{smcl}
{* 08oct2002}{...}
{hline}
help for {hi:gcause}{right:[P.Joly]}
{hline}
{title:Granger causality test}
{p 8 27}
{cmd:gcause}
{it:var1}
{it:var2}
[{cmd:if} {it:exp}]
[{cmd:in} {it:range}]
{cmd:,}
{cmdab:l:ags(}{it:#}{cmd:)}
[
{cmdab:ex:og(}{it:varlist}{cmd:)}
{cmdab:reg:ress}
]
{p}
{cmd:gcause} is for use with time-series data. You must {cmd:tsset}
your data before using this commands; see help {help tsset}.
{title:Description}
{p}
{cmd:gcause} performs a Granger causality test to investigate whether
lagged values of a variable, {it:var2}, help in forecasting another
variable, {it:var1}. See {cmd:Methods & Formulas} below.
{title:Options}
{p 0 4}
{cmd:lags(}{it:#}{cmd:)} is not optional.
It specifies the number of lags of {it:var1} and {it:var2} to include
in the regression.
{p 0 4}
{cmd:exog(}{it:varlist}{cmd:)} for lack of a better name, specifies
other conditioning variables which may enter the regression.
{it:varlist} may contain time-series operators which is particularly
useful to add other lagged variables to the test regressions.
{p 0 4}
{cmd:regress} requests the display of the output from the unrestricted
regression.
{title:Methods & Formulas}
{p}
Granger-causality tests are usually performed in the context of vector
autoregressions (VAR) or more specifically, individual equations
within VAR systems.
Individual equations in VARs are known as autoregressive distributed
lag (ADL) relationships and may be represented as
{p 4 4}
{bind: {it:p} {it:p}}{break}
y_{it:t} = {it:c}_1 + SUM {it:a_i} *y_{it:t-i} + SUM {it:b_i} *x_{it:t-i} + D_{it:t} + u_{it:t}{break}
{bind: {it:i}=1 {it:i}=1}
{p 4 4}
({it:t} = 1,...,{it:T} )
{p}
where y_{it:t} and x_{it:t} respectively refer to {it:var1} and
{it:var2} in {cmd:gcause}'s syntax diagram and D_{it:t} corresponds
to other variables that need to be controlled for, if any, specified
at {cmd:exog()}.
{it:p} is determined by {cmd:lags()}.
{p}
The null hypothesis that x_{it:t} does not Granger-cause y_{it:t}
amounts to testing whether {it:b_i} = 0 for {it:i} = 1,...,{it:p}.
The rationale for conducting such a test is simple. If event X is
seen as causing event Y, then event X should precede Y
(Hamilton, p.303).
The test statistic is calculated from the sum of squared residuals
(RSS) of the unrestricted equation (above) and restricted equation
{p 4 4}
{bind: }{it:p}{break}
y_{it:t} = {it:c}_0 + SUM {it:g_i} *y_{it:t-i} + D_{it:t} + e_{it:t}{break}
{bind: }{it:i}=1
using the formula for joint-significance tests given by
{p 4 4}
{it:F} = ({it:RSS}_0 - {it:RSS}_1)/{it:p}{break}
{bind: }{dup 18:{c -}}{break}
{bind: {it:RSS}_1 /({it:T} -2{it:p} -1)}
{p}
which is distributed as an {it:F} ({it:p},{it:T} -2{it:p} -1) variable.
{it:RSS}_0 ({it:RSS}_1) is the residual sum of squares of the
restricted (unrestricted) regression.
{p}
The above test in only valid asymptotically due to the presence of
a lagged dependent variable in the regression.
An asymptotically equivalent test is given by,
{p 4 4}
{it:F}_a = {it:T} ({it:RSS}_0 - {it:RSS}_1){break}
{bind: }{dup 17:{c -}}{break}
{bind: {it:RSS}_1}
which is distributed as a chi2({it:p}) variable.
{title:References}
{p 0 2}
Hamilton, J. D. (1994). {it:Time Series Analysis}. Princeton University
Press. 799 p.
{title:Acknowledgements}
{p}
Thanks to Carol Miu for helpful comments.
{title:Author}
Patrick Joly, Industry Canada
pat.joly@utoronto.ca
{title:Also see}
{p 0 19}
On-line: help for
{help test},
{help vecar} (if installed)
{p_end}