S2 Ap SDF Ccapm F 2019

Empirical Finance
Abhay Abhyankar
University of Exeter, U.K.
Abhay Abhyankar University of Exeter, U.K. (Universities of

Empirical
Somewhere
Finance
and Elsewhere) 1/4
Road Map
Financial markets & Trading, Data sources etc.
Asset Pricing: Consumption-based & Linear Factor Models.
Asset Pricing: Empirical Test Strategies.
Efficient Markets & Event Studies.
Present Value Models & Return Predictability.
Student Presentations.
1
Empirical Facts - Stock

0.1 versus Bonds
1900 20 40 60 80 2000
Equities, 9.4% per year Bonds, 4.8% per year
Stocks have paid o↵ considerably
Bills, 3.9% per year moreInflation,
than3.0% bonds
per year
$VNVMBUJWF3FUVSOTPO64&RVJUJFT
1900 = $1
#POET #JMMT BOE*OGMBUJPO oReal Terms
10,000
1,000 $851×
100
10
$7.5×
$2.9×
1
0.1
1900 20 40 60 80 2000
Equities, 6.3% per year Bonds, 1.8% per year
Bills, 1.0% per year
Source: Based on Dimson, Marsh, and Staunton (2002) and as updated

in Dimson, Marsh, and Staunton (2011b).
Prof. Abhay Abhyankar (University of Exeter) Empirical Finance UAB April 2017 7/1
Stocks versus Bonds
Why don’t all investors just buy equities?

Stocks versus Bonds
40.00%
20.00%
0.00%
1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
-20.00%
-40.00% S&P 500 3-month T.Bill 10-year T. Bond
Stocks versus Bonds ERP.book Page 92 Wednesday, December 21, 2011 9:06 AM
GDP growth andthestock

Rethinking market
Equity Risk Premium performance.
Growth in U.S. Real GDP, Real per Capita GDP, Real
Stock Price Return, Real Earnings, and Real Dividends
(lognormal scale)
1800 Real GDP = 100

1,000,000
100,000
10,000
1,000
100
10
1800 20 40 60 80 1900 20 40 60 80 2000
Real GDP Real per Capita GDP
Real Stock Price Return Real Dividends ×10
Real Earnings ×3
Source: Based on data from CRSP, Morningstar (Ibbotson), Robert Shiller, and
William Schwert.
The statistical danger thisexpected story points to is None
returns. mental beliefs abouttothevariation
corresponds economy. inWas it clear dividend growth, and non
expected
selection or survival bias. If you flip one
tional coin in which
bubbles” to people in 1945
prices (or 1871,
are high justoronwhenever the
the expectation of even higher future p
Price Dividend Ratio - Predictability ten times, the chance of seeing eight heads is
constant expected returns.
low. But if you flip ten coins ten times, the
sample starts) and throughout the period
the average return on stocks would be 8 percent
that
chance that the coin with the greatest Figure

number 2 gives another viewthat
greater than of ofthe sameIf facts
bonds? so, one(this
wouldis an interpretive view of the
of heads exceeds eight heads isresponse much larger. function toexpect
a dp shock
them towith
have no change
bought moreinstocks,
currentevendividends). Suppose you se
Does this story more closely capture decline thein
50-year considering
prices relative the risk described
to dividends, by the 17 What
as illustrated. per- would you forecast? In t
return on U.S. stocks? Brown,view, Goetzmann,you wouldand forecast
cent year-to-year
no change variation. But perhaps
in future prices, itbut
wasa decline in future divide
Dividend growth or Expected returns? What drives P/D?
Ross (1995) present a strong case that the uncer-
current view is exactly not in
thefact obvious in 1945, that rather than
reverse.
tainty about true average stock returns is much slipping back into depression, the U.S. would
larger than σ / T suggests. As they put it,
Use DY and PE for the S&P 500 index over the period 1872 to 2000 (annual data). experience a half century of growth never be-
“Looking
For upto-date data on these series and back over
for others the to
required history
replicateof
thethe London
graphs in
Classic
fore seen in human history.viewof
andnew price-divided
If so, much of the ratios.
or the New York stock markets can be extraor-
the paper http://www.econ.yale.edu/~shiller/data.htm equity premium was unexpected; good luck.
35.00 dinarily comforting to an investor—equities 12.00%
Now
appear to have provided a substantial premium
P/E Time varying expected returns
Price
29.59
30.00 AV.PE
over bonds, and markets appear to have recov-
D/P
10.00%
Regressions of returns on price/dividend ratios
25.00
ered nicely after huge crashes. . . . Less com-
AV.DP
We are not only concerned with the aver-
forting is the past history of other major mar- 8.00%
age return on stocks but whether returns are New fact

20.00
kets: Russia, China, Germany, and Japan. Each expected to be unusually low at a time of high
of these markets has had one or more major
6.00%
Dividend
15.00
14.54 prices, such as the present. The first and most
interruptions that prevent their inclusion in 4.65%
4.00% natural thing one might do to answer this ques-
10.00 long term studies” [my emphasis]. tion is to look at a regression forecast. To this
In addition, think of the things that didn’t 2.00%
end, table 4 presents regressions of returns on Classic view
5.00
happen in the last 50 years. There were no 1.17%
the price/dividend (P/D) ratio. Time
0.00 0.00%
1871 1891 1911 1931 1951 1971 1991
TABLE 4
OLS regressions of excess returns and dividend growth on VW P/D ratio
Rt→t+k = a + b(Pt /Dt) Dt+k/Dt = a + b(Pt/Dt)
Horizon k
(years) b σ( b ) R2 b σ( b ) R2
1 –1.04 (0.33) 0.17 –0.39 (0.18) 0.07

2 –2.04 (0.66) 0.26 –0.52 (0.40) 7 0.07
3 –2.84 (0.88) 0.38 –0.53 (0.43) 0.07
5 –6.22 (1.24) 0.59 –0.99 (0.47) 0.15
Notes: Rt→t +k indicates the k year return on the value weighted NYSE portfolio less the k year
return from continuously reinvesting in Treasury bills; b = regression slope coefficient
(defined by the regression equation above); σ(b) = standard error of regression coefficient.
Standard errors in parentheses use GMM to correct for heteroscedasticity and serial correlation.
Prof. Abhay Abhyankar (University of Exeter)

FEDERAL RESERVE BANK OF CHICAGO Empirical Finance UAB April
7 2017 10 / 1
What We Will Build On?
MV Portfolio Theory, CAPM, Arbitrage Pricing Theory (APT).
The Consumption Capital Asset Pricing Model (CCAPM) and its

extensions.
Efficient Market Hypothesis, Information Flow in Markets, Basic Ideas

about Market Microstructure.
Abhay Abhyankar University of Exeter, U.K. (Universities of

Empirical
Somewhere
Finance
and Elsewhere) 3/4
Empirical asset pricing - Introduction (4)
time line of discovery traditional

Portfolio theory
Mean-Variance frontier
Cochrane’s approach
CAPM
APT
Option pricing
contingent claims state preference
consumption-based modell
stochastic discount factor
Prof. Joachim Grammig, University of Tübingen, Department of Econometrics, Statistics and Empirical Economics 11
Asset pricing (Valuation of financial assets)
delay of risk of
account for
payoff payoff
⇒ risk correction
50 years US stocks: 9% average return (real) p.a.
1% real interest rate p.a. (treasury bills)
8% premium earned for holding risk
What is the risk that is priced?
Asset pricing
normative positive
how should the world work? how does the world work?
are the prices ”wrong”?
- trading opportunities?
- cost of capital
- non traded assets: ”fair” price
Basic : Prices equal discounted expected payoff

What probability measure?
Absolute Asset Pricing

exposure to ”fundamental” macroeconomic risk
Asset priced given other asset prices (e.g. option pricing)
Relative Asset Pricing
³ ´ ³ ´
E Ri = Rf + βi E m
(R {z f
) −R}
e.g. CAPM: |
¡ i m¢
cov R ,R
β i = var(Rm )
Market price of risk (factor)risk premium not explained
Basic pricing equation pt = Et(mt+1xt+1)
asset price stochastic payoff

at t discount (r.v.)
factor
(r.v.)
mt+1 = f (data
| , parameters
{z })
the model
Moment condition: Et(mt+1 xt+1) − pt = 0
1P → E()
use n WLLN
Generalized Method of Moments (GMM) to estimate parameters
From an utility maximising investor`s first order conditions we
obtain the basic asset pricing formula (1)
Basic objective: find pt, the present value of stream of uncertain payoff xt+1
dividend
xt+1 = pt+1 + dt+1
price of asset in t+1
Utility function period utility function
U (ct, ct+1) = u (ct) + βEt [u (ct+1)]

expected utility
consumption subjective discount factor
consumption level without asset purchase (other income)

ct = et − ptξ quantity of asset bought/sold
ct+1 = et+1 + xt+1ξ
Random variables: pt+1, dt+1, xt+1, et+1, ct+1, u (ct+1) Et [·] , E [· | Ft]
From an utility maximising investor`s first order conditions we
obtain the basic asset pricing formula (2)
max [U (ct , ct+1)] s.t.

(ξ)
ct = et − pt ξ; ct+1 = et+1 + xt+1 ξ
max {u (et − pt ξ) + βEt [u (et+1 + xt+1ξ)]}

(ξ)
£ 0 ¤
−pt · u0 (ct ) + β · Et u (ct+1) · x t+1 = 0
utility loss if investor buys discounted expected utility increase

another unit of the asset from extra payoff
0 £ 0 ¤
pt u (ct) = Et βu (ct+1 ) xt+1
∙ 0 ¸ Investor continues to buy
u (c )
pt = Et β 0 t+1 xt+1 or sell the asset until marginal
u (ct ) loss equals marginal gain.
No complete solution: endogenous variables

Turning off uncertainty we are in the standard two-goods case (1)
max [u (ct) + βu (ct+1)] s.t. ct = et − pt · ξ, ct+1 = et+1 + xt+1 · ξ
∂U (ct, ct+1) ∂u (ct) ∂u (ct+1)

= −pt · + β · xt+1 · =0
∂ξ ∂ct ∂ct+1
pt · u0 (ct) = xt+1 · βu0 (ct+1)

βu0 (ct+1)
pt = xt+1 ·
u0 (ct)
marginal valuation
of consumption
dct β · u0 (ct+1) pt opportunity cost to transfer
− = =
in t+1 in terms of dct+1 u0 (ct) xt+1 consumption from t to t+1
consumption in t
0 £ 0 ¤
ptu (ct) = Et βu (ct+1) xt+1
∙ 0 ¸
u (c )
pt = Et β 0 t+1 xt+1
u (ct)
We often use a convenient power utility function (1)
µ ¶
1 1−γ 1 1−γ
u (ct) = c lim c = ln (ct)
1−γ t γ→1 1−γ t marginal
µ ¶ rate of
−γ dct βu0 (ct+1) ct+1 −γ
u0 (ct) = ct = =β substitution
dct+1 u0 (ct) ct
utility u(ct)
10
parameter γ:
8
0.3 0.5 0.8
6
increasing concavity
of utility function
4
00 2 4 6 8 10 12 consumption (ct)
x
Prices, payoffs, excess returns
Price pt Payoff xt+1
stock pt pt+1 + dt+1
return 1 Rt+1
excess return 0 Ret+1 = Rat+1 − Rbt+1
one $ one period discount bond pt 1
risk-free rate 1 Rf
x
Payoff xt+1 divided by price pt ⇒ gross return Rt+1 = t+1
pt
Return: payoff with price one
1 = Et (mt+1 · Rt+1)
Zero-cost portfolio:
Short selling one stock, investing proceeds in another stock
⇒excess return Re
Example: Borrow 1$ at Rf , invest it in risky asset with return R.

Pay no money out of the pocket today → get payoff Re = R − Rf .
Zero price does not imply zero payoff.
The covariance of the payoff with the discount factor rather than its
variance determines the risk-adjustment
cov (m t+1 , xt+1 ) = E (mt+1 · xt+1 ) − E (m t+1 ) E (xt+1 )
pt = E (mt+1 · xt+1 )
= E (mt+1 ) E (xt+1 ) + cov (m t+1 , xt+1 )
f 1
R =
E (m t+1 ) Marginal utility declines
E (xt+1 ) as consumption rises.
pt = f
+ cov (mt+1 , xt+1 )
R Price is lowered if payoff
µ 0 ¶ covaries positively with
E (x t+1 ) u (ct+1 ) consumption. (makes consumption
pt = f
+ cov β 0 , x t+1 stream more volatile)
R u (c t )
¡ 0 ¢
E (x t+1 ) cov u (c t+1) , xt+1 Price is increased if payoff
pt = +β covaries negatively with
R f u0 (c t ) consumption. (smoothens
consumption) Insurance !
price in risk-neutral risk adjustment
world
Investor does not care about volatility of an individual asset, if he can keep a steady consumption.
All assets have an expected return equal to the risk-free rate, plus
risk adjustment
³ ´
i
1 = E m t+1 · R t+1
³ ´ ³ ´
1 = E (m t+1 ) E
+ cov R it+1 i
m t+1 , R t+1
f 1 1 ³ i ´ ³
R = ; 1 − f E R t+1 = cov m t+1 , R it+1
E (m t+1 )
³ ´ ³ R ´
i f f
E R t+1 − R = −R · cov m t+1 , R t+1 i
³ ´ µ 0 ¶
1 u (c )
E R it+1 − R f = − ³ u 0 (c ) ´ · cov β 0 t+1 , R it+1
E β t+1 u (ct )
u 0 (c t )
excess return
³ ´ ¡ 0 i
¢
i cov u (ct+1 ) , R t+1
E R t+1 − Rf = −
E (u 0 (ct+1 ))
Investors demand higher excess returns for assets that covary positively with consumption.
Investors may accept expected returns below the risk-free rate. Insurance !
The basic pricing equation has an expected return-beta
representation
³ ´ ³ ´
i f f i
E Rt+1 − R = −R · cov Rt+1, mt+1
³ ´ ¡ i ¢
i f cov Rt+1, mt+1 V ar (mt+1)
E Rt+1 − R = −
V ar (mt+1) E (mt+1)
Ã ¡ i ¢! µ ¶
³ ´ cov Rt+1, mt+1 V ar (mt+1)
E Rit+1 = Rf − ·
V ar (mt+1) E (mt+1)
asset specific quantity of risk price of risk for all assets

³ ´
i
Beta-pricing model: E R = Rf + βRi,m · λm
³ ´
ct+1 −γ c
With m = β ct and lognormal consumption growth t+1
ct
³ ´
The more risk averse the investors
E Ri = Rf + βRi,∆c · λ∆c or the riskier the environment,
λ∆c ≈ γ · V ar (∆ ln c) the larger the expected return
premium for risky (high-beta)
assets.
Marginal utility weighted prices follow martingales (1)
xt+1
Basic first order condition:
³ ³ ´ ´
ptu0(ct) = Et β 0
u (ct+1) (pt+1 + dt)
Market efficiency ⇔ Prices follow martingales (random walks)? NO!
Risk neutral investors u’( )=const.
or no variation in consumption
Required: β = 1 ⇐ OK short time horizon
no dividends
Then: pt = E(pt+1 )
pt+1 = pt + εt+1
if σ2(εt+1 ) = σ2 = Random Walk
³p ´
⇒ Returns are not predictable E t+1
pt =1
Marginal utility weighted prices follow martingales (2)
With risk aversion (but no dividends) and β=1
p̃t = E(p̃t+1 )
p̃t = p̃t · u0(ct)
Scale prices by marginal utility, correct for dividends and apply risk neutral
valuation formulas
Predictability in the short horizon?

consumption
does not change day by day
risk aversion
⇒ Random Walks successful ⇒ Predictability of asset returns (day by day)?
Technical analysis, media reports...
Some popular linear factor models
Factor pricing models return on wealth portfolio
CAPM : mt+1 = a + bRw

t+1
Free parameters
Compatible with utility maximisation ?
ICAPM : mt+1 = a + b0ft+1

parameter factors factors (macro, term spread, price-
earnings ratio help forecast
vector
conditional distribution of future
asset returns)
similar, but factors determined by principal

APT : component analysis of payoff covariance
matrix
Practice : just test m = b0f and don’t worry about derivations
The basic pricing equation implies a set of CONDTIONAL moment
restrictions
³ ´
pt = Et mt+1xt+1 {mt} and
³ ´ {xt} non i.i.d. ⇒
= E mt+1xt+1 | It Et (·) 6= E (·)
Information set (partially) not observed,

conditional density not known, conditional expectation cannot be computed
Conditioning down to coarser
information set ³ ´
pt = Et mt+1xt+1
³ ³ ´´
E (pt) = E Et mt+1xt+1 l.i.e.
³ ´
= E mt+1xt+1
SDF and the Mean-Variance frontier
Supressing the time subscripts for clarity

1 = E (mR) ! 1 = E (m)E (R) + Cov (mR)
= E (m)E (R) + ⇢mR (m) (R)
dividing both sides by E (m) (R) leads to
1 E (R) (m) E (R) R f (m)

= + ⇢mR = = ⇢mR
E (m) (R) (R) E (m) (R) E (m)
E (R) R f (m) 1
since, 1  ⇢  +1 : <= (note : = Rf )
(R) E (m) E (m)
Prof. Abhay Abhyankar (University of Exeter) Empirical Finance UAB April 2017 46 / 1
Systematic risk
Means and Variances of asset returns must lie in the wedge shaped
region in the figure.The boundary is the MV frontier.
All returns on the frontier are perfectly correlated with the SDF. The
frontier is generated by |⇢mR |=1.
Returns on the upper part have ⇢mR = 1 (and ⇢ = +1 with
consumption) so they are maximally risky (highest expected returns
per unit variance).
Returns on the lower part have ⇢mR = 1 (and ⇢ = 1 with
consumption) so they are perfect hedges against consumption risk.
since
E (R) R f (m)
<=
(R) E (m)
E (R mv ) R f (m)
=
(R mv ) E (m)
The highest Sharpe ratio is associated with portfolios that lie along
the Mean-Variance efficient(MV) frontier.
Importantly. the slope of the frontier is driven by the volatility of the
discount factor. It is limited by the volatility of the SDF.
Systematic risk
All frontier returns are perfectly correlated with each other as they are
perfectly correlated with the SDF.
Two Fund Separation-Any frontier return (not R f ) R mv can be
expressed as R mv =R f + a(R m R f ) for some constant a.
Since each return on the frontier is perfectly correlated with m, we can
find constants a, b, d, e such that m = a + bR mv and R mv = d + em.
This simply means that any MV portfolio contains all pricing
information in m. (for example, the market in CAPM proxies m)
Systematic risk
Given m, we can also construct a single-beta representation such that

expected returns are expressed in a single-beta model using the return
on any mean-variance efficient portfolio (except R f )
E (R i ) = R f + i,mv [E (R mv ) R f )]
Also holds forR mv ,E (R mv ) R f = = the price of risk.
The priced part of the return is perfectly correlated with the SDF
while the idiosyncratic part generates no return.
Discount factors, beta models and MV frontier are all related.
Consumption-based AssetPricing
Abhay Abhyankar
University of Exeter, U.K.
Consumption-based Asset Pricing
I We first turn to the family of consumption-based asset pricing
models.
1. Consumption-based models are theoretically motivated.
2. However, in the data, there is little support for these models.
3. Importantly not used in practice e.g. determining cost of
capital for firms.
I The standard CAPM also does not work well but variations
are widely used in industry and applications.
I Inspired by the ICAPM/ APT, several factor models like the
Fama and French model have been proposed.
1. Neither the ICAPM/APT tell us what the factors should be -
many possible candidates.
2. Empirically motivated models also unable to fully explain
cross-sectional variation in risk and return.
3. Currently no model works well across different assets.
I We will see, based on selected recent work, where the
literature is going.
Outline
1 The Stochastic Discount Factor

2 The Consumption Capital Asset Pricing Model
3 The Equity Premium Puzzle
4 Explanations for the Puzzle
5 Other approaches
Oliver Linton obl20@cam.ac.uk () F500: Empirical Finance Lecture 9: Intertemporal Equilibrium

January
Pricing
11, 2017 2 / 46
Intertemporal Optimization
Lucas (1978). Investors have a concave, positively sloped, time-invariant
utility function for consumption, and a constant rate of time preference δ.
They invest in risky assets fRi ,t gni=1 and consume the proceeds over time
f Ct g
Investors choose investment/consumption to maximize the discounted
expected utility of lifetime consumption
" # cont. value
∞ z }| {
Vt = E t ∑ δ U ( Ct + j )
j
= U (Ct ) + δ Et Vt +1
j =0
subject to a budget constraint

N
Wt +1 = (Wt Ct ) ∑ wit (1 + Ri ,t +1 )
i =1

January
Pricing
11, 2017 3 / 46
First order condition for each risky asset i, the so-called Euler equation
U 0 (Ct ) = δEt (1 + Ri ,t +1 ) U 0 (Ct +1 ) ,
where Et means expectational on information at time t.

De…ning
U 0 ( Ct + 1 )
Mt +1 = δ 0
U ( Ct )
and rearranging the series of …rst-order conditions (i = 1, . . . , n )
1 = Et [Mt +1 (1 + Ri ,t +1 )]

January
Pricing
11, 2017 4 / 46
Stochastic Discount Factor
The random variable Mt > 0 is called the stochastic discount factor or

pricing kernel. It is the (random) ratio of marginal utilities between each
“investment” date-state and “realized return” date-state, weighted by
pure time preference.
Pricing formula for any asset
Pt = Et [Mt +1 Xt +1 ] ,
where Xt +1 is the cash ‡ow in period t + 1 (e.g., Pt +1 + Dt +1 )

Relationship can be derived more generally from non-arbitrage
assumption: There does not exist a negative-cost portfolio with a
uniformly non-negative payo¤.

January
Pricing
11, 2017 5 / 46
Risk Neutral Expectation
Replace the true probability weights (denoted P ) in the basic pricing

expectation with “hypothetical” probability weights
P ∝ Mt P,
and taking expectations under these transformed probabilities gives
Pt = Et [Mt +1 Xt +1 ] = Et [Xt +1 ]
All assets have the same expected return under the transformed
probabilities.
This new hypothetical "probability” measure is called the equivalent
martingale measure or the risk neutral measure. It is very useful for
empirical derivatives pricing (not covered in this course).

January
Pricing
11, 2017 6 / 46
The Consumption Capital Asset Pricing Model Again
Adding and subtracting we obtain
1 = Et [(1 + Ri ,t +1 ) Mt +1 ]
= Et [(1 + Ri ,t +1 )] Et [Mt +1 ] + covt (Ri ,t +1 , Mt +1 )
Let R0t denote an asset such that
covt (R0,t +1 , Mt +1 ) = 0
(zero beta or risk free asset). Then
1 U 0 ( Ct + 1 )
Et [Mt +1 ] = = δEt
Et [1 + R0,t +1 ] U 0 (Ct )

January
Pricing
11, 2017 7 / 46
Then substituting in and rearranging we obtain for any asset i
Et [Ri ,t +1 R0,t +1 ] = covt (Ri ,t +1 , Mt +1 ) Et (1 + R0,t +1 )

u 0 ( Ct + 1 ) 1
= covt Ri ,t +1 , 0 h 0 i
u ( Ct ) δE
U (C t +1 )
t U 0 (C t )
An asset whose covariance with Mt is negative tends to have low returns

when the investor’s marginal utility of consumption is high ie when
consumption is low.
Require a large risk premium to hold it.

January
Pricing
11, 2017 8 / 46
Suppose there is an asset Rmt that pays o¤ exactly Mt then
Et [Rm,t +1 R0,t +1 ] = vart (Mt +1 ) Et (1 + R0,t +1 )
Therefore,
Et [Ri ,t +1 R0,t +1 ] = βim,t Et [Rm,t +1 R0,t +1 ]
covt (Ri ,t +1 , Rm,t +1 )

βim,t =
vart (Rm,t +1 )
This pricing model is called the consumption CAPM.

We can also, starting from 1 = E [(1 + Ri ,t +1 ) Mt +1 ] , derive an
unconditional version
cov(Rit , Rmt )
E [Rit R0t ] = βim E [Rmt R0t ], βim =
var(Rmt )
Note that βim 6= E βim,s .

January
Pricing
11, 2017 9 / 46
Note that the CCAPM model has:
cross-sectional predictions (relative risk premia are proportional to
consumption betas),
time-series predictions (expected returns vary with expected
consumption growth rates, etc.),
joint time-series/cross-sectional predictions.
The standard CAPM only has cross-sectional predictions.

January
Pricing
11, 2017 10 / 46
Need to specify U (.) in order to estimate betas from consumption data.
Simple elegant utility function is the CRRA class with risk aversion
parameter γ
1 γ
Ct 1
U ( Ct ) =
1 γ
Calculating the stochastic discount factor gives

γ
Ct + 1
Mt +1 = δ
Ct
mt +1 = log Mt +1 = log δ γgt +1 ; gt +1 = log (Ct +1 /Ct )

The "riskless" asset satis…es
γ
1 Ct + 1
1 + Rft = Et
δ Ct

January
Pricing
11, 2017 11 / 46
How to test the consumption CAPM?
Hansen and Singleton (1982) GMM conditional moment restriction
" #
γ
Ct + 1
Et (1 + Rit +1 )δ 1 =0
Ct
Do not need to specify dynamics for returns or consumption except

stationarity on consumption growth.
Convert to unconditional moment restriction and do GMM

January
Pricing
11, 2017 12 / 46
Let with Xt denoting all the data and θ = (δ, γ)
20 1 3
1 + R1,t +1 instruments
6B .. C Ct + 1 γ
7 z}|{
g ( Xt , θ ) = 4 @ . Aδ 15 Zt 2 Rp
Ct
1 + Rn,t +1
Then we have the unconditional moment restriction
E [g (Xt , θ )] = 0.
Estimate the parameters θ by the Generalized Method of Moments

(GMM) using p > 2 sample moments and quadratic form
T
1 |
GT ( θ ) =
T ∑ g ( Xt , θ ) ; min GT (θ ) WGT (θ )
θ
t =1
This is nonlinear in θ.
January
Pricing
11, 2017 13 / 46
Test whether overidentifying restrictions (p > 2) hold using the J-test.
|
jjGT (bθ )jjW opt = GT (bθ ) Wopt GT (bθ )
This is asymptotically chi squared (χ2p 2 ) under the null hypothesis that
the moments are correct.
Empirically the CCAPM model performs very poorly, see below. The
empirical failure of the consumption CAPM is among the most important
anomalies of asset pricing theory.

January
Pricing
11, 2017 14 / 46
The Equity Premium Puzzle
We make an assumption that log consumption growth and log equity
market return are jointly normal (can hold a little more generally in an
approximate sense like the Campbell log linearization) and that utility is
CRRA.
Then letting rit be logarithmic returns and ct = log Ct we have the linear
(in parameters) equation
1 2
0 = Et [ri ,t +1 ] + log δ γEt [∆ct +1 ] + σ + γ2 σ2c 2γσic ,
2 i
where (assuming conditional homoskedasticity):

σic = covt (ri ,t +1 , ct +1 )
σ2i = vart (ri ,t +1 )
σ2c = vart (ct +1 )

January
Pricing
11, 2017 15 / 46
The risk-free rate is determined endogenously in the model (setting
σ2i = σic = 0)
γ2 σ2c
Et [rft ] = log δ + γg
2
where g is the mean growth rate of consumption. Risk free rate depends
on impatience, risk aversion, consumption growth and volatility.
For any other asset i we have
σ2i
Et [ri ,t +1 rf ,t +1 ] = γσic γσic
2
This is a pricing equation for the risk premium in terms of covariation with
consumption growth.

January
Pricing
11, 2017 16 / 46
The Equity Premium Puzzle. Empirically, σic is very small relative to
the observed premium of equities over …xed income securities, hence this
implies a very high coe¢ cient of risk aversion γ.
The Risk Free Rate Puzzle. If γ is set high enough to explain observed
equity risk premia, it is too high (given average consumption growth) to
explain observed risk-free returns! The rate of pure time preference is
driven below zero.

January
Pricing
11, 2017 17 / 46
Mehra and Prescott (1985).
"Historically the average return on equity has far exceeded
the average return on short-term virtually default-free debt. Over
the ninety-year period 1889-1978 the average real annual yield on
the Standard and Poor 500 Index was seven percent, while the
average yield on short-term debt was less than one percent. The
question addressed in this paper is whether this large di¤erential
in average yields can be accounted for by models that abstract
from transactions costs, liquidity constraints and other frictions
absent in the Arrow-Debreu set-up. Our …nding is that it cannot
be, at least not for the class of economies considered. "

January
Pricing
11, 2017 18 / 46
No-Arbitrage Asset Pricing Term Structure Equity Premium
Mehra and Prescott (1985)
Does model produce equity premium of the size found in the data?
Data
Variable Mean Description
rts 1.0698 Real return on S&P 500
rt 1.0080 Real return on relatively riskless security
rte 0.0618 Equity premium
ct+1 /ct 1.0183 Per capita real consumption growth
Table: U.S. Annual Data 1889-1978 (Mehra and Prescott, 1985)
So equity premium in the data about 6.2%
UNIL-HEC Asset Pricing Basics 20 Sept. 2017

Mehra and Prescott (1985)
And in the model?

Assume functional form for period utility is CRRA, i.e.
u(ct ) = ct1−α /(1 − α)
Two interpretations of α
Coefficient of relative risk aversion
Inverse of the elasticity of intertemporal substitution
Consumption growth process calibrated to the data
In macroeconomics, α typically calibrated to value no larger than
about 5; Mehra and Prescott are more generous:
For values of α ∈ [0, 10] and β ∈ (0, 1), what values of the bond
return and the equity premium can the model generate?
E(rt ) ≤ 1.04 (vs. data: 1.008)
E(rte ) ≤ 0.0035 (vs. data: 0.0618)
⇒ “Equity Premium Puzzle”
Weil (1989)
Mehra and Prescott: “The equity premium puzzle may not be why
was the average equity return so high but rather why was the risk-free
rate so low.”
Weil (1989) follows up on this point
Shows that if
one fixes β (and other parameters) to a plausible value
relaxes the assumption that the coefficient of relative risk aversion be
the inverse of the intertemporal elasticity of substitution (EIS) by using
Epstein-Zin utility
considers plausible values of relative risk aversion and the EIS
then the bond return (“risk-free rate”) is a lot too high in the model
compared to the data.
⇒ “Risk-free rate puzzle”

Explanations for the Puzzle
A large number of explanations for the puzzle have been proposed. These
include:
a contention that the equity premium does not exist: that the puzzle
is a statistical illusion
modi…cations to the assumed preferences of investors, and
imperfections in the model of risk aversion.

January
Pricing
11, 2017 19 / 46
Statistical Illusion
The most basic explanation is that there is no puzzle to explain: that there
is no equity premium. Essentially, we don’t have enough statistical power
to distinguish the equity premium from zero.
Sample selection bias: US equity market is the most intensively studies in
equity market research. Not coincidentally, it had the best equity market
performance in the 20th century; others (e.g. Russia, Germany, and
China) produced a gross return of zero due to bankruptcy events.
Low number of data points: the period 1900–2005 provides only 105
independent years which is not a large number of years statistically.
Sample period choice: returns of equities (and relative returns) vary
greatly depending on which points are included. Using data starting from
the top of the market in 1929 or starting from the bottom of the market in
1932 (leading to estimates of equity premium of 1% lower per year), or
ending at the top in 2000 (vs. bottom in 2002) or top in 2007 (vs. bottom
in 2009 or beyond) completely change the overall conclusion.

January
Pricing
11, 2017 20 / 46
Is there an equity premium puzzle in the USA?
3-month tbill rate annualized

January
Pricing
11, 2017 21 / 46
S&P500 daily total return annualized

January
Pricing
11, 2017 22 / 46
Annualized moments of daily returns, log returns, and tbills 1954-2012
µ med σ IQR/1.3 ρ (1)

(1950-) log returns 0.0309 0.0504 0.0675 0.0484 0.02933
(1950-) returns 0.0832 0.1160 0.1549 0.1114 0.02883
(1954-) tbill 0.0475 0.0466 0.0019 0.0016 0.99940
(1954-) Risk premium 0.0254 0.0631 0.1568 0.1136 0.02524
The correlation between returns and log returns is 0.9997, but see how
di¤erent the mean and standard deviations are over this period
The correlation between returns and the tbill rate is -0.014 so almost
insigni…cant.
Does not include Dividends

January
Pricing
11, 2017 23 / 46
Nominal Returns in the US since 1900 (Dimson, Marsh,
and Staunton)

January
Pricing
11, 2017 25 / 46
Market frictions
If some or all assets cannot be sold short by some or all investors, then the
stochastic discount factor equation is much weaker:
E [(1 + Rit ) Mt ] 1
The inequality-version of the stochastic discount factor does not aggregate

across investors. Hence aggregate consumption is not directly relevant.

January
Pricing
11, 2017 27 / 46
Separating Risk Aversion and Intertemporal Substitution
The standard multiperiod von Neumann-Morgenstern utility function is

elegant but may not provide an accurate representation.
Multiperiod von Neumann-Morgenstern utility has a single parameter (the

risk aversion parameter) that governs both the elasticity of intertemporal
substitution and Arrow-Pratt risk aversion.

January
Pricing
11, 2017 28 / 46
Elasticity of intertemporal substitution
Consider an investor who consumes/saves in period zero and consumes in
period one with no risk. Suppose that the risk-free interest rate is Rf .
The elasticity of intertemporal substitution is de…ned as the percentage
change in optimal consumption growth for a percentage change in the
risk-free interest rate:
%∂(C1 /C0 )
EIS =
%∂(Rf )
In the CRRA case it is easy to show

1
EIS = .
γ
Note that EIS is an intertemporal concept with no connection to risk.

January
Pricing
11, 2017 29 / 46
The Epstein-Zin-Weil "utility" function
Separates EIS and RRA, it is de…ned recursively by
1 γ h i 1
θ
1 γ
1 γ θ
Ut = (1 δ ) Ct θ
+ δ Et Ut + 1
where δ is discount factor, γ is coe¢ cient of relative risk aversion ψ is the

elasticity of intertemporal substitution
1 γ
θ=
1 ψ1
The …rst-order conditions are more complex than in the vN-M case, but
one useful series of …rst-order conditions is
2( )θ 3
1 1 θ
Ct + 1 ψ 1
Et 4 δ (1 + Ri ,t +1 )5 = 1
Ct (1 + Rm,t +1 )
where Rm,t is the return on the market portfolio.

January
Pricing
11, 2017 32 / 46
Assuming that consumption and the return on the market portfolio are
jointly lognormal, and substituting gives
θ 1 θ 2 1
r f ,t +1 = log δ + σ2m 2
σc + Et [∆ct +1 ]
2 2ψ ψ
θ σ2i
Et [ri ,t +1 ] rf ,t +1 = σic + (1 θ ) σim
ψ 2
Consumption betas and market portfolio betas both a¤ect asset risk
premia.
With consumption data can test this model

January
Pricing
11, 2017 33 / 46
Other Asset Pricing Approaches
Habit models
Di¤erence model (Constantinides (1990))
∞
( Ct + j Xt + j ) 1 γ
1
U t = Et ∑ δj 1 γ
j =0
Ratio model (Abel (1990))

∞
(Ct +j /Xt +j )1 γ
1
Ut = Et ∑ δj 1 γ
j =0
Habit Xt , for example Xt some level of previous consumption. Gives

additional ‡exibility, but not very plausible.
Hyperbolic discounting (Laibson (1996))
∞
U (Ct ) + βEt ∑ δj U (Ct +j )
j =0

January
Pricing
11, 2017 39 / 46
Long run risks model Bansal and Yaron (2004)
Epstein Zin preferences
2( )θ 3
1 1 θ
Ct + 1 ψ 1
Et 4 δ (1 + Ri ,t +1 )5 = 1
Ct (1 + Rc ,t +1 )
where Rc is the gross return on an asset that delivers aggregate

consumption as its dividend each period (like, but not equal to, the market
portfolio). In logs with gt +1 = log Ct +1 /Ct and zt = log(Pt /Ct ), where
P is the price level
rc ,t +1 = κ 0 + κ 1 zt +1 zt + gt
θ
mt +1 = θ log δ gt +1 + (θ 1)rc ,t +1
ψ

January
Pricing
11, 2017 42 / 46
They specify dynamics for consumption and dividend growth rates
gt +1 = µ + xt + σt η t +1
gd ,t +1 = µd + φxt + ϕd σt ut +1
where the unobserved state variables (x is the "Long Run Risks")
xt +1 = ρxt + ϕe σt et +1
σ2t +1 = σ2 + ν1 (σ2t σ2 ) + σw wt +1
innovations et +1 , wt +1 , η t +1 , ut +1 are standard normal and iid and

mutually independent. The parameter ρ 1

January
Pricing
11, 2017 43 / 46
Captures the idea that news about growth rates and economic
uncertainty (i.e., consumption volatility) alters perceptions regarding
long-term expected growth rates and economic uncertainty
Asset prices will be fairly sensitive to small growth rate and
consumption volatility news.
Log linearizing, they obtain

January
Pricing
11, 2017 44 / 46
Innovation to the pricing kernel
consumption shock LRR shock shock to vol

z }| { z }| { z }| {
mt +1 Et mt +1 = λm,η σ t η t +1 λm,e σ t et + 1 λm,w σw wt +1
Equity premium
1
Et (rm,t +1 rf ,t +1 ) + vart (rm,t +1 ) = βm,e λm,e σ2t + βm,w λm,w σ2w
2
vart (rm,t +1 ) = ( β2m,e + ϕ2d )σ2t + β2m,w σ2w
Consumption shock is not priced

Risk return relationship
Et (rm,t +1 rf ,t +1 ) = τ 0 + τ 1 vart (rm,t +1 )

January
Pricing
11, 2017 45 / 46
Bansal et al. show that
I The model is capable of justifying the observed magnitudes of the
equity premium, the risk-free rate, and the volatility of the market
return and the dividend-yield.
I It captures the volatility feedback e¤ect, that is, the negative
correlation between return news and return volatility news.
I As in the data, dividend yields predict future returns and the volatility
of returns is time-varying.
I At plausible values for the preference parameters (IES and RRA), a
reduction in economic uncertainty or better long-run growth prospects
leads to a rise in the wealth–consumption and the price–dividend ratios.
There is a signi…cant negative correlation between price–dividend ratios
and consumption volatility.
I They show that about half of the variability in equity prices is due to
‡uctuations in expected growth rates, and the remainder is due to
‡uctuations in the cost of capital.

January
Pricing
11, 2017 46 / 46
Rare disasters
" −γ #
Ct +1
Et (Rt +1 ) − Rtf = covt , Rt + 1
Ct
I A small chance of a very low Ct +1 /Ct can drive the whole

covariance, raise Et Rt +1 despite reasonable γ, and despite samples
with small σ (∆ct +1 ).
I Objections:
1. Shouldn’t we see them more often? (Data controversy)
2. Beyond equity premium? To get return predictability, p/d volatility,
varying volatility, we need time-varying probabilities of rare disasters.
External measurement or dark matter?
3. We seem to need different time-varying probabilities for different
assets (Gabaix).
4. Correlation with business cycles? Probability of rare disasters
exogenously correlated with business cycles? Or causality from stocks
to recessions?
Balance sheets – debt – institutional / intermediated
finance
Intermediated markets
Securities
Intermediary
?
“Equity” “Debt”
Investor Investor
Other assets
I As people / intermediaries lose money, closer to default, they get

more risk averse
Intermediary-based Asset Pricing
• For the “Take Home” assignment we will be testing a
prominent intermediary-based asset pricing model.
• This class of models try and bridge the gap between
macroeconomics and finance.
• They emphasize the role of various agents in the financial
sector like hedge funds, pension funds, banks and related
institutions, broker- dealer networks etc.
• Later, we will look at some of these.

S2 Ap SDF Ccapm F 2019

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

S2 Ap SDF Ccapm F 2019

Uploaded by

Copyright:

Available Formats

Empirical Finance

Abhay Abhyankar University of Exeter, U.K. (Universities of

Financial markets & Trading, Data sources etc.

Asset Pricing: Consumption-based & Linear Factor Models.

Asset Pricing: Empirical Test Strategies.

Efficient Markets & Event Studies.

Present Value Models & Return Predictability.

Empirical Facts - Stock

Source: Based on Dimson, Marsh, and Staunton (2002) and as updated

Why don’t all investors just buy equities?

-40.00% S&P 500 3-month T.Bill 10-year T. Bond

GDP growth andthestock

1800 Real GDP = 100

chance that the coin with the greatest Figure

age return on stocks but whether returns are New fact

1 –1.04 (0.33) 0.17 –0.39 (0.18) 0.07

Prof. Abhay Abhyankar (University of Exeter)

MV Portfolio Theory, CAPM, Arbitrage Pricing Theory (APT).

The Consumption Capital Asset Pricing Model (CCAPM) and its

Efficient Market Hypothesis, Information Flow in Markets, Basic Ideas

Abhay Abhyankar University of Exeter, U.K. (Universities of

time line of discovery traditional

contingent claims state preference

stochastic discount factor

Basic : Prices equal discounted expected payoff

Absolute Asset Pricing

Market price of risk (factor)risk premium not explained

asset price stochastic payoﬀ

Moment condition: Et(mt+1 xt+1) − pt = 0

U (ct, ct+1) = u (ct) + βEt [u (ct+1)]

consumption level without asset purchase (other income)

max [U (ct , ct+1)] s.t.

ct = et − pt ξ; ct+1 = et+1 + xt+1 ξ

max {u (et − pt ξ) + βEt [u (et+1 + xt+1ξ)]}

utility loss if investor buys discounted expected utility increase

No complete solution: endogenous variables

max [u (ct) + βu (ct+1)] s.t. ct = et − pt · ξ, ct+1 = et+1 + xt+1 · ξ

∂U (ct, ct+1) ∂u (ct) ∂u (ct+1)

pt · u0 (ct) = xt+1 · βu0 (ct+1)

Example: Borrow 1$ at Rf , invest it in risky asset with return R.

Zero price does not imply zero payoﬀ.

asset specific quantity of risk price of risk for all assets

With risk aversion (but no dividends) and β=1

p̃t = p̃t · u0(ct)

Predictability in the short horizon?

⇒ Random Walks successful ⇒ Predictability of asset returns (day by day)?

Technical analysis, media reports...

CAPM : mt+1 = a + bRw

Compatible with utility maximisation ?

ICAPM : mt+1 = a + b0ft+1

similar, but factors determined by principal

Information set (partially) not observed,

Supressing the time subscripts for clarity

dividing both sides by E (m) (R) leads to

1 E (R) (m) E (R) R f (m)

Given m, we can also construct a single-beta representation such that

1 The Stochastic Discount Factor

Oliver Linton obl20@cam.ac.uk () F500: Empirical Finance Lecture 9: Intertemporal Equilibrium

subject to a budget constraint

Oliver Linton obl20@cam.ac.uk () F500: Empirical Finance Lecture 9: Intertemporal Equilibrium

U 0 (Ct ) = δEt (1 + Ri ,t +1 ) U 0 (Ct +1 ) ,

where Et means expectational on information at time t.