Regime Detection in financial time series.

Hidden Markov models analyse changes in data series in order to identify regime shifts. The purpose of the technique is therefore to monitor for changes in regimes in order to identify the difference between potential downturns or short term blips.

Given some large changes in housing markets recently, I thought it would be interesting to have a look at changes in the house price series for Sydney. House prices are also an interesting time series in this case since momentum is relatively persistent in house prices. I used CoreLogic Data, and as per usual, the code is available on github.

RegimeDetection1.png

 

While useful as a proof of concept, one thing that makes the results slightly more robust is calculating rolling year returns and using that as a time series. Unfortunately, using daily changes in prices means that downturns are very often described as regime changes - using longer periods allows you to identify more interesting turning points.

Certainly, if anyone has an accurate and long term house price data series, I would love to take a look at it!