Page 583 -

582 Chapter 14 Introduction to Time Series Regression and Forecasting

Can You Beat the Market? Part I

H ave you ever dreamed of getting rich quickly Forecasts based on past values of stock returns
by beating the stock market? If you think that are sometimes called “momentum” forecasts: If
the market will be going up, you should buy stocks the value of a stock rose this month, perhaps it has
today and sell them later, before the market turns momentum and will also rise next month. If so, then
down. If you are good at forecasting swings in stock returns will be autocorrelated, and the autoregressive
prices, then this active trading strategy will produce model will provide useful forecasts. You can implement
better returns than a passive “buy and hold” strat- a momentum-based strategy for a specific stock or
egy in which you purchase stocks and just hang onto for a stock index that measures the overall value of
them. The trick, of course, is having a reliable fore- the market.
cast of future stock returns.

TABLE 14.2 Autoregressive Models of Monthly Excess Stock Returns, 1960:M1–2002:M12

Dependent variable: Excess returns on the CRSP value-weighted index (2) (3)
(1)

Specification AR(1) AR(2) AR(4)

Regressors

excess returnt - 1 0.050 0.053 0.054
(0.051) (0.051) (0.051)

excess returnt - 2 –0.053 –0.054
(0.048) (0.048)

excess returnt - 3 0.009
(0.050)

excess returnt - 4 −0.016
(0.047)

Intercept 0.312 0.328 0.331
(0.197) (0.199) (0.202)

F-statistic for coefficients on- 0.968 1.342 0.707
lags of excess return (p-value) (0.325) (0.261) (0.587)

R2 0.0006 0.0014 –0.0022

Note: Excess returns are measured in percentage points per month. The data are described in Appendix 14.1. All regressions

are estimated over 1960:M1–2002:M12 (T = 516 observations), with earlier observations used for initial values of lagged
variables. Entries in the regressor rows are coefficients, with standard errors in parentheses. The final two rows report the

F-statistic testing the hypothesis that the coefficients on lags of excess return in the regression are zero, with its p-value in
parentheses, and the adjusted R2.