top of page

The Knowable Future

Python Code: Quantifying Forecastability Through Predictive Information 
 

Auto-Mutual Information Analysis

Python implementation for estimating horizon-specific pairwise auto-mutual information (AMI). AMI measures the predictive information shared between Yt and Y(t+h), corresponding to a declared single-observation information set. It is one forecastability diagnostic, not a general estimator of all predictive information available from the full history.

 

The code can be applied to levels or transformed series. Any transformation changes the quantity being analysed where it changes the target process. In particular, AMI computed on first differences describes predictive information in the differenced process rather than an adjusted estimate of forecastability in the original levels.

The resulting profiles characterise horizon-specific dependence. Model choice requires separate out-of-sample evaluation because stronger measured dependence does not by itself identify which forecasting method will exploit it best.

 

Outputs:

​

  • Per-series AMI profile charts showing the raw series alongside its AMI profile

  • Classification and triage recommendation for each series

  • Structured results CSV

​

Dependencies:

​

  • Python 3.9+

  • numpy

  • pandas

  • matplotlib

  • seaborn

  • scikit-learn

Algorithm 1: KSG* Estimation of
Auto-Mutual Information 

bottom of page