Hi All,


For those who are familiar with the Frankel and Romer (1999) study, the authors construct instruments for bilateral trade in a first stage, where they regress bilateral trade on geographic variables. Thereafter, they obtain the predicted values of trade based on geography for a country pair, and then aggregate over trading partners for a particular country to obtain the exogenous fraction of trade for a particular country. THis is then used as an instrument for the trade, to analyze the effect of trade on income.


Silva and Tenreyro (2006) suggested the use of a PPML type estimator as it is more robust to inclusion of 0 trade flows as well as heteroskedasticity in the trade data in the construction of the instrument.


One problem common to both approaches is that the instrument is "generated", in that it is itself measured with error as it is based on predicted values in its construction. This introduces complications in the interpretation of the standard error in the final stage.

One solution is to bootstrap the standard errors in the final stage. Is there any command that allows for boostrapping in combination with either the iv or 2sls command?


Many thanks,
CS