2008 Schweser Study Note Bk 2, Page 227 “If the t-tests indicate that any of the correlations computed are statistically significant, the AR model is not specified correctly. Lags corresponding to the significant correlations are included in the model and the correlations of the residuals are checked again. This procedure will be followed until all autoregressions are insignificant.” Does this mean we now use AR(2) model and test for the significance of autoregression again. And if autoregression is still significant, we use AR(3) model and so on? Thanks.
Yeah sleepy, that’s my understanding of the procedure. Pages 404-405 of 2008 CFAI Volume I have a nice summary of the suggested steps in time-series forecasting. The part you’re asking about is at the top of p.405 (steps 5 and 6).
Yes… we continue to add lags to the model (AR(2) … AR(3) … AR(4) … AR(n)) and re-test for the significance of the t-statistics of each lag… And continue this processes till all the autocorrelations are insignificant, upon which the Model is said to be ‘correctly specified’ and ‘usable’.