A multivariate version is available with @MVARCHTEST (http://www.estima.com/forum/viewtopic.php?f=7&t=928).
archtest.src- Procedure files (requires RATS 7.3 or later)
- (2.97 KiB) Downloaded 105 times
@ARCHTEST series start end
Options
LAGS=(maximum) number of lags [4]
SPAN=spacing between number of lags in a test sequence.
SPAN=1 will give a table of tests for all lags from 1 to the LAG option. The default is to test only the value of the LAGS option.
DFC=degrees of freedom correction for tests.
By default, 0. If <<series>> are residuals from an ARCH or GARCH, DFC should be the number of ARCH/GARCH parameters.
FORM=[F]/LM
Determines whether F-test or Lagrange multiplier form is used
[PRINT]/NOPRINT
TITLE=Title for output ["Test for ARCH in Series xxxx"]
Variables Defined
| %CDSTAT | F-statistic for test with LAG option lags |
| %SIGNIF | Significance level of %CDSTAT |
Example
- Code: Select all
*
* Hill, Griffiths, & Lim, Principles of Econometrics, 3rd edition
* ARCH-GARCH examples starting page 369
*
open data byd.dat
data(format=free,org=columns) 1 500 r
*
linreg r
# constant
*
set e = %resids
*
* LM regression for ARCH test
*
set esq = e^2
linreg esq
# constant esq{1}
*
* This is slightly different from the result in the text as it uses the
* full precision of the R-squared.
*
cdf(title="ARCH test") chisqr %trsquared 1
*
* The ARCHTEST procedure can handle the entire test given the residuals.
* The FORM=LM asks for the form given in the text, rather than an F test.
*
@archtest(lags=1,form=lm) e
*
* Estimating an ARCH model. This is done using the GARCH instruction.
* The number of lagged squared residual terms is given by the Q option.
* Because the instruction can be used for several series at once, you
* put the estimation range first (here, just the / for the full range),
* followed by the series.
*
* One thing to note is that the estimated parameters and (particularly)
* the standard errors can be somewhat different. All of these models are
* estimated using iterative methods and different software use different
* algorithms to find the "best" set of parameters. While the parameter
* values are generally quite close, the standard errors are often quite
* different. Choice of algorithm has more of an effect on the latter
* than the former.
*
garch(q=1) / r
*
* Estimating the GARCH model. The number of lagged variance terms is
* indicating with the P option.
*
garch(p=1,q=1,hseries=hgarch) / r
*
* Estimating the model with asymmetry
*
garch(p=1,q=1,asymm,hseries=htgarch) / r
*
* Estimating the model with asymmetry and "M" effects. %garchv is a
* series used internally by GARCH to represent the variance. The
* REG(ressors) option indicates that the mean model takes the form of a
* regression, rather than just a simple mean.
*
garch(p=1,q=1,reg,asymmetric,hseries=htgarchm) / r
# constant %garchv
*
* The TABLE instruction can be used to determine the maximum value taken
* by the three variance series. Graphing all of them on a common scale
* is a good idea.
*
table(noprint) / hgarch htgarch htgarchm
spgraph(vfields=3,footer="Figure 14.7 Estimated variances of ARCH models")
graph(hlabel="(a) GARCH(1,1)",max=%maximum)
# hgarch
graph(hlabel="(b) T-GARCH(1,1)",max=%maximum)
# htgarch
graph(hlabel="(c) GARCH-in-mean",max=%maximum)
# htgarchm
spgraph(done)