CODEWITHSUNDEEP
rfinanceeconomicsmidasr
Julian Bausch
Julian Bausch

Reputation: 1

Exploding Values with Midas prediction with external variables

By estimating a MIDAS model for realized variance I ran into the problem that the model with external variables generated exploding values for the forecasts.

My midas regression is the following:

tsx_var_vix <- ts(coredata(data_xts$VIX), frequency = 252)
# Convert the time series identifier to a time series object
tsy_var <- ts(coredata(data_xts[, ts_id]), frequency = 252)

# Create MIDAS models
nealmon_model <- midas_r(tsy_var ~ mls(tsx_var, 1:22, 1, nealmon),
                         start = list(tsx_var = c(1, -0.5)), weight_gradients = list())
nealmon_model_vix <-  midas_r(tsy_var ~ mls(tsx_var, 1:22, 1, nealmon) + mls(tsx_var_vix, 1:22, 1, nealmon) ,
                              start = list(tsx_var = c(1, -0.5), tsx_var_vix = c(1, -0.5)), weight_gradients = list())
harstep_model <- midas_r(tsy_var ~ mls(tsx_var, 1:20, 1, harstep),
                         start = list(tsx_var = c(1, 1, 1)), weight_gradients = list())
s = list())

# Create forecasts
end_sample <- floor(length(tsx_var) * 0.8)
out_sample_start <- end_sample + 1
forecast <- average_forecast(list(nealmon_model, harstep_model, nealmon_model_vix),
                             data = list(tsx_var = tsx_var, tsy_var = tsy_var, tsx_var_vix=tsx_var_vix),
                             insample = 1:end_sample, outsample = out_sample_start:length(tsx_var),
                             type = "rolling", show_progress = FALSE)

While the in sample fit was better for the model with the external variable the forecast generated the follwing results in MSE, which are clearly unusual: 6.087358e-05 6.158823e-05 1.348308e+03 #third is the estimation with an external regressor

Does somebody know where the problem comes from and how to adress it?

Upvotes: 0

Views: 24

Answers (0)

Related Questions