For France Telecom, this gave an alpha () value of -0.076 and a volatility (sigma, ) of 8.691, equating to a Black-Scholes ATM volatility of 26.5% ( = S (-1) ). Using this we can imply the default probability of France Telecom in our reducedform model this is the probability of the share price falling to zero and from this, given a fixed recovery rate of 40%, the Model implied CDS spreads. These probabilities and spreads are shown in Table 3 and Figure 14. For France Telecom this gives an implied Model 5y CDS of 131bp and a current market CDS spread of 40bp, as observed in the market. Figure 14 shows that we can get divergent absolute values between the model implied spreads and market spreads which mean we need to calibrate our model to market levels as we discuss in the next section (calibration is a common requirement of any model).
Table 3: France Telecom CEV Implied Default Probabilities and Spreads
Term 6M 1Y 2Y 3Y 4Y 5Y Default Probability Cum def prob, % 0.00% 0.04% 1.22% 4.02% 7.52% 11.12% Market CDS bp 20.7 20.7 25.4 34.3 36.1 39.6 Model CDS bp 0.0 2.4 35.2 77.2 109.0 130.4

Figure 14: France Telecom - CEV 'Model' Implied Spread versus Market Spread
x-axis: Maturity in years; y-axes: Spread (bp) Market CDS 50
40 30 20 10 0 0
Source: JPMorgan

140 120 100 80 60 40 20 0 -20

Model CDS (right ax is)

Source: JPMorgan

17

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Model Calibration As we will see in more detail later (in section 3), our CEV model can get absolute values that are constantly a factor different from current market spreads. This could be because either the model assumptions (such as the recovery rate) are different from the market assumptions or because the markets are fundamentally mis-pricing the difference between credit and equity. In order to take out any systematic difference of assumptions and to highlight market mis-pricing we calibrate our Model implied 5y CDS spreads. Looking historically at the Model versus Market 5y CDS spreads of France Telecom Figure 15 shows why the calibration is needed. We can see that the CEV Model implied 5y CDS spread tracked (and sometimes led) moves in the Market 5y CDS spread however the levels are a factor different. We calibrate by regressing the Model 5y spread to the Market 5y spread as shown in Figure 16. It is this Calibrated Model spread that we use as a relative value signal and in our back-testing. We discuss this in more detail in Section 4.
Figure 15: France Telecom: CEV Model Implied 5y Spread versus Market 5y Spread
bp

Figure 16: Calibration of France Telecom Model 5y Spread to Market 5y Spread


x-axis: Model implied 5y CDS spread (bp); y-axis: Market 5y CDS spread (bp)

200 150

60 50 40

50 40 30 20 10 0 0
Source: JPMorgan

100 50 0 Nov -05 Mar-06 Jul-06 Nov -06 Mar-07 Jul-07


Source: JPMorgan

30 CEV Model 5Y CDS Market 5Y CDS 20 10 0 Nov -07

y = 0.169x + 14.065 R 2 = 0.560

50

100

150

200

Having seen in this section why we chose our JPM CEV Model for its theoretical features and how this differs from other debt-equity models available, we now move on to the key question of whether this model practically works. We will see how it satisfies the necessary criteria for a credit-equity model we outlined at the start and some evidence of how well this would have performed as a historical trading signal. We will also look at how it outperforms other simpler credit-equity models in terms of spotting valuation signals and trade structure.

18

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

4. Does Our CEV Model Work?


What would it mean for our debt/equity model to work? We return to our requirements set out in the introduction (shown in the grey box). It must be theoretically sound It must highlight valuation anomalies between credit and equity markets It should provide profitable trading strategies with some consistency It is not too complex to implement and the inputs should be as simple as is feasible It must give credit/equity trade structure, hedge ratios and Greeks We feel that the theoretically sound and simplicity arguments have been answered in the previous sections. We must now answer whether the model can provide trading signals, trade structure and profitable trading results. In order to give some structure to the argument in this section, we outline the steps we will take to test our JPM CEV Model in Figure 17. The dark grey box around Step 3 indicates this is a more technical test which readers can skip without losing the overall flow of the argument.
Figure 17: Structure of Our CEV Model Testing
Aim: Use CEV Spreads v s actual CDS Spread as a trading signal.

1. Does CEV Spread deviation from actual Identify Trading Signal CDS Spread indicate mis-pricing? Compare CEV-implied Spread Lev els v ersus CDS Actual Spread Lev els Calibration is needed. 2. Are Calibrated CEV to CDS Level Deviations a Good Trading Signal? Is there Z-score mean-rev ersion on the calibrated lev els? Yes Construct Trade to Capture Signal 3. Can we construct trades to capitalise on these signals? Can w e replicate CEV mov es (w e can't trade the model) w ith Stock and Var Sw aps? Yes Test Whether Strategy 4. Does our trading signal work as a real is Profitable trading backtest? If w e trade on our z-score signal, is the strategy profitable. Yes No

Source: JPMorgan

19

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Our overall aim here is to use the differential between the CEV Spread and market CDS Spread as a signal that the markets are mis-priced and that there may be trading opportunities. We will work through some preliminary tests to then show whether using our JPM CEV Model signal would have been a profitable investment strategy. We therefore start with what the signal should be and whether we should be trading whenever our CEV Model Spread is different from the market-observed CDS Spread. Universe of Stocks / CDS For all of our back-tests we use 93 US companies and 131 European companies. These are the companies for which we have several years of CDS spread and an equity options market. The ideal universe is the crossover or high yield section of the credit markets, although for Europe we extend this to high grade. We have avoided survivorship bias by using past as well as present members of the relevant CDS indices. The public-to-private trend over the past 2-3 years in the high yield market has meant that the universe of HY companies with active CDS and outstanding shares has become smaller, somewhat restricting the choice part of the market for these strategies. We list the universe of companies we use in Appendix III.

1. Does CEV Spread Deviation from Actual CDS Spread Indicate Mis-Pricing?
1. Does CEV Spread deviation from actual CDS Spread indicate mis-pricing? Compare CEV-implied Spread Lev els v ersus CDS Actual Spread Lev els

In this first part of our argument we will see that CEV Spreads need to be calibrated against actual CDS Spreads to get a useful trading signal, as the comovement is in line but the levels can be different. From what we have seen so far in our description of our JPM CEV model, from the equity (stock) price and equity volatility levels we observe we will get as an output a CEV-Implied CDS Spread. Theoretically this level should be directly comparable to a CDS spread. For example in Figure 18 if a company is trading at 70bp (5y CDS) in the market and the CEV Implied CDS Spread (using equity stock and options levels) is 90bp, then the market CDS spread is too low (CDS is dear, buy protection). Conversely, if the CEV Implied CDS Spread is 50bp, then the market CDS spread is too high (CDS is cheap, sell protection). Our first thought is to use this pure CEV Spread minus CDS Spread as a signal. We will see that generally simple calibration is needed.
Figure 18: Example of a Cheap or Dear CDS Comparing CEV Model Spread to Market Spread
x-axis: CDS Maturity (years); y-axis: Spread (bp)

100 80 60 40 20 0 0
Source: JPMorgan

Market CDS Dear CDS Cheap CDS

90 70

Dear CDS: CEV Model CDS - Market CDS = 20bp

Cheap CDS: CEV Model CDS - Market CDS = -20bp

50

20

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

A Single Company Example We find that whilst the CEV Spread tracks the observed market CDS Spread, the level can be different and so calibration is needed to generate investment signals. Figure 19 shows this over time for France Telecom. We can see that the movements of CEV-implied Spread and CDS Spread are largely the same, but the CEV Spread on the right axis is of a magnitude of almost double that of the observed market CDS spread (check the axis levels). A way of seeing this difference in level is to regress the CDS spread against the CEV spread, as in Figure 20, with the CEV spread on the x-axis. We have a reasonable R-Sq of 56%, an intercept ( in our regression) of 14.065 and a slope () of 0.169. This means that the CDS spread is generally of magnitude (14.065 + (0.169 CEV Spread)). In other words the CDS level is generally 0.169 times smaller than the spread implied by the CEV model. The two move together but at different levels.
Figure 19: France Telecom Comparison of CDS Spread and CEVimplied Spread
bp

Figure 20: France Telecom Comparison of CDS Spread and CEV Spread
x-axis: CEV Implied Spread (bp); y-axis: Actual CDS Spread (bp)

700 600 500 400 300 200 100 0 Jan-02


Source: JPMorgan

CDS Spread (left ax is) CEV Spread (right ax is)

1200 1000 800 600 400 200 0

50 40 30 20 10 0 0 50 100 150 200 y = 0.169x + 14.065 R 2 = 0.560

Jan-03

Jan-04

Jan-05

Jan-06

Jan-07

Source: JPMorgan. Data over last two years.

The reason for a systematic difference of level between the CEV-implied Spread and market CDS Spread could be due to a number of factors: a) The market is pricing in a different recovery rate to that which we assumed. A full structural model will allow estimation of the recovery rate, but within our reduced form model this is not possible and so levels may be systematically different. b) The CDS market may be pricing higher probabilities of a company defaulting than the probability of the equity price falling to zero. This could be because defaults can occur due to company liquidity problems ("failure to pay") or because management can take a company into bankruptcy proceedings before the asset value reaches the debt value (equity equals zero). This would mean the default probabilities implied by market spreads are higher than those implied by our JPM CEV model from equity inputs. Again, this would mean a systematic difference of levels between the CEV-implied Spreads and the observed market CDS Spreads.

21

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

From Single Company to a General Need for Calibration This need for calibration is systematic across the companies we have looked at. If we perform the regression in Figure 20 for all companies in our universe (across Europe and the US) shown in Figure 21 we can see that majority of the companies have betas (gradient of the line regressing CEV Spread (x-axis) and CDS Spread (y-axis)) that are between 0 and 0.75. In other words, actual CDS spreads are generally lower (below 0.75 most of the time) than the CEV-implied Spreads. This means that we should only use the model with some calibration. Figure 22 incorporates the strength of the regression into the display by using the Beta / R of the regression. This has the effect of increasing magnitude of the Beta for low R-Squareds. We can see that even when we do this, we still get the majority of the Beta / R below 1. This shows that the CEV Spread is consistently above the CDS spread for the companies in our sample. Using a simple comparison of CEV-Implied Spread minus CDS Spread will therefore give systematic error terms. They way to normalise for this, is to calibrate the levels.
Figure 21: Buckets of Beta of CDS versus CEV Model CDS Regression Figure 22: Buckets of Beta / R of CDS versus CEV Model CDS (CEV Spread is the x-axis in the regression) Regression (CEV Spread is the x-axis in the regression)
Count of beta for all companies from regression of CEV Spread (x-axis) versus Count of beta / r for all companies from regression of CEV Spread (x-axis) CDS Spread (y-axis) versus CDS Spread (y-axis) 250 200 180 200 160 140 150 120 100 100 80 60 50 40 20 0 0

-0.95 -0.85 -0.75 -0.65 -0.55 -0.45 -0.35 -0.25 -0.15 -0.05 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95

Source: JPMorgan

Source: JPMorgan

How Calibration Works The way this calibration works is to use the historical beta of CEV-implied Spreads against market CDS Spreads to modify for any consistent difference in the relationship. In other words, if we had France Telecom CEV-implied CDS spread at 100bp and the market CDS level is 35bp this is not a signal of divergence if we take the calibrated levels. Using the historical calibration above in Figure 20 [CDS = (14.065 + (0.169 CEV Spread))] the calibrated market CDS level for a CEV Spread of 100bp is 31bp. So, an observed market CDS spread of 35bp is around the level that is implied by the equity markets, using our model. In practice, we use a rolling 12 months of data to calibrate our levels. Looking in more detail at the relationship between CEV-implied CDS Spreads and actual Market CDS Spreads in Figure 23 and Figure 24 we can see how these move together over time and how the correlation changes. Figure 23 shows that when we average the CEV-implied Spread and Market CDS spread across all names, we get a good broad tracking of spreads over time. The correlation (shown as an average over the next year) tends to increase over higher spread periods something comforting given where we are in the credit and equity cycles. Figure 24 shows how the correlation between the CEV and CDS spreads changes with CDS volatility. We can see that over more volatile periods the correlations tend to increase. Again this gives us more confidence that as volatility rises our model will actually get stronger.
22

-2.375 -2.125 -1.875 -1.625 -1.375 -1.125 -0.875 -0.625 -0.375 -0.125 0.125 0.375 0.625 0.875 1.125 1.375 1.625 1.875 2.125 2.375

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Figure 23: On aggregate CEV spread moves are similar to CDS spreads with a correlation higher when spreads are wider ...
Left axis: CEV-implied and Market CDS Spread (bp); Right axis: average annual correlation between them, over the next year. Av erage CDS Av erage CEV Av erage annual correlation (RHS) 0.5 600

Figure 24:...and average correlation shows a tendency to increase when CDS volatility increases
x-axis: CDS volatility observed in given time periods (bucketed); y-axis: CEV to CDS spread correlation over time period.

0.5 0.4 0.3 0.2 0.1 0 0% 10% 20% 30% 40% 50% 60% 70%

500 400 300 200 100 0 2000 2001 2002 2003 2004 2005 2006 2007
Source: JPMorgan

0.4 0.3 0.2 0.1 0

Source: JPMorgan

Having seen that we need to calibrate our CEV Spreads to compare them to market levels, we now need to consider whether this calibrated level gives us a good signal. We turn to this next step in our argument by testing whether there is a meanreverting pattern by looking at Z-scores.

2. Are Deviations from our Calibrated CEV-to-CDS Level Good Trading Signals?
2. Are Calibrated CEV to CDS Level Deviations a Good Trading Signal? Is there Z-score mean-rev ersion on the calibrated lev els?

In the next step of the argument, we will see that calibrated CEV-implied spreads and CDS spreads mean-revert. They can therefore be used as trading signals. We have seen in Step 1 of our argument that a straight CEV Spread comparison to the CDS Spread is unlikely to be an effective trading signal due to systematic difference in the levels. However, by calibration we can see that our CEV Spreads do track market CDS Spreads. That leads us to our next test of whether a divergence in this calibrated spread relationship is a good trading signal. One way we will test whether there is a tradable relationship is to see if there is any mean-reversion in the relationship between the calibrated CEV Spread and the CDS Spread. We will use Z-scores (see grey box) to capture (statistical) divergence of the relationship between calibrated CEV Spreads and CDS Spreads. The Z-score represents the number of standard deviations that a point lies away from the mean of a distribution. We calculate this as: Z-score = (xi ) / where, xi is the ith value of x, in our case this corresponds to x at a specific time value i. is the Mean and is the standard deviation. This can be used to measure how far a particular CDS Spread is from the calibrated CEV-implied Spread in standard deviations.

23

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Example of Z-Score Mean-Reversion for a Single Company Once we have calibrated our CEV-implied level, we can compare this to the market observed CDS spread at every point in time. We can calculate a CDS rich / cheap signal, which is just the CDS Spread minus the Calibrated CEV-implied Spread. For example, Figure 25 shows this for France Telecom over a two year period from September 2005, with the Z-score shown as well. In order to show that this a tradable signal, we need to know that there is some mean-reversion. This will mean that if the CDS Spread is very different from the calibrated CEV-implied Spread at some point in time, this relationship will come back in line. One way of testing this is to observe the Z-score over a period and then re-observe the Z-score three months later, based on the original calibration parameters. Then we can measure the out-of-sample change in Z-score (three months being a reasonable trade horizon). If there is a pattern of mean-reversion between CDS Spread and calibrated CEV-implied Spreads we should see a negative relationship between the Z-score at one point and the change over the subsequent period. For example, if the Z-score is 3.0 (i.e. the CDS spread is trading much higher than the CEV implied level), then if there is some mean-reversion the Z-score will decrease in the next period to, say, 1.8. This would be a Z-Score change of -1.2 (1.8-3.0). Hence a negative relationship between the Z-score and subsequent change in Z-score.
Figure 25: Z-Score Example for France Telecom
CDS rich / cheap (= CDS Spread Calibrated CEV implied Spread, bp), left axis; Rolling 12m Z-score, right axis

Figure 26: Z-Score over Subsequent Periods


x-axis: Z-score period 1; y-axis: Change in out-of-sample Z-score over subsequent 3m period. 14 y = -1.129x - 0.6724 2 12 R = 0.2744

25 15 5 -5 -15 -25 Sep-05

CDS rich/cheap Z-Score (right axis)

6 4 2 0 -2 -4 -6

10 8 6 4 2 0 -6 -4 -2 -2 0 -4 -6 -8 2 4 6

Sep-06

Sep-07

Source: JPMorgan

Source: JPMorgan. Data last 2 years.

Figure 26 shows this comparison of Z-scores and changes in Z-scores over subsequent periods for France Telecom (we show the Z-score at one point in time on the x-axis and the subsequent change in the Z-score over the next three months on the y-axis). The negative slope shows that high Z-scores (showing a large calibrated CEV Spread to CDS Spread divergence) fall over the next period and low Z-scores rise. This is an illustration that the relationship between CDS Spreads and calibrated CEV-implied Spreads for a single company mean-revert, but this pattern is also true if we look across our entire universe of companies.

24

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Z-Score Mean Reversion Across All Companies Figure 27 shows the comparison of Z-scores across all European and US stocks (i.e. repeating Figure 26 for each company). Figure 27 shows the slope of the regression line of the relationship between Z-score and subsequent change in out-of-sample Zscore (all done on the calibration regression of CDS Spreads to CEV-implied spreads). The fact that most of the data points are below 0 means that the betas of this regression are generally negative which means that Z-scores tend to mean-revert. This shows that there is a strong bias for the relationship between the calibrated CDS Spreads and CEV-implied Spreads to mean-revert, suggesting that the model will have good predictive power.
Figure 27: Beta of Change in Z-Score vs Z-Score is Mostly Negative Showing Mean Reversion
Frequency of Regression beta of 3-month z-score change v z-score

350 300 250 200 150 100 50 0 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0

Source: JPMorgan

The results in this section show that once we have calibrated our CEV Spread, the relationship between the CEV Spread and the market CDS spread is mean-reverting. When a companys CDS trades much too wide (high) or tight (low) to the equityimplied CEV Spread, this is likely to be a temporary movement and it should meanrevert. This means we could look at trading this signal profitably. The argument requires two more steps to show that our JPM CEV Model works as an investment framework: Can we construct trades to capitalize on our CEV Spread signal (in other words can we replicate the CEV Spread with tradable instruments)? Does trading our model signals work in a back-test? The first point will be a bit more technical in terms of creating a replicating basket for our CEV Spread from Stocks and Variance swaps. We therefore put it in a grey box below. Those who wish to move straight to the overall back-test of our results can skip Section 3 below and still follow the overall flow of the argument.

25

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

3. Can we construct trades to capitalise on these signals? Can w e replicate CEV mov es (w e can't trade the model) w ith Stock and Var Sw aps?

3. Can we Construct Trades to Capitalise on These Signals?


This section shows how we can replicate CEV Spreads with a basket of Stocks and Variance swaps. Those short of time or patience can skip this section and jump to the final back-test in Section 4 without loss of continuity. Up to this point we have shown that we can generate promising trading signals from using a Z-score signal on our calibrated CEV Spread versus CDS Spread relationship. But we cannot trade a CEV Spread (or change in CEV Spread) directly it is derived from our model. So we will only be able to capitalise on the CDS Spread and calibrated CEV Spread coming back in line if we can actually create a basket of tradable products that will mirror the move in the CEV Spread. It is to this that we now turn. We will show that we can construct a basket of Stock and Variance Swaps, using the sensitivity parameters (Greeks) given by our model, which mirrors moves in the CEV Spread. This will allow us to construct trades to capitalise on our trade signals. Since the CEV model can be used to give theoretical CDS spreads, we can use the model to also give us sensitivities to the underlying share price and volatility surface. This is done through a numerical approximation for small shifts in the underlying market variables: the Stock price (S) and the Volatility Surface (V). Since the CEV model is fitted against the whole volatility surface, the concept of Vega or sensitivity to implied volatility must be handled with some care. By contrast, a fixed strike stock option only has a sensitivity to the implied volatility used to price that particular option, rather than a shift in the whole volatility surface. Since our CEV model uses a range of option strikes and maturities to calculate an implied spread, it has sensitivity to shifts in the entire volatility surface. We use shifts in Variance Swaps levels as a proxy for this, as we will see below. Replication of Our CEV Spread With Stocks and Variance Swaps Given our CEV-implied Spread will change when the stock price (S) or volatility (V) change, we can write the formula for the change in CEV-implied Spread (CDSCEV Spread) as:
CDS CEV ( S , V ) = S . CDS CEV CDS CEV + V. S V

[1]

where, CDSCEV (S,V): Change CEV implied CDS spread for a change in Stock Price (S) and Volatility (V) S: Change in Stock price CDSCEV / S: Change in CEV-implied CDS spread for a change in S V: Change in Volatility CDSCEV / V: Change in CEV-implied CDS spread for a change in V Sensitivity to Changes in Volatility It is not easy to generalise the sensitivity of our CEV spread to the volatility surface as there are many ways the volatility surface can change, most notably steepening of the term structure and / or of the skew. We consider the most important sensitivity to be how CDS would behave for a parallel shift in the volatility surface.

26

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

We can use (equity) Variance Swaps to capture a parallel shift in the volatility surface, as Variance Swaps can be thought of as a weighted average volatility across the skew of a particular maturity. This average is weighted more to the ATM volatility (forward) and hence will consequently have some exposure to the price change of the underlying (see Variance Swaps, Granger et al, Nov 2006). We can write the sensitivity of a variance swap to volatility and spot moves as:
Var ( S , V ) = S . Var Var + V. S V

[2]

The delta (sensitivity of the variance to changes in stock price) of the variance will be related to the steepness of the skew and the vega will be close to 1. Substituting the volatility surface vega (the V term) from [2] into [1] we obtain:
Var CDS CEV CDS V CEV CDS CEV ( S , V ) = S . S Var S V + Var CDS CEV V . Var V

[3]

This gives us a formula that allows us to predict the change in the CEV-implied CDS Spread based on tradable instruments Variance Swaps and Stocks and their sensitivities to Stock price moves and Volatility moves. The key point is that this allows us to reliably recreate the theoretical P+L from trading the CEV spreads with a combination of Variance Swaps and Stock. This is the missing step which will now allow us to trade the signal in our CDS Spread versus CEV-implied Spread calibrated relationship. Does our Tradable Replication Work to Mirror CEV-implied Spread Changes? Figure 28 shows an example for France Telecom of how our replicating basket has changed in line with the actual CEV spread changes. The strong relationship confirms that we can construct a way to trade the CEV Spread against the CDS spread.
Figure 28: France Telecom Replicating Basket Mirrors CEV Spread Well
x-axis: 1-month change in France Telecom CEV spread. y-axis: 1-month change in France Telecom replicating portfolio (spreads, bp)
y = 0.7732x + 4.9638 R2 = 0.7479 600 500 400 300 200 100 0 -600 -400 -200 -100 -200 -300 -400 0 200 400 600

Source: JPMorgan

27

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Looking across all the names in our universe of companies in Figure 29, we can see that the betas are concentrated around 1.0. This shows that the replicating basket estimates the right level of the CEV-implied Spread change for a given change in Stocks and Variance. Factoring in the strength of the correlation (between the replicating basket and the realised CEV spread change) by dividing the Beta by R ( / R) in Figure 30 the results only shift slightly higher showing that correlations are generally good (they don't alter the beta much).
Figure 29: Beta of Realised CEV Spread versus Predicted Spread from Figure 30: Beta / R of Realised CEV Spread versus Predicted Spread Replicating Basked (Histogram) from Replicating Basked (Histogram)
x-axis: Buckets of Beta of realised CEV Spread versus Replicating Basket Predicted Spread regression (x-axis in regression is replication basket, 1m of data) for each company for each year; y-axis: Count x-axis: Buckets of Beta / R of realised CEV Spread versus Replicating Basket Predicted Spread regression (x-axis in regression is replication basket, 1m of data) for each company for each year; y-axis: Count

400 300 200 100 0 -5.0 -4.5 -4.0 -3.5 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

300 250 200 150 100 50 0 -5.0 -4.5 -4.0 -3.5 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
Source: JPMorgan

Source: JPMorgan

The reasons the replicating basket is not 100% correlated with the actual CEV Spread move (for a given move in Stock price and Variance Swap level) are due to non-parallel shifts in the volatility surface and second-order effects (convexity) in our JPM CEV Model for moves in Stock and Variance. Overall, we can see that we can create a basket of Stocks and Variance Swaps that replicate the change in CEV-implied Spreads. This gives us a mechanism to trade our CEV-implied Spread change. Since we have shown that calibrated CEVimplied Spreads and actual CDS Spreads mean-revert, we can trade this signal. The final step is to create a trading back-test using our CEV vs CDS signal, to see how a strategy using this signal would perform.

28

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

4. Does our Trading Signal Work as a Real Trading Backtest?


4. Does our trading signal work as a real trading backtest? If w e trade on our z-score signal, is the strategy profitable.

We find that our JPM CEV Model provides profitable trading signals, with good results over time, particularly in higher volatility periods and for real outliers. The final step we need to take is to test whether our CEV vs market CDS signal works as an investment strategy. To construct a back-test based on our model we use the following trading rule as our base case parameters: i) For each day, we calculate the CEV-implied Spread (from Stock price and Option price inputs) and identify outliers by filtering companies with a Z-score >2 and where the calibrated spreads are well correlated to the market spread (R-Sq > 0.6). We stress test these assumptions later. ii) If CEV-implied Spread > market CDS Spread, construct a trading basket to: - Buy 10m 5y CDS protection - Buy stock and sell variance swaps, in the sizes given by the CEV model Greeks (our replicating basket). If CEV-implied Spread < market CDS Spread, do the reverse trades. iii) Hold the trading basket for six months and then close out. iv) Calculate the P+L on each trading basket (mark-to-market the CDS, Stock and Variance Swap). We will typically express this as an average monthly P+L on the CDS notional traded by adding the P+L from all trades over a month, divided by the notional traded to normalise for the notional exposure committed each month. 4.1 Overall Results When we look at the overall results of our trading strategy in Table 4 we can see good performance from our valuation signal. The average P+L per trade per month was 302,252 for Europe + US, which is 3.02% of the 10m CDS notional per trade per month. This gives an annualised Information Ratio of 1.96 for Europe + US (1.70 for Europe and 1.70 for the US).
Table 4: Overall Results from March 2001 to November 2007
z-score > 2, R-Sq > 0.6 Average Monthly P+L (on 10m ntnl) St Dev of Monthly P+L (on 10m ntnl) No. of Trades No. of Positive P+L Trades Max Monthly P+L (on 10m ntnl) Min Monthly P+L (on 10m ntnl) Down Std Dev Monthly P+L (on 10m ntnl) Up Std Dev Monthly P+L (on 10m ntnl) IR (annualised monthly) Sortino (annualised monthly) Success Rate (% profitable trades)
Source: JPMorgan

All 302,252 534,517 1,729 1,106 2,040,952 -583,880 145,960 520,742 1.96 7.17 64%

Europe 223,373 456,345 1,025 694 2,040,952 -572,091 183,110 430,744 1.70 4.23 68%

US 425,635 869,727 704 412 3,056,097 -798,223 205,088 874,217 1.70 7.19 59%

29

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

The trades were profitable 64% of the time overall (the Success Rate in Table 4, which is 68% for Europe and 59% in the US). Results are shown before bid-offer costs. Assuming a bid-offer of 10bp on CDS and 2 vegas on variance swaps, we estimate round-trip costs to be approximately 100,000 per 10million CDS notional trade basket. In table 4 this will reduce the Average Monthly P+L to 202,252, giving an information ratio of 1.31. As well as the Information Ratio (Average of Monthly P+L / St Dev of Monthly P+L) we look at the Sortino Ratio which shows the Average Monthly P+L / Downside St Dev of Monthly P+L (i.e. the standard deviation of the negative P+L points only). Information Ratios treat all volatility (St Dev) the same which penalises strategies that have some very high profitable trades, which are beneficial to investors. We find that our CEV Model can throw up trades that have a significant upside, so we look at the Sortino Ratio as well, which shows only the volatility of the negative P+L. Our overall Sortino Ratio of 7.17 is comfortably high and the higher value in the US is due to there being more very large positive P+L trades in the US which are ignored in the Information Ratio. Comparing a Monthly Trading Strategy to Analysis Per Trade The results above show how our trading rule would have done as a monthly trading strategy. That is, for each month that we have any trades we take an average of the P+L of the trades in that month expressed as a % of notional. We do this to allow comparison to other trading strategies or portfolio manager performance, which are generally reported on a monthly basis. However, this will have the effect of underweighting trades in months where there are a lot of signals and therefore a lot of trades. It also has the effect of reporting a lower volatility as we are taking standard deviation of monthly P+Ls which are themselves averages of trades in that month. An alternative approach would be to look at the statistics based on each trade. For example, the average P+L per trade and Information Ratio using the average and standard deviation of P+L per trade. The key results looking across each trade using our base-case parameters of Z-score > 2.0 and R-Sq > 0.6 are shown below in Table 5. For the remainder of this section we stick to showing the analysis as a monthly trading strategy.
Table 5: Per Trade Overall Results from March 2001 to November 2007
z-score > 2, R-Sq > 0.6 Average Monthly P+L (on 10m ntnl) St Dev of Monthly P+L (on 10m ntnl) No. of Trades No. of Positive P+L Trades Max Monthly P+L (on 10m ntnl) Min Monthly P+L (on 10m ntnl) Down Std Dev Monthly P+L (on 10m ntnl) Up Std Dev Monthly P+L (on 10m ntnl) IR (non-annualised) Sortino (non-annualised) Success Rate (% profitable trades)
Source: JPMorgan

All 254,232 1,050,597 1,729 1,106 9,546,682 -4,470,875 565,003 1,024,117 0.24 0.45 64%

Europe 162,305 833,142 1,025 694 5,017,068 -4,038,649 536,361 756,758 0.19 0.30 68%

US 388,074 1,292,955 704 412 9,546,682 -4,470,875 596,541 1,296,211 0.30 0.65 59%

Looking at the back-test on a per trade basis will tend to overstate the volatility as in practice you will have diversification from putting on many trades at each point. Overall the results are good; we next turn to how the strategy has done over time.

30

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

4.2 Performance Over Time Looking at how the strategy performs over time in Figure 31 we can see that it has made a positive P+L in 45 out of 62 months in which we traded since March 2001 (we only trade when we have signals of Z-score > 2 and R-Sq > 0.6 which means there are some months in which we put on no trades). Overlaying the credit spreads as an indicator of the credit cycle in Figure 31, we can see some bias to better performance when spreads are higher (in the 2002 - 2003 period). This is unsurprising as most trading strategies that look to capitalise on value signals and mean-reversion will do better in periods when volatility and therefore opportunity are greater (we saw this before in Figure 24).
Figure 31: P+L Per Month Over Time
Left axis: P+L per month on 10m CDS notional in each trade; Right axis: Credit Spreads (bp) 2,500,000

P/L per month per 10mm notional Credit Spreads (right ax is, bp)

150

2,000,000 1,500,000 1,000,000 500,000 0 -500,000 -1,000,000 Jan-01 Jul-01 Jan-02 Jul-02 Jan-03 Jul-03 Jan-04 Jul-04 Jan-05 Jul-05

100

50

-50 Jan-06 Jul-06 Jan-07 Jul-07

Source: JPMorgan. Timeline shows date trade was initiated.

The number of trades we make is not cyclical in the same way, as we can see in Figure 32 which shows the number of trades we open each month. We show the equity implied volatility to give a different indicator of the cycle through time. Whilst higher volatility periods do seem to give more trades (as in mid-2002), there are other periods of lower volatility in January 2004 for example where the model had lots of trade signals but not necessarily such profitable ones.
Figure 32: Number of Trades Per Month
Left axis: P+L per month on 10m CDS notional in each trade; Right axis: Credit Spreads (bp) 120
100 80 60 40 20 0 Jan-01 Jul-01 Jan-02 Jul-02 Jan-03 Jul-03 Jan-04 Jul-04 Jan-05 Jul-05 Jan-06 Jul-06 Jan-07 Jul-07 20 10 0 No. of trades per month Equity implied v olatility , 6m Euro Stox x (%, right ax is) 50 40 30

Source: JPMorgan. Timeline shows date trade was initiated.

31

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

This weaker performance can happen when technicals in each market mean CDS Spreads or Equity Volatility (or Stock) diverge and stay that way. This was also a feature of the more recent period when LBO risk had caused credit-equity models to perform less well as a credit spread widening was sometimes accompanied by large rises in equity value. These abrupt changes in leverage are generally an example of where structural models have the upper-hand against reduced-form models. 4.3 Comparing Our JPM CEV Model With Other Simpler Models An important analysis is to look at how our JPM CEV Model compares with other, simpler strategies. As we have shown, although the JPM CEV Model involves simple inputs and can be constructed with tradable instruments, it is clearly more complicated to implement than just comparing observed stock prices and CDS spreads in a linear plot (regression). Figure 33 shows this regression of stock price against CDS spread for British Airways, which could be used to spot trade opportunities when the stock price / CDS spread relationship is outside its normal parameters. Whilst there are technical arguments against such regressions such as the problem of regressing stationary (credit spread) and non-stationary (equity price growth) variables we also need to see whether our JPM CEV Model outperforms other simpler models. In other words, is it worth the effort?
Figure 33: Example of Regression of British Airways Stock Price Against CDS Spread Changes
x-axis: Stock Price (); y-axis: 5y CDS Spread (bp)

180 160 140 120 100 80 60 300 350 400 450 500 550 600
Source: JPMorgan

y = -0.2682x + 223.83 R 2 = 0.8658

Table 6 shows the summary of the performance comparison of our JPM CEV model to a Stock Price Regression model (as in Figure 33), a Variance Regression (like Figure 33 but with implied volatility or variance as the x-axis) and a Multiple (linear) regression of both Stock Price and Variance against CDS Spreads. We can see that the CEV model outperforms on the key statistics, with a much higher Information Ratio and Average Monthly P+L figure. The Variance regression is also a fair strategy, but Stock Price or the Multiple regression are weaker trading signals.

32

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Table 6: Comparison of CEV Performance to Regression Strategies


Average Monthly P+L (on 10m ntnl) St Dev of Monthly P+L (on 10m ntnl) No. of Trades No. of Positive P+L Trades Max Monthly P+L (on 10m ntnl) Min Monthly P+L (on 10m ntnl) Down Std Dev Monthly P+L (on 10m ntnl) Up Std Dev Monthly P+L (on 10m ntnl) IR (annualised monthly) Sortino (annualised monthly) Success Rate (% profitable trades)
Source: JPMorgan

CEV 302,252 534,517 1,729 1,106 2,040,952 -583,880 145,960 520,742 1.96 7.17 64%

Price Regression 39,192 448,578 5,013 2,940 1,277,781 -971,112 287,345 288,412 0.30 0.47 59%

Var Regression 257,551 758,485 2,717 1,843 3,356,311 -3,646,453 850,176 550,768 1.18 1.05 68%

Multi Regression 29,097 355,068 6,676 3,952 868,258 -1,017,527 254,941 196,728 0.28 0.40 59%

Figure 34 and Figure 35 show the comparison for the Information Ratio and Monthly P+L Per 10m notional, clearly indicating the better performance of our JPM CEV Model compared to other strategies. We can also see the next best strategy is the Variance regression model and that any strategy that uses Stock Price (Price Regression and Multi-Linear Regression) performs poorly.
Figure 34: Information Ratios Across Strategies
Annualised Information Ratios

Figure 35: Monthly P+L Per 10m Notional Across Strategies


Monthly P+L () Per 10m Notional CDS Traded

2.50 2.00 1.50 1.00 0.50 0.00

1.96 1.18 0.30

350,000 300,000 250,000 200,000 0.28 150,000 100,000 50,000 0

302,252 257,551

39,192

29,097

CEV

Price Regression

Var Regression

Multi Regression
Source: JPMorgan

CEV

Price Regression

Var Regression

Multi Regression

Source: JPMorgan

In our view, this is a further justification that our JPM CEV Model is a better model to use. It has more theoretical justification, allows accurate construction of trade strategies (has a way to give trade Greeks) and also performs better as an investment signal. 4.4 Stress Testing Our Trade Signal Filters Varying the Z-Score Filter We finish our back-testing by looking at stress-testing some of our investment signal parameters. In particular, we have restricted our trading to opening a credit-equity basket trade only when the (calibrated) relationship between CEV-implied Spread and actual market CDS levels is sufficiently out of line: the Z-score must be at or above 2 (that is 2 standard deviations out of line). We also ensure that the signal is a real one by restricting our trades to where the R-Sq of the regression is above 0.6, in other words, there is a reasonably significant historical relationship between the CEV-implied Spread and the market CDS Spread. We stress-test the stability of our results by varying these two filters.
33

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

In Table 7 we show the impact of using a different Z-score filter, starting from 0 (meaning we trade every divergence) up to a Z-score of 4. We keep the restriction constant that the R-Sq is above 0.6. The higher the Z-score filter the more the strategy is focused on significant divergences from the historical relationship between CEV-implied Spread and CDS Spread. We see that the Information Ratios are generally increasing with Z-score (Z-score of 3 is an anomaly) and the Success Rates improve with higher Z-Scores.
Table 7: Z-Score Stress Testing
Z Score Average Monthly P+L (on 10m ntnl) St Dev of Monthly P+L (on 10m ntnl) No. of Trades No. of Positive P+L Trades Max Monthly P+L (on 10m ntnl) Min Monthly P+L (on 10m ntnl) Down Std Dev Monthly P+L (on 10m ntnl) Up Std Dev Monthly P+L (on 10m ntnl) IR (annualised monthly) Sortino (annualised monthly) Success Rate (% profitable trades)
Source: JPMorgan

0 25,555 261,613 17,341 9,041 529,050 -775,939 173,020 167,227 0.34 0.51 52%

1 84,921 391,587 6,955 3,905 1,275,268 -903,605 244,967 284,280 0.75 1.20 56%

2 302,252 534,517 1,729 1,106 2,040,952 -583,880 145,960 520,742 1.96 7.17 64%

3 476,244 933,308 357 249 4,407,827 -1,120,490 384,101 903,120 1.77 4.30 70%

4 819,453 804,566 100 84 2,643,382 -75,960 0 798,898 3.53 84%

Figure 36 shows this relationship for these key metrics. This indicates that our JPM CEV Model is particularly strong in highlighting real divergences of credit and equity markets. We would really focus it on higher Z-scores and chose 2 as our base case in order to balance the number of trades (Count) and Information Ratios to give us more frequent trading opportunities than we would get with a higher Z-Score.
Figure 36: Stress Test of Different Z-Score Signals
x-axis: Z-score threshold; y-axes: Information Ratio (left axis) and Success Rate (%, right axis)

Figure 37: Stress Test of Different R-Sq


x-axis: Z-score threshold; y-axes: Information Ratio (left axis) and Success Rate (%, right axis)

4.00 3.50 3.00 2.50 2.00 1.50 1.00 0.50 0.00 0


Source: JPMorgan

IR (annualised monthly ) Success Rate (% profitable trades, right ax is)

IR (annualised monthly )
3.53 100% 80%

2.00 1.80 1.60 1.40 1.20 1.00

Success Rate (% profitable trades, right ax is) 1.96 1.93 1.86 1.77

100% 80% 60% 40% 20% 0%

1.96 0.75 0.34

1.77

60% 40% 20% 0%

4
Source: JPMorgan

0.2

0.4

0.6

Varying the R-Squared Filter Table 8 and Figure 37 show a similar stress test varying the R-Squared filter. By only trading when the R-sq is above a certain level, we attempt to ensure there is some significance to the relationship between the CEV-implied Spread and the market CDS Spread. If the R-Sq is very low, it indicates the relationship is poor and so any signal or divergence from the relationship may be spurious.

34

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

We find that whilst the Average Monthly P+L increases as the R-Squared filter increases justifying our decision to add this filter the Information Ratios and Success Rates vary little as we increase the R-Sq. We do find the Sortino Ratios increase, showing that the R-Sq filter helps to minimise the downside volatility by taking out spurious signals.
Table 8: R-Sq Stress Testing
R-Sq Average Monthly P+L (on 10m ntnl) St Dev of Monthly P+L (on 10m ntnl) No. of Trades No. of Positive P+L Trades Max Monthly P+L (on 10m ntnl) Min Monthly P+L (on 10m ntnl) Down Std Dev Monthly P+L (on 10m ntnl) Up Std Dev Monthly P+L (on 10m ntnl) IR (annualised monthly) Sortino (annualised monthly) Success Rate (% profitable trades)
Source: JPMorgan

0 140,771 252,170 12,906 8,494 882,067 -554,538 129,676 217,454 1.93 3.76 66%

0.2 197,230 368,189 5,836 3,801 1,799,673 -425,709 116,066 347,213 1.86 5.89 65%

0.4 230,980 452,499 3,247 2,099 1,900,129 -437,559 113,532 457,018 1.77 7.05 65%

0.6 302,252 534,517 1,729 1,106 2,040,952 -583,880 145,960 520,742 1.96 7.17 64%

To conclude our back-test, we are satisfied that our JPM CEV Model provides a successful investment signal and outperforms other linear models. This along with its theoretical soundness and ability to construct useful trade baskets from tradable instruments demonstrates its use in our view. In a further note we will focus more on investment signals and trade construction, as well as introducing our analytics reports.

35

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

5. Conclusions
In this note we introduced a new framework for investing across equities and credit the JPM CEV Model. We have shown how this framework allows us to: Identify companies where the credit (CDS) spread is too high or too low given the pricing in the equity markets. It does this by calculating an implied CDS spread from the equity markets using the inputs of the stock price and implied volatility, using a CEV model to imply a default probability. We analysed how we could profit from these mis-pricings by using tradable instruments: a company's stock, equity variance swaps and credit default swaps. The model provides both trade structure and sensitivities (Greeks). Finally, we have seen that it is a reliable trading signal and saw how our model would have given a profitable investment strategy historically.

Whilst this is unlikely to be the final word on credit-equity investing, we do think it gives us a useful investment framework which is theoretically sound, practically tradable and empirically shown to work. We will be using this in our future analysis and trade ideas across both credit and equities. This note has focused on introducing our framework for credit-equity investing. We will look to follow this with a more practically orientated note showing how we use our trading signals. We will also give more detail on how we deal with trade construction and the Greeks (sensitivities) of our trades involving stock, variance swaps and CDS. We also plan to make available a set of regular reports with the outputs of our model to highlight where we see investment opportunities across credit and equities markets.

36

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Appendix I: The Merton Model Option Pricing Formula


In this appendix we outline the mathematics of the Merton Model for pricing a corporate bond or stock as an option on the companys assets11. The Underlying Asset Value Process The fundamental premise of the Merton model is that both debt (credit) and equity are themselves derivatives on a company's asset value (we will denote this company asset value by V). This asset value process V is stochastic and has a drift () and volatility component (). We will need to describe how this asset value V changes (dV) over time. The process describing the change in asset value (dV) is geometric Brownian motion: dV = Vdt + VdZ (1)

where, dZ represents a generalized Wiener process (a random process). Relating Bonds (Credit) and Shares to Company Asset Value The fundamental accounting identity of a firms balance sheet states that a firms asses must equal its equity (share price, S) plus its liabilities (i.e. debt, D): V=S+D (2)

Using the simplifying assumption that all debt (i.e. all bonds) matures at the same time T we can calculate the price of the bonds of a company (BT) and equity (ST) at maturity T as function of the company asset value VT and debt level DT. BT = min(DT,VT) ST = max(VT - DT,0) (3a) (3b)

(3a) and (3b) show that both bonds (credit) and equities have the same underlying source of uncertainty, the company's asset value V. We will look to solve for the share price of a company, St. Creating a Riskless Portfolio of Asset Value and Shares We will look to calculate the companys share price St as a function of the asset value of the company. Using (1) above, we can see that the change in asset value dV is a function of both V and t. Given this, we can use Its lemma to show how a derivative of V changes over time in our case this derivative is S (the share price) and its change is therefore dS.
S S S 1 2 S 2 2 V VdZ V+ dt + dS = + 2 V 2 t V V

(4)

11

For more information readers are encouraged to read Schnbucher (2003). We have relied on the clarity of his explanation.
37

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Where, S = the value of the shares of the company V = the asset value of the company = the drift of V = the volatility of V dZ = a random (Wiener) process Or in discrete time (using t as the discrete time period):

S = V + 2V 2 t + VZ + 2 V 2 V t V

1 2S

(5)

As the share price S is a function of V and time, S = f(V,t), the random component (the Wiener process Z) is the same in both S and V. By choosing a portfolio of the asset value of the company V and its shares S a derivative of V we can eliminate this random component. [Even though we cannot buy the asset value of the company directly using tradable instruments, we could in theory create this synthetically using equation (2) by using bonds and shares of the company.] We will set up this portfolio as: -1: Shares (the derivative)

S : Assets V

If we represent the value of this portfolio as then, by definition:


= S + S V V

(6)

A change in the value of the portfolio over time interval t is:

= S +

S V V

(7)

Combining V from (1) and S from (5) with equation (7), and a bit of cancelling gives:

S 1 2 S 2 2 V 2 t t 2 V

(8)

Because equation (8) has no random element its payoff is known, i.e. it is riskless. So, must earn the riskless rate of return r, over any period. This means:

= r t

(9)

Substituting from (6) and from (7) and swapping sides / signs gives:

38

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

S 1 2 S 2 2 S + V t = r S V t t 2 V 2 V

(10)

Canceling t and re-arranging gives:

S S 1 2 S 2 2 + rV + V = rS t V 2 V 2

(11)

This is the Black-Scholes partial differential equation. Any derivative S (the value of the shares) of the asset value of the firm must satisfy this equation. We can now use this to value our share price and bonds.
Calculating the Value of the Shares and Bonds as Options We can solve the Black-Scholes equation with the boundary conditions as the value of the share at option maturity T. In (3b) before, we saw that value of the shares (ST) at time T has the payoff of a call option on the company's asset value with strike DT (the value of the debt at T):

ST = max(VT - DT,0)

(see 3b before)

Solving the Black-Scholes equation with this boundary condition gives: S = V0N(d1) - DTe-rtN(d2) Where,
d1 = ln(Vo / DT ) + (r + ( 2/ 2))T

(12)

T
ln(Vo / DT ) + (r ( 2/ 2))T

(12a)

d2 =

(12b)

We have therefore solved for the value of the shares S given the asset value V0, the volatility of the asset value , the value of the debt at T (DT) and the risk free rate r. Using put-call parity we can now give the value of the bonds of the company B. S - (DTe-rt - B) = V0 - DTe-rt B = V0 - S (13, put-call parity) (14)

This is exactly what the Merton Model allows us to do. We can calculate the value of the shares (S) of a company or its bonds (B) by knowing its asset value (V), asset value volatility (), outstanding debt (D), interest rates and time. Once we know the bond price (B) we can simply calculate the bond spread, which is usually considered in the market as the credit risk compensation of the bond and most comparable to a CDS spread.

39

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Appendix II: The CEV Model from Equity Inputs to Credit Spreads
The aim of the JPM CEV model is to imply a credit spread from equity market inputs. It does this using a reduced-form Merton approach and a CEV option pricing framework. The JPM CEV model aims to calculate a cumulative default probability of a company over a given time horizon (e.g. 5 years) as implied by the equity market. Once we have an implied default probability, we can use standard CDS pricing models to calculate the spread on a par CDS given this default probability and an assumption about recovery rates. It is clear that the main modeling challenge that needs to be overcome is to infer a default probability distribution from equity market levels. This is the main aim of the JPM CEV model.
Figure 38: Schematic of JPM CEV Model
1. Market Inputs

A simplified outline of the JPM CEV model steps is shown in Figure 38. This appendix deals with the mechanics and mathematics of the model.

Equity price

Equity implied option v olatilities

1. Market Inputs
The Constant Elasticity of Variance (CEV) process describes the evolution of a share price S as: dS = Sdt + SdZ where, = The drift of the share price = The volatility (standard deviation) parameter of the share price = An elasticity parameter This contrasts with a Black-Scholes (B-S) model of the evolution of a share price S as: (1)

2. Solve for and

This creates an equity process consistent w ith market inputs

3. Calculate Default Probability

Using equity process, calculate probability equity price = 0, i.e. default

4. Calculate CDS Spread

Use default probability and recov ery rate to calculate CDS spread

dS = Sdt + SdZ

(2)

5. Investment Signal

Compare market CDS spread to CEVimplied CDS spread for signal


Source: JPMorgan

In the CEV model, we will need to estimate both and in order to derive this process. In the standard B-S model, we only need an estimate for , the volatility. As in a B-S model, we could take these parameters from historical observations or we can imply them from the traded market prices of options (in the equity markets as standard we refer to this implied parameter as implied volatility). For our CEV model, we also imply the values of and from the market and use equity price and options prices to do this. As we have two parameters ( and ) we will need more than one option price in order to solve for these two parameters. The inputs we use to do this are the options prices in the market. These are shown in Figure 39 (this shows the implied volatilities, we input the corresponding options prices into the CEV model). We use the 70% -110% strikes for both 6m and 1y maturity options.

40

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Figure 39: CEV Model Inputs


B-S Vols for Option Prices at 70-110 Strikes, for 6m to 1y Maturity

30% 28% 26% 24% 22% 20% 60


Source: JPMorgan

28.9% 6M 26.1% 27.7% 25.6% 23.8% 22.5% 70 80 90 100 110 23.8% 22.1% 1Y

A standard Black-Scholes option pricer gives us the price of European calls given these observed volatilities.

2. Solve for and


We now have the prices for European-style options. To calculate the values of and implied by these, we need to use the price of a call (or we could use the price of a put) in a CEV model. The price of a European call and put under the CEV model where < 1 is given as12: c = S0e-qT[1-2(a,b+2,c)]-Ke-rT 2(c,b,a) p = Ke where,
a= [ Ke ( r q )T ] 2(1 ) (1 ) 2 v
-rT

(3a) (3b)

[1- (c,b,a)]-S0e (a,b+2,c)

-qT 2

b=

1 , 1

c=

S 2(1 ) (1 ) 2 v

with v =

2
2(r q)( 1)

[e 2( r q )( 1)T 1]

where,
S0: the current share price q: the dividend yield T: the time to maturity r: the risk-free rate K: the option strike 2(z,k,v): the cumulative probability that a variable with a non-central 2 distribution with non-centrality parameter v and k degrees of freedom is less than z.

12

See Computing the Constance Elasticity of Variance Option Pricing Formula, M Schroder, Journal of Finance (March 1989) for a derivation of this. We use the notion in Hull for clarity. Readers can see Hull for the CEV formula when >1, which we skip for brevity.
41

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Fitting the Model For a standard Black-Scholes model, we only imply a value of (the volatility) so only need one option price to do this. For our CEV model we need at least two as we need to solve equations (3a) and (3b) for both a value of and for . The fitting algorithm we use is as follows:

We take 6-month and 1-year expiry calls and puts struck between 70% and 110% of the share price at intervals of 10%. We fit the CEV model option prices to the observed option prices in a non-linear least-squares sense. For example, if the underlying share price is 100, we use the 6-month and 1-year options taking strikes of 70, 80, 90, 100 (put options) and 110 (call option). This gives us 10 observed options prices. We use an initial estimate for the starting values of and and price each of the options under the CEV model. We then compute the sum of the squared differences between the CEV prices and the observed market prices. We then iterate selecting new values of and to reduce this sum of squares until a minimum is found. The results of this process are values of and that best fit the current market option prices. This describes the process of the underlying share price and hence the probability distribution for the share price for different points in time13.

3. Calculating the Default Probability


We saw that we could fit the values of and from the options prices observed in the market. This allows us to fit a probability distribution (a non-central chi-squared distribution) of the share price of the firm. In our reduced-form model, we define the probability of default as the probability that the share price hits zero over a time horizon T we are considering (i.e. 1y for a 1y credit default swap). Recalling the option pricing formulae in (3a) and (3b), we can see the probability of default over time horizon T is given by: [Prob of Default Prior to T] = Pr(ST = 0) = [1-2(c,b,a)] using the distribution with the and we inferred. This allows us to calculate a cumulative probability of default and survival probability (1 - Cum Prob of Default) for any time horizon T. If TCDS represents the maturity of the CDS we wish to value, we calculate a probability of default for each CDS cashflow i, for i = 1 to TCDS , where is the year fraction of the cashflows of the CDS (e.g. 0.25 for a standard CDS that pays cashflows every quarter). We bootstrap the CDS curve in that we use the conditional default probability (hazard rate) implied by the CEV model for each period when calculating the spread to a given maturity.

13

This probability distribution is a non-central chi-squared probability distribution, not the standard normal distribution that we are used to using in the Black-Scholes framework.
42

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

4. Calculating the CDS Spread


We are now able to calculate a default and survival probability for each CDS cashflow. This is not quite enough to price our CDS. Equation (4) shows the pricing formula for a par credit default swap in discrete time. The variable we are missing is R, the recovery rate. In order to help price under standard market assumptions (to best approximate a market CDS) we use a recovery rate of 40% and solve the CDS pricing equation so find ST, which is the CDS spread for a par CDS with maturity T.
Sn.

i =1

i.Psi.DFi + Accrual on Default = (1 R ).

( Ps
i =1

(i 1)

Psi ).DFi

(4)

PV(Fee Leg) Where, Sn = Spread for protection to period n i = Length of time period i in years Psi = Probability of Survival to time i DFi = Risk-free Discount Factor to time i R = Recovery Rate on default

PV(Contingent Leg)

Accrual on Default = Sn.

2 .( Ps
i =1

(i 1) Psi ).DFi

5. Using the JPM CEV-Implied CDS Spreads as an Investment Signal


Our CEV model gives us a term structure of default probabilities. With a given recovery rate assumption we can then imply the spread on a par CDS. In practice we calibrate the model values against the market CDS spreads to derive an investment signal using a Z-score. For more details on the steps taken to calibrate and derive our investment signal using this Z-score measure, see Section 4 in the main body of the note.

43

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Appendix III: Universe of Companies For Back-test Results


Table 9: European Companies Used in Back-test (Source: JPMorgan)
Company Name TELEKOM AUSTRIA SOLVAY FORTUM OYJ STORA ENSO OYJ-R SHS UPM-KYMMENE OYJ ACCOR ALCATEL-LUCENT ALSTOM CAP GEMINI CARREFOUR CASINO GUICHARD PERRACHON COMPAGNIE DE SAINT-GOBAIN ELECTRICITE DE FRANCE FRANCE TELECOM GAZ DE FRANCE GECINA HAVAS LAFARGE LVMH MOET HENNESSY LOUIS VUI MICHELIN (CGDE)-B PERNOD-RICARD PEUGEOT PPR PUBLICIS GROUPE RALLYE RENAULT REXEL RHODIA SA - REGR SANOFI-AVENTIS SUEZ THOMSON (EX-TMM) VALEO VEOLIA ENVIRONNEMENT VINCI VIVENDI BAYER BAYERISCHE MOTOREN WERKE CONTINENTAL DAIMLER DEUTSCHE LUFTHANSA DEUTSCHE TELEKOM Company Name E.ON ENBW ENERGIE BADEN-WUERTTEMB FRESENIUS SE-PFD HEIDELBERGCEMENT HENKEL KGAA-VORZUG INFINEON TECHNOLOGIES LINDE METRO PROSIEBEN SAT.1 MEDIA AG-PFD RWE SIEMENS THYSSENKRUPP TUI VOLKSWAGEN HELLENIC TELECOMMUN ORGANIZA EDISON ENEL FIAT SEAT PAGINE GIALLE TELECOM ITALIA ARCELORMITTAL AKZO NOBEL EUROPEAN AERONAUTIC DEFENCE KONINKLIJKE AHOLD KONINKLIJKE DSM KONINKLIJKE KPN PHILIPS ELECTRONICS STMICROELECTRONICS VNU WOLTERS KLUWER NORSK HYDRO ASA TELENOR ASA ENERGIAS DE PORTUGAL PORTUGAL TELECOM SGPS ALTADIS ENDESA GAS NATURAL SDG IBERDROLA REPSOL YPF SOL MELIA TELEFONICA Company Name UNION FENOSA ELECTROLUX AB-SER B SAS AB TELIASONERA AB VOLVO AB-B SHS ABB LTD ADECCO CIBA SPECIALTY CHEMICALS CLARIANT ALLIANCE BOOTS BRITISH AIRWAYS BRITISH AMERICAN TOBACCO BT GROUP CABLE & WIRELESS CADBURY SCHWEPPES CENTRICA COLT TELECOM GROUP COMPASS GROUP CORUS GROUP PLC. DIAGEO DSG INTERNATIONAL EMI GROUP EXPERIAN GROUP LTD FKI GALLAHER GROUP PLC GKN HANSON PLC. IMPERIAL CHEMICAL INDS IMPERIAL TOBACCO GROUP INMARSAT INTERNATIONAL POWER INVENSYS ITV KINGFISHER LADBROKES MARKS & SPENCER GROUP MORRISON <WM.> SUPERMARKETS NATIONAL GRID PEARSON RANK GROUP REED ELSEVIER Company Name REUTERS GROUP SAINSBURY (J) TATE & LYLE TESCO UNILEVER UNITED UTILITIES VODAFONE GROUP WPP GROUP

44

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Table 10: US Companies Used in Back-test (Source: JPMorgan)


Company Name AMERISOURCEBERGEN CORP AES CORP AK STEEL HOLDING CORP ADVANCED MICRO DEVICES AMKOR TECHNOLOGY INC AMR CORP AMERICAN TOWER CORP-CL A ARVINMERITOR INC ALLIED WASTE INDUSTRIES INC AMERICAN AXLE & MFG HOLDINGS ALLEGHENY ENERGY INC BEAZER HOMES USA INC AVIS BUDGET GROUP INC CROWN CASTLE INTL CORP CLEAR CHANNEL COMMUNICATIONS CHESAPEAKE ENERGY CORP CHARTER COMMUNICATIONS-CL A CELESTICA INC CUMMINS INC CMS ENERGY CORP COOPER TIRE & RUBBER CABLEVISION SYSTEMS-NY GRP-A CITIZENS COMMUNICATIONS CO DILLARDS INC-CL A DR HORTON INC ECHOSTAR COMMUNICATIONS - A DYNEGY INC-CL A EASTMAN KODAK CO EL PASO CORP FORD MOTOR CO FLEXTRONICS INTL LTD FOREST OIL CORP CORNING INC GENERAL MOTORS CORP GOODYEAR TIRE & RUBBER CO HCA Inc HARRAH'S ENTERTAINMENT INC STARWOOD HOTELS & RESORTS HOST HOTELS & RESORTS INC HERTZ GLOBAL HOLDINGS INC Company Name IDEARC INC IKON OFFICE SOLUTIONS INC IRON MOUNTAIN INC J.C. PENNEY CO INC KB HOME LEAR CORP LENNAR CORP-CL A L-3 COMMUNICATIONS HOLDINGS LEVEL 3 COMMUNICATIONS INC LYONDELL CHEMICAL COMPANY MEDIACOM COMMUNICATIONS-CL A MASSEY ENERGY CO MGM MIRAGE MIRANT CORP MOSAIC CO/THE NAVISTAR NALCO HOLDING CO NRG ENERGY INC NORTEL NETWORKS CORP OWENS-ILLINOIS INC PRIDE INTERNATIONAL INC PARKER DRILLING CO POLYONE CORPORATION QWEST COMMUNICATIONS INTL RITE AID CORP REYNOLDS AMERICAN INC ROYAL CARIBBEAN CRUISES LTD R.H. DONNELLEY CORP RELIANT ENERGY INC RADIOSHACK CORP SANMINA-SCI CORP SERVICE CORP INTERNATIONAL SMITHFIELD FOODS INC ISTAR FINANCIAL INC SIX FLAGS INC SAKS INC SOLECTRON CORP. STANDARD-PACIFIC CORP SMURFIT-STONE CONTAINER CORP TENET HEALTHCARE CORP Company Name TRIAD HOSPITALS INCO. TRW AUTOMOTIVE HOLDINGS CORP TESORO CORP TXU CORP DOMTAR CORP UNISYS CORP UNUM GROUP UNITED RENTALS INC VISTEON CORP WINDSTREAM CORP WILLIAMS COS INC UNITED STATES STEEL CORP XEROX CORP

45

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Bibliography
Papers
The Merton Framework: Theory, Implementation and Testing On the Pricing of Corporate Debt: The Risk Structure of Interest Rates, R Merton, The Journal of Finance, Vol. 29 (May 1974)

EDFTM Credit Measure and Corporate Bond Pricing, KMV, O Vasicek (Nov 2001) Using Equities to Price Credit, JPMorgan, M King (Sep 2001) Response to JPMorgans Paper, Using Equities to Price Credit, KMV, J Bohn (Nov 2001) CreditGrades Technical Document, RiskMetrics, C Finger (May 2002) Using Equity Prices to Trade Corporate Bonds: An Evaluation, JPMorgan, P Rappoport (Aug 2002) Modeling Default Risk: Modeling Methodology, Moodys KMV, P Crosbie, J Bohn (Dec 2003) Reduced Form vs Structural Models of Credit Risk: A Case Study of Three Models, Moodys KMV, N Arora, J Bohn, F Zhu (Feb 2005)
The CEV Formula for Option Pricing and Empirical Support The Valuation of Options for Alternative Stochastic Processes, J Coxx and S Ross, Journal of Financial Economics (1976)

Computing the Constance Elasticity of Variance Option Pricing Formula, M Schroder, Journal of Finance (March 1989) The Constant Elasticity of Variance Model and Its Implications For Option Pricing, S Beckers, The Journal of Finance (June 1980) gives empirical evidence of the leverage effect and support for a CEV process with alpha < 1. The Stochastic Behaviour of Common Stock Variances, A Christie, Journal of Financial Economics (1982).

Books
Credit Derivatives Pricing Models, P Schnbucher, chpt 9, (2003) Understanding Credit Derivatives and Related Instruments, A Bomfim, chpt 17 (2005) Options, Futures and Other Derivatives, J Hull, 5th edition chpt 20 on CEV models and chpt 26 on equity/credit pricing (2003)

46

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

47

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

48

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

49

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

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 analysts 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 Credit Research Ratings: Ratings System: JPMorgan uses the following sector/issuer portfolio weightings: Overweight (over the next three months, the recommended risk position is expected to outperform the relevant index, sector, or benchmark), Neutral (over the next three months, the recommended risk position is expected to perform in line with the relevant index, sector, or benchmark), and Underweight (over the next three months, the recommended risk position is expected to underperform the relevant index, sector, or benchmark). JPMorgans Emerging Market research uses a rating of Marketweight, which is equivalent to a Neutral rating. Valuation & Methodology: In JPMorgans credit research, we assign a rating to each issuer (Overweight, Underweight or Neutral) based on our credit view of the issuer and the relative value of its securities, taking into account the ratings assigned to the issuer by credit rating agencies and the market prices for the issuers securities. Our credit view of an issuer is based upon our opinion as to whether the issuer will be able service its debt obligations when they become due and payable. We assess this by analyzing, among other things, the issuers credit position using standard credit ratios such as cash flow to debt and fixed charge coverage (including and excluding capital investment). We also analyze the issuers ability to generate cash flow by reviewing standard operational measures for comparable companies in the sector, such as revenue and earnings growth rates, margins, and the composition of the issuers balance sheet relative to the operational leverage in its business.
JPMorgan Credit Research Ratings Distribution, as of September 28, 2007 Overweight EMEA Credit Research Universe IB clients* 24% 60% Neutral 58% 69% Underweight 19% 60%

Represents Ratings on the most liquid bond or 5-year CDS for all companies under coverage. *Percentage of investment banking clients in each rating category.

Analysts Compensation: The 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. The firms overall revenues include revenues from its investment banking and fixed income business units.

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 Corporations Characteristics and Risks of Standardized Options, please contact your JPMorgan Representative or visit the OCCs website at http://www.optionsclearing.com/publications/risks/riskstoc.pdf. Legal Entities Disclosures U.S.: JPMSI is a member of NYSE, FINRA and SIPC. J.P. Morgan Futures Inc. is a member of the NFA. JPMorgan 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. Registered in England & Wales No. 2711006. Registered Office 125 London Wall, London EC2Y 5AJ. 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) is regulated by ASIC and J.P. Morgan Securities Australia Limited (ABN 61 003 245 234/AFS Licence No: 238066) is a Market Participant with the ASX and regulated by ASIC. Taiwan: J.P.Morgan Securities (Taiwan) Limited is a participant of the Taiwan Stock Exchange (company-type) 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
50

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007

Stock Exchange and Surabaya Stock Exchange and is regulated by the BAPEPAM. Philippines: J.P. Morgan Securities Philippines Inc. is a member of the Philippine Stock Exchange and is regulated by the Securities and Exchange Commission. Brazil: Banco J.P. Morgan S.A. is regulated by the Comissao de Valores Mobiliarios (CVM) and by the Central Bank of Brazil. Mexico: J.P. Morgan Casa de Bolsa, S.A. de C.V., J.P. Morgan Grupo Financiero is a member of the Mexican Stock Exchange and authorized to act as a broker dealer by the National Banking and Securities Exchange Commission. Singapore: This material is issued and distributed in Singapore by J.P. Morgan Securities Singapore Private Limited (JPMSS) [mica (p) 030/09/2007 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 (18146-x) which is a Participating Organization of Bursa Malaysia Securities Bhd and is licensed as a dealer by the Securities Commission in Malaysia. Pakistan: J. P. Morgan Pakistan Broking (Pvt.) Ltd is a member of the Karachi Stock Exchange and regulated by the Securities and Exchange Commission of Pakistan. 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 JPMSLs Policies for Managing Conflicts of Interest in Connection with Investment Research which outline the effective organisational and administrative arrangements set up within JPMSL for the prevention and avoidance of conflicts of interest with respect to research recommendations, including information barriers, and 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 2005 (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 fr 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. Japan: There is a risk that a loss may occur due to a change in the price of the shares in the case of share trading, and that a loss may occur due to the exchange rate in the case of foreign share trading. In the case of share trading, JPMorgan Securities Japan Co., Ltd., will be receiving a brokerage fee and consumption tax (shouhizei) calculated by multiplying the executed price by the commission rate which was individually agreed between JPMorgan Securities Japan Co., Ltd., and the customer in advance. Financial Instruments Firms: JPMorgan Securities Japan Co., Ltd., Kanto Local Finance Bureau (kinsho) No. [82] Participating Association / Japan Securities Dealers Association, The Financial Futures Association of Japan. 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. Pakistan: 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 analysts 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 non-U.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. Other Disclosures last revised November 5, 2007.

Copyright 2007 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.

51

Jonny Goulden (44-20) 7325-9582 jonathan.m.goulden@jpmorgan.com Peter S Allen (44-20) 7325-4114 peter.allen@jpmorgan.com

European Corporate Research 19 November 2007