Prediction Intervals
Moderators: EViews Gareth, EViews Moderator
Prediction Intervals
Hello
I have a linear regression model. Given forecasts for my variables I have created forecasts of my y variable. On these forecasts obviously I would like a prediction interval. In eviews all I can see atm is getting a confidence interval of +/ 1/.96*se around my forecasts. How can I get a prediction interval (where the interval widens over time). I'm doing static forecasting.
Thank you
I have a linear regression model. Given forecasts for my variables I have created forecasts of my y variable. On these forecasts obviously I would like a prediction interval. In eviews all I can see atm is getting a confidence interval of +/ 1/.96*se around my forecasts. How can I get a prediction interval (where the interval widens over time). I'm doing static forecasting.
Thank you

 Fe ddaethom, fe welon, fe amcangyfrifon
 Posts: 11811
 Joined: Tue Sep 16, 2008 5:38 pm
Re: Prediction Intervals
By default, the forecast graph includes prediction intervals.
Follow us on Twitter @IHSEViews
Re: Prediction Intervals
At the moment when I am forecasting an equation the output graph shows a confidence interval that is labelled +/ 2 SE
and the confidence interval does not widen through time.
What am I doing incorrectly ?
and the confidence interval does not widen through time.
What am I doing incorrectly ?

 Nonnormality and collinearity are NOT problems!
 Posts: 3337
 Joined: Wed Sep 17, 2008 2:25 pm
Re: Prediction Intervals
Why do you think the confidence intervals should widen with time?
Re: Prediction Intervals
This article explains it well
https://www.otexts.org/fpp/2/7
"A common feature of prediction intervals is that they increase in length as the forecast horizon increases. The further ahead we forecast, the more uncertainty is associated with the forecast, and so the prediction intervals grow wider. However, there are some (nonlinear) forecasting methods that do not have this attribute."
https://www.otexts.org/fpp/2/7
"A common feature of prediction intervals is that they increase in length as the forecast horizon increases. The further ahead we forecast, the more uncertainty is associated with the forecast, and so the prediction intervals grow wider. However, there are some (nonlinear) forecasting methods that do not have this attribute."

 Nonnormality and collinearity are NOT problems!
 Posts: 3337
 Joined: Wed Sep 17, 2008 2:25 pm
Re: Prediction Intervals
What the text says is true in some models and not in others. In particular, it is not generally true in models lacking dynamics.
Re: Prediction Intervals
It is however made clear here and in other sources that for 1 step ahead you can estimate the prediction interval using the residual standard deviation but not for multi step.
"To produce a prediction interval, it is necessary to have an estimate of the standard deviation of the forecast distribution. For onestep forecasts for time series, the residual standard deviation provides a good estimate of the forecast standard deviation. But for all other situations, including multistep forecasts for time series, a more complicated method of calculation is required. These calculations are usually done with standard forecasting software and need not trouble the forecaster (unless he or she is writing the software!)."
So my question is does eviews use this more complicated formula . (If so the confidence interval will change over time.......)
"To produce a prediction interval, it is necessary to have an estimate of the standard deviation of the forecast distribution. For onestep forecasts for time series, the residual standard deviation provides a good estimate of the forecast standard deviation. But for all other situations, including multistep forecasts for time series, a more complicated method of calculation is required. These calculations are usually done with standard forecasting software and need not trouble the forecaster (unless he or she is writing the software!)."
So my question is does eviews use this more complicated formula . (If so the confidence interval will change over time.......)

 Nonnormality and collinearity are NOT problems!
 Posts: 3337
 Joined: Wed Sep 17, 2008 2:25 pm
Re: Prediction Intervals
Perhaps you should post the specific model you are using and exactly how you made the forecast.
Re: Prediction Intervals
At the moment I want to focus on the simple multivariate ols regression using forecasts for variables to create a forecast for my dependent variable

 Nonnormality and collinearity are NOT problems!
 Posts: 3337
 Joined: Wed Sep 17, 2008 2:25 pm
Re: Prediction Intervals
Earlier, you asked what you are doing incorrectly. Maybe you are doing something incorrectly, or maybe you are doing everything right. For someone to advise you, you pretty much need to show enough detail so that they can reproduce your steps.
Re: Prediction Intervals
Yes , what am I doing incorrectly such that I am not getting the correct prediction intervals, only +2 SE?
As explained already I am asking about the simple case of forecasting from an ols equation in eviews given future data for your repressors .Not sure how much more detail you want from there that's all you need to replicate my steps .
As explained already I am asking about the simple case of forecasting from an ols equation in eviews given future data for your repressors .Not sure how much more detail you want from there that's all you need to replicate my steps .

 Fe ddaethom, fe welon, fe amcangyfrifon
 Posts: 11811
 Joined: Tue Sep 16, 2008 5:38 pm
Re: Prediction Intervals
Perhaps another way of tackling this would be for you to provide a definition of "prediction interval" and how it is different from the bands you are seeing.
Follow us on Twitter @IHSEViews

 Nonnormality and collinearity are NOT problems!
 Posts: 3337
 Joined: Wed Sep 17, 2008 2:25 pm
Re: Prediction Intervals
mvictor96 wrote:Yes , what am I doing incorrectly such that I am not getting the correct prediction intervals, only +2 SE?
As explained already I am asking about the simple case of forecasting from an ols equation in eviews given future data for your repressors .Not sure how much more detail you want from there that's all you need to replicate my steps .
If you are forecasting from an OLS regression with homoskedastic errors with forecasts based on given future data for the regressors, then + 2 SE gives the correct prediction intervals.
Re: Prediction Intervals
Ok , thanks for clearing this up
Who is online
Users browsing this forum: No registered users and 8 guests