stsb2

This is the homepage for structural time series, round 2. It implements a grammar defined over structural time series blocks. You can read more about the theory here.

This library is useful because it allows you to easily express interpretable yet complex time series models by reasoning about them as collections of semantically-meaningful objects, not individual temporally-linked sample nodes. The following are all valid stsb2 models:

import stsb2.stsb as sts

...

# model 1
rw = sts.RandomWalk(t1=100,)

# model 2
loc = sts.MA1(t1=100,)
position = sts.RandomWalk(loc=loc)

# model 3
trend = sts.GlobalTrend(t1=100,)
seasonal = sts.GlobalTrend(t1=100,).cos()
noise = sts.AR1(t1=100,)
sgt = seasonal + trend + noise

# model 4
log_vol_1 = sts.RandomWalk(t1=100)
log_vol_2 = MA1(t1=t1,)
vol = sts.changepoint(log_vol_1, log_vol_2, frac=0.6).exp()
price = sts.RandomWalk(t1=t1, loc=0.0, scale=vol, ic=0.0).exp()

You can see more examples on the project Gitlab.

Installation and license

pip install stsb2

If you just want the source code:

curl -O https://davidrushingdewhurst.com/stsb2/package/stsb2.tgz

or on the project Gitlab.

This code is released under GNU GPL 2. For inquiries about obtaining a proprietary license, you can contact me.

Documentation:

It’s here.