El 01/05/11 23:07, Anutechia Asongu escribió:
> Thanks Micheal,
> In addition to the concern over what optimal lags to choose. When
> observations are less than 60, the AIC and FPE best to optimise lags
> that fit the model to the data structure. However, when observations are
> above 60, the HQC is best. The SIC or BIC underestimate lags with
> increase in degrees of freedom( See: Khim(2004):
> http://ideas.repec.org/a/ebl/ecbull/v3y2004i33p1-9.html)

I am not sure this last sentence is correct (although I see it is 
correctly interpreted from the work of Khim). In fact HQC and BIC are 
criteria that obtain a consistent estimator for p (the order of the 
VAR). So, when T (or T-k) grows the probability of underestimate p 
should decrease.

With respect to this question it is very detailed the explanation given 
in chapter 4 of "New Introduction to Multiple time Series" by Helmut 
Lütkepohl (Springer, 2005). Let me copy here some conclusions that may 
be appropiate given the debate that has emerged in this list: (page 151)

"... It is worth emphasizing that the foregoing results do nor 
necessarily mean that AIC and FPE are inferior to HQ and SC [BIC]. Only 
if consistency is the yardstick for evaluating the criteria, the latter 
two are superior under conditions of the previous section. So far we 
have not considered the small sample properties of the estimators. In 
small samples, AIC and FPE may have better properties (chose the correct 
order more often) than HQ and SC. Also, the former two criteria are 
designed for minimizing the forecast error variance. Thus, in small as 
well as large samples, models based on AIC and FPE  may produce superior 
forecasts although they may not estimate the orders correctly. In fact, 
Shibata(1980) derived asymptotic optimality properties of AIC and FPE 
for univariate processes. He showed that, under suitable conditions, 
they indeed minimize the 1-step ahead forecast MSE asymptotically".


-- 
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.ea3.ehu.es




Reply via email to