This article dives into the specifics of overparametrized models for aggregate market forecasting when regime changes are accounted for. We document the sensitivity of the out-of-sample analysis of Kelly et al. (2023) to sub-periods and bandwidth parameters. For the average investor, our results show that focusing on periods of 15 years can generate very heterogeneous returns, especially for small bandwidths. Large bandwidths yield much more consistent outcomes, but are far less appealing from a performance standpoint. All in all, our findings tend to recommend cautiousness when resorting to large linear models for stock market predictions.
fr
en