CODEWITHSUNDEEP
rtime-seriesmodel-comparison
user155417
user155417

Reputation: 55

model selection with unrestricted model in midasr package

I have a monthly time series and two weekly time series and I want to use the MIDAS regression using the midasr package in R. Furthermore, I am using the unrestricted model, where six lags of the monthly variable and one lag of every weekly variable is included as stated in the following: 

    unrestricted_model <- midas_r(monthly ~ mls(monthly,1:6,1) + mls(weekly_1,0:7,4) + mls(weekly_2,0:7,4), start = NULL)

Since many lags are insignificant, I want to do a model selection in the next step. The code demos of the User guide provided by the authors of the package present the usual steps of the model selection procedure very well, however they use the different algorithms as "nealmon" or "almonp" with specific starting values in their regression and later in the model selection. Now comes my question, how can I do the model selection when I have a unrestricted model? It seems to me that expand_weight_lags does only work with the algorithms and and not with the unrestricted model.

Upvotes: 1

Views: 372

Answers (1)

anguyen
anguyen

Reputation: 477

In the output of the function midas_r_ic_table, the AIC and BIC for both the restricted and unrestricted models is part of the output. Using a modified example from the documentation, in order to fit the restricted model use

midas_r(y ~ mls(x, 0:7, 4))

Or for the restricted version

midas_r(y ~ mls(x, 0:7, 4, nealmon) , start = list(x = c(1, -0.5)))

Where the starting list is the starting values of the exponential Almon lag coefficients

Upvotes: 0

Related Questions