EGARCH

[puf]

Hi,
I would like to do a rolling window forecast with EGARCH model. The following code:

Code: Select all

wfopen 28.csv
series logr = dlog(abs(prc)/cfacpr)
series true_vol = logr^2
series forecasted_garch4
scalar estimation_window = 400
scalar length = @obsrange
for !i=estimation_window to length-2
   smpl @first+!i-estimation_window @first+!i
   equation kamo4.ARCH(egarch) logr
   smpl @first+!i+1 @first+!i+1 
   kamo4.forecast r se var
   forecasted_garch4(!i+2)=var(!i+2)
next
smpl @first+estimation_window+1 @last
scalar num_obs = @obssmpl
scalar RMSE= @sqrt((@sumsq(forecasted_garch4-true_vol))/num_obs)

works for most of the stocks (27 out of 30), but for three of them it gives the following error message:

Overflow in "do_kamo4.forecast r se var"

could you please fix this problem?

[EViews Gareth]

I'm not sure this is a bug. The estimation on that particular stock failed - it couldn't converge. Therefore it couldn't forecast particularly well and led to an overflow.

[puf]

Hi,
I do not think that you are right. The error message says that what failed was just forecasting, not estimation itself. And indeed, when you delete from the code the lines for forecasting, the code works, there is no error message.

please tell me what the problem is (and eventually fix it)

[EViews Gareth]

The problem is that your estimation did not converge. Therefore you cannot forecast from it. When an estimation does not converge, EViews does not produce an error, rather it just prints out in the estimation output that it didn't converge.