Calcutta Portfolio Backtest
Would any betting strategy have made money?
Historical backtest of the 2024 & 2025 auctions · generated 2026-06-10 (US Eastern)
Bottom line: No price-rule strategy made money with statistical
confidence. A few posted positive two-year averages (mid-price +4.0%,
buy-cheapest +1.3%), but every CI runs from roughly
−55% to +73% and the two years flip signs.
The standout — human-intel/sandbaggers +77%, 71% hit — rests on
just 7 picks (CI ≈ [−45%, +192%]) and two flight wins: a lead to
track, not a proven edge. Expected ROI of a disciplined one-team-per-flight price
rule is near zero; realistic single-year range −50% to +30%.
The one robust, repeatable finding is a don't: buying cheap longshots in the richest flights
lost −78% (−100% in 2025 — zero of six cashed).
-10.9%
Random portfolio mean ROI
32%
Random portfolios that profited
100%
Name match rate (both years)
Verified mechanics
- Payout structure. Within each flight, 70% of the payout pool goes to the winner
and 30% to the runner-up; 10% of the pot is directed to the shootout. Per-flight payouts reconcile to
0.9×pot to the dollar.
- Tie-aware payouts. Winner ties (9 flights) pool 1st+2nd money split evenly;
runner-up ties (5 flights) split 2nd money. Handled explicitly — they materially move payouts.
- Match rate 100% (108/108 in 2024,
90/90 in 2025) via surname-pair matching with a fuzzy
fallback (caught Shirely→Shirley).
Strategy battery — combined 2024 + 2025
One team per flight unless noted. "Combined" = equal-dollar across all 33 picks
pooled over both years. CI = across-year block bootstrap (20,000 resamples) — deliberately
wide to reflect that we have only two years. Buy-all shown price-weighted;
Rich-6 strategies use the 6 highest-pot flights (our 2026 slate logic applied retroactively).
Monte-Carlo: random portfolios
Drawing one random team per flight, 50,000 times, the combined mean ROI is
-10.9%, with a 95% range of
[-56%, +40%]. A random portfolio
turned a profit 32% of the time.
Roughly a third of random portfolios made money, so a
two-year positive result from any named strategy is not evidence of an edge.
2026 slate logic, applied retroactively
- Rich-6 Cheapest (cheapest team in the 6 richest flights):
−55% (2024), −100% (2025), −78% combined.
- Rich-6 Favorite (favorite in the 6 richest flights):
−41% then +63%, +11% combined — best hit rate among rich-6
plays; the swing across years reflects variance, not skill.
- Human-intel / sandbaggers (Williford, Copeland, Wood+Estes, Rekenthaler+Stafford):
+10% then +127%, +77% combined, 71% hit rate — the best result in the
battery. But 7 picks over two years, dominated by two flight wins (Williford 2024,
Copeland & Wood+Estes 2025); CI ≈ [−45%, +192%]. Williford flipped from flight
winner (2024) to 5th (2025) — exactly the sandbagger volatility the intel doc flags.
Verdict for the partner
- Expected ROI is near zero. Best point estimates for the
stronger strategies are 0% to +5%, inside the noise.
- Realistic single-year range: −50% to +30%.
- No edge is statistically established. Every positive strategy's CI includes
large losses.
- One robust rule, and it's a "don't": never buy cheap longshots in rich flights.
- Watch, don't bank: human intel/sandbaggers (+77% on 7 picks) is the only signal
worth tracking forward, not yet bankable on two years of data.
Data limitations
- Prices exist for 2024 & 2025 only — full P&L is a two-year backtest.
2021–2023 are results-only context.
- Tiny sample: 33 one-team-per-flight picks per strategy. Bands are wide by necessity.
- Surname matching across files (100% here, but rests on within-year uniqueness).
- 'Value = price/(pot/6)' is mathematically identical to buy-cheapest (pot/6 is
constant within a flight); we added a seed-implied fair-value variant as a genuine value test.