Transcript WCS Academy
Welcome to the course
Financial Econometrics II
Course objectives
Financial data analysis, based on
Asset pricing models
Econometric methods
Active knowledge
Try it yourself!
Share personal vision
Avoid common pitfalls
2
What you already know: Asset pricing models
Risk-return trade-off
Only systematic risk is priced!
Idiosyncratic risk is reduced by diversification
What are the risk factors?
CAPM: market
APT: many other
3
What you already know: Econometric methods
Regression
Coefficients
Standard errors
R-squared
Potential problems
Outliers, heteroscedasticity, endogeneneity, etc.
Packages
Eviews, Excel
4
Plan for today
Specifics of financial data
The efficient market hypothesis
Tests for return predictability
5
Asset prices: S&P500
6
Asset prices
Convenient for graphical analysis, but..
Non-stationary
Properties of the stochastic process change over time
Must be corrected for dividends, stock splits
Should be normalized to compare dynamics over time
and across securities
7
Asset returns: S&P500 (monthly)
8
Asset returns: discrete
Net return:
Rt = (Pt-Pt-1)/Pt-1
Gross return:
1+ Rt
Adjusting for dividends (total returns):
Rt = (Pt+Dt-Pt-1)/Pt-1
Adjusting for inflation (real returns):
1+Rt(real) = (Pt/Pt-1)*(CPIt-1/CPIt)
Excess returns (risk premia):
Zt = Rt – RF,t
9
Asset returns: discrete
k-period return:
1+ Rt(k) = (1+ Rt)*…*( 1+ Rt-k+1)
Annualized return
Geometric average (effective return): (1+ RA)k = 1+ Rt(k)
Arithmetic average: RA = (1/k) j=0:k-1 Rt-j
Portfolio return: RP,t = i wi Ri,t
wi: portfolio weights summing up to 1
10
Asset returns: continuously compounded
Log return:
rt = ln(1+ Rt) = ln(Pt/Pt-1) = ln(Pt) – ln(Pt-1)
k-period return:
rt(k) = ln(Pt/Pt-k) = j=0:k-1 rt-j
Annualized return: rA = j=0:k-1 rt-j
Portfolio return:
rP,t = ln(1+ RP,t) ≠ i wi ri,t
But rP,t ≈ i wi ri,t for small RP,t
11
Asset returns: simple vs. log
Simple returns:
Exact
Easy to aggregate stocks into portfolio
Log returns:
Easy to aggregate over time
Closer to normal distribution
– Don't violate limited liability
12
Market microstructure effects
Which prices to use to measure returns?
Average vs. close
Bid vs. ask
Can you make real profit out of paper returns?
Transaction costs, including bid-ask spread
Liquidity
Price impact
13
What do we know about stock returns?
Characteristics of the distribution
Mean (center):
μ = E[R]
Variance (spread): 2 = E[(R-μ)2]
Skewness (symmetry):
S = E[(R-μ)3/3]
Kurtosis (tail thickness): K = E[(R-μ)4/4]
15
Normal distribution
X ~ N(μ, 2)
S=0
K=3
QQ-plot: standardized empirical quantiles vs.
theoretical quantiles from specified distribution
2 ( K 3) 2
n
Jarque-Berra test for normality: JB S
6
4
n : # observations
JB has an asymptotic chi-square distribution with 2 degrees of
freedom
16
Histogram of S&P500 monthly returns
17
Normal QQ-plot for S&P500 monthly returns
18
S&P500: daily returns
19
Normal QQ-plot for S&P500 daily returns
20
Stylized facts about stock returns
Monthly returns
Appoximately normal
Daily returns
Non-normality
– Asymmetry: usually, S < 0 for the indices
– Thick tails: K > 3
Volatility сlustering
– Esp. at daily frequency
21
How can we model stock returns?
Traditional approach:
Explain asset prices by rational models
Only if they fail, resort to irrational investor behavior
Behavioral finance models
22
What is an efficient market?
Informational efficiency (Fama, 1969):
asset prices accommodate all relevant information
History of prices / all public variables / all private information
Price movements must be random!
Otherwise one can forecast future price and make arbitrage profit
Prices should immediately respond to new information
Why is it important?
Stock markets: portfolio management
Corporate finance: choice of the capital structure
23
The efficient market hypothesis
EMH: stock prices fully and correctly reflect all relevant
information
Pt+1 = E[Pt+1 |It] + εt+1
Rt+1 = E[Rt+1 |It] + et+1
The error has zero expectation and is orthogonal to It
E[Rt+1 |It] is normal return or opportunity cost implied by some model
24
Different forms of ME
Weak (WFE):
I includes past prices
Semi-strong (SSFE):
I includes all public info
Strong (SFE):
I includes all (also private) info
25
What if the EMH is rejected?
The joint hypothesis problem: we simultaneously test
market efficiency and the model
Either the investors behave irrationally,
or the model is wrong
One may not necessarily earn on this
Ex-ante expected profit is within information acquisition and
transaction costs
26
If the EMH is not rejected, then…
the underlying model is a good description of the
market,
fluctuations around the expected price are unforecastable,
due to randomly arriving news
there is no place for active ptf management…
technical analysis (WFE), fundamental analysis (SSFE),
or insider trading (SFE) are useless
the role of analysts limited to diversification, minimizing taxes and
transaction costs
or corporate policy:
the choice of capital structure has no impact on the firm’s value
still need to correct market imperfections (agency problem, taxes)
27
How realistic is it?
The Grossman-Stiglitz paradox:
there must be some strong-form inefficiency left
to provide incentives for information acquisition
Operational efficiency:
one cannot make profit on the basis of info,
accounting for transaction costs
28
Different types of tests
Tests of informational efficiency:
Finding variables predicting future returns
Statistical significance
Tests of operational efficiency:
Finding trading rules earning positive profit taking into account
transaction costs and risks
Economic significance
29
Different types of models
Constant expected return:
Et[Rt+1] = μ
Tests for return predictability
CAPM:
Et[Ri,t+1] – RF = βi(Et[RM,t+1] – RF)
Tests for mean-variance efficiency
Multi-factor models
30
Topics covered in this course
Time series analysis of asset returns
Predictability at different horizons
Event study analysis
– Speed of stock price adjustment in response to news announcements
Cross-sectional analysis of asset returns
CAPM and market efficiency
Return anomalies and multi-factor asset pricing models
Performance evaluation of mutual funds
Performance persistence, survivorship bias, dynamic strategies,
gaming behavior, etc.
31
How can we model stock returns?
Is this market efficient / is price predictable?
33
Martingale and MDS
Let {Yt} be a sequence of random variables
Let {It} be a sequence of info sets, It I (universal info set)
(Yt, It) is a martingale if
It It+1 (filtration)
Yt It (Yt is adapted to It)
E[|Yt|]<
E[Yt|It-1] = Yt-1 (martingale property)
Example: random walk
Yt = Yt-1 + εt, εt ~ IID(0, σ2)
It = {εt, εt-1, …}
34
Martingale and MDS
Law of iterated expectations:
E[Yt|It-2] = E[E[Yt|It-1] |It-2] = E[Yt-1|It-2] = Yt-2
In general, E[Yt|It-k] = Yt-k
(Xt, It) is a martingale difference sequence (MDS) if
E[Xt|It-1] = 0
Construct MDS from a martingale: Xt = Yt - E[Yt|It-1]
Can we apply it to stock prices/returns?
Under EMH, stock prices are martingales after detrending
the unexpected stock returns are MDS: E[Rt+1 - E[Rt+1 |It] |It] = 0
35
The random walk hypotheses
Random walk with drift:
Δln(Pt) = μ + εt or rt = μ + εt
RW1: independent and identically distributed
increments, εt~IID(0, σ2)
Any functions of the increments are uncorrelated
Unrealistic
RW2: independent increments
Allows for heteroskedasticity
Test using filter rules, technical analysis
RW3: uncorrelated increments, cov(εt, εt-k) = 0, k>0
Allows for dependence on higher moments (e.g., GARCH)
Test using autocorrelations, variance ratios, regressions
36
Autocorrelation tests
Assume that rt is covariance stationary
Test that autocorrelation is 0
For a given lag: Fuller’s test
Asymptotically, T k ~ N(0,1)
For m lags:
Box-Pierce statistic: Q ≡ T Σk=1:mρ2(k) ~ 2(m)
Ljung-Box statistic (finite-sample correction):
Q’ ≡ T(T+2) Σk=1:mρ2(k)/(T-k) ~ 2(m)
37
Autocorrelation tests: results
CLM, Table 2.4: US, 1962-1994
The CRSP stock index has significantly positive first-order
autocorrelation at D, W, and M frequency
Economic significance: 12% (R2 = square of 1) of the variation in
daily value-wtd CRSP index is predictable from the last-day return
The equal-wtd index has higher autocorrelation
Predictability declines over time
38
Variance ratios
Intuition: how to aggregate volatility over time?
Usually, T rule for standard deviation
Example: compute the variance of a 2-day return:
var(rt + rt-1) = var(rt) + var(rt-1) + 2 cov(rt, rt-1) = {assuming covariance
stationarity and no autocorrelation}= 2 var(rt)
Variance ratio: VR(2) ≡ var(rt(2)) / 2var(rt) = 1 + 1
In general,
VR(q)≡Var[rt(q)]/(qVar[rt]) = 1 + 2k=1:q-1 (1-k/q) k
Under RW1, VR=1
Test statistic:
3Tq
(q) (VR (q) 1)
~ asy N (0,1)
2(2q 1)(q 1)
39
Variance ratios: results
CLM, Tables 2.5, 2.6, 2.7: US, 1962-1994, weekly data
Indices: VR(q) rises with time interval (positive autocorrelation),
predictability declines over time, is larger for small-caps
Individual stocks: weak negative autocorrelation
How to reconcile this contradiction?
Table 2.8: size-sorted portfolios
Large positive cross-autocorrelations,
large-cap stocks lead small-caps
40
Regression analysis: ARMA models
Testing for long-horizon predictability:, Rt+h(h)=a+bRt(h)+ut+h,
Non-overlapping horizons: too few observations
Overlapping horizons: autocorrelation ρ(k)=h-k => use HAC s.e.
Results from Fama&French (1988): US, 1926-1985
Negative autocorrelation (mean reversion) for horizons from 2 to 7 years,
peak b=-0.5 for 5y
Poterba&Summers (1988): similar results based on VR
Critique:
Small-sample and bias adjustments lower the significance
Results are sensitive to the sample period, largely due to 1926-1936 (the
Great Depression)
41
Interpretation: Behavioral finance
Investor overreaction
Assume RW with drift, Et[Rt+1] = μ
There is a positive shock at time τ
The positive feedback (irrational)
traders buying for t=[τ+1:τ+h] after
observing Rτ>μ
SR (up to τ+h): positive
autocorrelation, prices overreact
LR (after τ+h): negative
autocorrelation, prices get back to
normal level
Volatility increases
42
Interpretation: Market microstructure
Non-synchronous trading
Low liquidity of some stocks (assuming zero returns for
days with no trades) induces
negative autocorrelation (and higher volatility) for them
positive autocorrelation (and lower volatility) for indices
lead-lag cross-autocorrelations
Consistent with the observed picture (small stocks are
less liquid), but cannot fully explain the magnitude of
the autocorrelations
43
Interpretation: More complicated model
Time-varying expected returns:
Et[Rt+1] = Et[RF,t+1] + Et[RiskPremiumt+1]
Changing preferences / risk-free rate / risk premium
Decline in interest rate => increase in prices
If temporary, then positive autocorrelation in SR, mean reversion in
LR
44
Conclusions
Reliable evidence of return predictability at short
horizon
Mostly among small stocks, which are characterized by low liquidity
and high trading costs
Weak evidence of return predictability at long horizon
May be related to business cycles (i.e., time-varying returns and
variances)
45