Testing for predictability in conditionally heteroskedastic stock returns

The difficulty of predicting stock returns has recently motivated researchers to start looking for more powerful tests, and the current paper takes a step in this direction. Unlike existing tests, the test proposed here exploits the information contained in the heteroskedasticity of returns, which is expected to lead to higher power, a result that is confirmed by our results. In order to also maintain good size accuracy, subsample critical values are used.