{smcl}
{* 2010-06}{...}
{cmd: help xtmoralb}
{hline}

{title:Title}

{phang}
{cmd:xtmoralb} {hline 2} sub-system LIML Dynamic Panel Data Estimation
{p_end}

{title:Syntax}

{pstd}Moral-Benito (2011): sub-system LIML Dynamic Panel Data{p_end}

{p 8 17 2}
   {cmd: xtmoralb} {depvar} {indepvars} {ifin}
   [{cmd:,} {opt om:ethod(#)} {opt rob:ust} {opt tim:e}]
{p_end}


{title:Description}

{pstd}
{cmd:xtmoralb} implements the maximum likelihood estimator proposed in Moral-Benito (2011) for dynamic panel data models with fixed-effects and predetermined (or endogenous) regressors. 
This estimator can be interpreted as the likelihood-based (LIML) counterpart of GMM estimators for panel data models such as Arellano-Bond. 
The maximum likelihood estimator implemented by {cmd:xtmoralb} alleviates the many weak instruments biases of Arellano-Bond in finite samples without resorting to additional stationarity assumptions (e.g. system-GMM). 
This command accepts time series operators and it uses Mata functions so that you need at least Stata 9.
{p_end}

{title:Options}

{phang}
{opt om:ethod(#)} allows you to select the numerical method employed for the likelihood 
maximization in case you face convergence problems. The four options are: 
(1)Broyden-Fletcher-Goldfarb-Shanno. (2)Newton-Raphson.(3)Davidon-Fletcher-Powell.
(4)Berndt-Hall-Hall-Hausman. Default method is Broyden-Fletcher-Goldfarb-Shanno.
{p_end}

{phang}{opt tim:e} includes a set of time dummies in the model (i.e. cross-sectional correlation among the units in the panel is allowed).{p_end}

{phang}{opt rob:ust} computes robust standard errors with sandwich formula.{p_end}

{title:Example}

{p 8 14 2}{cmd:. use http://www.moralbenito.com/panel_growth.dta, clear}{p_end}
{p 8 14 2}{cmd:. tsset cid time} (or with the delta option for ten-year periods: tsset cid year, delta(10)){p_end}
{p 8 14 2}{cmd:. xtmoralb gdp L.gdp investment}{p_end}
{p 8 14 2}{cmd:. xtmoralb gdp L.gdp investment, omethod(2) time}{p_end}

{title:Notes}

You need to {manhelp tsset TS} data before using {cmd:xtmoralb}. 
You also need to specify delta() option with tsset if you do not have subsequent time-series observations (i.e. delta(3) if you have years 1987-1990-1993). 
If your panel is not balanced, the command will balance your data within mata but without changing your original data in Stata. 
For big samples in the cross-sectional dimension (i.e. big N) convergence takes a long time. 
Note also that the bigger the N the more similar is the estimator to Arellano-Bond. 
{pstd}


{title:References}

{phang}
Moral-Benito, E. 2011. "Dynamic Panels with Predetermined Regressors: Likelihood-based Estimation and Bayesian Averaging with an Application to Cross-Country Growth." (*) 
Available at http://www.moralbenito.com
(*) Previously circulated under the title "Panel Growth Regressions with General Predetermined Variables: likelihood-based estimation and Bayesian Averaging"


{title:Also see}

{psee}
Manual:  {bf:[TS] tsset}, {bf:[XT] xtreg}
{p_end}

{psee}
Online: {manhelp tsset TS}, {manhelp xtreg XT}
{p_end}

{title:Author}

Enrique Moral-Benito
email: {browse "mailto:enrique.moral@gmail.com":enrique.moral@gmail.com}
web: {browse "http://www.moralbenito.com":http://www.moralbenito.com}