Research Report: Kalshi KXBTC15M Momentum Continuation Sweep

Short Disclaimer

Historical simulation only. No statement herein implies future profitability. Past results do not guarantee future performance.

Intro / Thesis

This report covers a completed 100-variant parameter sweep on Kalshi’s 15-minute Bitcoin markets (series ticker KXBTC15M). The core idea was momentum continuation: buy YES when spot BTC is trending up on short timeframes and buy NO when it is trending down, with entries restricted to the final 1–7 minutes before expiry.

The sweep focused on two risk parameters — price floor and price ceiling — while holding the entry, exit, and edge-feed logic constant. The goal was to understand how position entry filtering by contract price affects net P&L, drawdown, Sharpe, and win rate across a wide range of market conditions.

Variant and Strategy Explanation

Each variant is a modified version of the base strategy with different allowed contract price bands. The base strategy:

• Loops every 10 seconds
• Uses Coinbase BTC-USD data refreshed every 5 seconds
• Enters YES when spot is above both the 20-period SMA and 1-hour VWAP, with positive 5-minute change and positive 1-minute velocity
• Enters NO on the mirror setup: spot below both SMA-20 and VWAP, with negative 5-minute change and negative 1-minute velocity
• Requires a spread of 0.03 or less
• Restricts entries to the final 1–7 minutes before expiry
• Uses fixed order size of 50 contracts, max position 150
• Exits all positions at +$4.40 unrealized P&L (take profit) or -$11.00 (stop loss)

The sweep varied two parameters:

• price_floor: 0.05, 0.09, 0.14, 0.18, 0.23, 0.27, 0.32, 0.36, 0.41, 0.45
• price_ceiling: 0.55, 0.59, 0.64, 0.68, 0.73, 0.77, 0.82, 0.86, 0.91, 0.95

This gave 100 total cells, all of which completed successfully. Each successful variant has been saved as a runnable Turbine strategy.

Top Results

The top eight variants by net P&L all cluster in the upper-middle range of price ceilings (0.73–0.86) and lower price floors (0.05–0.14). The top performer used a floor of 0.09 and ceiling of 0.77.

Top variant metrics:

Several observations stand out. First, the top results are tightly packed — the spread between rank 1 and rank 8 is only about $116, or roughly 3.3% of the top P&L. Second, a price ceiling of 0.73–0.77 appears most frequently in the top tier, paired with low floors. Third, the higher-ceiling variants (0.86) show higher win rates but lower Sharpe and lower total P&L per trade — they trade more often but with thinner average edge.

Bottom Results

The lowest-ranked variants cluster around the 0.55–0.59 price ceiling with either very low floors (0.05) or moderately high floors (0.45). The bottom eight by net P&L:

The most striking bottom performer is rank 96: floor 0.05 with ceiling 0.55 produced a 36.4% win rate and $457 max drawdown — far worse than any other variant. This combination allowed many cheap, low-quality entries while capping upside. It traded 1,899 times but barely outperformed the much more selective rank 100, which traded only 503 times.

It’s worth noting that even the worst variant was profitable in absolute terms. The sweep didn’t find a losing parameter combination — it found a wide range of profitability with meaningful differences in efficiency.

Conclusion

The parameter sweep produced a clear pattern: ceilings around 0.73–0.77 combined with floors between 0.05 and 0.14 delivered the best balance of total P&L, Sharpe, and drawdown. Ceilings at or below 0.55 materially constrained profitability, and floors above 0.32 tended to reduce trade count without improving quality enough to compensate.

However, the robustness statistics require caution. The permutation test produced a p-value of 0.010, meaning the edge-feed timing beat 99.0% of 100 time-scrambled re-sweeps. While that passes the conventional 0.05 threshold, the deflated Sharpe is 0.996, which sits above the expected max Sharpe of 0.605 — but not dramatically so. The neighborhood degradation of 0.022 suggests that nearby parameter cells performed similarly, which is a mild positive for stability.

The raw ROI figures are extremely high by any standard. This is typical of short-dated binary markets with fixed take-profit and stop-loss rules, where a small per-trade edge compounds across thousands of small trades. Those figures should not be extrapolated to live performance.

The sensible read: the strategy family shows a consistent edge in simulation, but the top result is not strongly distinguishable from its neighbors. The signal is plausible but not overwhelming. Any live deployment should start small and treat this as an unproven hypothesis in a noisy market.

Long Disclaimer

This report is a historical simulation study conducted for research purposes only. It is not investment advice, a recommendation to trade any specific market, or a guarantee of future results.

All performance figures are derived from backtesting using historical market data and a simulated trading engine. They do not represent actual trading activity. Backtested results are subject to numerous limitations, including but not limited to:

• Survivorship and selection bias: The strategy parameters were chosen after observing results, which can inflate apparent performance.
• Execution assumptions: The simulation assumes fills at specified prices and does not account for liquidity constraints, partial fills, or order book depth.
• Slippage and fees: Unless explicitly stated, results may not include realistic transaction costs