Big Data Asset Pricing

Lecture 2: A Primer on Empirical Asset Pricing

Lasse Heje Pedersen

AQR, Copenhagen Business School, CEPR

https://www.lhpedersen.com/big-data-asset-pricing

The views expressed are those of the author

Electronic copyand
available at: https://ssrn.com/abstract=4068783
not necessarily those of AQR
Overview of the Course
Lectures
I Quickly getting the research frontier
1. A primer on asset pricing
2. A primer on empirical asset pricing
3. Working with big asset pricing data (videos)
I Twenty-first-century topics
4. The factor zoo and replication
5. Machine learning in asset pricing
6. Asset pricing with frictions

Exercises
1. Beta-dollar neutral portfolios
2. Construct value factors
3. Factor replication analysis
4. High-dimensional return prediction
5. Research proposal
Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 2
Overview of this Lecture

I Basic questions in empirical asset pricing

I How to make factors
I Factor models: how to use factors
I Regressions:
I Time series
I Cross-sectional
I Fama-MacBeth
I Predictability: Regressions and event studies
I Empirical methods
I Frequentist vs. Bayesian vs. ML

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 3
 Basic Questions in Empirical Asset Pricing

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 4
Basic Questions in Empirical Asset Pricing
I State prices empirically:
I Does arbitrage exist?
I (If not) can we find m empirically?
I Can we find mimicking portfolio for m? Beta relation?
I Does a theory-implied m work for prices or returns?
I Risk and expected return:
I How to measure risk and expected return?
I What is the relation between risk and expected return?
I Do expected returns vary in the
I cross section? If so, which assets outperform? Why?
I time series? If so, when is the expected return high?
I Variation in cond. expected returns also called “predictability”
I Investors:
I How do investors behave and why?
I How do their trades affect prices and expected returns?
I What is the optimal portfolio?
Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 5
 How to Make Factors

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 6
How to Make a Tradable Factor

I Collect data on excess returns, rti

I Collect data on some characteristic (or signal), sti
I Make sure that the set of securities is free of selection bias
I Make sure that sti was known before the beginning of period t
(no look-ahead bias)
I Clean the data (without introducing biases)
I Go long stocks with high signals, short stocks with low signals
X i ,L X i ,S
i i
ft+1 = xt rt+1 − xt rt+1
i i

with weights of the longs, xti ,L , and shorts, xti ,S

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 7
How to Make a Factor: Choosing the Weights
I Dollar-neutral, $1 long and $1 short
I Decile (or quintile/tercile) sorts: Long top, short bottom
I Value-weighted
I Capped-value-weighted (See Jensen et al. (2022))
I Equal-weighted
I Risk-weighted
I Fama-French factors
I Rank-weighted
I Signal-weighted
I Beta-neutral
I Leverage long- and short-side to β = 1 (Frazzini and Pedersen
(2014))
I Hedge with market
I Beta-dollar neutral: ensure both, if possible (not BAB)
I Risk-adjusted factor
IFirst construct factor in one of the above ways
IThen re-scale it to keep it at constant ex-ante volatility
Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 8
Typical Table for Factor Construction

Tercile sorts

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 9
Well-Known Factors
I Fama-French factors: see next slides
I “Everywhere factors”: factors that work across asset several
classes (e.g., stocks, bonds, FX, commodities, real estate, ...)
I Value and momentum (Stattman (1980), Jegadeesh and
Titman (1993), Asness et al. (2013))
I Betting against beta, BAB, (Frazzini and Pedersen (2014))
I Time series momentum (Moskowitz et al. (2012))
I Carry (Koijen et al. (2018))
I Other well-known factors
IAccruals (Sloan (1996))
IShort-term reversal (Jegadeesh (1990))
I Long-term reversal (De Bondt and Thaler (1985))

I Quality-minus-junk, QMJ, (Asness et al. (2018))

I Environmental, social, governance, ESG, (Pedersen et al.

(2021))
I Liquidity risk (Pastor and Stambaugh (2003), Acharya and

Pedersen (2005))
Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 10
 Fama-French Factors

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 11
Fama-French Three-Factor Model
I Fama and French (1993) three-factor model
1. MKT-RF: market
2. SMB (small-minus-big): size
3. HML (high-minus-low): value
I Construction (lots of details – see paper and problem set):
MKT-RF = All stocks, value weighted
1
SMB = (Small Value+Small Neutral+Small Growth)
3
1
− (Big Value+Big Neutral+Big Growth)
3
1 1
HML = (Small Value+Big Value) − (Small Growth+Big Growth)
2 2
Median ME
|
Small Value Big Value
70th percentile of B/M−
Small Neutral Big Neutral
30th percentile of B/M−
Small Growth Big Growth

I What is the idea of this construction?

Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 12
Fama-French Five-Factor Model
I Fama and French (2015) five-factor model
1. MKT-RF: market – as in FF3
2. SMB: size – similar to FF3, but different construction
3. HML: value – as in FF3
4. RMW (robust-minus-weak): profitability
5. CMA (conservative-minus-aggressive): low investment
I Note that “investment” is really the percentage change in
total assets over the past year, which also captures equity
issuance and redistributions, debt changes, etc., not just “real
investment” such as capex
I Construction:
I SMB: here they try to make it neutral to HML, RMW, CMA
I RMW= 12 (Small Robust+Big Robust)- 12 (Small Weak+Big Weak)
I CMA= 12 (Small Conservative+Big Conservative)- 12 (Small Aggressive + Big Aggressive)
I No attempt to make HML, RMW, CMA uncorrelated
I Correlation of especially HML and CMA
Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 13
Fama-French Motivation: Present Value Relation
I Define k as the “internal rate of return” given by DDM
∞  i 
X Et dt+τ
pt =
(1 + k)τ
τ =1

I Accounting variables:
I “Clean surplus accounting relation,” Bt = Bt−1 + et − dt
I So p can be written in terms of earnings and asset growth
∞  
X i
Et et+τ − ΔBt+τ
pt =
τ =1
(1 + k)τ

I or residual income model, w/residual income, RIti = et − kBt−1

∞  i 
X Et RIt+τ
pt = Bt +
τ =1
(1 + k)τ

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 14
Fama-French Motivation: Present Value Relation, cont.
I Fama-French motivate their 3- or 5-factor models as follows:
I Look at DDM+accounting versions:

∞  i  ∞   ∞  i 
X Et dt+τ X i
Et et+τ − ΔBt+τ X Et RIt+τ
pt = = = Bt +
τ =1
(1 + k)τ τ =1
(1 + k)τ τ =1
(1 + k)τ

I Low p means high future returns, k, holding everything else

fixed (e.g., for given expected dividend growth in the left eqn.)
I High B/p (or low p/B) means high future returns holding
everything else fixed
I High profitability (i.e., high earnings e or residual income)
means high expected return holding everything else fixed
I High asset growth (ΔB), e.g., high investment, means low
returns holding everything else fixed
I FF factors show efficiency – correct?
IPositive factor returns, E (FFti ) > 0, consistent with PV logic
I FF factors “price” other test assets
Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 15
Fama-French Motivation: Ceteris Paribus?

∞  i  ∞   ∞  i 
X Et dt+τ X i
Et et+τ − ΔBt+τ X Et RIt+τ
pt = = = B t +
τ =1
(1 + k)τ τ =1
(1 + k)τ τ =1
(1 + k)τ

I Arguments above are “...holding everything else fixed ”

I Meaning of “...holding everything else fixed ” in
I PV logic: vary an endogenous/exogenous variable, consider the effect of
an endogenous variable, holding other endogenous variables fixed (e.g., p)
I normal economics: vary one exogenous variable, consider the effect of an
endogenous variable, holding other exogenous variables fixed
(what is taken to be endogenous/exogenous depends on the context)
I If e exogenous, what does a higher Et (et+τ ) mean for k in
I PV logic:
I normal economics:

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 16
Fama-French Motivation: Ceteris Paribus?
∞  i  ∞   ∞  i 
X Et dt+τ X i
Et et+τ − ΔBt+τ X Et RIt+τ
pt = = = B t +
τ =1
(1 + k)τ τ =1
(1 + k)τ τ =1
(1 + k)τ

I Arguments above are “...holding everything else fixed ”

I Meaning of “...holding everything else fixed ” in
I PV logic: vary an endogenous/exogenous variable, consider the effect of
an endogenous variable, holding other endogenous variables fixed (e.g., p)
I normal economics: vary one exogenous variable, consider the effect of an
endogenous variable, holding other exogenous variables fixed
(what is taken to be endogenous/exogenous depends on the context)
I If e exogenous, what does a higher Et (et+τ ) mean for k in
I PV logic: p unchanged, k higher
I normal economics: p adjusts, k is unchanged if risk is unchanged
I PV logic is what a believer in inefficiency might use
I Treat prices as exogenous (i.e., ignore that prices are endogenous or
assume prices have “random” variation)
I If so, where can we hope to find high returns? FF factors
Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 17
Fama-French Motivation: Rationality? Market Efficiency?

Does PV relation imply FF factors are signs of efficiency? No!

I Math + unconventional “...everything else fixed ” 6⇒ efficiency
I The PV relation just based on accounting identities
I Accounting identities always hold, no need for empirics
I PV holds for efficient or inefficient market when k is free
I In rational market, expected returns are driven by risk
I Really - just risk
I Saying: “in an efficient market, profitable firms must have high expected
returns” sounds like the wrong answer in an MBA exam...
I We must identify the rational risk of profitable firms in order to say that
they should have high or low expected returns

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 18
Fama-French Factors: Alternative Motivations

I A potential description of the tangency portfolio

I Does the Fama-French factor model “work”?
I Do the factors have positive returns?
I Do other portfolios have zero alpha relative to FF?
I If so, we learn the “building blocks” of the tangency portfolio
I Whether rational or not
I A short-hand way to capture existing knowledge about returns
I E.g., suppose someone comes with a new factor with positive
average excess return
I Is this factor truly new or just a way to repackage old stuff?
I We can check by regressing on FF

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 19
 Factor Models: Time Series Regressions

See Cochrane (2009) ch. 12 for more details and test statistics

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 20
Testing α = 0 Using Time Series Regressions
I We have N “test assets,” e.g. 25 Fama-French portfolios
I Portfolios are easy to work with: balanced panel
I Portfolios reduce noise, although reduce information too
I Want to test whether these assets are priced by tradable
factors
I Run time series regression for each asset i
X
rti = αi + β i ,k ftk + εit
k

I So we can test for “pricing,” for each asset i

αi = 0
I or that that all alphas are jointly zero (“GRS test,” Gibbons
et al. (1989)):
α=0
IElectronic
Betas also interesting
copy available –at:show factor exposures
https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 21
Typical Table for Time Series Regression
I Test asset: QMJ (and its parts, profitability, safety, growth)
I We test whether it is priced by MKT, SMB, HML, UMD
QMJt = α + β MKT MKTt + β SMB SMBt + β HML HMLt + β UMD UMDt + εt

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 22
Interpreting Alphas for Tradable Factors
rti = αi + β i ft + εit

I An excess return, rti , can be seen as “self-financing strategy”

I long $1 of the risky asset
I financed by borrowing $1
I Therefore, any linear combination of excess returns is a
self-financing strategy, e.g.
rti − β i ft = αi + εit
which has expected return
E (rti − β i ft ) = αi

I A self-financing strategy with no factor exposure must have αi = 0

if f prices the assets (as discussed before)
I Note that this argument does not hold for
Itotal returns (here r f gets messed up)
I non-tradable
Electronic factorsat: https://ssrn.com/abstract=4068783
copy available
c Lasse Heje Pedersen 23
Interpreting Alphas for Non-Tradable Factors
I Suppose that f is a non-traded factor such as GDP growth or
liquidity risk
I In this case, rti − β i ft is not a trading strategy
I As shown before, we cannot expect α = 0 in a time series
regression because the risk price λ is not E (f )
I The problem is even worse if f is related to the pricing kernel,
but its scale is unclear, e.g., there exists unknown constants a
and b s.t. mt = a + bft
I For example, if ft is a measure of liquidity computed via
bid-ask spreads, then the scale is somewhat arbitrary (e.g.,
should commissions be added?)
I E.g., what happens to α when a is higher?
I i ] = βi λ
However, we are still interested in testing E t [rt+1 t t
I I.e., do riskier assets have higher expected return?
I How to test?
IFind tradable mimicking factor(s), and then use these in time
series regression
I Use copy
Electronic instead a cross-sectional
available regression - see next slides
at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 24
 Factor Models: Cross-Sectional Regressions

See Cochrane (2009) ch. 12 for more details and test statistics

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 25
Two-Pass Cross-Sectional Regression
I Want to use a cross-sectional regression to test E[rti ] = β i λ
I Cannot use time series when factor is non-traded
I Can use either time series or cross-section for traded factor

Two-pass method:
1. First estimate betas, β̂ i , using time series regressions:
rti = αi + β i ft + εit
P
and expected returns, E(rti ), as sample counterpart, E T (rti ) := 1
T
i
t rt

2. Estimate factor premia, λ, via regression across assets

ET (rti ) = c + β̂ i λ + u i

where c is an intercept (which can be excluded) and u i is noise

I Tests of interest:
Iis the risk price significantly positive, λ > 0 ?
Iis the intercept zero as it should be, c = 0 ?
I are all the pricing errors zero, E T (r i ) − β̂ i λ = 0 for all i ?
Electronic copy available at: https://ssrn.com/abstract=4068783
t
c Lasse Heje Pedersen 26
Fama-MacBeth Regression: In General
We want to estimate b ∈ RK for the panel
i
rt+1 = b sti + εit+1

1. Each time t, get estimates b̂t , by running a cross-sectional regression of

i
rt+1 on sti :
i
rt+1 = bt sti + εit+1

where bt is the regression coefficient and uti is noise

2. Compute the time series average of b̂t over the T time periods:
1 X
b̂ = b̂t
T t

3. Estimate of the standard error of b̂ based on the idea that returns have
low correlations over time (or using more advanced ways - see
Cochrane): s
1 1 X
σ(b̂) = √ (b̂t − b̂)2
T T t
Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 27
Fama-MacBeth Regression: To Estimate Beta Pricing
Want to test E[rti ] = β i λ
1. Each time t, get an estimate of the risk premium, λ̂t :
I estimate betas, β̂ti , using time series regressions
I use past data, e.g., the 5 past years
I note that β̂ti plays the role of sti in the previous slide
I or alternatively use full-sample betas
I i
Then run a cross-sectional regression of realized returns, rt+1 ,
on betas:
i
rt+1 = ct + β̂ti λt + ut+1
i

where λt is the slope coefficient, ct is the intercept (can be

excluded), and uti is noise
2. Estimate the risk premium as the time series average of the slopes:
1 X
λ̂ = λ̂t
T t

3. Estimate of the standard error of λ̂, e.g., as:

s
1 1 X
σ(λ̂) = √ (λ̂t − λ̂)2
T T t
Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 28
 Factor Models as Risk Models

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 29
Factor Models to Capture Risk
I A specification of the variance-covariance matrix, Var(rt ), is also called a
“risk model”
I A factor model, rt = α + βft + εt , implies a risk model
Var(rt ) = βVar(ft )β 0 + Var(εit )
I This can be implemented in different ways:

1. (Observed factors) Start with observed factors ft

I these should be important for Var(rt )
I estimate factor model and its risks using time series regression
2. (PCA) Do as in 1, but using principal components
3. (BARRA) Start with observed characteristics (e.g., industry,
B/M, etc.) interpreted as factor loadings, βt , and consider
factor returns, ft+1 , to be unobserved in
i
rt+1 = ct + βti ft+1 + ut+1
i

I i
Each t, run a cross-sectional regression of rt+1 on βt , where
the slope coefficients are interpreted as the realized factor
returns, ft+1
I Estimate the factor risk, Var(ft ), and idiosyncratic risk,
i
Var(ε
Electronic copy t ), using time
available series
at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 30
 Return Predictability

See Cochrane (2011), Presidential Address: Discount Rates

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 31
Return Predictability

I An interesting issue is how expected returns change over time

and across assets
I In other words, can returns be predicted?
I If so, how can returns be predicted?
I This issue can be addressed using predictive regressions

i
rt+1 = a + b sti + εt+1
where the signal sti is known ex ante

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 32
Return Predictability: Time Series I
I Take a single asset, typically an overall market like the equity
market, and consider predictive regression:

rt+1 = a + b st + εt+1
I Suppose rt+1 is annualized, st = D
Pt is annual dividend yield
t

I What do you think b is? What is the interpretation of

I b = 0?
I b = 1?
I b ∈ (0, 1)?
I b > 1?

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 33
Return Predictability: Time Series II
I Example from Cochrane (2011):

I Discuss the magnitude of

I predictive coefficient b
I its statistical significance, t(b)
I the time-varying in expected returns, σ[Et (rt+1 )]
I explanatory power, R 2

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 34
Return Predictability: Time Series III
I Comments on time series predictability
I Out-of-sample predictability is much weaker, and small for
most predictors of the market
I Welch and Goyal (2008), update Goyal et al. (2021)
I b may suffer from Stambaugh (1999) bias and other biases
I See Campbell (2017) ch. 5.4 for responses
I Predictive regression on the dividend yield can be understood
via present value relation
I See Campbell (2017) ch. 5.1-3 and Cochrane (2011)
I A pervasive phenomenon (cf. “everywhere factors”), from
Cochrane (2011):

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 35
Return Predictability: Cross Section

I 1 , ..., r N
Suppose that we have many assets, rt+1 t+1
I Each asset has its own signal, sti , at each time
I The predictive regression is now a panel:

i
rt+1 = a + b sti + εt+1
I We can run using
I a pooled regression (be careful with standard errors, see
Petersen (2009)), or
I Fama and MacBeth (1973) as described previously, where
I b̂t can be interpreted as the profit of a trading strategy
I b̂ as the average profit
√
I Sharpe ratio times T provides simple t-statistic
I see Pedersen (2015) ch. 3.4

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 36
Predictability: Link to Factors and Trading Strategies

I Note the economic equivalence of

1. a multivariate predictive regression
2. a long short factor
I More broadly, Pedersen (2015) ch. 3.4 states

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 37
Event Studies: Event Time vs. Calendar Time
I Another way to study predictability is using an event study
I An event study is used to analyze
I stock returns before and after an event, e.g.,
I a stock split, merger, earnings announcement, issuance
I note: returns before the event obviously have selection bias (but still

interesting), while returns after are predictive

I and also the firm fundamentals around the event

I How to make an event study:

Ifirms are aligned in “event time” rather than calendar time
Icompute average (abnormal) return at each event time, e.g., 1
month after the event, 2 months after, and so on
I A factor rebalance can be used as an “event”
I That means that we have an event each time period

I Notes on event studies

I If many firms have events at the same time, then biases can arise

and standard errors in an event study might be wrong

I Good idea to also check the result in calendar time, e.g.:
I
Event time: look at performance for firms the months after they
issue shares (each firm is counted equally)
I Calendar time: go long/short firms based on their issuance; look
Electronic copy
time available at: https://ssrn.com/abstract=4068783
series of returns (each time period counted equally)
c Lasse Heje Pedersen 38
Event Studies: Post-Earnings Announcement Drift
I PEAD: a classic example of using an event study to document
predictability, Ball and Brown (1968)
I Here from Bernard and Thomas (1989):

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 39
Event Studies: Example of Factor Return in Event Time
I Example from Asness et al. (2021)

Notes: an event study tracking historical and future returns to

value portfolios having different levels of valuation spreads.
Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 40
Other Signs of Predictability: Excess Volatility and Bubbles

I Excess volatility: Cochrane (2009) page 396 says it so well:

I Bubbles: several recent papers, e.g., Giglio et al. (2016)

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 41
Views on Predictability

Old school view (1960s and 1970s)

I Stock returns are relatively unpredictable
I i.e., expected returns do not move around
I so price changes must be due to news about cash flows
I and market efficient in simplest sense: random walk hypothesis
I CAPM works pretty well, although low-beta effect discovered
in the 1970s
Modern view of the facts (1980s-)
I Stock returns are predictable
I in the cross section by stock characteristics (value, mom, etc.)
I in the time series
I I.e., expected returns move around
I CAPM works poorly: SML is relatively flat and, more broadly,
market betas explain little of the cross-section of exp. returns
Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 42
Views on Predictability, continued
Interpretations of modern facts
I Rational
I Markets can be efficient even though expected returns vary
I Expected returns driven by risk, but risk is not CAPM beta
I Joint hypothesis problem
I New models of utility (habit, long-run risk), macro-finance, etc.
I Behavioral
I Markets are inefficient
I Mispricings (bubbles and crashes) drive expected returns
I Frictions (more in Lecture 6)
I Leverage constraints cause the low-beta effect
I Market/funding liquidity risk affect expected returns
I in the time series (high ER during crises, esp. illiquid assets)
I in the cross section
I Replication crisis (more in Lecture 4)
IModern facts are driven by data mining
IPredictability weaker out of sample, especially time-series
predictability of the market
I Maybe
Electronic copytheavailable at:was
old school https://ssrn.com/abstract=4068783
not all wrong?
c Lasse Heje Pedersen 43
 Further Empirical Methods

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 44
Standard Errors and Test Statistics

I Getting the “right” standard errors an important issue

I Not enough time to cover this topic in this course
I Some well-known methods
I OLS
I Bootstrap
I Newey and West (1987)
I Fama and MacBeth (1973)
I Clustering (by time and/or firm)
I Overview of several approaches in Petersen (2009)

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 45
Three Tricks for Causality

1. Instrumental variables
2. Difference in differences
3. Regression discontinuity

I Not the topic of this class – see

I Book by Angrist and Pischke (2008)
I Short Ph.D. class at Copenhagen Business School by Kasper
Meisner Nielsen

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 46
Frequentist Stats, Bayesian Stats, and Machine Learning
I Three tool sets
1. Frequentist statistics (e.g., book by Cochrane (2009))
I Maximum likelihood estimation (MLE), generalized method of
moments (GMM), etc.
2. Bayesian statistics (e.g., free book by Gelman et al. (2020))
I Full Bayes, empirical Bayes, hierarchical models, etc.
3. Machine learning (e.g., free book by Hastie et al. (2009))
I Penalized regressions, trees, neural networks, etc.
I All are useful in different situations
I Different views on data-generating process (DGP), see
Breiman (2001)
Model view Goal
1. Frequentist There is a true DGP Estimate true parameters
2. Bayesian DGP is random Compute posterior
3. ML What DGP? Get an algo to work

I Learn about the details of these in other classes

IWe later consider some applications (frequentist in this lecture,
Bayesian
Electronic copy in Lecture at:
available 4, ML in Lecture 5)
https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 47
Links to Books/Notes on Empirical Asset Pricing

I General empirical asset pricing

I “Financial Decisions and Markets: A Course in Asset Pricing,”
Campbell (2017)
I “Asset Pricing,” Cochrane (2009)
I “Empirical Dynamic Asset Pricing: Model Specification and
Econometric Assessment,” Singleton (2006)
I More applied
I “Efficiently Inefficient,” Pedersen (2015)
I “Empirical asset pricing: The cross section of stock returns”
Bali et al. (2016)
I Lecture notes
I John Cochrane
https://www.johnhcochrane.com/empirical-asset-pricing
I Ralph Koijen and Stijn Van Nieuwerburgh:
https://www.koijen.net/phd-notes-empirical-asset-pricing.html

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 48
References Cited in Slides I
Acharya, V. and L. H. Pedersen (2005). Asset pricing with liquidity risk. Journal of Financial Economics 77,
375–410.
Angrist, J. D. and J.-S. Pischke (2008). Mostly harmless econometrics. Princeton university press.
Asness, C., A. Frazzini, and L. H. Pedersen (2018). Quality minus junk. Review of Accounting Studies,
forthcoming .
Asness, C., J. Liew, L. H. Pedersen, and A. Thapar (2021). Deep value. The Journal of Portfolio
Management 47 (4), 11–40.
Asness, C. S., T. J. Moskowitz, and L. H. Pedersen (2013). Value and momentum everywhere. The Journal of
Finance 68 (3), 929–985.
Bali, T. G., R. F. Engle, and S. Murray (2016). Empirical asset pricing: The cross section of stock returns. John
Wiley & Sons.
Ball, R. and P. Brown (1968). An empirical evaluation of accounting income numbers. Journal of accounting
research, 159–178.
Bernard, V. L. and J. K. Thomas (1989). Post-earnings-announcement drift: delayed price response or risk
premium? Journal of Accounting research 27, 1–36.
Breiman, L. (2001). Statistical modeling: The two cultures (with comments and a rejoinder by the author).
Statistical science 16 (3), 199–231.
Campbell, J. Y. (2017). Financial decisions and markets: a course in asset pricing. Princeton University Press.
Cochrane, J. H. (2009). Asset pricing: Revised edition. Princeton university press.
Cochrane, J. H. (2011). Presidential address: Discount rates. The Journal of Finance 66, 1047–1108.
De Bondt, W. F. and R. Thaler (1985). Does the stock market overreact? The Journal of finance 40 (3), 793–805.
Fama, E. F. and K. R. French (1993). Common risk factors in the returns on stocks and bonds. Journal of
Financial Economics 33 (1), 3–56.
Fama, E. F. and K. R. French (2015). A five-factor asset pricing model. Journal of financial economics 116 (1),
1–22.
Electronic copy available at: https://ssrn.com/abstract=4068783
c Lasse Heje Pedersen 49
References Cited in Slides II
Fama, E. F. and J. D. MacBeth (1973). Risk, return, and equilibrium: Empirical tests. Journal of political
economy 81 (3), 607–636.
Frazzini, A. and L. H. Pedersen (2014). Betting against beta. Journal of Financial Economics 111 (1), 1 – 25.
Gelman, A., J. B. Carlin, H. S. Stern, D. B. Dunson, A. Vehtari, and D. B. Rubin (2020). Bayesian data analysis.
Chapman and Hall, Free electronic version.
Gibbons, M. R., S. A. Ross, and J. Shanken (1989). A test of the efficiency of a given portfolio. Econometrica,
1121–1152.
Giglio, S., M. Maggiori, and J. Stroebel (2016). No-bubble condition: Model-free tests in housing markets.
Econometrica 84 (3), 1047–1091.
Goyal, A., I. Welch, and A. Zafirov (2021). A comprehensive look at the empirical performance of equity premium
prediction ii. Available at SSRN 3929119 .
Hastie, T., R. Tibshirani, and J. Friedman (2009). The Elements of Statistical Learning: Data Mining, Inference,
and Prediction. Springer series in statistics. Springer.
Jegadeesh, N. (1990). Evidence of predictable behavior of security returns. The Journal of finance 45 (3), 881–898.
Jegadeesh, N. and S. Titman (1993). Returns to buying winners and selling losers: Implications for stock market
efficiency. The Journal of finance 48 (1), 65–91.
Jensen, T. I., B. T. Kelly, and L. H. Pedersen (2022). Is there a replication crisis in finance? Journal of Finance,
forthcoming .
Koijen, R. S., T. J. Moskowitz, L. H. Pedersen, and E. B. Vrugt (2018). Carry. Journal of Financial
Economics 127 (2), 197–225.
Moskowitz, T. J., Y. H. Ooi, and L. H. Pedersen (2012). Time series momentum. Journal of financial
economics 104 (2), 228–250.
Newey, W. K. and K. D. West (1987). A simple, positive semi-definite, heteroskedasticity and
autocorrelationconsistent covariance matrix. Econometrica 55 (3), 703–708.
Pastor, L. and R. F. Stambaugh (2003). Liquidity risk and expected stock returns. Journal of Political
Economy 111, 642–685.

Electronic copy available at: https://ssrn.com/abstract=4068783

Pedersen, L. H. (2015). Efficiently inefficient. Princeton University Press.
c Lasse Heje Pedersen 50
References Cited in Slides III

Pedersen, L. H., S. Fitzgibbons, and L. Pomorski (2021). Responsible investing: The esg-efficient frontier. Journal
of Financial Economics 142 (2), 572–597.
Petersen, M. A. (2009). Estimating standard errors in finance panel data sets: Comparing approaches. The Review
of financial studies 22 (1), 435–480.
Singleton, K. J. (2006). Empirical dynamic asset pricing: model specification and econometric assessment .
Princeton University Press.
Sloan, R. G. (1996). Do stock prices fully reflect information in accruals and cash flows about future earnings?
Accounting review , 289–315.
Stambaugh, R. F. (1999). Predictive regressions. Journal of financial economics 54 (3), 375–421.
Stattman, D. (1980). Book values and stock returns. The Chicago MBA: A journal of selected papers 4 (1), 25–45.
Welch, I. and A. Goyal (2008). A comprehensive look at the empirical performance of equity premium prediction.
The Review of Financial Studies 21 (4), 1455–1508.

Electronic copy available at: https://ssrn.com/abstract=4068783

c Lasse Heje Pedersen 51