Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

From |
Søren Møller-Larsson <soren_ml@hotmail.com> |

To |
<statalist@hsphsun2.harvard.edu> |

Subject |
st: Xtabond2 system gmm h(2) or h(3) |

Date |
Sun, 13 May 2012 12:28:43 +0200 |

Dear statalisters Stata 12.1 (mac version). I am using the most recent version of -xtabond2- (system GMM) on a panel data set including 46 countries (groups) in order to find shared determinants of FDI. The dependent variable is FDI inflow as a share of GDP. T=20 and data is almost balanced (above 95%). Here is an example of an estimation with push factors only. realgrowth_g7 realinterest_g7 vix are all treated as exogenous. xtabond2 fdi L.fdi growth_g7 interest_g7 vix yrdum*, gmm(fdi,laglimits(2 2) col) iv(growth_g7 interest_g7 vix) iv(yrdum*, equat(level)) h(3) twostep robust (i am aware that h(3) is standard) I have two questions: 1) The weighting matrix h(3) is standard in xtabond2, however, h(2) is used in DPD, Dornik et. al. estimator in ox. I am aware of the mathematical differences as outlined in Roodman (2009) "How to Xtabond2". What is the implication of using one over the other? Roodman does not explain why the one is preferred to the other. If I use h(3) coefficients in all explanatory variables (except lagged FDI) are inflated and so is standard errors making inference impossible. However, if I use h(2) all coefficients and standard errors behave nicely and inference is indeed possible. Is it justifiable to use h(2) instead of h(3)? Also, by using h(2) instead of h(3) the coefficient estimate on L.fdi is within the bracketing LSDV-OLS range. Using h(3) makes the coefficient estimate well below the LSDV estimate, so maybe the model is miss-specified? All my specification test perform very nicely, and for the above estimation, instrument count is 25. 2) . According to Roodman (2009) "How to Xtabond2" year dummies are treated as predetermined and should be treated as in the above command and is only used as instruments of the level. Many other papers assume year dummies to be strictly exogenous. Can year dummies be treated as exogenous, and should they? Any answer will be much appreciated. Thanks in advance Best regards Soren * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/

- Prev by Date:
**Re: st: Estimating multiple regression on panel dataset by year, saving and testing results.** - Next by Date:
**st: Computing the proportion of significant variables after running numerous regressions** - Previous by thread:
**st: Scatterplot matrix** - Next by thread:
**st: Computing the proportion of significant variables after running numerous regressions** - Index(es):