A1787
Title: Monitoring the predictability of stock returns under nonstationary volatility
Authors: Matei Demetrescu - TU Dortmund University (Germany)
Fabian Schmidt - TU Dortmund University (Germany) [presenting]
Robert Taylor - University of Essex (United Kingdom)
Abstract: The predictability of stock returns is most likely episodic in nature. To exploit upcoming pockets of predictability, one must detect the point in time when stock returns become predictable. Such real-time monitoring of predictability entails the repeated application of predictability tests as new data become available. Therefore, in addition to dealing with so-called predictive regression endogeneity, one must account for the multiple testing issues inherent to monitoring. Moreover, stock returns typically exhibit time-varying volatility, and ignoring such data features typically results in spurious detection of predictability. For this reason, a real-time monitoring procedure is proposed that takes uncertain persistence and time-varying volatility into account. The strategy is based on a CUSUM procedure originally proposed for bubble monitoring but with two essential modifications. First, it is applied to specific moment conditions under the null, and second, it is adjusted as-you-go to take possible unconditional changes in volatility into account. The adjustments are nonparametric in nature and do not require any specific assumptions for the volatility path. Monte Carlo simulations show the procedure to work reliably for various patterns of volatility changes, and an application to an S&P 500 dataset illustrates its practical application.
