#230 Scheweser Notes latest version (for 2010): For auto correlation (AC), notes said the standard error is 1/root T where T = no. of observation In first example given on page 230 , there’s sample of 102 observation on a AR1 model, each of the t-statistics with calculated using T = 102. (one intercept, one lagged independent var) However, second example on page 231 (using a log AR1 model) with 40 observations, the standard error was calculated using T = 39. (one intercept, one lagged independent var) Additionally, in the third example (linked to example 2) to correct for seasonality where an additional lag of dependent variable is added to original model as a independent variable, T = 38. Why was the second example T-1 in calculating the standard error and T without adjustment was used in first example? Second question is, if I correct for one seasonality, do I remove one observation? If so, why?
An AR1 model regresses the variable on the change from one period to the next. If you have T obersvations over time then if you convert them to new observations recognizing the change from period to period, it would be T-1 observations. If you already have them in AR1 format, it would be T. Thats what I recall.