We show that  Kolmogorov complexity and such its estimators as
universal codes (or data compression methods) can be applied for
hypothesis testing in a framework of classical mathematical
statistics. The methods for identity testing and nonparametric
testing of serial independence for time series are described.