i applied an ARDL model to assess the impact of macroeconomic factors on default rate. How can i validate the model other than dividing the data between training and testing? (Because my data is already small)
Which are serial correlation, functional form, normality and heteroscedasticity? I tested them.
What I mean by validation is for example we take 75% of the observation and we extract the model from them. Then we test the model on the 25% to see if the model estimated is performing well. But my data is very small so I can't divide it. There is any other method for validation ?
what i understand from your question is that you want to test for the sensitivity analysis of your results to different sampling sizes. you could either incorporate a structural break test like chow test in the analysis to see the date of break where from you can divide your sample and run the ARDL separately for two data periods as decided by test. or you can just use other estimation methods like FMOLS, DOLS, CCR etc to ensure the robusteness of your ARDL results.