(When) Do persistent predictors predict stock returns directionally?
Loading...
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Alternative Title(s)
Abstract
This paper addresses the question of whether the direction of stock price movements can be predicted, even if only temporarily, by lagged financial predictors. We argue that, analogously to the usual problems faced by predictive regressions for stock returns, directional models like probit or logit type regressions tend to be severely oversized in the presence of highly persistent lagged predictors when employing standard inference. To robustify against uncertain persistence of the underlying time series we resort to estimation and testing of a specific auxiliary regression. Following Demetrescu et al. (2022, J. Econometrics 227, 85-113), we employ tailored instrumental variables for this purpose, deployed in a rolling window approach. Based on a fixed-regressor wild bootstrap implementation, we are able to pin down periods of highest predictability for each considered predictor irrespective of its degree of persistence, while controlling for the effect of repeated testing. Monte Carlo simulations for a variety of empirically relevant data generating processes are conducted to confirm the reliability of the testing scheme proposed here. For the sign of monthly returns of the S&P500 index, we find the inflation rate, the dividend yield, dividend payout ratio, net equity expansion, and term spread, to exhibit significant predictive power only within historical pockets, primarily between 1940 and 1980. Regarding daily returns of individual stocks of the DJIA, the (lagged) stock return itself, a moving average of closing prices, the divergence between price and moving average, and the stochastic oscillator exhibit periods of significant predictive power, though without a clear temporal pattern.
Description
Table of contents
Keywords
sign predictability, binary dependent variables, predictive regression endogeneity, uncertain persistence
Subjects based on RSWK
Endogene Variable, Regressionsanalyse, Aktienkurs, Prognose, Monte-Carlo-Simulation, SP500