Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Save to My Library
Look up keyword
Like this
0Activity
0 of .
Results for:
No results containing your search query
P. 1
Extra Slides on Asset Pricing for Day 2

Extra Slides on Asset Pricing for Day 2

Ratings: (0)|Views: 3 |Likes:
Published by Vladimir Gusev

More info:

Published by: Vladimir Gusev on Dec 21, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

12/21/2011

pdf

text

original

 
11/10/111
Predicting Betas: Correcting for Mean Reversion
 
In order to correct for mean reversion 
 – 
 
Because the average beta is 1 
 
 We calculate the following: 
© 2008 Patrick J. Kelly 97
13132 
×
+
×
=
ta HistoricBe Beta Adjusted 
Estimating beta for illiquid stocks
 
 Assume that the stock is not traded in dates t
1
, t
2
,…
 
 What would be the prices in the non-traded days?
 – 
 
Same as in the last trading day, thus zero return
 
Large return in the first trading day after the break 
 – 
 
Predicted by the market model: R 
t
=
β
*R 
M,t
 
 
In the first trading day after the break, the stock 
s return will accumulate allidiosyncratic noise over the non-traded period
 
Betas will be biased!
 
Dimson (1979): regression including lead and lag values of the market index
 
j,t
=
α
j
+
Σ
l=-l1:l2
β
j,l
M,t+l
+
ε
j,t
 – 
 
 True beta is a sum of all lead-lag betas:
Σ
l=-l1:l2
β
j,l
Estimating beta for illiquid stocks
 
 The “trade-to-trade” approach:
 – 
 
Compute stock returns from the last traded day to the next tradedday (if necessary - over 2, 3 or n
t
days) – 
 
Compute market returns for the same periods – 
 
Run a regression with matched multi-period returns:
 
j,nt
=
α
j,
n
t
+
β
j
M,nt
+
Σ
t=0:nt-1
ε
j,t
 – 
 
How to control for heteroscedasticity?
 
 WLS: the variances are proportional to n
t
 
 
OLS regression with data divided by 
n
t
(R 
j,nt
/
n
t
 ) =
α
j
*
n
t
+
β
j
(R 
M,nt
/
n
t
 ) + ( 
Σ
t=0:nt-1
ε
j,t
/
n
t
 )
 What if the stock has a large weight in the index?
 
Endogeneity problem
 – 
 
In the extreme, when the index is dominated by one stock, thisstock will have beta of 1 by construction – 
 
In Russia, this is a problem for Gazprom, Lukoil, …
 
How to solve it?
 – 
 
Usual way: exclude this stock from the index
 –
 
 Theoretical
solution: use IV approach with other blue chips (orindustry indices) as instruments
 The evidence
CAPM
© 2011 Patrick J. Kelly 101
CAPM predictions
 
α
should be zero.
 – 
 
If not, there may be missing factors.
 
β
is the only relevant factor
 
Relation between
β
and returns is linear
 
Over long periods the return on the market is greater thanthe risk free return
 – 
 
In general, riskier stocks should earn higher return on average
 
Market portfolio is mean-variance efficient
 
If you cross-sectionally regress
β
on risk premia, estimatesshould equal the average market risk premium ( 
λ
 )
© 2011 Patrick J. Kelly 102
 
11/10/112
Related findings- Fama French 1992 - Size and Beta
© 2011 Patrick J. Kelly 103
4TJoFn
TeI
ARuPRnOaASzFPooFmoSzath,SoSoME(DwthPRn((AJy1toDm1
PooafomyyTbnfothsz(MEpctmshoandeadmninJoyt(= 11unaNsooCANA MEaNDsothmthCCMPdarrmaaototh1szpoounthNbnEszdeisuvdino1
f
poounpn,Boinvdsoemwh2to5yom hyru(aaaeeninJoytWeuoyNsothmthCCMPdarrmtoeahth3bnTewgmhyruothrun1pooathccafoJyoyttoJoyt+1Tpnfuthfu(Jy1toDm1smeopnrufoepooTpapnO(hainaohtaeathsumothsofmaroomhyruothcapomhruothvuwgpoooNAMEa(a1NDsoTaruithtmeaothmhyewgpooruinpTas zoapooithtmeaomhyaolnMEfosointhpooatheoJoe ywhMEdmninmoodaTanmosopmhfothszpoointhsmeszdevefm7to1Tanmosofothszpooinszde2a3ibw1a4athanmfothla7szdeibw1a2TAcumshwsacfoewgszde(MEpooTArwshwsacfoewgpooothsoine:gALwf2f3f4f5f6f7f8f9HgPAAMohyRu(nPA1213121313131212121211SmME1517151716151513161514ME21212141313161613131311ME31211131117121113131207ME41212111510131014111309ME51213141314141111121110ME61110151211121211101010ME71009121210111112061307ME81110101312120911101009ME90909081011101209080805LME0810091109090810070705
 Testing CAPM
 
 Approach 1: time series regression
 – 
 
 We observe the series of returns of the asset and of the marketindex – 
 
 We estimate beta (in-sample) and check whether other variables(const, D/P, P/E, B/M, …) explain remaining time variation in theasset
s return
 
 Approach 2: cross-sectional regression
 – 
 
 We check whether observed betas are (linearly) related to stock returns (as in the Security Market Line)
 Time series test
 
Run time series regression for N assets:
 
i,t
-R 
F
=
α
i
+
β
i
(R 
M,t
-R 
F
 ) +
ε
i,t
(+
δ
i
i,t-1
 )
 
Usually, N portfolios
 – 
 
Grouped by industry / size / … – 
 
 To reduce idiosyncratic variation present in individual stocks
 
H
0
:
α
i
=0 ( 
δ
i
=0) for any i=1,…,N
 – 
 
 There are no systematic deviations from the return predicted by theCAPM
 
 Assuming that R 
i,t
~ IID Normal, estimate by OLS
 – 
 
F-test for the joint significance of 
α
and
δ
coefficients
Results
 
Early tests, up to 80s
 – 
 
Did not reject CAPM – 
 
But: betas are unstable over time
 
Gibbons, Ross, and Shanken (1989)
 – 
 
Data: US, 1926-1982, monthly returns of 11 industry portfolios and VW-CRSP market index – 
 
For each individual portfolio, standard CAPM is not rejected – 
 
But the joint test rejects CAPM
 
CLM (1997, Table 5.3)
 – 
 
Data: US, 1965-1994, monthly returns of 10 size portfolios, VW-CRSP market index – 
 
 Joint test rejects CAPM, esp. in the earlier part of the sampleperiod
Fama-French (1993)
 
Run CAPM regressions using 342 months of data for 25 sizeand book-to-market portfolios
 
Strongly reject CAPM. F=1.91 with p-value 0.004
© 2010 Patrick J. Kelly 107
Intuition: Tests of Mean Variance Efficiency 
 
Cochrane (2001, Chapters 2 and 5) shows that any expectedreturn can be related to any mean-variance efficient portfoliolying on the efficient frontier:
 
 This structure suggests a natural test for efficiency:
 
 Test whether
α
is non-zero.
 – 
 
 This is the basic intuition of most tests of portfolio efficiency 
 
Differences across tests are mostly econometric refinements
© 2011 Patrick J. Kelly 108
 E r
i
( )
=
r
 f 
+
i
,
mv
E r
mv
( )
!
r
 f 
"#$%
 E r
i
( )
!
r
 f 
=
+
i
,
mv
E r
mv
( )
!
r
 f 
"#$%
 
11/10/113
 Test of Mean Variance Efficiency 
 
Gibbons, Ross, Shenken (1989)
 – 
 
 Where
 f 
is a return based factor or portfolio return on the mean- variance efficient frontier,
E
(f)
is the sample mean of the factor,and
σ
(f)
the sample standard deviation,
is the number of observations,
 N 
is the number of test assets,
1
is the number of factors,
α
is a vector of the intercepts from the N test-assetregressions, and
Σ
is the cross test asset residual covariance matrix,such that
© 2011 Patrick J. Kelly 109
!
 N 
!
1
 N 
1
+
 E 
( )
 ˆ
!  
( )
"#$$%&''
2
()**+,--
!
1
 ˆ
.
ˆ
/
!
1
 ˆ
~
 N 
,
!
 N 
!
1
 E 
!
[ ]
=
"
Mean of 
é
 
Cross-sectional tests
 
Main idea:
 
i,t
=
γ
0
+
γ
1
β
i
+
ε
i,t
(+
γ
2
i,t
 )
 
H
0
: asset returns lie on the SML
 – 
 
 The intercept is the risk-free rate:
γ
0
= R 
F
  – 
 
 The slope is the market risk premium:
γ
1
= mean(R 
M
-R 
F
 ) > 0 – 
 
 There are no additional effects:
γ
2
= 0
 
 We don
t know true betas!
 – 
 
 We first need to estimate them
measurement error – 
 
Usually, use predicted betas (estimated over the
 previous 
period)
 
 The realized returns may be very different from the expectedreturns (in SML)
 – 
 
 We need to estimate this cross-sectional regression over many periods (to measure
average 
effect)
 Two-stage procedure
 
Recursive procedure: for each
(e.g., every month), repeat
 – 
 
Time-series 
regression over the previous K periods to estimate beta
i,t-k 
=
α
i
+
β
i
M,t-k 
+
ε
i,t-k 
, for k=1:K 
 
 – 
 
Cross-sectional 
(CS) regression
i,t
-R 
F
=
γ
0
+
γ
1
β
i
+
ε
i,t
(+
γ
2
β
2i
+
γ
3
σ
i
)
 
Fama-MacBeth approach (first in 1973)
 – 
 
Running CS regression for each month
, we get the time series of coefficients
γ
0,t
,
γ
1,t
, – 
 
Compute mean and st. deviation of 
γ
s from these time series:
No need for s.e. of coefficients in the cross-sectional regressions!Shanken
s correction for the estimation error in betas
 – 
 
 Assuming normal IID returns, t-statistic =
 √ 
 T
γ
avg 
/
σ
γ
 )
Results
 
Fama-MacBeth (1973): estimate CS regression
 
i,t
-R 
F
=
γ
0
+
γ
1
β
i
+
γ
2
β
2i
+
γ
3
σ
2
ε
,i
+
ε
i,t
 – 
 
γ
1
> 0: positive effect of beta – 
 
γ
2
= 0: no non-linear effects of beta – 
 
γ
3
= 0: no impact of idiosyncratic risk 
 
Cochrane (2001)
 – 
 
Sort all NYSE stocks into 10 size deciles, add government andcorporate bonds – 
 
Run separate time-series regressions to estimate 12 portfolio betas – 
 
Regress sample average portfolio returns against estimated betas – 
 
Compare
fitted
SML with the one predicted by the CAPMCAPM does well in explaining stock vs. bond returns, but poorly inexplaining large vs. small-cap stocks
 
Cross-Sectional Approach: Fama-McBeth (1973)
 
 Tests if high beta is associated with high returns and vice versa.
 
General format:
 – 
 
 Time-series regressions to get betas for test portfolios
 
Usually beta sorted porttflios
 – 
 
Monthly cross-sectional regressions to test if high beta is associated with high return
 
Cochrane (1999) shows that GMM panel regressions areidentical under some assumptions
© 2011 Patrick J. Kelly 114

You're Reading a Free Preview

Download
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->