Raj Sachdeva
New Member
Hi David, as part of the back testing how would we calculate confidence limits for PD? Should the proportion method he used or the continuous variable method?
Regards
Raj
Regards
Raj
Confidence Limits for PD
- Thread starter Raj Sachdeva
- Start date
Hi @Raj Sachdeva I tend to think of backtest as distinct from confidence intervals, although they have inference in common. I think there are pros and cons, I'm sorry I am not familiar with your terms by I *assume* proportion refers to binomial/analytical and continuous refers to parametric. In the FRM, at least, the most natural approach would be analogous to VaR backtest; i.e., as both are treated like a binomial distribution (and making the dreadful i.i.d. assumption).
So, if we are okay to assume (a series of) defaults can be characterized a binomial, then the standard error elegantly reduces to SQRT[p*(1-p)/n]
see discussion here https://forum.bionicturtle.com/thre...stic-and-confidence-interval.5318/#post-22826
as n increases, which in practice it would, we could follow Jorion's (Chapter 6) just assume this binomial is approximated by a normal (binomial --> normal, nice!), such that the confidence interval is really elegant: +/- (z deviate)* SQRT[p*(1-p)/n]. Thanks,
So, if we are okay to assume (a series of) defaults can be characterized a binomial, then the standard error elegantly reduces to SQRT[p*(1-p)/n]
see discussion here https://forum.bionicturtle.com/thre...stic-and-confidence-interval.5318/#post-22826
as n increases, which in practice it would, we could follow Jorion's (Chapter 6) just assume this binomial is approximated by a normal (binomial --> normal, nice!), such that the confidence interval is really elegant: +/- (z deviate)* SQRT[p*(1-p)/n]. Thanks,
Similar threads
