mboeck11
commented
9 months ago This happens because BGVAR yields no stable posterior draws. I would suggest that you estimate the model in stationary terms or play around with the eigen command.
Nevertheless, I added a if-condition in the new version to print an appropriate error message.
Dear Author: I have an error as below.
summary_bgvar <- summary(model_bgvar_with_config)
Error in 1:d1 : argument of length 0
My time series is quarterly data (72 points) and there are 100 series. My pc has 8GB of ram. It seems after the run everything is empty, a list of nothing. This happens if I set my draw number high. What can I do to get a result? I tried 1000 and it worked. Could I know why?
I have my config set as follows:
BGVAR_data_start = "2003Q1"
BGVAR_data_end = "2021Q4"
BGVAR_data_start_formatted = "2003-01-01"
Model configuration
BGVAR_lag = 1
BGVAR_prior = "MN"
BGVAR_draw = 5000
BGVAR_burnin = 5000
BGVAR_SV = TRUE
BGVAR_hold.out = 0
BGVAR_thin = 1
BGVAR_trend = FALSE
BGVAR_hyperpara = NULL
BGVAR_eigen = T
BGVAR_verbose = TRUE
BGVAR_expert = NULL
BGVAR_Ex = NULL
#################################################################### Start estimation of Bayesian Global Vector Autoregression.
Prior: Minnesota prior.
Lag order: 1 (endo.), 1 (w. exog.)
Stochastic volatility: enabled.
Number of cores used: 1.
Thinning factor: 1. This means every draw is saved.
Estimation of country models starts...
Estimation done and took 9 mins 15 seconds.
Start stacking:
Stacking finished.
Computation of BGVAR yields 0 (0%) draws (active trimming).
Needed time for estimation of bgvar: 9 mins 16 seconds.
Error in 1:d1 : argument of length 0