Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Econ 219BPsychology and Economics:
Applications(Lecture 10)
Stefano DellaVigna
March 31, 2004
Outline
1. CAPM for Dummies (Taught by a Dummy)
2. Event Studies
3. Event Study: Iraq War
4. Attention: Introduction
5. Attention: Oil Prices
1 CAPM for Dummies (Taught by
a Dummy)
1.1 Summary
• Capital Asset Pricing Model: Sharpe (1964) andLintner (1965)
• Tenet of Asset Pricing II
Assumptions:
• All investors are price-takers.
• All investors care about returns measured over oneperiod.
• There are no nontraded assets.
• Investors can borrow or lend at a given riskfree in-terest rate (Sharpe-Lintner version of the CAPM).
• Investors pay no taxes or transaction costs.
• All investors are mean-variance optimizers.
• All investors perceive the same means, variances, andcovariances for returns.
Implications:
• All investors face the same mean-variance tradeofffor portfolio returns
• All investors hold a mean-variance efficient portfolio.
• Since all mean-variance efficient portfolios combinethe riskless asset with a fixed portfolio of risky assets,all investors hold risky assets in the same proportionsto one another.
• These proportions must be those of the market port-folio or value-weighted index that contains all riskyassets in proportion to their market value.
• Thus the market portfolio is mean-variance efficient.
1.2 Mean-Variance Optimization
• Assume investors care only about mean (positively)and variance (negatively)
• (Can motivate with normally distributed assets)
• Mean-variance analysis with one riskless asset andN risky assets.
• The solution finds portfolios that have minimum vari-ance for a given mean return, Rp.
• These are called “mean-variance efficient” portfoliosand they lie on the “minimum-variance frontier”.
• Define:
— R as the vector of mean returns for the N riskyassets and Rf as the return of the riskless asset
— Σ as the variance-covariance matrix of returns
— w as the vector of portfolio weights for the riskyassets
— ι as a vector of ones and 1 − w0ι is the weightin the portfolio for the riskless asset
• Rewrite maximization as min Variance s.t. given re-turn Rp:
minw
1
2w0Σw s.t. (R−Rfι)
0w = Rp −Rf
Lagrangian:
L(w, λ) = 1
2w0Σw + λ(Rp −Rf − (R−Rfι)
0w)
First Order Conditions:
∂L(w, λ)∂w
= Σw − λ∗(R−Rfι) = 0
∂L(w, λ)∂λ
= Rp −Rf − (R−Rfι)0w∗ = 0
Rearranging,
w∗ = λ∗Σ−1(R−Rfι)
Rp −Rf = (R−Rfι)0w∗
Solve for λ?
Substitute for w∗ in the second equation using the firstequation
Rp −Rf = (R−Rfι)0λ∗Σ−1(R−Rfι)
λ∗ =Rp −Rf
(R−Rfι)0Σ−1(R−Rfι)
Consequently,
w∗ =⎛⎝
³Rp −Rf
´(R−Rfι)
0Σ−1(R−Rfι)
⎞⎠ ³Σ−1(R−Rfι)´
Implications:
• w∗i increasing in return of asset i Ri
• w∗i decreasing in variance of asset i σ2i (see Σ)
• Different portfolio choices with different risk aver-sion?
— Only Rp varies
— more risk-averse —> lowerRp —> hold fewer riskyassets, more riskless assets (w∗ lower)
— Everyone holds same share of risky assets: if writedown wi/wj, the parenthesis disappears
1.3 Asset Pricing Implications
• Assume that w is a vector of weights for a mean-variance efficient portfolio with return Rp.
• Consider the effects on the variance of the porfolioreturn for very small change in the weights of twoassets wi and wj such that dRp = 0.
dV ar(Rp) = 2Cov(Ri,Rp)dwi + 2Cov(Rj,Rp)dwj
• Must be dV ar(Rp) = 0, or initial portfolio was notoptimal.
• Substituting,
2Cov(Ri,Rp)dwi = 2Cov(Rj,Rp)
Ã(Ri −Rf)
(Rj −Rf)
!dwi
Ri −Rf
Cov(Ri,Rp)=
Rj −Rf
Cov(Rj,Rp)
• Use relationship for mean-variance return (j = p):
Ri −Rf
Cov(Ri,Rp)=
Rp −Rf
V ar(Rp)
Ri −Rf =Cov(Ri,Rp)
V ar(Rp)(Rp −Rf)
• Write for market return Rm (which is mean-varianceefficient under null of CAPM):
Ri −Rf =Cov(Ri,Rm)
V ar(Rm)(Rm −Rf)
• Cov(Ri,Rm)/V ar(Rm) is the famous Beta!
• Test of CAPM in a regression:
Rit −Rf = αi + βim(Rmt −Rf) + εit
• Jensen’s αi should be zero for all assets. (rejected indata)
• Point of all this: stock return of asset i depends oncorrelation with market.
• High correlation with market —> higher return tocompensate for risk
1.4 Implications for Event Studies
• Assume an event (merger announcement, earningannouncement) happened to company i
• Want to measure effect on stock return i
• Can just look at Rit before and after event?
• Better not. Have to control for correlation with mar-ket
• Should look at³Rit −Rf
´− βim(Rmt −Rf)
• Otherwise bias.
• In reality two deviations from CAPM:
1. Control for both α and β
2. Neglect Rf
• Typical estimation of abnormal return:
— Run (daily or monthly) regression:
Rit = αi + βiRmt + εit
for days (-150,-10) prior to event
— Obtain α̂i and β̂i
— Abnormal return is
ARit = Rit − α̂i + β̂iRmt
— Use this as dependent variable
2 Event Studies
• Examine the impact of an event into stock prices:
— merger announcement —> Mergers good or bad?
— earning announcement —> How is company do-ing?
— campaign-finance reform —> Effect on compa-nies financing Reps/Dems
— election of Bush/Gore —> Test quid-pro-quo parties-firms
— Iraq war (later in class) —> Effect of war
• How does one do this?
• Three main methodologies:
1. Regressions
2. Deciles
3. Portfolios
• Illustrate with earning announcement literature
• Event is earning surprise st,k
• Methodology 1. Run regression:r(h,H)t,k = α+ φst,k + εt,k
• Details:
— Use abnormal returns as dependent variable r
— (For short-term event studies, can also use netreturns rt,k − rt,m)
— Look at returns at multiple horizons: (0,0), (1,1),(3,75), etc.
— Worry about cross-sectional correlation: clusterby day
— Can add control variables to allow for time-varyingeffects, size-related effects
• Identification:
— time-series (same company over time, differentannouncements)
— cross-sectional (same time, different companies)
• Issues:
— Do you know event time?
∗ earning surprise?
∗ legal changes
— Need unexpected changes in information
• Methodology 2. Create deciles (Fama-French)
• Sort event into deciles (quantiles):
— Decile 1 d1t,k: Bottom 10% earnings surprises
— Decile 2 d2t,k: 10% to 20% earnings surprises
— etc.
• Estimate average return decile-by-decile
• Equivalent to running regression:
r(h,H)t,k =
10Xj=1
φjdjt,k + εt,k
• Details:
— Use buy-and-hold returns
— Worry about correlation of standard errors
• Issues:
— Plus: Non-linear specification
— Minus: Cannot control for variables
• Finance uses (abuses?) this ‘decile’ methodology
• Examples:
— Small firms and large firms — deciles by size
— Growth vs. value stocks — deciles by book-marketratios
• Methodology 3. Form portfolios
• Aggregate stock of a given category into one port-folio
• Observe its daily or monthly returns
• Idea: can you make money with this strategy??!
• Examples:
— Size.
∗ Form portfolio of companies by decile of size
∗ Hold for one/2/10 years
∗ Does a portfolio of small companies outper-form a portfolio of large companies?
— Momentum
∗ Form portfolio of companies by measure ofpast performance
∗ Hold for one/2/10 years
∗ Do stocks with high past returns outperformother stocks?
• Big difference from methodology 2:
— Now there is only one observation for time period(day/month)
— Have aggregated all the small firms into one port-folio
• Details:
— Run regression of raw portfolio returns on marketreturns as well as other factors:
rsmallt = α+ βrt,m + β2rt,2 + β3rt,3 + εt,k
— Standard Fama-French factors:
∗ control for market returns rt,m
∗ control for size ‘factor’ rt,2
∗ control for book-to-market ‘factor’ rt,3
— Idea: Do you obtain outperformance of an eventbeyond things happening with the market, withfirms size, and with book-to-market?
• Issues:
— Pluses: Get rid of cross-sectional correlation –now only have one ebservation per time period
— Minus: Cannot control for variables
3 Event Study: Iraq War
• See Additional slides
4 Attention: Introduction
• Attention as limited resource:
— Satisficing choice (Simon, 1955)
— Heuristics for solving complex problems (Gabaixand Laibson, 2002; Gabaix et al., 2003)
• In a world with a plethora of stimuli, which ones doagents attend to?
• Psychology: Salient stimuli (Fiske and Taylor, 1991)
b
c
d
e
a
f
j
g
h
i
0.25.15.2.07.33
0
3
1
4.45.5.05
5.75.24.01
1
-5.85.01.03.01.1
-1
5.65.35
2.05.95
2.15.01.14.7
-2.05.85.05.05
-1.55.45
0
1.3.15.3.1.15
-2.01.75.24
5.25.14.25.01.35
4.15.5.2.15
-5.01.65.3.01.03
-2.05.2.75
-4.15.5.2.15
-4
5.5.5
-2.25.7.05
1.95.01.04
-3.75.25
4.3.05.65
-5.03.01.01.95
-4.7.04.25.01
-3.1.9
0.5.4.1
1
1.15.04.8.01
4.85.15
-4
0.33.67
3.25.1.65
3
1
-2.55.05.3.05.05
-2.65.05.1.2
4.01.03.01.15.8
-7.05.6.2.05.1
3.35.01.15.14.35
-8.49.5.01
-5
3.01.3.03.01.65
1.15.45.05.3.05
-4.52.33.1.05
-2.6.4
-2.05.05.07.33.5
0
-1.4.6
3.85.01.03.01.1
3.01.01.18.8
-2.95.05
-4.75.25
-1.05.9.05
-1
-1
13.15.85
-5.85.05.09.01
3
-5.6.07.33
5.1.55.1.1.15
3.15.75.1
-5.01.2.79
-1.1.75.15
0.25.5.25
-2.33.67
-4.25.4.05.2.1
-5.33.15.52
5.15.05.2.6
5.6.25.14.01
-1.66.29.05
2
-6
0
-1.33.67
2.15.33.01.01.5
-5.45.55
3.1.8.04.05.01
-3
-5.25.55.2
3
2.3.3.4
-4.25.2.05.5
-4
5.65.35
0
2
0
4
1
-2
4
-3
-1
5
4.1 Attention to Non-Events
• Remember Huberman and Regev (2001)?
• Timeline:
— October-November 1997: Company EntreMedhas very positive early results on a cure for cancer
— November 28, 1997: Nature “prominently fea-tures;” New York Times reports on page A28
— May 3, 1998: New York Times features essen-tially same article as on November 28, 1997 onfront page
— November 12, 1998: Wall Street Journal frontpage about failed replication
• In a world with unlimited arbitrage...
• In reality...
Figure 5: ENMD Closing Prices and Trading Volume 10/1/97-12/30/98
5
10
15
20
25
30
35
40
45
50
55
10/1/97 10/31/97 12/3/97 1/6/98 2/6/98 3/11/98 4/13/98 5/13/98 6/15/98 7/16/98 8/17/98 9/17/98 10/19/98 11/18/98 12/21/98
Pri
ce [
$]
November 28,1997
May 4,1998
November 12,1998
• Which theory of attention explains this?
• We do not have a theory of attention!
• However:
— Attention allocation has large role in volatile mar-kets
— Media is great, underexplored source of data
• Suggests successful stategy on attention papers:
— Do not attempt geneal model
— Focus on specific deviation
5 Attention: Oil Prices
• Idea here: People do not think of indirect effects thatmuch
• Josh’s slides.