Hi allI'm trying to estimate the forward libor instantaneous correlations in a BGM model from time series.The method I use is the following:1) I download from my provider the time series of market instruments (depo, futures, swaps)2) For each day in the past I bootstrap the discount curve and calculate the desired forward libor curve3) I consider the series of daily log returns, i.e. log( F_k(t+1) / F_k(t) ) and calculate the corresponding correlation matrix, which should correspond to the desired instantaneous correlation I am looking for.However -unlike terminal correlation- the instantaneous correlation computed in this way seems to be pretty unstable, and pretty much depending on the bootstrap type and on the market instruments used to do it. In particular, if I use deposit rates the estimate seems to be quite unrealistic (meaning that "adjacent" forwards have very low correlation).Does anyone have any suggestion on how to improve the estimation procedure?Thanks in advance. Regards.Baoh

This stuff with log of returns is for terminal correlation.Isn't there an approximate direct relationship between instantaneous and terminal correlations?If yes, why not to use it?