LR unconditional coverage

[Dox Wang]

hi, does any know about the concept of LR,
LR=-2ln(1-p)^T-N*p^N) + ...... it said H0: accurate model H1: inaccurate model
we reject H0 if LR>3.84, so which means if LR<3.84, we can accept the Exception Number of the back test. the Number could be 1 2 3 4 5 6 7, but 8 which made LR>3.84 we reject.
so my confusion is that when we dealing with Market risk VaR, it said:
Green zone:0-4
Yellow zone:5-9
Red zone:10
so at this point, how can i understand LR if exception Number is 5, we accept model, but it's in Yellow zone, K is > 3.
thanks

 

[ShaktiRathore]

The red zone is the zone where the model gets rejected. If the no of exceptions are >=8 than the LR ration goes>3.84 which led us to conclude that the model is not significant and thus reject it at the given said level of confidence. So we are x% confident that the model does not does well in explaining the results. If exception is 5 than we do not reject the model on the face of it since LR<3.84 so that we accept the model backed by our back-testing. Its in the yellow zone so that is a chance that model is not that accurate. The zones here are referring to the number of exceptions that can turn out in the model if the no of exceptions are up to 4 that there the model is correct as is also supported by the LR value<3.84. For the no of exceptions 5--8 the model is also correct by LR ratio. The yellow zone indicates that chances of model rejection are higher than green zone so be cautious before using the model but red zone bans the model altogether. So yellow zone accommodates 5-8 exceptions up-to which we can accept the model. But beyond that the skepticism increases that model is incorrect with the increase in the number of exceptions as supported by LR ratio >3,84 thus for N=8,9 the model is invalidated bu there is degree of skepticism as to whether the model is actually invalid (which is partially verified that model is indeed invalid). However as the number of exceptions increases the red zone indicator together with our LR conclusion concludes that model is in fact invalid and not proper to use.

thanks