What Is Feature engineering?

Feature engineering is the design, transformation, and preparation of raw data into model inputs. In systematic trading, it commonly converts price, volume, funding, order-book, fundamentals, or alternative data into lagged, rolling, normalised, or cross-sectionally ranked features.

Good feature engineering defines the calculation, data source, timestamps, missing-value handling, and applicable trading universe. Every transformation must use only point-in-time data and be fitted on the training portion of each evaluation fold. A feature is not validated merely because it is well engineered: its incremental predictive and economic value should be assessed with feature ablation and out-of-sample testing.

Research and literature

• Maximilian Christ, Nils Braun, Julius Neuffer, and Andreas W. Kempa-Liehr, Time Series FeatuRe Extraction on basis of Scalable Hypothesis tests, Neurocomputing 307, 2018.