A regression-based valuation model applying the sabermetrics "Moneyball" question to IPL auction data: does price track performance, or reputation?
IPL franchises spend tens of millions of rupees per season bidding for players, but auction prices are driven by reputation, recent form, and bidding-war dynamics as much as by underlying on-field output. A franchise that can systematically identify players who are underpriced relative to their output gains a real competitive edge — the same data-driven scouting advantage made famous by the Oakland A's in baseball.
Two genuinely independent real sources, merged: ball-by-ball performance data from Cricsheet.org, and IPL auction prices (2013–2022) from a Kaggle dataset. There's no shared player-ID registry between the two — Cricsheet uses "DA Warner," Kaggle uses "David Warner." Built a surname + year + initial disambiguation layer to join them, and reported the match rate (620 of 969 records, 64.0%) honestly rather than hiding the gap.
Built a composite performance score per player-season (standardized batting runs + standardized bowling output), then regressed performance on log(auction price) to get each player-season's price-implied expected performance. The residual — actual minus expected — is z-scored across all eligible player-seasons to flag outliers in either direction.
A methodology correction made mid-build, not after: the first version used a naive performance ÷ price ratio, which is mathematically dominated by the denominator and surfaces "cheap players," not genuinely undervalued skill. Replaced it with the regression-residual approach above — the actual mechanism real sabermetrics value-over-replacement metrics use.