I've got a negative correlation between forward rates differences (calculated from libor fixings) using 10 yr history.ex, corr(diff(forward(30,60)),diff(forward(60,90))) =-0.32 is it possible? thanks

I want to be sure I understand what you did.You had 10 years of data (what frequency?) of 30, 60 and 90 day LIBOR rates. You computed the 30-60 day forward rate, and the 60-90 day forward rate; then you took the differences of each of these series.If this is correcct, the likely explanation for the negative correlation is noise. If there is any noise in the 60-day rate, it will have opposite effects on the 30-60 day forward rate and the 60-90 day forward rate. This will induce apparent negative correlation.Underlying economics says that the 30-60 day and 60-90 day forward rate will have strong positive correlation. But If the noise is large compared to the standard deviation of interest rates, it could make the overall correlation negative. This is much more likely with daily data than longer-term, because it will have smaller standard deviation of actual moves with roughly the same potential for noise.I also suggest looking for some outliers. A single 60 day yield with a massive error, say 3.09 instead of 0.0309, could cause a negative correlation for the entire ten years.

thanks Aaron.i checked for outliers, and tried to remove them, but it cause only mor negative correlation (probably it was not outliers at all, but real jumps in interest rates).can you advise easy methods to smooth data before calculate correlation?thanks

thanks Aaron.i checked for outliers, and tried to remove them, but it cause only mor negative correlation (probably it was not outliers at all, but real jumps in interest rates).can you advise easy methods to smooth data before calculate correlation?thanks

It depends on what you're trying to do, but using a longer time interval is a general-purpose technique. You can use overlapping ones if you want.