Research Report: Kalshi BTC 15-Minute Momentum Confirmation Sweep

Short Disclaimer

This is a historical simulation research report. Past simulated performance does not guarantee future results. All findings are based on backtesting only and carry significant risk of overfitting.

Intro / Thesis

We tested whether requiring stronger momentum confirmation—specifically, raising the velocity thresholds on BTC and ETH from zero to higher values—improves trade quality on Kalshi's BTC 15-minute binary markets (KXBTC15M). The intuition: a bare-minimum threshold of "velocity greater than zero" captures plenty of weak, choppy moves that reverse quickly. By demanding stronger, synchronized directional momentum across both BTC and the correlated ETH asset, we expected fewer noise entries, a higher win rate, and better net PnL. We then swept both the momentum thresholds and the entry price floor and ceiling to map the full interaction surface.

Variant and Strategy Explanation

The base strategy trades KXBTC15M contracts in the final five minutes before expiry. It enters only when:

• Price is between a configurable floor and ceiling (avoiding tails)
• BTC 5-minute change is positive (buy yes) or negative (buy no)
• BTC spot is above (yes) or below (no) its 1-hour VWAP
• BTC 1-minute velocity exceeds a positive threshold (yes) or falls below a negative threshold (no)
• ETH 1-minute velocity confirms the same direction

A hard stop-loss exits any position with -$4.50 unrealized PnL, and all remaining positions are liquidated 5 seconds before expiry.

For this sweep, we tested 100 combinations: 10 price floor values (0.05 to 0.45) × 10 price ceiling values (0.55 to 0.95). Each successful variant is saved as a runnable Turbine strategy.

Top Results

The top 8 variants are effectively identical in performance—they share the same total PnL, win rate, Sharpe, and drawdown. They all use a ceiling of 0.59 and vary only in the floor (0.05 through 0.36). The winner sits at floor 0.05 / ceiling 0.59:

The tight clustering around the 0.59 ceiling is the dominant pattern—floors up to 0.36 don't change the result at all because the signals simply aren't firing in those price bands with the same frequency. Once the floor rises to 0.41, PnL drops off, and at 0.45 it collapses across all ceilings.

Bottom Results

The worst performers cluster at high ceilings (0.86–0.95) with a floor of 0.45:

At these higher ceilings, the strategy enters many more trades (100+ vs. 69 for the winner) but with dramatically worse risk-adjusted returns. The wide entry band allows many marginal setups that the tight 0.59 ceiling filtered out. Drawdowns balloon to nearly 4× the winner's, and Sharpe collapses to 0.10–0.17. Interestingly, win rates don't crater—they hover around 56–62%—but the magnitude of losses on bad trades overwhelms the winners.

Conclusion

The sweep reveals a clean pattern: tight entry bands beat wide ones. The 0.59 ceiling acts as a strong filter, likely because contracts priced above 60 cents in the final five minutes are already heavily favored or disfavored, leaving little edge for a momentum signal to exploit. Lower floors (0.05–0.36) don't degrade performance because the momentum conditions simply don't trigger there often, so the floor is non-binding in practice.

However, the robustness statistics demand serious caution:

• The deflated Sharpe is 0.90, which sits above the expected max Sharpe of 0.38 under the null of no skill. This is encouraging but not overwhelming.
• The permutation test p-value is 0.002, meaning the strategy's edge-feed timing beat 99.8% of time-scrambled versions. This offers some evidence the signal structure isn't pure noise.
• But: the winner has only 7 distinct PnL days, and the deflated Sharpe of 0.90 still falls below the 0.95 threshold we'd want for high confidence. The top results are consistent with selection noise and overfitting given the 100-cell sweep. No variant should be described as validated or strong.

In plain terms: the winner looks good on paper, but the sample is thin, and the sweep was wide enough that some lucky parameter combinations are expected. Forward testing would be required before any confidence is warranted.

Long Disclaimer

This report is a historical simulation study conducted for research purposes only. All performance figures—PnL, ROI, Sharpe ratios, win rates, and drawdowns—are derived from backtesting on past market data and do not represent live trading results. Backtested performance is subject to survivorship bias, look-ahead bias, and overfitting, particularly when derived from parameter sweeps.

The deflated Sharpe ratio and permutation test provide statistical context, but neither can eliminate the risk that the identified strategy configurations are artifacts of the specific historical period tested. A deflated Sharpe below 0.95 indicates that the top-performing parameter set is not distinguishable with high confidence from the luckiest result in a skill-less sweep. The permutation test's p-value tests only the edge-feed timing relative to scrambled versions; it does not validate the strategy against out-of-sample data or against alternative signal constructions. The market price series was not permuted in that test, so price-conditional rules are not directly tested by the permutation design.

The fact that top variants share only 7 distinct PnL days means daily Sharpe estimates rest on very few independent observations, making them unstable and potentially misleading.

This report does not constitute investment advice, a recommendation to trade any instrument, or a prediction of future performance. Trading binary options and event contracts involves significant risk of loss. Past simulated results, even with favorable statistical tests, are not reliable indicators of future outcomes. Any decision to deploy capital using these strategies is solely the responsibility of the trader and should be preceded by thorough forward testing in a paper or minimal-size live environment.