IPO Return Prediction: A Systematic Review on Methods, and the Case for Auditable Scorecards

Abstract

Abstract—Predicting the direction of an initial public offering (IPO) first-day return is difficult because the firm has little history and disclosure sits in a single prospectus. This review synthesises recent predictive modelling of IPO outcomes by horizon, method family, and three evaluation dimensions: validation design, data leakage and class imbalance. Of the 42 studies read at full text from Scopus and IEEE Xplore, a two-condition rule on outcome and model retains 37; evaluation quality is deliberately not a condition, so studies never evaluated out-of-sample are reported rather than screened out. Ensembles and textual methods outperform linear baselines in-sample, yet only six of the 37 are validated in time order. The binding weakness is a missing comparator: no benchmarks against a majority-class rule (always predicting the more common outcome), so most models cannot be shown to beat a naive baseline; a minimum future work should report. For retail-dominated emerging markets, we identify the auditable credit-style scorecard as an open direction: a Weight-of-Evidence points table that is deterministic, so a subscriber or regulator can reproduce and verify it, a property interpretable but stochastic rule systems do not provide.

Keywords—initial public offering, machine learning, first-day return, systematic literature review, data leakage, class imbalance, credit scorecard, Weight of Evidence, auditability