Hi,
I'm trying to develop a program that can produce forecast confidence intervals for model averages, using bootstrapped residual series to generate data for estimation. My method is currently:
1. Estimate the models on the original data Y and extract the residuals and fitted values.
2. Resample the residuals n times (e.g. 1000 times) and generate n new Y* series by adding the resampled residuals to the fitted values.
3. Re-estimate the models n times on Y*.
4. In each n re-estimation, generate the forecast ahead using the original Y series but with the newly estimated coefficients.
5. Save the forecasts and repeat to build up a distribution of forecasts.
My problem is in step 4. I've managed to estimate the models on the generated data but I can't figure out a way to then apply those coefficients to a different data set (the actual/original data). Forecasting with the generated series is incorrect as each repetition's forecasts will originate at a different starting point and the distribution will be much wider than what it should be.
I thought there might be a way to code this using a 'model' with different scenarios but it seems that scenarios can only apply to exogenous variables. Is there a way to apply an estimated model to different data for forecasting? Hoping there is a much simpler routine built in than needing to save the coefficients and manually specifying the equations which can get messy with different model classes (E.g. MIDAS, VAR, etc).
Sem-parametric Bootstrapping - Repeated Model Simulation
Moderators: EViews Gareth, EViews Moderator, EViews Jason, EViews Matt
-
- Fe ddaethom, fe welon, fe amcangyfrifon
- Posts: 13401
- Joined: Tue Sep 16, 2008 5:38 pm
Re: Sem-parametric Bootstrapping - Repeated Model Simulation
If you've estimated an equation, and then forecast from that equation, the forecast will use whatever data is currently in the series. So if you where to do something like:
Not sure if that helps or not.
- Generate Y and X
- Estimate equation
- Change data in X
- Forecast from equation
Not sure if that helps or not.
Follow us on Twitter @IHSEViews
Who is online
Users browsing this forum: No registered users and 2 guests