Survivorship in the training universe
Question
You train and backtest using only stocks that remain listed today. Why can this distort the measured value of an ML signal?
Short answer
The universe excludes firms that failed, merged or delisted, using future survival information to choose historical observations. This can change returns, feature distributions and apparent predictability. Reconstruct membership and investability at each historical date, including delisting outcomes when relevant. Assess missing-data and vendor-coverage rules as well; adding a chronological split does not correct a future-selected universe.
Original question and answer written by SelfMadeQuant.com.
See the hint and the full worked solution in the question bank (free account).
Open the worked solution