EViews.com

I was wondering if someone can assist me with a VAR I am using to forecast interest rates. I am using 12 variables (lagged four periods) to forecast the yield on non-financial commercial paper rates. The data is monthly and runs back to January 2003. To create my forecast I insert the 12 explanatory variables and my commercial paper rate as endogenous variables. (I am not sure if this is right though, shouldn't the explanatory variables be exogenous? And if so and I want to include four lags, do I have to type in each coefficent in the Exogenous box?). The R-square is .9999 ("var4" in the attachment) and it seems like the model should work well. I solve the model and use a static solution. But the forecast makes no sense at all. If you open "group1" it shows the actual yield and the forecast. The model shows the rate going to -1.67% two peridos forward, which is actually impossible. It then shows the rate shooting up to 2.36%. Am I not controlling for something correctly in my model or have I set it up wrong? Is part of the issue I don't have any exogenous variables? I am guessing I am missing some type of setting but can't seem to figure it out. Any suggestions anyone has would be greatly appreciated.

You have 72 observations. You also have 14 variables, each with 5 lags, plus an intercept. That's 71 variables for 72 observations. Basically, you're just getting random noise.

I see. It fits alot better when I use only two or three variables. (But still goes negative). Is there a rule of thumb for the number of coefficents vs. observations?

If you plot the variable of interest, you'll see that you're essentially projecting out trends. It's not surprising that they go negative.