Normality - Transformation/Box Cox won't work

#1
Hi Guys,

I hope you can help me. I'm looking at default risk change of companies who have/don't have Investment Covenants & issued Corporate Bonds. I'm using the Black-Scholes-Merton Model to calculate default risk.

My Problem being, the data isn't normally distributed. I can't do parametric test. In addition the standard errors of my regression (linear) are not normally distributed which leave me with Coefficients that can't be interpreted. I'm currently trying to get the depend variable transformed into a normally distributed variable. I tried windsorizing with EXP, SQRT and LN Transformation (with and without Reflection) Link Transformation + a Box Cox Transformation in minitab Link Box Cox.

I'm attaching an Excel file with the data: Data
Colum A has the Default Risk Change Column B the Dummy Variable for Issuers who had Investment Covenants
I'm hoping that someone can either confirm that a transformation is not possible and that I have to use non parametric tests or give me a pointer as to what I'm doing wrong.

Best regards

Nokita
 

hlsmith

Not a robit
#2
Can you post a q-q plot of the untransformed residuals and then the Box-Cox transformed variable's residuals? Also, is this just univariate or if there is a dependent and independent variable, which did you transform?