I’m going over notes one final time and trying to do some memorization work now. Plan to sacrifice stats and am using all free time to cover remaining 9 topics…
Stats is straight forward. Problem is in retaining all of it.
Yea, I guess the point is that sacrificing 1 section (the one I am weakest on) in order to spend my remaining time on others is probably better…
… just before you let it go…here are some quick points to earn. 1) Look at the autocorrelations in the bottom of the table, if any has a t-stat > about 2, you know there is serial correlation. Ignore the model. 2) Look at b1. If it is b1 =1, you know you have a unit root. Time series is a random walk. 3) If it is a random walk, you cannot model it as AR(1), unless you turn it into a non random walk series. Stock prices rises over time, so they are not covariance stationary, so don’t model them as AR(1). Turn them into covariance stationary by looking at the *change* of stock price today from yesterday. That will always be covariance stationary, because the changes don’t just increase or decrease forever. 4) If they give you a problem with a *dependent* variable which is 0 or 1, you cannot use normal linear regression. You should use logit or probit. Just know that you select that as ananswer. 5) Look at the autocorrelations in the bottom of the table. If you see one with t_stat > about 2, you know there is a seasonal lag. Don’t trust the model. Get rid of the lag, by making it a coefficient in the equation, i.e., add another independent variable for that lag. Then check again autocorrelations in the bottom of the table to see no t_stat > 2. Now, you have adjusted for the seasonal lag. 6) Know quickly the mean level of the equation, by simply computing i as b0/(1-b1). That’s why if b1 =1, your equation (your series) is not covariance stationary. 7) If your equation says ln Sales, etc. Make sure at the end you compute the right sales amount, by doing e^(ln Sales). That should get you one or two points quickly.
dreary that was awesome. seriously, for some reason, that was totally awesome. like they part “Turn them into covariance stationary by looking at the *change* of stock price today from yesterday. That will always be covariance stationary, because the changes don’t just increase or decrease forever.” like DUH! i knew to first difference, but just accepted it since I thought it would let you test via dickey-fuller but hey, thats so simple and OF COURSE thats why we first difference! cause rates of change cant increase or decrease forever!
And if there’s a graph, look for variations above and below the Standard deviation. If it’s evenly distrubuted, then it’s covariance stationary. Otherwise, it’s not…