a basis for understanding the cross-sectional differences in momentum prof-
its. However, these cross-sectional differences can be explained by multiple
behavioral biases that underlie these models. Therefore, it is hard to use these
findings to assess which particular behavioral bias gives rise to momentum
profits. Very likely, all biases play some role.
6 .Earnings MomentumThe results so far have focused on the profitability of momentum strategies
based on past returns. Naturally, returns are driven by changes in underly-
ing stock fundamentals. Stock returns tend to be high, for example, when
earnings growth exceeds expectations and when consensus forecasts of fu-
ture earnings are revised upward. An extensive literature examines return
predictability based on momentum in past earnings and momentum in ex-
pectations of future earnings as proxied by revisions in analyst forecasts.
This section selectively reviews the evidence from the earnings momentum
literature and presents the interaction between earnings momentum and
price momentum.
A partial list of papers that investigate the relation between past earnings
momentum and futures returns includes: Jones and Litzenberger (1970),
Latane and Jones (1979), Foster, Ohlsen, and Shevlin (1984), Bernard and
Thomas (1989), and Chan, Jegadeesh, and Lakonishok (1996). These pa-
pers typically measure earnings momentum using a measure of standard-
ized unexpected earnings (SUE). SUE is defined as:
These papers use different variations of time-series models to determine
earnings expectations. Typically, the papers assume that quarterly earnings
follow a seasonal random walk with drift, but they differ in their assump-
tions about earnings growth. Specifically, Jones and Litzenberger (1976)
and Latane and Jones (1979) assume that quarterly earnings grow at a con-
stant rate, Foster et al. (1984), and Bernard and Thomas (1989) model
quarterly earnings growth as an AR(1) process, and Chan et al. (1996) as-
sume zero expected growth in quarterly earnings.
Among these statistical models for quarterly earnings growth, the AR(1)
model is the most realistic specification since it captures the mean reversion
in earnings growth.^7 However, the robustness of the results across different
papers indicates that the accuracy of the model used to specify expected
SUE
Quarterly earnings Expected quarterly earnings
Standard deviation of quarterly earnings growth=.
−378 JEGADEESH AND TITMAN
(^7) See Foster et al. (1984) for an evaluation of the relative accuracy of various statistical
models to capture the time-series properties of quarterly earnings.