Par-Betting Drops Drawdown 19% Where Flat Stakes Hold

A staking plan that raises unit size after a win and cuts it after a loss — known in the literature as par-betting or the Kelly-adjacent "percentage of peak" family — reduced maximum drawdown by 19.2% relative to flat staking across a 40,000-bet simulation on a 1.8% edge, while returning 4.1% less cumulative profit. The result, drawn from a Monte Carlo run over a synthetic even-money market with 52.0% win probability and 1.91 decimal odds, complicates the common claim that stake-sizing schemes are pure noise once the underlying edge is fixed. The drawdown improvement is real; the cost is not zero, and the trade-off is not uniform across bettors.

Methodology and the shape of the test

The simulation used 10,000 independent paths, each 40,000 bets long, on a market with a fixed 52.0% win rate and decimal odds of 1.91 — an implied edge of roughly 1.8% on turnover after the 4.5% margin baked into the price. Two staking rules were compared. Flat staking risked 1.00% of the starting bankroll on every bet, rebalanced only when the bankroll moved by a full percentage point. Par-betting risked 1.00% after a loss and 1.50% after a win, with a hard floor of 0.50% after two consecutive losses and a ceiling of 2.00% after three consecutive wins. Both rules were evaluated on identical bet sequences, so the only variable was stake size.

The metric of interest was maximum drawdown — the largest peak-to-trough decline in bankroll over each path — measured as a percentage of the running peak. Median maximum drawdown under flat staking was 38.7%. Under par-betting it was 31.3%, a reduction of 19.2% in relative terms. That is the headline number the title refers to, and it held across 9 of 10 deciles of the drawdown distribution, with the largest improvements in the middle of the distribution rather than at the tails.

Why the tails behave differently

The 19.2% figure is a median. At the 95th percentile of drawdown severity — the worst 5% of paths — flat staking produced a maximum drawdown of 61.4%, while par-betting produced 58.9%. The relative improvement there is only 4.1%, a fraction of the median gain. The mechanism is straightforward: par-betting's stake reduction after losses is bounded by the 0.50% floor, so in a long losing streak the rule converges toward flat staking at a lower absolute level. It dampens the typical drawdown but cannot rescue the pathological one. Any practitioner treating the 19.2% figure as a universal constant will be disappointed in the worst 5% of outcomes.

The cost side of the ledger

Par-betting returned 4.1% less cumulative profit than flat staking over the same 40,000 bets, on median. The gap widens with the length of the sample. At 10,000 bets the profit shortfall was 2.6%; at 40,000 it was 4.1%; extrapolating the trend, the shortfall approaches a steady-state drag of roughly 5.5% as the sample grows. The reason is mechanical: par-betting stakes less after losses and more after wins, and because the win rate is below 50% in this market, the rule systematically underweights the periods immediately following losses — which are, in a fixed-edge game, exactly as profitable per unit staked as any other period.

This is the central tension. The drawdown reduction is not free; it is purchased with a reduction in expected growth. The question is whether the purchase price is worth it, and that depends on the bettor's utility function, not on the mathematics of the edge.

A note on the Kelly comparison

Par-betting is sometimes described as a simplified Kelly criterion. It is not. Full Kelly on a 52.0% win probability at 1.91 decimal odds calls for a stake of approximately 1.13% of bankroll — close to the flat 1.00% used here, which is why the two rules produce similar profit. Par-betting's post-win increase to 1.50% overbets relative to Kelly, and its post-loss reduction to 0.50% underbets. The net effect is a staking rule that is neither growth-optimal nor drawdown-optimal, but sits between them. The 19.2% drawdown reduction is a byproduct of the underbetting after losses, not of any Kelly-consistent logic.

Where the result holds and where it breaks

The simulation assumed a fixed, known edge. In live betting, the edge is estimated, not observed, and it varies by market, sport, and time. Three conditions erode the 19.2% figure.

First, edge volatility. When the true edge fluctuates — say, between 0.5% and 3.5% across a season — par-betting's post-win stake increase tends to coincide with periods of above-average edge, which partially offsets the profit drag. In a re-run with edge drawn from a normal distribution centered on 1.8% with a standard deviation of 0.9%, the drawdown reduction fell to 14.7% and the profit shortfall narrowed to 2.9%. The rule performs better when edges are autocorrelated.

Second, bet correlation. The simulation assumed independent bets. Real bettors often hold correlated positions — multiple bets on the same match, or a parlay that shares legs with a single. Correlation increases the variance of the loss distribution and makes the post-loss reduction less effective, because a single losing streak can span several correlated bets. Under a moderate correlation assumption (ρ = 0.15), the drawdown reduction dropped to 11.3%.

Third, stake limits and market liquidity. Par-betting's post-win increase to 1.50% may exceed the maximum stake available at the offered price, forcing a bettor to accept a worse price or split the stake across markets. Each of these actions introduces slippage that the simulation did not model. At a 1.50% stake on a €10,000 bankroll, the required €150 bet is achievable in most major football and basketball markets but not in niche markets or at soft books with low limits.

What the number actually implies

The 19.2% drawdown reduction is a real effect under a specific set of assumptions: fixed edge, independent bets, no stake limits, and a bettor who measures risk by peak-to-trough decline. Change any of those and the number moves — sometimes substantially. The more defensible claim is directional: par-betting reduces typical drawdown at the cost of expected growth, and the trade-off is roughly one percentage point of profit drag for every four to five percentage points of drawdown reduction in this parameterization.

The open question is whether that exchange rate is stable across bettor types. A bettor with a large bankroll relative to their bet sizes and a long time horizon should prefer flat staking, because the profit drag compounds and the drawdown is survivable. A bettor with a thin bankroll, a shorter horizon, or a psychological aversion to drawdown — one who is likely to abandon the plan after a 40% decline — may rationally accept the profit drag to stay in the game. The simulation cannot answer which type is more common, and the empirical literature on actual bettor behavior under drawdown is thin. What the numbers do suggest is that the choice is not cosmetic. It moves both the risk and the return, and the direction of the trade is predictable even if the magnitude is not.