[R-SIG-Finance] Does autoarfima from rugarch only work with external regressors in "full" mode?

[Maximilian Bredendiek]

 Dear all,

I want to determine the optimal lag length for an ARMA model, using the
"autoarfima" function from the rugarch package.

I am not interested in all possible lag combinations, but only in those of
consecutive orders. Thus, I call the function with method = c("partial").

However, I found that the external regressors I provide are only included
when setting the mode to "full". I tried the partial mode with and without
external regressors for lags up to 10 and got the same results w.r.t. to
AIC.

Here is a minimum working example (Regr is a vector of returns, exRegr a
matrix of external regressors with the same number of rows as Regr):

1) autoarfima(Regr, ar.max = 2, ma.max = 2, criterion = c("AIC"), method =
c("full"), external.regressors = exRegr)

Optimal Parameters
------------------------------------
         Estimate  Std. Error  t value Pr(>|t|)
mu       0.001050    0.000641  1.63898 0.101217
mxreg1   0.010120    0.000326 31.01558 0.000000
mxreg2   0.000207    0.000319  0.65083 0.515156
mxreg3  -0.000422    0.000318 -1.32807 0.184156
mxreg4  -0.000503    0.000319 -1.57776 0.114621
mxreg5   0.000092    0.000318  0.28860 0.772885
mxreg6   0.045095    0.028073  1.60634 0.108199
mxreg7   0.023944    0.029951  0.79945 0.424027
mxreg8  -0.293903    0.133907 -2.19482 0.028176
mxreg9  -0.301839    0.064925 -4.64906 0.000003
mxreg10  0.317559    0.029646 10.71167 0.000000
mxreg11 -0.001181    0.000896 -1.31891 0.187200
mxreg12 -0.001620    0.000895 -1.81045 0.070225
mxreg13  0.000365    0.000897  0.40702 0.683997
sigma    0.014096    0.000222 63.39932 0.000000

2) autoarfima(Regr, ar.max = 2, ma.max = 2, criterion = c("AIC"),
method = c("partial"), external.regressors = exRegr)

Optimal Parameters
------------------------------------
       Estimate  Std. Error  t value Pr(>|t|)
ar1    0.040781    0.022287   1.8298 0.067286
sigma  0.018892    0.000298  63.4193 0.000000

Do you know why this is the case?

Can you tell me how to run the function in "partial" mode and
including external regressors? The "full" mode takes too much
resources on my machine...

Thank you!

	[[alternative HTML version deleted]]