We tested whether Bitcoin 15-minute prediction markets on Kalshi exhibit predictable mean-reversion behavior. The core hypothesis: prices in these short-duration markets tend to overreact early in the contract window, and buying when the price dips below a defined entry band — then exiting once it returns inside that band — captures that reversion profitably.

We ran 100 fully automated variants exploring different entry bands, price floors, and risk guardrails. The markets traded are Kalshi’s KXBTC15M series (BTC, 15-minute expiry). The strategy type is pure mean reversion, polling every 60 seconds. This report covers what worked, what didn’t, and why the band matters more than the floor.

Variant and Strategy Explanation

All variants share a common structure derived from a base YAML strategy definition:

Platform: Kalshi
Market: KXBTC15M (BTC 15-minute binary contracts)
Strategy: Mean reversion
Loop interval: 60 seconds
Risk controls: Max position of 10 contracts, price ceiling of 0.85 (no buying above that), and a cooldown period of 30 seconds between trades

The base configuration used a price floor of 0.15 and an entry band of 0.30 to 0.60. Entry logic: if the contract price falls below the entry band, the strategy buys. Exit logic: if the price rises back into or above the entry band, the strategy exits the position.

From this base, we tested 100 variants by adjusting two key parameters:

Entry band lower bound — values sampled around the base 0.30, including 0.20, 0.25, 0.30, 0.35, and 0.40
Price floor — values sampled around the base 0.15, including 0.05, 0.10, 0.15, and 0.20

Some variants also adjusted the upper bound of the entry band (e.g., 0.55, 0.60, 0.70). The combination space covered wide vs. narrow bands and low vs. moderate floors. Each successful variant has been saved as a runnable Turbine strategy.

The research question was straightforward: which combinations of entry band tightness and floor protection produce the best risk-adjusted returns in historical simulation, and which reliably lose money?

Top Results

The best-performing variants share a clear signature: entry bands that start at 0.30 or lower, capped at 0.55, with a price floor of 0.20. The top 10 variants all posted Sharpe ratios above 1.0 and ROI figures exceeding 500%.

Rank 1 (Top Performer)

Label: BTC mean reversion · band 0.30–0.55 · floor 0.20
ROI: 655.4%
Sharpe: 1.18
Total PnL: $65.54
Total Trades: 76
Win Rate: 52.6%
Max Drawdown: -19.58%

Rank 2

Label: BTC mean reversion · band 0.25–0.55 · floor 0.20
ROI: 586.9%
Sharpe: 1.08
Total PnL: $58.69
Total Trades: 74
Win Rate: 51.4%

Rank 3 (tied with Rank 4)

Label: BTC mean reversion · band 0.30–0.55 · floor 0.10 (and 0.15)
ROI: 573.8%
Sharpe: 1.09
Total PnL: $57.38
Total Trades: 88
Win Rate: 45.5%

Ranks 5–8

The next tier widens the entry band lower bound to 0.20–0.25 or keeps band 0.30–0.55 while reducing the floor to 0.05–0.15. These variants still produce strong returns (ROI 505–555%, Sharpe 1.0–1.28) but trade more frequently and have slightly lower win rates (43–51%). The Rank 6 variant (band 0.20–0.55, floor 0.20) shows the highest Sharpe in the group at 1.28 with a softer drawdown of -17.08%.

What the top group tells us: The band matters most. A lower entry bound of 0.30 or below captures overreactions cleanly. Capping the band at 0.55 rather than 0.60 keeps the exit discipline tight. The floor provides moderate improvement — 0.20 edges out lower floors on risk-adjusted returns, but the difference is not enormous.

Bottom Results

The worst-performing variants consistently used entry bands starting at 0.35 or 0.40 — waiting too long for a deeper dip that either doesn’t come or doesn’t revert enough before contract expiry.

Rank 100 (Worst Performer)

Label: BTC mean reversion · band 0.35–0.60 · floor 0.05
ROI: -113.3%
Sharpe: -0.56
Total PnL: -$11.33
Total Trades: 80
Win Rate: 20.0%
Max Drawdown: -21.03%

Ranks 98–99

Label: BTC mean reversion · band 0.35–0.60 · floor 0.10/0.15
ROI: -86.5%
Sharpe: -0.43
Win Rate: 21.6%

Ranks 94–97

Band 0.35–0.70 and band 0.40–0.60 variants produced similarly poor outcomes, with ROIs ranging from -27.7% to -72.3%, Sharpe ratios deeply negative, and win rates stuck around 18–24%. The band 0.35–0.70 variants traded fewer times (60–66 trades) but still bled capital, suggesting the wide band simply fails to capture enough mean-reversion events before expiry.

What the bottom group tells us: Starting the entry band at 0.35 or higher is too conservative. By the time price falls that far in a 15-minute window, the remaining time is often insufficient for reversion. The floor setting doesn’t rescue these variants — even a 0.05 floor can’t fix bad entry timing. The market overreactions that drive this strategy’s edge occur above these levels.

Conclusion

Historical simulation across 100 variants confirms that the original thesis holds — but only within specific parameter ranges. Mean reversion in KXBTC15M markets is real and exploitable when you buy early enough. Entry bands bounded by 0.30 (lower) and 0.55 (upper) consistently produced strong risk-adjusted returns. Dropping the lower bound to 0.20 or 0.25 maintained profitability and slightly improved Sharpe in some cases, at the cost of more frequent trading.

The price floor is a secondary factor. A floor of 0.20 modestly improved drawdown characteristics in the top variants, but floors as low as 0.05 still performed well when paired with the right band. The floor is insurance, not the engine.

The strategy breaks down when entry bands start at 0.35 or above. Those variants waited for deeper dips that either didn’t occur or didn’t reverse in time. Win rates collapsed below 25%, turning a consistently positive edge into a reliable loser.

For practitioners reviewing these results: the saved Turbine strategies represent concrete, runnable implementations of the top variants. They are not predictions. They are historical parameterizations that have been stress-tested across the full variant space. The narrow band of success — 0.25–0.30 entry, 0.55 exit, any reasonable floor — suggests a genuine market inefficiency rather than curve-fitting, but only live trading can confirm whether this pattern persists.

Long Disclaimer

This report contains historical simulation results generated by Turbine’s automated research framework. All performance figures — ROI, Sharpe ratio, total PnL, win rate, and maximum drawdown — are derived from backtesting against historical market data and do not represent live trading results.

No future performance is implied or guaranteed. The strategies described herein may perform differently, including incurring substantial losses, in live market conditions. Market dynamics change. Liquidity conditions, fee structures, and execution characteristics in live trading may differ materially from the assumptions embedded in these simulations.

This report does not constitute investment advice, a solicitation, or a recommendation to buy or sell any financial instrument. Prediction markets involve significant risk, including total loss of principal. Past simulated results are not indicative of future outcomes.

All trading strategies are subject to market risk, execution risk, and model risk. The variants described have not been tested out-of-sample or in walk-forward analysis beyond the historical period covered. Users of Turbine, or any trading strategy generated by it, assume full responsibility for their trading decisions and outcomes.

Each successful variant identified in this report has been saved as a runnable strategy within the Turbine platform for research continuity and potential future evaluation only.

This report is generated from historical simulations. Backtests can be wrong or incomplete, and live trading can differ materially because of liquidity, fees, slippage, latency, market resolution, outages, and data quality. Do your own review before running any strategy.