Title
suchowtest performs successive Chow tests on cross section data
Syntax
suchowtest depvar [indepvars] [if] [in] [weight] [, options]
options Description ------------------------------------------------------------------------- Model thresv(varname) indicates the threshold variable stub(string) designates a string name from which new variable names will be created fpctile(#) specifies the lower bound percentile of the threshold variable lpctile(#) specifies the upper bound percentile of the threshold variable step(#) indicates the step by which we want to move the sample's break point sig(#) designates the significance level we want to set for the p-value of the Chow test
Reporting nographs suppress the display of graphs after the estimations are performed
Additional Options regress_options In addition to the options listed above, all options of the regress command can be used ------------------------------------------------------------------------- aweights, fweights, iweights, and pweights are allowed; see weight. by is not allowed with suchowtest; see [D] by for more details on by. indepvars and the thresv(varname) option may contain factor variables; see fvvarlist. depvar, indepvars and the thresv(varname) option may contain time-series operators; see tsvarlist.
Description
suchowtest performs successive Chow tests on cross section data. Habitually, when we are doing the Chow test, we split the sample of study in two subsamples using an exogenous break point. Unlike the previous method, the command suchowtest finds the break point endogenously. We do not have to supply a break point. If there is a threshold, the command finds it by using the information given by the data. If there is no break point the command will inform us too. This method of finding thresholds appears to be more reasonable in cases where the researcher does not know a priori the breaking point. The theory behind the command suchowtest is provided by Berthelemy and Varoudakis (1996).
Options
+-------+ ----+ Model +------------------------------------------------------------
thresv(varname) indicates the threshold variable. To form this option, you put inside the brackets the variable name representing the threshold variable. You must specify this option in order to get a result. Hence this option is not optional.
stub(string) designates a string name from which new variable names will be created. To form this option, you put inside the brackets a string name. Then new variable names will be created from this string. You must specify this option in order to get a result. Hence this option is not optional.
fpctile(#) specifies the lower bound percentile of the threshold variable that must be included in the search for a break point. The default value of this option is 10. Hence the search for the break point starts at the 10th percentile of the threshold variable.
lpctile(#) specifies the upper bound percentile of the threshold variable that must be included in the search for a break point. The default value of this option is 90. Hence the search for the break point starts at the 10th percentile and goes up to the 90th percentile of the threshold variable.
step(#) indicates the step by which we want to move the sample's break point. The default value of this option is 1. This means that the sample's break point is moved forward by one observation every time.
sig(#) designates the significance level we want to set for the p-value of the Chow test. The default value of this option is 0.10. This means that we want the p-value of the Chow test to be significant at most at the 10% level.
+-----------+ ----+ Reporting +--------------------------------------------------------
nographs suppress the display of graphs after the estimations are performed. This option is used when we do not want to display the graphs after the estimations are done.
+--------------------+ ----+ Additional Options +-----------------------------------------------
regress_options: noconstant, hascons, tsscons, vce(vcetype), level(#), beta, eform(string), etc. See [R] regress. All options of the regress command can be used.
You can use all the options of the command regress. To use them, enter them in the same manner that you would do with the regress command.
Saved results
suchowtest saves the following in r():
Scalars r(maxchowfh) Chow Test (F-Test) r(maxpvchowtest) P-Value of the Chow Test r(maxbreakpt) Value of the break point parameter r(maxobsvalue) Observation number corresponding to the break point r(qlstat) Maximum of the QL Statistic
Macros r(chowfh) Variable containing all the Chow F-Statistics r(chowstatpv) Variable containing all the P-Values of the Chow Test r(breakptpar) Variable containing all the Break Point Parameters r(qlvariable) Variable containing all the QL Statistics
Examples
Before beginning the estimations, we use the set more off instruction to tell Stata not to pause when displaying the results.
set more off
We load the data we are going to use and describe them. The description shows that we have cross section data which represent the average from 1975 to 2004.
use http://fmwww.bc.edu/repec/bocode/s/suchowtestdata.dta, clear
describe
We estimate a standard conditional convergence growth regression in which the real GDP per capita growth rate is regressed on initial real GDP per capita, stock market capitalization (financial development) and human capital. We use financial development as threshold variable with the option thresv(). We also specify the option stub() in which we put the string "sct". Note that these two options are required.
suchowtest croisspibt lninitgdppc lnstmktcap lnyr_sch_sec, thresv(lnstmktcap) stub(sct)
The estimation firstly gives some statistics concerning the threshold. The first statistic is the observation number at which the break point occurs. The second is the maximum of the QL statistic. The third is the Chow test and its p-value and the last one is the value of the threshold variable at the break point. Secondly, the estimation provides the OLS regression below the break point parameter. Thirdly, the estimation gives the OLS regression above the break point parameter. The estimation also offers three graphs. The first titled "QL STATISTIC" draws the QL statistic against the break point parameter. The Greene vertical line represents the observation at which the break point occurs. The second titled "P-VALUES OF THE CHOW TEST" graphs the p-values of the Chow test against the break point parameter. The Greene horizontal line is the significance level of the p-value of the Chow test. The third titled "PV. CHOW TEST AND QL STAT." is the combination of the two previous graphs. The left y-axis graphs the p-values of the Chow test while the right y-axis provides the QL statistic.
The estimation also generates variables containing the previously provided statistics. These variables contain: The break point parameter, the QL statistic, the p-values of the Chow test and the Chow F-statistic. To see these variables, we type:
describe sct_*
We illustrate the use of options fpctile(#) and lpctile(#). By default the search for the break point starts at the 10th percentile and goes up to the 90th percentile of the threshold variable. Now we extend the search range of the threshold. We start from the 5th percentile and goes up to the 95th percentile.
suchowtest croisspibt lninitgdppc lnstmktcap lnyr_sch_sec lngconsgdp lnopenwb, thresv(lnstmktcap) stub(sct) fpctile(5) lpctile(95)
We show the use of the option sig(). We enter the value 0.01 as the significance level of the p-value of the Chow test.
suchowtest croisspibt lninitgdppc lnstmktcap lnyr_sch_sec lngconsgdp lnopenwb, thresv(lnstmktcap) stub(sct) sig(0.01)
The command suchowtest indicates that there is no break point at this significance level. It suggests us to increase the significance level to augment the chance of obtaining a threshold. Hence we increase the significance level to 0.05.
suchowtest croisspibt lninitgdppc lnstmktcap lnyr_sch_sec lngconsgdp lnopenwb, thresv(lnstmktcap) stub(sct) sig(0.05)
If we do not want to display the graphs, we type:
suchowtest croisspibt lninitgdppc lnstmktcap lnyr_sch_sec lngconsgdp, thresv(lnstmktcap) stub(sct) nographs
In addition to the options cited above, all options of the regress command can be used. For instance we show here how to compute robust standard errors with the option vce(robust).
suchowtest croisspibt lninitgdppc lnstmktcap lnyr_sch_sec lngconsgdp vtotopen, thresv(lnstmktcap) stub(sct) vce(robust)
References
Berthelemy, J. C. and A. Varoudakis: 1996, "Economic Growth, Convergence Clubs, and the Role of Financial Development", Oxford Economic Papers 48(2), 300-328.
Author
Diallo Ibrahima Amadou, zavren@gmail.com
Also see
Online: help for regress, chowreg (if installed)