Forecasting with ANN

[ateeb]

Hello,

I am forecasting 12 steps ahead with an AR model with rolling window using the following code:

"clear rhatar rhatar2 rhatar3 rhatar4 rhatar5 rhatar6 rhatar7 rhatar8 rhatar9 rhatar10 rhatar11 rhatar12
do regend=2018:2,2019:2
boxjenk(ar=||1,2||,define=ar2) Dsalsn regend-60 regend
uforecast(static,equation=ar2) rhatar regend+1 regend+1
uforecast(static,equation=ar2) rhatar1 regend+2 regend+12
compute rhatar2(regend+2)=rhatar1(regend+2)
compute rhatar3(regend+3)=rhatar1(regend+3)
compute rhatar4(regend+4)=rhatar1(regend+4)
compute rhatar5(regend+5)=rhatar1(regend+5)
compute rhatar6(regend+6)=rhatar1(regend+6)
compute rhatar7(regend+7)=rhatar1(regend+7)
compute rhatar8(regend+8)=rhatar1(regend+8)
compute rhatar9(regend+9)=rhatar1(regend+9)
compute rhatar10(regend+10)=rhatar1(regend+10)
compute rhatar11(regend+11)=rhatar1(regend+11)
compute rhatar12(regend+12)=rhatar1(regend+12)
end do"

It works perfectly fine. I want to know if there is a way to write it in a parsimonious way?

Second, what would be a way to write such code for a neural network with 2 lags of the same variable itself. I am asking this because I am unable to define model as above (with the AR) for the neural network so that I can use the uforecast function.

Thank you in advance

Dr. Ateeb Syed

[TomDoan]

You can create a VECT[SERIES] for the different horizon forecasts.

dec vect[series] rhatarf(12)
clear rhatarf
do regend...
...
do h=1,12
compute rhatarf(h)(regend+h)=rhatar1(regend+h)
end do h
end do regend

That looks like it does what you want. However, are you sure you want to do multi-step static forecasts? The 12 step static "forecast" treats 11 step out data as known which doesn't seem to make much sense.