Random Walk Tests and Variance Ratios Fin250f: Lecture 4.1 Fall 2005 Reading: Taylor, chapter
Outline Variance ratios Autocorrelation sampling theory
Types of Random Walks e(t) IID: No volatility persistence e(t): expectation zero, zero correlation No linear predictors Might be nonlinear predictors Allows for volatility prediction
Random Walks and Market Efficiency Classic implications Price forecasting hopeless Technical analysis useless Modern thoughts/reminders Dynamic strategies that Increase expected returns Increase risk Still consistent with market efficiency
Variance Ratio Tests
Variance Ratios: Random Walk Test Test: VR(N)=1 Two problems Distribution of VR? Which N?
Distribution of VR(N): Asymptotic
Matlab Examples vratio vratiotest nmcvratio bsvratio
Longer Time Horizons Weekly Lo and Mackinlay(1988) Strong rejections on weekly equal weighted index (not value weighted) Few rejections for individual stocks Stale prices and nontrading?
Longer Time Horizons Monthly Poterba and Summers(1988) Weak positive correlations (not sig) Annual Weak negative long range correlations (not sig)
Autocorrelations For r(t) IID Autocorrelations are asymptotically distributed N(0, 1/n) n=sample size 95% confidence bands /sqrt(n) pacf.m
Autocorrelations: Small sample issues
Autocorrelations: Variance