# Without referring to your notes (quant)......

True or false? You can use a DW statistic for testing for serial correlation on a regression equation that uses lagged values. Include a reason in your answer.

false. keep it simple soddy!!! it’s MC after all. errm, there is a reason. ultimately you end up checking the significance of the auto-correlations.

is when DW is significant in both a linear and log-linear model???

Q. A model is of the specification MA(q). How many autocorrelations will be significantly different from zero if the model is correctly specified.

q?

What’s wrong with this? b0 = CovX,Y/Standard Deviation X

q was correct CP.

that would be b1. b1=Covxy/varx b0 = Mean Y Value - Mean X * b1 calculated above.

Question. What is the variance of the prediction error? (no peeking at notes!)

sf^2 = see^2 * {1+1/n + (Xi-xBar)^2/(n-1)sx^2}

CP, do we need to memorize this formula? What is your opinion? I had ignored it, thinking it is too complicated.

actually do not think there is a need. I do not have trouble remembering the formula. Breaking it down into 3 parts 1 1/n (Xi-XBar)^2/(n-1)*Sx^2 helps me keep it in mind. if you referred to searched for Level II posts by mvwt9 -> he had used the Confidence interval method - and selected the one slightly higher… or something like that, instead of memorizing this formula.

Thanks CP and i will search for mvwt9 post.

Nice, well done. I don’t think we need to memorize it either, but I just did. Here’s another one, which I’m sure you’ll get. What is the standard error of the autocorrelations of the residuals? (it’s not a big equation either)

davidyoung@sitkapacific.com Wrote: ------------------------------------------------------- > Nice, well done. I don’t think we need to memorize > it either, but I just did. > > Here’s another one, which I’m sure you’ll get. > > What is the standard error of the autocorrelations > of the residuals? (it’s not a big equation either) Square root of n, where n is the number of lagged variables.

it is 1/sqrt(T) where T = # of observations of the sample. (# of time periods for which the time series regression is being performed).

Nice. Ok, ok, so here’s an easy one. Calculate adjusted R2.

1 - ((n-1/n-k-1) - (1-r^2)) think that’s it

True or false? You can use a DW statistic for testing for serial correlation on a regression equation that uses lagged values. Include a reason in your answer. why is this false? DW is used for serial correlation, isnt it?

You use DW for serial correlation. But the question shows that it is a time series model which by itself mostly has serial correlation. You will need to check the significance of auto correlations to determine whether it is a problem or not.