Skip to content

readme.md

Latest commit

 

History

History
175 lines (133 loc) · 6.37 KB

readme.md

File metadata and controls

175 lines (133 loc) · 6.37 KB

IPL Auction Analysis: Where Do the Crores Go?

The Big Picture

IPL teams spent ₹687 Cr on disappointing players from 2012-2025. Using ball-by-ball data to compute WAR (Wins Above Replacement), we identify which purchases wasted the most money.

Top Disappointments (by Wasted Money)

Total wasted: ₹728 Cr across 315 disappointing players

Year Player Team Paid Predicted Actual Wasted
2015 Yuvraj Singh DC ₹16.0Cr 7.3 -5.3 ₹15.5Cr
2018 Axar Patel PBKS ₹12.5Cr 5.9 -3.1 ₹11.1Cr
2014 Shane Watson RR ₹12.5Cr 9.9 0.9 ₹11.1Cr
2014 Rohit Sharma MI ₹12.5Cr 5.7 -3.3 ₹11.0Cr
2022 Ravindra Jadeja CSK ₹16.0Cr 7.5 -1.4 ₹10.9Cr
2018 Benjamin Stokes RR ₹12.5Cr 9.5 1.5 ₹9.9Cr
2018 Chris Morris DC ₹11.0Cr 8.5 0.6 ₹9.8Cr
2014 Shikhar Dhawan SRH ₹12.5Cr 3.4 -4.3 ₹9.4Cr
2023 Sam Curran PBKS ₹18.5Cr 6.3 -1.2 ₹9.3Cr
2022 Kane Williamson SRH ₹14.0Cr -2.7 -9.9 ₹8.9Cr
2018 Rohit Sharma MI ₹15.0Cr 1.2 -6.0 ₹8.9Cr
2018 Manish Pandey SRH ₹11.0Cr -2.2 -9.2 ₹8.6Cr
2022 Virat Kohli RCB ₹15.0Cr -2.8 -9.6 ₹8.3Cr
2022 Venkatesh Iyer KKR ₹8.0Cr 2.4 -7.5 ₹8.0Cr
2014 Gautam Gambhir KKR ₹12.5Cr 0.3 -5.9 ₹7.7Cr

Team Efficiency

Team N Spent WAR WAR/Cr
KKR 67 ₹235Cr 350 1.49
SRH 73 ₹236Cr 302 1.28
RR 66 ₹231Cr 293 1.27
RCB 66 ₹263Cr 332 1.26
CSK 55 ₹248Cr 293 1.18
PBKS 73 ₹275Cr 324 1.18
LSG 20 ₹97Cr 105 1.09
DC 78 ₹303Cr 309 1.02
MI 63 ₹255Cr 238 0.93

Backtest Results

How well can we predict player performance at auction time using only prior data?

Year N R² Rank ρ RMSE
2009 112 -0.30 0.07 5.8
2010 129 0.08 0.37 5.4
2011 102 0.21 0.44 5.1
2012 119 0.30 0.60 5.2
2013 109 0.31 0.61 5.6
2014 132 0.09 0.48 5.1
2015 105 0.16 0.54 5.6
2016 111 0.27 0.53 5.1
2017 115 0.13 0.43 5.5
2018 120 0.33 0.53 4.7
2019 123 0.14 0.33 5.4
2020 112 0.24 0.33 4.9
2021 133 0.34 0.47 4.4
2022 158 0.35 0.54 5.0
2023 143 0.12 0.41 5.5

Methodology

WAR Calculation

From 3,137 player-season records of ball-by-ball data:

Batting WAR = (runs - expected_runs - dismissal_penalty) / RUNS_PER_WIN

  • Expected runs = balls_faced × replacement_strike_rate
  • Dismissal penalty = dismissals × RUNS_PER_DISMISSAL (6.0)

Bowling WAR = (replacement_runs - runs_conceded + wicket_bonus) / RUNS_PER_WIN

  • Replacement runs = overs × replacement_economy
  • Wicket bonus = wickets × RUNS_PER_WICKET (6.0)

Replacement levels: 15th percentile SR (batting), 80th percentile economy (bowling)

Empirically Estimated Constants

We estimate WAR constants directly from IPL data using year-specific (contemporaneous) estimation:

Constant IPL Mean Range Derivation
RUNS_PER_WICKET 5.8 5.0-7.4 OLS regression: innings_runs ~ wickets_lost + overs_batted
RUNS_PER_WIN 16.0 11-20 Mean victory margin per season

Constants are estimated separately for each IPL season to capture year-specific game dynamics (e.g., 2023 had higher runs per wicket than 2022). The overall pooled IPL estimates are RUNS_PER_WICKET=5.72 and RUNS_PER_WIN=16.04.

Note on symmetry: RUNS_PER_DISMISSAL and RUNS_PER_WICKET are conceptually distinct (batting opportunity cost vs. bowling value), but the innings-level regression provides a single estimate for the marginal effect of a wicket on team total. We apply this symmetrically: a dismissed batter costs their team potential runs; a bowler taking a wicket saves their team runs. This is a modeling choice that simplifies the framework while remaining empirically grounded.

Wasted Money Calculation

For each player:

  1. Predict WAR using only info available at auction (lagged IPL WAR, T20I history)
  2. Compare to actual WAR
  3. Wasted = (predicted - actual) × price_per_WAR, capped at price paid

Implied market price: ₹1.01 Cr per WAR

Prediction Model

XGBoost with expanding window training:

  • Features: lag1/2/3 IPL WAR, T20I WAR, career totals, trends
  • Training: all prior auction+outcome data
  • Mean rank correlation: 0.28 (13 years evaluated)

Repository Structure

ipl_auction/
├── data/
│   ├── acquisitions/         # Auction price data
│   └── perf/                 # IPL/T20I WAR by season
├── scripts/
│   ├── auction/              # Scraping, assembly
│   ├── perf/                 # Performance data processing
│   ├── prediction/           # WAR prediction, backtest
│   ├── retrospective/        # Hedonic regression
│   └── shared/               # Utilities
├── tabs/                     # Output tables (CSV)
│   ├── disappointments.csv   # Sorted by wasted money
│   ├── team_efficiency.csv
│   └── retroactive_*.csv
└── .github/workflows/        # Auto-update README

Quick Start

# Full pipeline
make all

# Just analysis
python3 scripts/prediction/06_comprehensive_backtest.py
python3 scripts/prediction/07_economic_analysis.py

# Update README tables
python3 scripts/update_readme_tables.py

Data Sources

Source Coverage
Kaggle, Wikipedia, IPL official Auction prices 2008-2026
Kaggle ball-by-ball Match data 2008-2025
World Bank/RBI CPI for inflation adjustment

Limitations

  1. Missing years: 2024 auction data not available
  2. Selection bias: ~19% of auctioned players never fielded
  3. Private information: Teams have scouting data we don't

References

  • Rosen, S. (1974). Hedonic prices and implicit markets. JPE.
  • Scully, G. W. (1974). Pay and performance in MLB. AER.