The Thermal Optimal Path methods originates from Non-parametric Determination of Real-Time Lag Structure between Two Time Series: the “Optimal Thermal Causal Path” Method, D.Sornette and W.-X. Zhou (2004).
It enables dynamic lead/lag analysis between two time series. Borrowed from the statistical physics literature on directed polymers, it is suited for short and noisy time series.
- Implementation of the original Thermal Optimal Path method:
- Compiled with Numba's JIT to achieve significant speedup on typical use cases
- Partition function computation
- Average path computation
- Mean squared error model for correlated and anti-correlated time series
Jupyter notebooks provide example uses cases.
The master branch on GitHub is the most up to date code.
Modified BSD (3-clause), as found in the LICENSE file.
We welcome feedback about usability and suggestions for improvements.