Fitting trend/bias in an AR(1) timeseries

[roualdes]

@lorcan, you might be interested in a model that I’ve been working on – it’s certainly not a final product yet. It’s a Stan version of bsts’s semilocal linear trend model. I have some code on GitHub that might help get you started: GitHub - roualdes/semilocal-linear-trend: Semilocal linear trend model in Stan.

If you end up fitting the Stan version, let me know what you think. Thanks.

Also, here’s a discourse post about my efforts thus far:

Semilocal Linear Trend Model, opinions/suggestions welcome

I’ve been playing around with a Stan version of the semilocal linear trend model, as found in the R package bsts. The model both automatically scales and matches bsts’s fit fairly well. Still, I’m hoping to get some feedback about: starting points mu[1] and nu[1], choice of priors (I offer my thoughts in line), and possible ways to eliminate the few and sporadic divergences, as the simple solution of increase adapt_delta is already used. data { …