<p>We develop statistical tools for comparing two predictors in a time series setting. We define test statistics for testing superior predictive performance and derive asymptotic distributions of the test statistics. The test statistics are accompanied with graphical tools: an image of <i>p</i>-values and a graph colored with <i>p</i>-values. The graphical tools make it possible to analyze the performance of predictors: we get insight into how the performance varies in different time periods and in different regimes of various explanatory variables. We show that a kernel regression predictor performs better in some cases than typical GARCH-predictors.</p>