Julia Cains
2010-Feb-15 06:31 UTC
[R] Exponential Fitting to Credit default data - A theoretical query
Dear R helpers I am working on the credit risk default data and am referring to "An introduction to Credit Risk Modelling" by Christian Bluhm. The literature affirms that 'the default frequencies grow exponentially with decreasing credit worthiness'. I have a data of rating wise default frequencies. The idea is to fit exponentail of the type Y = a * exp( b*X ) where Y is (dependent variable) default frequency and X is rating class code. So e.g. suppose I have rating classes AAA, AA, A.... and so on coded as 1, 2, 3 etc. and suppose the observed default frequencies are say 0.00011, 0.0029, 0.0083 respectively, then my values of X and Y are as follows. X Y 1 0.0001 2 0.0029 3 0.0083 .............................. .............................. etc and to this data I am fitting Y = a * exp(b*X) My quereis are (1) is there any R function which will find out the estimated values of a and b instead of taking logaritham and converting both sides into linear equation and solve for a and b; (2) How do I find out out Statistically that this is the best fit? e.g. had it been linear regression, the R^2 gives me some idea about the fitted linear line. In otehr words, is there any way of comparing the observed values of Y against the estimated values of Y i.e. values obtained from the equation Y = a * exp(b*X). Will the Chi-Square will be a good choice? I was trying the t-test as given below and am not sure whether I am right to do so. For each rating class, I have the observed value of Y and estimates value. Say for AAA, the observed value is 0.0001 whereas the estimated value is say 0.0017. I have tested the following hypothesis Ho : Y(AAA) = 0.0017 H1 : Y(AAA) <not equal> 0.0017 So this is a two tailed case and I have applied the 't' test as tcal (AAA) = [Y(obse) - Y(estim)]/(s / sqrt(n)). Probelm is in some cases, I have observed significant difference ie. the t(calculated value) falling in Critical region. Thus, the need to find out whetehr my fit is correct or not. Thanking you in advance Regards Julia Cains, Brisbane ****************************************************** Only a man of Worth sees Worth in other men ****************************************************** [[alternative HTML version deleted]]