I was playing around with the nonparametric entropy pooling approach
from Meucci to integrate external views (EntropyProg() in
While in theory the approach looks very appealing, I realized that in
practice the entropy pooling algorithm often does not converge due to
the fact that the views are "incompatible" with the scenarios that I
generated from bootstrapping historical data (no feasible solution).
To get solutions I either had to shift the views towards the data, or
the other way round, but both approaches do not really make sense in
this context as the prior should be independent of the views.
Anybody out there who had similar experiences and wants to share his thoughts?