Turbine Research Report: Momentum Confirmation on Kalshi BTC 15-Minute Markets

Short Disclaimer

This is historical simulation research only. Past performance does not indicate future results. These findings describe backtested outcomes, not live trading profits.

Intro / Thesis

We tested whether demanding stronger, synchronized momentum signals improves entry quality on Kalshi's KXBTC15M markets. The core idea: strip away VWAP and price-level complications, and trade pure directional momentum. A position only opens when three conditions align—BTC's 5-minute change direction, BTC's 1-minute velocity, and ETH's 1-minute velocity all point the same way. The strategy enters in the final five minutes before expiry, when prices sit between 0.40 and 0.60, betting that late momentum persists through settlement.

The question was whether tightening price bounds around fair value (raising the floor, lowering the ceiling) filters out bad entries, or whether it simply chokes off opportunity. We swept price floors from 0.05 to 0.45 and ceilings from 0.55 to 0.95 across 100 variants, all using the same 10-second loop interval and Coinbase feed for BTC and ETH velocity data.

Variant and Strategy Explanation

Every variant shares the same skeleton. The strategy polls Coinbase every 10 seconds for BTC-USD (5-minute change, 1-minute velocity) and ETH-USD (1-minute velocity). It trades KXBTC15M contracts on Kalshi, max 30 positions, with a hard stop at -$4.50 unrealized P&L and forced exit under 5 seconds to expiry.

What changed across the 100 variants: only the price floor and ceiling. The base DSL uses 0.40 and 0.60; our sweep tested floors at 0.05, 0.09, 0.14, 0.18, 0.23, 0.27, 0.32, 0.36, 0.41, and 0.45, paired with ceilings at 0.55, 0.59, 0.64, 0.68, 0.73, 0.77, 0.82, 0.86, 0.91, and 0.95. Each successful variant is saved as a runnable Turbine strategy.

The momentum logic itself never changed. Buy YES when BTC is up over 5 minutes, BTC velocity is positive, and ETH velocity is positive. Buy NO when all three flip negative. No thresholds on velocity magnitude—just direction. The bet is that correlated-asset confirmation catches momentum that single-asset signals miss, and that price bounds keep you out of expensive or cheap lottery tickets where edge is thin.

Top Results

The clear winner: floor 0.41, ceiling 0.59. This tight band around 0.50 produced $129.45 total P&L, 431.5% ROI, 56.1% win rate across 100 trades, with a 0.69 Sharpe and -$19.61 max drawdown. Notably, this variant traded less frequently than neighbors—only 100 trades versus 102 for the cluster just below it.

Ranks 2 through 8 are essentially identical: floor 0.05 through 0.32, all with ceiling 0.59. Each returned $125.28, 417.6% ROI, 54.8% win rate, 102 trades, 0.65 Sharpe, -$23.78 drawdown. The ceiling of 0.59 dominates the top of the leaderboard. The marginal analysis confirms this: mean P&L across all floor values peaks at ceiling 0.59 ($122.24 mean, $129.45 max), then degrades steadily as you widen toward 0.95.

The pattern is stark. A 0.59 ceiling with any floor below 0.36 works. Push floor to 0.45 and results collapse. Push ceiling above 0.59 and you bleed P&L, win rate, and Sharpe.

Bottom Results

The worst performers share two profiles: wide ceilings (0.86–0.95) with any floor, and the single case of a tight ceiling (0.55) with a low floor.

Rank 100: floor 0.45, ceiling 0.95. Only $17.89 P&L, 59.6% ROI, 178 trades, 0.12 Sharpe, -$51.89 drawdown. The 0.45 floor alone is restrictive, but pairing it with 0.95 ceiling is catastrophic—you're buying expensive YES and cheap NO with the same momentum signal, and the signal doesn't discriminate well at extremes.

Ranks 99 and 97–96 show the wide-ceiling decay: 0.91, 0.86, 0.82 ceilings with 0.45 floor produce $34.65 to $54.68 P&L, Sharpe 0.23–0.35, drawdowns near -$44 to -$49. More trades, worse outcomes. The strategy is working harder for less.

Rank 98 is the anomaly: floor 0.05, ceiling 0.55. Tight ceiling but $36.25 P&L, 120.8% ROI, brutal 34.9% win rate. The 0.55 ceiling apparently cuts off too many valid entries, leaving only bad ones.

The 0.95 ceiling cluster (ranks 87–89) with low floors shows another pathology: 200 trades, ~$57 P&L, 0.33 Sharpe, -$52 drawdown. High trade count, low edge per trade, deep drawdowns. The strategy bleeds on transaction costs and noise.

Conclusion

The sweep produces a clear, narrow prescription: trade KXBTC15M with a 0.59 price ceiling and any floor from 0.05 to 0.41. The 0.41/0.59 variant sits at the sweet spot—tightest band, best Sharpe, lowest drawdown, highest P&L.

However, the statistical validation is weak. Two warnings flag the winner directly. First, only 7 distinct PnL days underpin the daily Sharpe—far below the 10-day threshold for reliable estimation. Second, and more seriously, the deflated Sharpe is 0.76, below the 0.95 threshold that would distinguish the winner from random luck across 100 trials. The expected maximum Sharpe from noise alone is 0.39; 0.76 is better than that, but not convincingly so.

The permutation test offers one brighter note: p = 0.002 against 964 time-scrambled edge-feed resamples. The edge-feed timing—the specific sequence of momentum readings—beat 99.8% of shuffled versions. But this test did not touch market prices. It validates that the Coinbase feed sequence matters, not that the strategy extracts true alpha from KXBTC15M pricing.

Taken together, these results are consistent with selection noise and overfitting. The 0.59 ceiling pattern is real in this historical sample, but the magnitude of outperformance may inflate from mining 100 parameter combinations. The top variants are not validated as strong or