Predict function VARX

[alexsuarez94]

I have a penalized VARX model after applying cv.BigVAR() function and I want to perform a 10 -step forecast but the function predict returns the following error: "Error in 2:nrow(Z) : argument of length 0".

I have, 30 endogenous variables and 15 exogenous variables so I am specifying:
predict(Model_results, n.ahead = 10, newxreg = exogenous_variables[1:10, 1:15])
There is no problem when I launch forecast with n.ahead = 1.

Any suggestion of what could be the problem?

Thank you so much!

[wbnicholson]

Could you post the specifications that you passed to cv.BigVAR?

[alexsuarez94]

Could you post the specifications that you passed to cv.BigVAR?

Model <- constructModel(Y = as.matrix(train), p=1, struct="Basic", gran=c(50, 10), verbose=TRUE, RVAR = FALSE, h = 1, MN = TRUE, VARX=list(k = 30, s = 2), intercept = TRUE, C = mn_prior, recursive = FALSE)
cv.BigVAR(Model)

where Y is a 45x144 matrix, and mn_prior is a vector of 30 values [0, 1]. The rest of the arguments are in their default values.

Thank you so much!

[wbnicholson]

It appears that there was an issue with the Minnesota Prior component of the multi-step forecast. I just pushed an update that should fix it.

[kaijennissen]

How do I create forecasts for a VARX?
When I try to predict a VARX I get the following error message:

Error in if (contemp) { : argument is of length zero

The model was created using the following code:

mod1 <- constructModel(Y = Y, p = 12, struct = "Basic", gran = c(50, 10), verbose = TRUE, h = 3, cv = "Rolling", VARX = list(k = 4, s = 1), intercept = TRUE)
results <- cv.BigVAR(mod1)
predict(results, n.ahead = 3, newxreg = new_X)

Y is a 226x15 matrix with 4 endogeneous and 11 exogeneous variable (i.e. monthly dummies).
For new_X I've tried to supply a 3x11 matrix containing only the dummy variables as well as a 3x15 matrix where the values for the endogenous variables are NA. In both cases, the above mentioned error occured.

[wbnicholson]

This issue should have been fixed in the latest development version. Let me know if it still persists or if you encounter other issues. I will submit the updated package to cran shortly.

[kaijennissen]

You're right. The error does no longer exist in development version. Thanks

[lenkiefer]

When I try to run predict with n.ahead >1 I get an error:

Error in predictMS(matrix(fcst, nrow = 1), Y[(nrow(Y) - p + 1):nrow(Y), : object 'g' not found

If I look at BigVARSupportFunctions.R at line 2257 there appears to be a misplaced g that is causing the error.

[wbnicholson]

Thanks, this has been fixed

[jonlachmann]

I am getting the same error as @kaijennissen when creating a model as follows

T1 <- floor(nrow(Y)/3)
T2 <- floor(2*nrow(Y)/3)

VARX <- list()
VARX$k <- 2 # indicates that the first two series are modeled
VARX$s <- 2 # sets 2 as the maximal lag order for exogenous series

model <- constructModel(Y, p=4, struct="Basic", gran=c(50,10), verbose=FALSE, VARX=VARX, T1=T1-3, T2=T2, h=3)
bigvar <- cv.BigVAR(model)
predict(bigvar, n.ahead = 1)

The matrix Y has the structure

str(Y)
 Time-Series [1:227, 1:3] from 1957 to 2014: -0.888 3.877 -4.121 -10.506 2.631 ...
 - attr(*, "dimnames")=List of 2
  ..$ : NULL
  ..$ : chr [1:3] "GDPGrowth" "TSpread[-1]" "TB10YS[-1]"

The last row with predict gives the error

Error in if (contemp) { (main.R#109): argument is of length zero

It might be that I am doing something wrong, but if that is the case I cant see what.

[wbnicholson]

Do you still encounter this issue in the github build? I will correct this issue in the cran version in the very near future.

[jonlachmann]

Do you still encounter this issue in the github build? I will correct this issue in the cran version in the very near future.

Hi! I tried building the package from source from the files here at github, works like a charm! Thanks!

[leosaenger]

Hi -- I am not having the above issue with Error in if (contemp) { (main.R#109): argument is of length zero (downloading from the GitHub worked great to resolve this!), but I am getting the error Error in rbind(Y, pred) : number of columns of matrices must match (see arg 2). I am running

predict(results, n.ahead = 2, newxreg = as.matrix.data.frame(holdout))

Where I have 12 endogenous variables and 2 exogenous variables. Running this with n.ahead = 1 works great. Varying n.ahead and ensuring holdout is formatted correctly doesn't seem to help much. I'm not sure what's going on here -- very sorry if it's an obvious mistake, and thanks for your work on this package!

[wbnicholson]

Could you post a reproducible example (including your specification in constructModel)? I can't seem to replicate your error.