Structural Time Series research using Stan

Dear all,

I just finished a working paper on how to estimate real estate price indices using structural time series. Its challenging as (commercial) real estate transactions are; (1) heterogeneous, (2) rare and (3) plagued by omitted variables. The models are estimated using Stan. I put it on SSRN just now (as of writing its still under review there);


If you have any comments, questions, or if you know a good journal (apart from the Real Estate journals) let me know!


I read your paper and became fascinated. I am a novice in R and Stan. But, I am learning and try to replicate your methodologies to produce a repeat sales index for thin land market in Korea. Could upload or send me the Stan code for the methodology? My e-mail is

I just looked at the conclusion:

In our applications Gibbs sampling, by using the program JAGS (Plummer, 2003), could take up to to 10 hours to converge, while NUTS only needed 10 – 20 seconds.

:-) I was largely motivated to learn more about sampling when I was a natural language processing/machine learning researcher trying to fit the Dawid/Skene noisy coder model to human labels for machine learning tasks. The models could take a day or two to fit in BUGS, but would often crash with a core dump partway through. When we finally had Stan up and running, I marginalized the discrete parameters as the original paper did for EM and the same model fit robustly in Stan in 10 or 20 minutes. I’m sure it would’ve fit faster in BUGS if you’d used their 0 (or 1—I can never recall) trick to do the marginalization.

Also for more info, one of the authors of the paper co-authored an entire book on the topic of real estate modeling. It looks like an interesting mix of finance, decision theory, and stats.