SAS/ETS software provides several procedures that estimate models that have endogeneity. Endogeneity usually occurs for three reasons: omitted variables, measurement error in regressors, and simultaneity. In dynamic models, endogeneity is even more relevant, because regressors might be correlated with the error term not only from the current time period but from preceding periods as well. The following procedures support models that have endogeneity.
The MODEL and TMODEL procedures include the following features related to endogeneity:
nonlinear regression analysis of single equations
nonlinear regression analysis of systems of simultaneous equations
support for general-form models that have endogeneity
a variety of estimation methods to handle endogeneity, including the following:
(nonlinear) two-stage least squares (2SLS)
iterated two-stage least squares (IT2SLS)
(nonlinear) three-stage least squares (3SLS)
iterated three-stage least squares (IT3SLS)
generalized method of moments (GMM)
iterated generalized method of moments (ITGMM)
full-information maximum likelihood (FIML)
The SYSLIN procedure includes the following features related to endogeneity:
linear regression analysis of single equations
linear regression analysis of systems of simultaneous equations
a variety of estimation methods to handle endogeneity, including the following:
(nonlinear) two-stage least squares (2SLS)
(nonlinear) three-stage least squares (3SLS)
iterated three-stage least squares (IT3SLS)
limited-information maximum likelihood (LIML)
minimum expected loss (MELO)
general K-class estimators
full-information maximum likelihood (FIML)
The SIMLIN procedure performs simulation and multiplier analysis of simultaneous systems of linear regression models that have endogeneity.
The QLIM procedure includes the following features related to endogeneity:
test of endogeneity for a list of regressors in the model
overidentification test for the validity of instrumental variables
ability to estimate models that have endogeneity by adding regressions of endogenous regressors on exogenous regressors and instrumental variables
ability to estimate structural models that contain one endogenous variable by using full-information maximum likelihood (FIML)
ability to estimate structural models that contain multiple endogenous variables by using simulated maximum likelihood
The PANEL procedure uses instrumental variable regressions to estimate both static and dynamic panel models that have endogeneity:
Hausman-Taylor and Amemiya-MaCurdy estimation for static panel models
One-step, two-step, or iterative generalized method of moments (GMM) for dynamic panel models