Dear All,

I am trying to implement on Stata 14.0 the same model adopted in Remolona, Scatigna and Wu (2008). Specifically, I would like to perform the two-stage least squares estimation technique reported at pages 9 and 10.

From my understanding, the dependent variable predicted in the first stage (lambda_R_t+1 in the equation (5), p. 10) is different from the endogenous independent variable instrumented in the second stage (lambda_R_t+1 – S_*_t-1 in the equation (4), p. 9), i.e. the instrumented variable in the second stage is equal to the predicted variable in the first stage less another variable.

However, I am unsure about how to repeat this procedure in Stata. Two alternative strategies came to my mind:

1. As far as I know, the ivregress command requires the predicted variable to be the same as the instrumented variable, so it does not seem suitable for my case.

2. On the other hand, performing the instrumental variables regression “by hand" with the two distinct reg commands (one for each step, while taking the difference mentioned above) would return incorrect standard errors. Even if applying some subsequent correction to the variance-covariance matrix, I am still not sure whether this strategy would be consistent in the presence of the aforementioned variables.

I would greatly appreciate if someone could shed some light on this. Thanks in advance.

Best,

Daniele Veggiato


References:

Remolona, E. M., Scatigna, M., & Wu, E. (2008). The dynamic pricing of sovereign risk in emerging markets: Fundamentals and risk aversion. The Journal of Fixed Income, 17(4), 57-71.
Available online at https://www.cass.city.ac.uk/__data/a...914/Wu-140.pdf