I was trying to recover some cross-correlations between two series and I noticed that the command "xcorrgm" and the genr function "corr" do not give exactly the same estimations. I suppose they are using different formulas so I tried to look for the differences in the source (describe.c file) but I really do not find any difference, do you? I think they should give the same result so if you run scalar xy3=corr(y,x(-3)) you should obtain the same result as the xcorrgm gives for lag 3. Now I see that the same occurs with the correlogram. For example scalar x3=corr(x,x(-3)) is not exactly the same as the third coefficient that gives corrgm d_Ct <script> open data3-6 diff Ct Yt xcorrgm d_Ct d_Yt scalar c0=corr(d_Ct,d_Yt) ;# this is the only one that coincides scalar cm1=corr(d_Ct(-1),d_Yt) scalar c1=corr(d_Ct,d_Yt(-1)) scalar c2=corr(d_Ct,d_Yt(-2)) corrgm d_Ct scalar cx1=corr(d_Ct,d_Ct(-1)) scalar cx2=corr(d_Ct,d_Ct(-2)) </script> -- Ignacio Diaz-Emparanza DEPARTAMENTO DE ECONOMÍA APLICADA III (ECONOMETRÍA Y ESTADÍSTICA) UPV/EHU Avda. Lehendakari Aguirre, 83 | 48015 BILBAO T.: +34 946013732 | F.: +34 946013754 www.et.bs.ehu.es

On Mon, 3 Mar 2008, andreas.rosenblad(a)ltv.se wrote: You can't. This is not documented yet because I'm not yet sure that it's useful, or should be retained. It can't fall within the usual "smpl" framework, because it may create a superset rather than a subset of the original data, depending on the value of n. I tend to think that this is better handled via the resample() function as applied to matrices. Allin.