ARAUTOLAGS Procedure

@ARAutoLags computes an information criterion for various lags of AR processes using the Burg or Yule estimates of the partial autocorrelations. @MAAutoLags does a similar "quick" calculation for choosing a pure MA process. @BJAUTOFIT and @GMAUTOFIT do more time-consuming calculations (they do exact maximum likelihood fits) for choosing full ARMA models. (@GMAUTOFIT is for seasonal ARMA models).

@ARAutoLags( options ) series start end

Parameters

series	series to analyze
start, end	range of series to use (by default, the maximum possible)

Options

METHOD=[YULE]/BURG

Method used to compute the autocorrelations. YULE is more commonly used (it's the standard "textbook" calculation), but BURG more accurate if the process is nearly non-stationary.

MAXLAGS=maximum number of lags to consider [25]

CRIT=[AIC]/BIC/CAIC/HQ

Criterion to use:

AIC is (uncorrected) Akaike information criterion.

BIC is the Bayesian or Schwarz criterion.

HQ is Hannan-Quinn.

CAIC is the AIC-corrected for degrees of freedom

[PRINT]/NOPRINT

Show output

TABLE/[NOTABLE]

TITLE=title for table ["'criterion' analysis of 'series'"]

Show full table of results (not just best)

Variables Defined

%%AUTOP

number of parameters selected (INTEGER)

Example

* Replication file for Bauwens and Laurent(2005), "A New Class of

* Multivariate Skew Densities, With Application to Generalized

* Autoregressive Conditional Heteroscedasticity Models," JBES, vol 23,

* 346-354.

* Exchange rate example

open data txch.xls

data(format=xls,org=columns,left=2) 1 3065 date eurusd yenusd gbpusd $

r_eurusd r_yenusd r_gbpusd

* The returns series exist from the start of the data set.

* The authors use a 1 lag AR for the euro and yen---it's not clear how

* they came to that decision on the yen, which seems to have almost no

* noticeable serial correlation. BIC picks zero for all, but we'll stick

* with the choices in the paper.

@arautolags(maxlags=10,crit=bic,table) r_eurusd

@arautolags(maxlags=10,crit=bic,table) r_yenusd

@arautolags(maxlags=10,crit=bic,table) r_gbpusd

Sample Output

In the output, small is good. The chosen (starred) model is the AR(0) in each case, though the choice between 0 and 1 for R_EURUSD is a very close one.

Schwarz/Bayesian IC Lag Analysis for R_EURUSD

Lags IC

0 -0.679*

1 -0.678

2 -0.676

3 -0.673

4 -0.671

5 -0.668

6 -0.666

7 -0.664

8 -0.661

9 -0.659

10 -0.656

Schwarz/Bayesian IC Lag Analysis for R_YENUSD

Lags IC

0 -0.531*

1 -0.528

2 -0.526

3 -0.525

4 -0.522

5 -0.520

6 -0.517

7 -0.515

8 -0.512

9 -0.510

10 -0.507

Schwarz/Bayesian IC Lag Analysis for R_GBPUSD

Lags IC

0 -0.925*

1 -0.922

2 -0.922

3 -0.919

4 -0.917

5 -0.916

6 -0.914

7 -0.912

8 -0.910

9 -0.907

10 -0.905