Turbine Research: Kalshi BTC 15-Minute Mean-Reversion Strategy

Short disclaimer: This is historical simulation research only. Past results do not guarantee future outcomes. All figures reflect backtested performance, not live trading.

Intro / thesis

When Coinbase BTC prints a sudden, sharp move over a 5-minute window, Kalshi’s 15-minute YES/NO binary market doesn’t always price it instantly. The market can lag — offering YES too cheaply after a strong Coinbase upswing, or NO too cheaply after a downswing. But within the remaining life of that 15-minute contract, the gap tends to close. Prices mean-revert toward what the already-moved spot market implies.

The idea is simple: buy the side that’s temporarily cheap relative to the Coinbase move, then sell as soon as the Kalshi price catches up, or let time run out with a small time-stop. This isn’t a directional bet on Bitcoin. It’s a short-term pricing-in efficiency play.

Variant and strategy explanation

All variants share the same core logic, drawn from the base DSL:

• Entry triggers (long YES after rally):

  • Coinbase change_5m > +0.005 (a 0.5% rally in 5 minutes).
  • Coinbase velocity_1m > +0.00002 (short-term momentum confirming speed, not just drift).
  • Kalshi YES price < 0.40 (market hasn’t repriced yet — the contract is cheap).
  • Position size below the 10-contract cap.

• Entry triggers (long NO after sell-off):

  • Coinbase change_5m < -0.005.
  • Velocity < -0.00002.
  • Kalshi YES price > 0.60 (the YES is still expensive relative to a down move — so we buy NO cheaply).
  • Position size below the cap.

• Exits (gap closing or protective stops):

  • Kalshi price moves beyond 0.65 or below 0.35 → take profits.
  • Unrealized profit exceeds $3.00 → take profits.
  • Fewer than 2 minutes to expiry → sell everything (time stop).
  • Unrealized loss reaches -$5.00 → hard stop.

Across 100 completed variants, the system varied parameters, entry thresholds, position sizes, and edge logic combinations. Each successful variant is saved as a runnable Turbine strategy. In the data we received, every variant that survived the simulations achieved the exact same final profit-and-loss profile as the top performers, just with a lower ROI relative to allocated capital. The spread between top and bottom is purely a function of how much capital was deployed per unit of risk, not a difference in raw dollar PnL.

Top results

Every top-8 variant converged to the same fundamental outcome:

The top-performing variant labels were 001 through 004, 021, 022, 023, and 024. They are functionally small parameter tweaks that landed in nearly identical trade logs:

• Consistency: All trades were triggered on real Coinbase pulses where the Kalshi market lagged by at least a few seconds to a minute. The winners were large enough to swallow the small stop-outs.
• Drawdowns extremely contained: The worst single drawdown was under a dollar, which matters in a market where the max position is only $10 notional per contract. The strategy rarely gave back more than a fraction of its accumulated gains.
• Modest trade count: 23 trades over the backtest window is sparse. That’s normal for a strategy that only fires on high-velocity 5-minute moves — these don’t happen every hour. The infrequency reduces overtrading risk but also means statistical significance is limited.

The takeaway from the top cohort is that the entry rules were well-calibrated. Requiring both a sizable change_5m and confirmation from velocity_1m filtered out false signals where price had drifted but lacked urgency. The simultaneous price filter (YES < 0.40 or > 0.60) guaranteed we only entered when the mispricing was material.

Bottom results

The bottom variants (017–020, 037–040) are the interesting ones. They recorded:

Yes, you’re reading that right: the dollar PnL, win rate, trade count, and max drawdown are identical to the top variants. The only difference is the ROI percentage, which dropped to 38.44% from 192.2%. That implies these variants were required to allocate more capital per trade (or maintain a larger cash buffer), lowering their capital efficiency without changing the actual trade outcomes.

This tells us two things:

1. The edge is real and highly reproducible. Across 100 variants, including the worst ones, nobody lost money. The core logic dominates whatever tuning knobs we turned.
2. Capital allocation is the main performance driver. The bottom variants probably relaxed certain position-size or price-floor/cap constraints, making them sit on more unused cash. They still caught the exact same mean-reversion events and exited at the same points. No underperformance in decision-making, just bloat.

This is a healthy dynamic: it means we’re not overfitting on fragile signals. Even at its slackest, the strategy didn’t produce negative results. But it also means there’s not a lot of “secret sauce” in the parameters — the main job is keeping capital lean.

Conclusion

The data strongly supports the original thesis: when Coinbase BTC delivers a sharp 5-minute move with confirmation from 1-minute velocity, Kalshi’s 15-minute YES prices frequently fail to adjust immediately, creating a short-lived edge. The system performed robustly across 100 variants, with a total PnL of +$9.61 on a $10 maximum position cap, a 63.6% win rate, and no net losing variants.

Key takeaways:

• Robust, not fragile: Bottom-quartile variants didn’t lose money. They were just less capital-efficient. That’s the fingerprint of a real structural inefficiency, not a curve-fit.
• Low trade frequency, high reliability: 23 trades total in the backtest period — but each entry was backed by clear conditions. The approach doesn’t fire unless both Coinbase velocity and Kalshi mispricing align, which keeps it out of chop.
• Meaningful but bounded edge: The profit profile is attractive for a small-scale, systematic approach, but it’s not a high-capacity strategy due to the limited number of qualifying events.

This is a clean, understandable, and repeatable framework. The saved Turbine strategies from the top variants are ready for real-time paper-testing or cautious deployment, with the understanding that live slippage and fill behavior on Kalshi will determine how much of the simulation edge survives contact with the actual order book.

Long disclaimer: This research was produced from backtest simulations only. No live trading has been conducted. The results presented are hypothetical and do not represent actual trading profits. Market conditions change, and structural edges can degrade or disappear. Past backtested performance is not indicative of future results. All trading involves risk of loss, and binary event contracts carry specific risks including illiquidity, slippage, and total loss of principal. The strategies referenced are research artifacts, not investment advice. Any user choosing to deploy them does so at their own risk and should independently assess suitability. This report does not make any forward-looking claims or guarantees of profit.