BTC Rebate MM

Short disclaimer

This analysis uses historical simulation data only and does not forecast future performance. All trades were executed in a test environment on historical Kalshi data for the KXBTC15M market. Actual live trading will differ.

Intro / thesis

We investigated whether a pure market-making strategy on the Kalshi KXBTC15M (a crypto 15-minute binary outcome market) can achieve positive total P&L by combining post-only limit orders with the platform’s Volume Incentive Program maker rebates. The core hypothesis: if quote parameters are tuned to balance fill rate against adverse selection, the spread capture plus rebate stream should consistently outpace the occasional losses from being picked off when the market moves against resting quotes.

We swept 100 parameter variants to find the sweet spot—testing quote width, order depth, refresh behavior, and time-to-expiry gates. The results confirm the thesis, with several variants producing substantial total P&L and remarkably high win rates. This report lays out how the strategy works, what drove the top results, and where the bottom variants fell short.

Variant and strategy explanation

All variants rest on a common base structure defined in DSL:

• Strategy type: spread_capture on KXBTC15M.
• Risk controls: Max position of 50 contracts, price floor at 0.05, price ceiling at 0.95, max loss capped at $25.
• Execution loop: 10-second interval between quote evaluations.
• Base parameters (overridden per variant):
  • spread_floor: Minimum distance between bid and ask as a percentage of price.
  • order_count: Number of price levels placed on each side.
  • post_only: true: Orders are always maker-only, never cross the spread.
  • refresh_on_fill: true: When one side fills, quotes are immediately refreshed to maintain depth.

Each variant tweaks one or more of the following:

• Quote width (spread_floor): Ranges from very tight (0.01) to wide (0.05), changing how aggressively the strategy competes for fills and how much spread it earns per fill.
• Order depth (order_count): More levels mean higher potential inventory accumulation but also more rebate-generating fills if resting orders are hit.
• Refresh behavior: Whether quotes are replaced on fill, which affects how quickly the strategy gets back to full depth after an execution.
• Time-to-expiry gating: Some variants limit quoting to specific windows before expiry to avoid thin, volatile markets or to concentrate on periods of higher trading activity.

The goal was to find a parameter set where the combination of (spread earned + maker rebates) exceeds losses from adverse price moves against filled positions, expressed in total P&L, ROI%, Sharpe ratio, and max drawdown.

Top results

The leading variants cluster into two clear performance tiers. The top four are nearly identical in outcomes, and the next four form a higher-volume, higher-P&L group with a larger drawdown.

Key takeaways from the top tier:

• Extraordinary win rates: All top variants sustained a 99.02% win rate across tens of thousands of trades. The strategy almost never takes a directionally losing bet—it just earns spread and rebates.
• High ROI, controlled drawdown: The first four variants turned a $15 starting risk (max loss $25 implied base) into a $254.76 total P&L with a max drawdown of only -$10.35. That is a very smooth equity curve.
• Higher volume, higher absolute P&L: Variants 097–100 ramped up trade count to 62,377 and hit $407.22 total P&L. The trade-off: max drawdown increased to -$17.26, and Sharpe dipped slightly to 1.10, but the overall risk-adjusted return remained strong.
• Parameter sensitivity: The near-identical clustering among 089–092 and 097–100 suggests that small parameter variations around the sweet spot produce nearly indistinguishable results, indicating a robust parameter region rather than a fragile single-point optimum.

Each of these successful variants is saved as a runnable Turbine strategy, so they can be deployed live with the exact same parameterization that produced these historical results.

Bottom results

The worst-performing variants still earned positive P&L but with significantly lower totals and lower Sharpe ratios. There are two notable bottom clusters:

Why these trailed:

• Lower trade count: Variants 001–004 executed only ~9,900 trades over the test period versus ~38,000 for the top tier. Fewer fills mean less spread and rebate capture, directly capping P&L.
• Lower win rate: At 96.57%, these variants still win the vast majority of trades, but the slightly higher loss rate eats into returns. The 2.5 percentage-point drop in win rate versus the top tier translates into a sharp drop in total P&L—from $254 to $42.
• Weaker Sharpe: Sharpe ratios of 0.69–0.80 indicate meaningfully lower return per unit of risk. The drawdowns are smaller in absolute terms, but the return erosion is disproportionate.
• Parameter weakness: These bottom variants likely used wider spreads that discouraged fills or overly restrictive time-to-expiry gates that reduced quoting windows. Less quoting activity compounds into less rebate income, and wider spreads increase the per-trade spread but don’t compensate for the lost volume.

Conclusion

Historical simulation confirms that post-only market making on KXBTC15M can be a reliable positive-expectancy strategy. The top parameter sets produce exceptional risk-adjusted returns: total P&L of $254–$407, ROI% exceeding 1,600%, Sharpe ratios above 1.1, and win rates topping 99%, all while keeping max drawdown under $17 on a $25 risk framework.

The research clearly demonstrates that:

• Balance is critical: Too few trades (from overly conservative parameters) kneecaps P&L even with good per-trade economics. Too many trades (from overly aggressive quoting) can inflate drawdowns.
• A stable optimum exists: The dense cluster of top variants means the strategy is not fragile—practitioners have a workable range of parameters, not a single brittle point.
• Lowest variants still profit: Even the worst performers were solidly green, indicating the core thesis holds across a wide parameter band, just with lower efficiency.

These findings are purely based on historical backtest data. They do not guarantee future results, and live market conditions—liquidity shifts, rebate program changes, volatility regimes—can alter the outcome materially.

Long disclaimer

This report is a historical simulation research output only. Nothing in it constitutes investment advice, a trading recommendation, or a prediction of future performance.

Key limitations:

• Simulated fills: The backtest assumes all post-only orders are filled exactly as placed with no latency, queue position, or partial-fill effects that occur in live markets. Actual fill rates may be lower.
• Rebate assumptions: The Volume Incentive Program terms are assumed constant throughout the test period. Changes to maker rebate rates would directly impact net P&L and are not modeled here.
• Market impact: The simulation does not account