Predicting equity premium out-of-sample by conditioning on newspaper-based uncertainty measures: A comparative study

In finance, the use of newspaper-based uncertainty measures has grown exponentially in recent years. For instance, a growing number of researchers have used the newspaper-based U.S. economic policy uncertainty (EPU) index suggested in Baker et al. (2016) as a predictor in their model to forecast the variable of interest out-of-sample. Likewise, inspired by the approach suggested in Baker et al. (2016), several other newspaper-based uncertainty measures have been introduced, such as indices measuring geopolitical risk (GPR) and monetary policy uncertainty (MPU). This study evaluates the relative out-of-sample predictive power afforded by more than fifty different newspaper-based uncertainty measures with regards to predicting excess returns on the S&P 500 index one-month ahead using data from 1985$m$1 through 2020$m$12. Our predictive model accounts for salient data features, namely, predictor endogeneity and persistence. Furthermore, we evaluate the evidence of conditional as well unconditional predictive ability as outlined in Giacomini and White (2006), and also explore whether any identified level of gains from a statistical viewpoint lead to gains from an economic viewpoint. We find that newspaper-based uncertainty measures linked with certain components of the equity market volatility (EMV) tracker suggested in Baker et al. (2019) help improve the accuracy of one month ahead point predictions relative to the benchmark the most. In contrast, EPU, GPR and MPU indices, which are more frequently used by researchers are much less successful.