Figs. 3 and 4 allow us to visualize the evolution of the out-of-sample performance of some of the (combinations of) variables over time. The performance is measured by the difference between the cumulative sum of squared errors (SSE) generated by the sample mean and the cumulative SSE generated by a given set of variables. The out-of-sample performance is measured by where the mean and the prediction are calculated over the period from s0 to t. An increase in the line indicates that the model provides a better prediction than the prevailing mean. We have selected the most relevant predictors in the figures. Two results are worth emphasizing. First, as Fig. 3 illustrates, the out-of-sample measures involving the market return, market variance, SII, and TR jump up during the subprime crisis, reflecting some instability in the relation between these predictors and the subsequent market return. The out-of-sample performance of market return, market skewness, average variance, AC, and SII are even negative for relatively long periods of time. For the (value-weighted and equal-weighted) average skewness, the out-of-sample measure increases in a smooth way (approximately, from 1983 to 2006), which reflects the stable relation between this variable and the subsequent market return. For both predictors, the subprime crisis results in a lower out-of-sample performance. Second, in the two-variable case (Fig. 4), adding the market return improves the out-of-sample performance of the two average skewness measures. Market return and value-weighted average skewness are clearly the best pair of predictors, with an almost continuously increasing out-of-sample performance. For the market variance and skewness, the SII, and TR, adding the market return does not improve the out-of-sample performance.