Embedded Files
Stata Command: oga
Stata Command: oga
Stata Command: oga.ado
This command executes estimation and inference for high-dimensional regressions without imposing the sparsity restriction.
Installation:
. ssc install oga
Help:
. help oga
Reference: Cha, Jooyoung, Harold D. Chiang, and Yuya Sasaki. "Inference in High-Dimensional Regression Models without the Exact or Lp Sparsity." Review of Economics and Statistics, Forthcoming. Paper.
Title
oga -- Executes estimation and inference for high-dimensional regressions without imposing the sparsity restriction.
Syntax
oga depvar indepvarlist1 indepvarlist2 [if] [in] [, dimension(integer) folds(integer) repdml(integer) cstar(real) cluster(varname) ]
Description
oga performs estimation and inference for high-dimensional regression models without imposing a sparsity assumption, based on the methodology of Cha, Chiang, and Sasaki. The estimation procedure combines the orthogonal greedy algorithm (OGA), the high-dimensional Akaike information criterion (HDAIC), and double/debiased machine learning (DML).
Options
dimension(integer) specifies the number of variables for indepvarlist1 whose coefficients are to be displayed in the output table. The default is dimension(1). The value must be a positive integer no greater than the total number of variables in indepvarlist1 and indepvarlist2.
folds(integer) sets the number K of folds used for cross-fitting in the double/debiased machine learning (DML) procedure. The default is folds(5). The value must be an integer greater than 1.
repdml(integer) sets the number of resampling repetitions used for finite-sample adjustment in the double/debiased machine learning (DML) procedure. The default is repdml(5). The value must be a positive integer.
cstar(real) specifies the tuning parameter C* for the high-dimensional Akaike information criterion (HDAIC). The default is cstar(2).
cluster(varname) specifies the variable used to define clusters. If this option is not specified, the command is executed without clustering.
Usage Examples
Estimation of the partial effect of d on y controlling for 100 variables:
. oga y d x1 ... x100
Cluster-robust standard error by the clustering variable state:
. oga y d x1 ... x100, cluster1(state)
Estimation of the partial effects of d1, d2, d3 on y controlling for 100 variables:
. oga y d1 d2 d3 x1 ... x100, dimension(3)
etc.
Stored results
oga stores the following in e():
Scalars e(N) observations e(dimension) number of indepvarlist1 e(folds) number of folds for the cross fitting e(repdml) number of resampling for DML e(cstar) tuning parameter C*
Macros e(cluster) clustering variable e(cmd) oga
Matrices e(b) coefficient vector e(V) variance-covariance matrix of the estimators
Functions e(sample) marks estimation sample
Reference
Cha, Jooyoung, Harold D. Chiang, and Yuya Sasaki. Inference in High-Dimensional Regression Models without the Exact or Lp Sparsity. Review of Economics and Statistics. Link to Paper.
Authors
Jooyoung Cha, Vanderbilt University, Nashville, TN. Harold D. Chiang, University of Wisconsin, Madison, WI. Yuya Sasaki, Vanderbilt University, Nashville, TN.
oga -- Executes estimation and inference for high-dimensional regressions without imposing the sparsity restriction.
Syntax
oga depvar indepvarlist1 indepvarlist2 [if] [in] [, dimension(integer) folds(integer) repdml(integer) cstar(real) cluster(varname) ]
Description
oga performs estimation and inference for high-dimensional regression models without imposing a sparsity assumption, based on the methodology of Cha, Chiang, and Sasaki. The estimation procedure combines the orthogonal greedy algorithm (OGA), the high-dimensional Akaike information criterion (HDAIC), and double/debiased machine learning (DML).
Options
dimension(integer) specifies the number of variables for indepvarlist1 whose coefficients are to be displayed in the output table. The default is dimension(1). The value must be a positive integer no greater than the total number of variables in indepvarlist1 and indepvarlist2.
folds(integer) sets the number K of folds used for cross-fitting in the double/debiased machine learning (DML) procedure. The default is folds(5). The value must be an integer greater than 1.
repdml(integer) sets the number of resampling repetitions used for finite-sample adjustment in the double/debiased machine learning (DML) procedure. The default is repdml(5). The value must be a positive integer.
cstar(real) specifies the tuning parameter C* for the high-dimensional Akaike information criterion (HDAIC). The default is cstar(2).
cluster(varname) specifies the variable used to define clusters. If this option is not specified, the command is executed without clustering.
Usage Examples
Estimation of the partial effect of d on y controlling for 100 variables:
. oga y d x1 ... x100
Cluster-robust standard error by the clustering variable state:
. oga y d x1 ... x100, cluster1(state)
Estimation of the partial effects of d1, d2, d3 on y controlling for 100 variables:
. oga y d1 d2 d3 x1 ... x100, dimension(3)
etc.
Stored results
oga stores the following in e():
Scalars e(N) observations e(dimension) number of indepvarlist1 e(folds) number of folds for the cross fitting e(repdml) number of resampling for DML e(cstar) tuning parameter C*
Macros e(cluster) clustering variable e(cmd) oga
Matrices e(b) coefficient vector e(V) variance-covariance matrix of the estimators
Functions e(sample) marks estimation sample
Reference
Cha, Jooyoung, Harold D. Chiang, and Yuya Sasaki. Inference in High-Dimensional Regression Models without the Exact or Lp Sparsity. Review of Economics and Statistics. Link to Paper.
Authors
Jooyoung Cha, Vanderbilt University, Nashville, TN. Harold D. Chiang, University of Wisconsin, Madison, WI. Yuya Sasaki, Vanderbilt University, Nashville, TN.
Page updated
Google Sites
Report abuse