Long-Horizon Predictability: A Cautionary Tale

Abstract

Disclosure: AQR Capital Management is a global investment management firm that may or may not apply investment techniques or methods of analysis similar to those described herein. The views expressed here are those of the authors and not necessarily those of AQR.

Should stock returns predictability be ‘hooked on’ long‐horizon regressions?

“…In particular, predictability is assessed by regressing stock returns on variables that can act as leading indicators. Nevertheless, for reasons related to the statistical inference emanating from the predictive regressions and the robustness of the conducted pseudo‐forecasting exercises, the debate on stock returns predictability is still open (e.g., see Boudoukh, Israel, & Richardson, 2018; Rapach & Wohar, 2005).…”

“…The current consensus in the existing literature, regarding the derived statistical inference from predictive regressions, could be summarized as follows: First, the Stambaugh (1999) bias that stems in the presence of small samples and persistent predictors, such as earnings‐price and dividend‐price ratios, leads to substantial size distortions, when testing the null hypothesis of no‐predictability (see also Boudoukh et al, 2018). Second, the Berkowitz and Giorgianni (2001) criticism on the inherent inconsistency of the linear predictive regression structure.…”

“…Rationally, such linear framework should imply ‘ predictability at all horizons or predictability at no horizon ’ (Rapach & Wohar, 2005: p. 328); since, in this context, long‐horizon forecasts are extrapolations of short horizon forecasts. Third, as Valkanov (2003) argues, the utilization of overlapping observations intensifies the problem of serial‐correlation and leads to non‐robust standard‐errors and misleading inferences; see also Boudoukh et al (2018). For the above reasons, approaches that lead to more robust inferences than predictive regressions have been proposed by several authors, such as Boudoukh, Richardson, and Whitelaw (2008), among others.…”

“…For empirical support on the use of the two measures of fundamental value see, inter alias , Fama and French (1988), Campbell and Shiller (1988, 1998) and Rapach and Wohar (2005). Nonetheless, recent studies, such as Boudoukh et al (2018), cast doubts on the reliability of such predictor variables, in the context of predictive regressions, concluding that the potential of long‐horizon return predictability should be considered with scepticism.…”