Pool forecasting

[beckistuta]

Dear all

I am trying to forecast t+1 and t+3 ROE of 47 firms using pooled time-series cross-sectional regression. However eView fails to product the forecast value. Would you mind to shed some lights on any step I did wrong? Attach please find my eView workfile for your reference.

Basically I first pool the cross-sectional identifiers into pool "ni" and "bv". then I calculate the ROE of each firm by roe? = ni? / bv? (You may notice that a number of firms have missing ni and bv in some years. This is deliberately arranged.)

Then I pool the calculated ROE of all firms into pool "ROE_2010".

Afterward I estimate the pooled time-series cross-sectional regression for all the firms in workfile by: ROE (i,t) = alpha + beta* ROE (i,t-1) + error (i,t). Set the sample solution as 1981 2008. Notice that I suppose today is 2010 and I would like to forecast the ROE from 2011 to 2013.

As eView guide suggested, I make the model in order to forecast the future ROE. So I click Proc --> Make a model. And then you will notice there are 47 endogenous variables in Eq1. (See Model 1).

Then I click "Solve". Set the simulation as deterministic and dynamics solution. I set the solution sample as "2011 2013". Then after I click OK, the error message pop up "ROE_Bethlehem_Steel_Dead" already exists. I randomly check those series with _0 as suffix and then I found that eView didn't run the forecasts at all.

So that's the problem. Deeply appreciate and welcome your comments. Thanks in advance.

[EViews Gareth]

Unfortunately EViews 6 just gives that somewhat cryptic error message. EViews 7 gives a better error message, stating that some of your variable names are too long for the solve to work. Series names should be no longer than 20 (ish) characters, and some of yours are a bit longer. If you rename those series to be shorter, then re-specify the pool, estimate the pool, then make the model, it should work.

[beckistuta]

Hi Gareth

Many thanks for your reply. I have shortened the name as you suggested. I ran the forecasting under the soultion smaple "2011 2013" again but I still encountered error. Please refer to below error message for your reference.

Model: Untitled
Date: 04/04/11 Time: 18:18
Sample: 2011 2013
Solve Options:
Dynamic-Deterministic Simulation
Solver: Broyden
Max iterations = 5000, Convergence = 1e-08

Parsing Analytic Jacobian:
0 derivatives kept, 0 derivatives discarded

Scenario: Baseline
Solve begin 18:18:47
2011 NA generated for ROE_INCO
Solve terminated - Unable to compute due to missing data in "ROE_INCO = @COEF(1) + @COEF(2) * ROE_INCO(-1)"

Besides forecast values in 2011 were recorded in the "_0" series. but the values are the same as those in 2010....

I have tried restrain my sample by "smpl 2010 2013" but same error message encountered. May I have your valuable comments again?

Many thanks

[EViews Gareth]

Could you post the shortened version?

[beckistuta]

Dear Gareth, here you go. thanks!

[EViews Gareth]

The problem is that for a bunch of your cross-sections, data is not available for 2010. Thus it cannot compute a forecast for 2011.

[beckistuta]

yes I know many of the cross-section does not have 2010 value. the reason is either I have to elminate those -ve book value, or they are not listed on Dow in respective time period. I have to forecast those listed on Dow in 2010 (which means those having actual data in 2010). therefore may you advise how I can do that?

[EViews Gareth]

It is going to be complicated. You might want to switch to a panel workfile rather than using pools. Then you can just use an equation object and use the equation's built-in forecasting tools, rather than having to deal with models.

viewtopic.php?f=7&t=74

viewtopic.php?f=7&t=3784

[beckistuta]

many thanks gareth!