Use to EViews to Determine Optimal Lag for Variables & Trend

[kjd31]

1. Is it possible to select and independent variable and then have EViews determine the optimal lag each dependent variable should use? Like determining that for Variable A a 3-month lag correlates much higher than a no-lag does? If so, how?

2. Separately, is it possible to upload decades of historical data and let EViews discover hidden trends and then build its own forecasting model based on the data? I am thinking along the lines of some big data programs like SPSS Modeler by IBM and just wanted to know what functionalities like this Eviews has.

[EViews Gareth]

You can use Stepwise regression to choose independent variables for you. If you enter a list of lagged variables as the selection variables, it will choose which lags are best.

For forecasting, the automatic exponential smoothing models built into EViews 8 can do auto-univariate forecasting for you.

[kjd31]

Great call. But for some variables the regression is using both variable(-1) and variable(-2). Is there anyway to avoid that and have it use only 1 version of the variable - be it lagged 1 month vs. 2 month vs. no lag? I want it to pick the ONE that correlates highest.

Thanks for such quick responses to all my q's.

[kjd31]

It's also spitting out a regression where the forecast is literally identical to the actual values and the R^2=1.0. It can't be that good of a regression... why is it doing this? Thanks again

[EViews Gareth]

No, you cannot have it choose only one lag of each regressor. If you only have a single regressor then you could, simply by telling it to choose one variable only.

Hard to tell what is causing your forecast to be like that. At a guess you included the dependent variable as one of the regressors. That would give an R^2 equal to 1.0 and would give a darn good forecast!

[kjd31]

"For forecasting, the automatic exponential smoothing models built into EViews 8 can do auto-univariate forecasting for you."

How do I set up an "auto-univariate" forecasting? Just run a dynamic model with the dependent variable as an explanatory variable?

[EViews Gareth]

http://www.eviews.com/EViews8/ev8ecets_n.html