motivation nonlinear models generalized linear models other models final considerations Using Stata to estimate nonlinear models with high- dimensional Using Stata to estimate nonlinear models with fixed effects Paulo high-dimensional fixed effects Guimaraes motivation nonlinear Paulo Guimaraes 1 , 2 models generalized linear models 1 Banco de Portugal other models 2 Universidade do Porto final considerations Portuguese Stata UGM - Sept 15, 2017 Paulo Guimaraes Using Stata to estimate nonlinear models with high-dimensional
motivation nonlinear models generalized linear models other models final considerations Using Stata to estimate ”More and more data”? nonlinear models with high- dimensional fixed effects Paulo Guimaraes • availability of microdata for researchers is increasing fast motivation • easy to gain access to very large data sets nonlinear models • these ”large data sets” open up research possibilities generalized • they also pose many technical challenges linear models other models • an important limitation is the lack of tools to efficiently final explore large data sets considerations Paulo Guimaraes Using Stata to estimate nonlinear models with high-dimensional
motivation nonlinear models generalized linear models other models final considerations Using Stata to estimate The response nonlinear models with high- dimensional fixed effects • Stata made significant improvements to respond to the Paulo need to work with larger data sets Guimaraes • introduction of Mata motivation • Stata MP nonlinear • increase in Stata limits models • faster code for many ados generalized linear models • plugins other models • and the Stata community also offered contributions final • parallel - by George Vega Yon considerations • ftools - by Sergio Correia • gtools - by Mauricio Caceres Bravo Paulo Guimaraes Using Stata to estimate nonlinear models with high-dimensional
motivation nonlinear models generalized linear models other models final considerations Using Stata to estimate What about regressions for nonlinear models with high- high-dimensional data? dimensional fixed effects • Stata has significantly expanded methods for Paulo Guimaraes panel/longitudinal data • but it still lacks command for dealing with regressions with motivation multiple fixed effects nonlinear models • many user-written packages for linear regression: generalized • areg by Amine Ouazad linear models • reg2hdfe by Paulo Guimaraes other models • gpreg by Johannes F. Schmieder final considerations • felsdvreg by Thomas Cornelissen • reghdfe by Sergio Correia • reghdfe is the gold standard! • it is very fast, allows weighs, and it handles multiple fixed effects and interactions Paulo Guimaraes Using Stata to estimate nonlinear models with high-dimensional
motivation nonlinear models generalized linear models other models final considerations Using Stata to estimate What about nonlinear regression nonlinear models with high- models with multiple fixed effects? dimensional fixed effects Paulo Guimaraes • There are theoretical challenges motivation • are the relevant parameters identifiable? nonlinear models • does the solution exist? • is the incidental parameter problem ”biting”? generalized linear models • and there are technical challenges ... other models final • what algorithms to use? considerations • are the approaches computationally feasible? • are algorithms fast enough for large data sets? Paulo Guimaraes Using Stata to estimate nonlinear models with high-dimensional
motivation nonlinear models generalized linear models other models final considerations Using Stata to estimate But there is hope for many nonlinear models with high- nonlinear models dimensional fixed effects Paulo Guimaraes • reghdfe does a great job for linear regression motivation • makes possible estimation of nonlinear models by iterative nonlinear algorithms based on linear regression models generalized • a good example are Generalized Linear Models - can be linear models efficiently estimated by Iteratively Reweighted Least other models Squares final considerations • another example are nonlinear models that can be estimated recursively using linear regressions Paulo Guimaraes Using Stata to estimate nonlinear models with high-dimensional
motivation nonlinear models generalized linear models other models final considerations Using Stata to estimate GLM - Generalized Linear Models nonlinear models with high- • GLM models can be estimade by IRLS as dimensional fixed effects Paulo ) − 1 ( X ′ W ( r − 1) X = X ′ W ( r − 1) z ( r − 1) Guimaraes motivation • Examples of GLM models are: nonlinear models • Poisson regression • logit regression generalized linear models • probit regression other models • cloglog regression final • negative binomial considerations • gamma • All of these (and more) can be estimated by IRLS • It is a simple matter to add hdfes! • poi2hdfe is an example for Poisson with 2 hdfes Paulo Guimaraes Using Stata to estimate nonlinear models with high-dimensional
motivation nonlinear models generalized linear models other models final considerations Using Stata to estimate Some examples nonlinear models with high- dimensional fixed effects Paulo Guimaraes motivation nonlinear • Example 1 - Poisson regression with 2 hdfes models • Example 2 - cloglog with 2 hdfes generalized linear models other models final considerations Paulo Guimaraes Using Stata to estimate nonlinear models with high-dimensional
motivation nonlinear models generalized linear models other models final considerations Using Stata to estimate Regression with peer effects nonlinear models with high- dimensional fixed effects • a regression with peer effects (Arcidiacono et al, 2012) Paulo Guimaraes can be written as motivation Y = X β + D α + γ WD α + ϵ nonlinear models • the regression is non linear generalized linear models • estimation can be implemented by alternating between other models estimation of β, γ and estimation of α final considerations • conditional on α the problem becomes linear • easy to add other fixed effects Paulo Guimaraes Using Stata to estimate nonlinear models with high-dimensional
motivation nonlinear models generalized linear models other models final considerations Using Stata to estimate An example of peer regression nonlinear models with high- dimensional fixed effects Paulo Guimaraes motivation nonlinear • Example - regression with peer effects models generalized linear models other models final considerations Paulo Guimaraes Using Stata to estimate nonlinear models with high-dimensional
motivation nonlinear models generalized linear models other models final considerations Using Stata to estimate Conclusion nonlinear models with high- dimensional fixed effects Paulo • it is possible to add fixed effects to some nonlinear models Guimaraes • Poisson regression is probably the easiest application motivation nonlinear • but we should worry about existence of a solution models • ability to estimate does not translate into consistency of generalized linear models estimators other models • should understand better how long a panel needs to be final considerations • estimation on large data sets likely to be a slow process Paulo Guimaraes Using Stata to estimate nonlinear models with high-dimensional
Recommend
More recommend
