The RATS Software Forum

Respected Sir
Can I forecast Y (the observed variable) using the FORECAST option for FAVAR model. If not then what would be the way out. Please guide me

Regards
Deepika

Yes. The Y's and factors make a self-contained model that can be forecast using standard methods. However, again, that's not what this is designed to do. The Y's have been standardized (mean zero, variance one) and so aren't in their natural levels.

respected sir

to test the forecasting ability of my favar model, i was trying to run procedure runtheil. the codes which i wrote were as follows:
theil(model=favar,setup) 1 7 2014:12
estimate(outsigma=v,residuals=resids4) * 2012:12
theil(print)
do time=2013:01,2016:03
kalman
theil(print)
end do time
theil(dump,window="olsvar")
however this gives me an error saying that favar is unidentifiable. could you please help me in this?

Regards
Deepika

I can tell you how it will forecast. Badly. It's not designed to forecast. It's a model set up to determine the "lower bound" on how monetary policy affects other variables in the economy. The factors are chosen to minimize the effect of monetary policy, not to forecast it.

Respected Sir

My objective is to compare the forecasting performance of VAR, BVAR and FAVAR models. So i have to show that econometrically. Can you please help me in checking my theil code in the previous post. I shall be highly obliged

Regards
Deepika

First of all, unrestricted VAR's don't forecast well. That's been known for over 30 years. The BBE FAVAR isn't designed to create forecasts and isn't the slightest bit designed to produce rolling sample forecasts. If you want to compare forecasting performance, you need to do some research and come up with methods which can actually be used for what you want.

Respected Sir

i am sorry for asking too many questions. actually i am looking to replicate "Forecasting Indian Macroeconomic Variables Using Medium-Scale VAR Models", Goodness C . Aye , Pami Dua , and Rangan Gupta, Current Trends in Bayesian Methodology with Applications, Edited by Satyanshu K. Upadhyay, Umesh Singh, Dipak K. Dey, and Appaia Loganathan, Chapman and Hall/CRC 2015, Pages 37–57. in this paper they compute one to twelve month ahead relative RMSE's for different variables of interest. In eviews this entire exrecise has to be done manually which is very time consuming. Since i have computed the same for VAR and BVAR models in RATS, i intended to do the same for FAVAR also but somehow i get an error. i could understand that model=favar is what the program is not identifying. do you think i should use the FORECAST option or obtain relative RMSE manually. Could you please provide a hint. i am quite confused.

Regards
Deepika

It would have been helpful to know what the "end game" was when you first asked, since the BBE paper is a dead end for that. The VAR and the BVAR are linear models---even if you didn't use the Kalman filter, it still takes only a fraction of a second to re-estimate the model at each time period in a rolling sample. Even if the BBE model were designed for forecasting (which it isn't), it's a complicated non-linear model which takes quite a while to estimate. You can't simply take the full sample estimates of the factors and use them for forecasting experiments since the factors would include information about the future values.

Respected Sir

So does it imply that the only option available is to do it manually even in RATS

Regards
Deepika

Also does it imply that the two step approach would be more useful in generating forecasts.

Regards
Deepika

Yes. If you want to do rolling samples, you have to re-estimate the model manually.

The two-step approach would certainly be quicker. Whether it would be better for forecasting is unclear since the model isn't designed to forecast. (Note again that in order to apply PC, it standardizes all the variables, including the ones that you want to forecast).

Respected Sir

Please three more issues but regarding the RATS favar program
1) why does the program only transforms the key variables and not all? Does it imply that we don't have to bother about stationarity of other variables or we have to transform other variables ourselves and let RATS do transformation of key variables, but why?
2) why is the program not providing impulse response of y variable that is the own shock. It could also have some meaningful implication
3) if i were to extract factors first and then run VAR then do i make use of PRINFACTOR procedure?

Regards
Deepika

Deepika wrote:Respected Sir

Please three more issues but regarding the RATS favar program
1) why does the program only transforms the key variables and not all? Does it imply that we don't have to bother about stationarity of other variables or we have to transform other variables ourselves and let RATS do transformation of key variables, but why?

Everything has been transformed already. The key variables are just the ones for which it produces special output.

Deepika wrote: 2) why is the program not providing impulse response of y variable that is the own shock. It could also have some meaningful implication

Because that's not the point of the paper.

Deepika wrote: 3) if i were to extract factors first and then run VAR then do i make use of PRINFACTOR procedure?

No. @PRINFACTORS is to create the factors. If you've done them some other way, then you don't use it.

Respected Sir

In continuation with my above mails, i was trying to find codes for Structural Factor-Augmented VARs (SFAVARs) and the Effects of Monetary Policy
Francesco Belviso / Fabio Milani, BE Journal Of Macroeconomics, Issue3, 2006. I think it helps to overcome the problem of identification of factors in the BBE model

Regards
Deepika

Respected sir

i am trying to use FAVAR technique based on "MEASURING THE EFFECTS OF MONETARY POLICY: A FACTOR-AUGMENTED VECTOR AUTOREGRESSIVE (FAVAR) APPROACH", BEN S. BERNANKE & BOIVIN (2005).

The author , in the paper , suggest that "We standardize the monetary shock to correspond to a 25-basis-point innovation in the federal funds rate"

Who i can find this assumption in the programm related in this paper ????