0
$\begingroup$

I have performed a logistic regression to estimate the default probability of a dataset of firms based on some basic balance-sheet ratios. I have winsorized all the ratios at the 1st and 99th percentile.

My question is: now that I want to use my estimated model on a new dataset for forecasting purposes, should I also winsorize this dataset as I did for the calibration dataset?

For example: if firm X in the forcasting dataset has a turnover of 1.1 billion but the 99% percentile in the forcasting dataset is 1 billion, should I change the turnover value for firm X to 1billion.

Thanks a lot.

$\endgroup$
1
$\begingroup$

Leaving aside whether the original winsorizing is a good idea (I suspect not), the answer is yes. The estimated coefficients are for winsorized variables and would have to be different to work with I transformed data.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.