Summer of Econ

One of the priorities for July of 2016 is to add to econ-related stuff to rstan(arm) prior to our presentations at the AAEA meeting August 2 and the New York Federal Reserve August 12. Much of this would also be useful for the book Jim Savage is writing. Some possibilities include:

Instrumental Variables, Systems of Equations, and Selection Models

I now have a feature/ivreg branch that starts to implement this for an IV model, where the syntax of stan_ivreg on that of ivreg in the AER package. For systems of equations, we want to base the syntax on the systemfit function in the systemfit package. For selection models, we want to base the syntax of stan_selection on that of selection in the sampleSelection package.

Multinomial Probit Models

There is an rstanarm branch for this called feature/mnp. The syntax of stan_mnp is like that of the mnp function in the MNP package but the normalization being used to estimate the model is different. The vignette on that branch explains things in more detail. I think the Stan code is mostly ready to go (although it needs to be converted to the new syntax), so the bulk of the remaining work would be on the R code, testing, etc.

A General State Space Model

There is an rstanarm branch for this called feature/arima. At first I just implemented an ARIMA model where the syntax of stan_arima was based on the arima function in the base package. But then I decided it might be better to have a general state space model and base the syntax of stan_statespace on the dlmodeler package, which in turn is based on the notation in Durbin and Koopman.

A Dynamic Stochastic General Equilibrium Model

This is more for the presentation at the New York Federal Reserve than the AAEA meeting. I did some work on this in 2015 with Carlos for a conference, but I understand the model(s) somewhat better know. This requires first merging a branch into Stan Math that exposes and utilizes more of Eigen.

Spatial Econometrics Models

We have virtually no work on this yet, but I don't think it would be too difficult to incorporate a spatial weights matrix into the linear predictors for our existing .stan programs. Presumably, we would base the syntax on that in the spdep package.

Beta Regression Models (Implemented)

This would be pretty simple and also useful for the mi package. Basically, we need a betareg.stan file that models proportions using R syntax that is similar to that in the betareg function of the betareg package. Jasper may have some old code for this.