Generally speaking, statistical models are validated by measuring how well/bad they fit the data, for example, studying the residuals and performing statistical tests. Predictive power can be estimated by using the model to predict known outcomes (using out-of-sample data).