Training ML models on tails

Research and presentation material on the topic of instance filtering in machine learning.

Files

Formal analysis & graphs: some plots and the simple gain contest analysis between linear and quadratic generators Rmd source
Toy model: a simple illustration of the impact of filtering Rmd source
Simulations: simulations on the impact of filtering Rmd source
Portfolio backtesting: boosted tree-based portfolio construction (out-of-sample) Rmd source
Data in R format: 1,182 firms & 108 features over 162 months
Slides, working paper and published paper

The data and code are meant for research and pedagogical use only.