คุณควรเดิมพันจริง ๆ เท่าไร? การวางเดิมพันแบบ Kelly ทดสอบกับ value bet จริง 152 รายการ

Our accumulator study ended with one survivor: value signals, played as singles. That raises the only staking question that matters: ควรลงเงินต่อเดิมพันเท่าไร? We tested it on our real record — 152 settled value bets (84 won, 68 lost; 55.3% ที่อัตราต่อรองเฉลี่ย 2.43, จุดคุ้มทุน 41.2%) — and then stress-tested every scheme the honest way.

The seduction: what actually happened

Starting bankroll 1,000 lei, compounding, on the real historical sequence:

Read that table and every instinct says: bet more. This is exactly how bettors talk themselves into oversized stakes after a good month — so we ran the month again. Five thousand times.

The truth: 5,000 alternative universes

Monte Carlo, same 152 bets, each winning with its fair probability (i.e., assuming our measured edge is real):

Two things demolish the "bet more" instinct:

1. The median barely moves. Tripling the stake from 1% to 3% moves the typical outcome from 1,056 to 1,091 — a 3% improvement for 3× the risk. All the extra profit in the historical table above was luck multiplied, not edge multiplied.
2. The tail explodes. At 1% flat, a 30%+ drawdown is a 2% event. At 3%, it is 77% — near-certain, even with a real edge. And a 30% drawdown is where real humans abandon good processes. (In the scenario where our edge is an illusion — never excludable at 152 bets — 3% staking ends below water in 59% of universes.)

Why 3% is the cliff edge

Our average full-Kelly fraction across these bets is 2.9%. Flat 3% staking is therefore full Kelly — the theoretical maximum growth rate if the edge estimate is exactly right. Edge estimates from 152 bets are never exactly right, and overestimated-edge Kelly means systematic overbetting: less growth and more risk. That is why professionals bet a fraction of Kelly, not Kelly.

The doctrine we actually run

1. Flat 1% per signal (≈ quarter-Kelly for our edge sizes) — keeps ~95% of the achievable median at a tenth of the ruin risk
2. Daily exposure cap ~10% — our busiest days produce 10–16 simultaneous signals; uncapped, that is a sixth of the bankroll on one afternoon
3. Half-Kelly is a promotion to be earned, not chosen: only after ~500 settled bets with a confidence interval on CLV that excludes zero
4. Never raise stakes after losses — an 8% drawdown is part of the process, not evidence against it

The most useful sentence in this entire study: position size cannot create an edge, but it can absolutely destroy one. Find the edge first (here is how we measure ours), then bet small enough to survive being wrong about it.

Stress-test your own record — sample size, luck probability, the works — with our free tipster audit tool.