Momentum Tight Band: Price-Entry Filtering on BTC 15-Minute Markets

Short Disclaimer

This report is a historical simulation study only. Nothing in it constitutes trading advice or a prediction of future results. Past simulated performance does not guarantee real-world outcomes.

Intro / Thesis

We tested whether tightening the entry price band from 40–60 to 45–55 on a pure momentum strategy for BTC 15-minute binary markets (KXBTC15M on Kalshi) would meaningfully improve signal quality. The intuition was straightforward: a price of 45–55 suggests the market has already moved decisively away from the coin-flip range of 50, yet still offers room for further price discovery as expiry approaches. Combine that with three simultaneous momentum confirmation signals — BTC 5‑minute price change, BTC 1‑minute velocity, and ETH 1‑minute velocity — and you should, in theory, trade fewer low-conviction positions while preserving upside on the strongest setups.

The full backtest sweep across 100 price-floor and price-ceiling combinations lets us see exactly how sensitive the thesis is to the band's exact borders.

Variant and Strategy Explanation

All variants share the same core rule structure, derived from the base strategy DSL. The strategy operates on KXBTC15M markets with a 10‑second loop interval, a $30 maximum position, and a hard stop-loss that liquidates the entire position when unrealized PnL reaches -$4.50.

An entry triggers only when all of these conditions hold simultaneously:

• Time to expiry ≤ 5 minutes
• Market price is within the chosen floor–ceiling band
• BTC 5‑minute price change is positive (for YES entries) or negative (for NO entries)
• BTC 1‑minute velocity is positive (for YES) or negative (for NO)
• ETH 1‑minute velocity is positive (for YES) or negative (for NO)

Each trade is sized at $10. Positions are force-closed when time to expiry drops below 5 seconds.

The sweep varied two parameters: risk.price_floor (from 0.05 to 0.45) and risk.price_ceiling (from 0.55 to 0.95). This creates 100 distinct strategies — 100 completed, no failures. The base thesis focused on the 0.45–0.55 band; the sweep deliberately explores what happens when the band expands, narrows, or shifts.

Every successful variant is saved as a runnable Turbine strategy, identified by a unique slug and ID.

Top Results

The eight best-performing variants are practically identical in outcome, simply reflecting that slight shifts in the price floor (from 0.05 up to 0.36) don't hurt performance as long as the ceiling remains at 0.59. All share:

• Total PnL: $97.23
• ROI: 324.1%
• Win rate: 66.7%
• Max drawdown: -6.96
• Sharpe (ex‑ante): 1.03
• Total trades: 57

Robustness data forces a sober read on these numbers. The permutation test scrambled edge-feed timing across 976 sweeps and found a p‑value of 0.001 for the best observed net PnL of $97.23. That sounds promising at first glance. However, the neighborhood degradation statistic is 20.8% — meaning small changes in parameters produce materially different results. More importantly, the deflated Sharpe is 0.99, while the expected maximum Sharpe from random screening is 0.33. These are modest numbers after adjusting for data mining, and the deflated Sharpe sits only modestly above the expected ceiling for a strategy born from a wide multi-parameter sweep.

There’s also a data-thinness warning: the winning cell produced only 7 distinct PnL days, well below the recommended threshold of 10 for reliable daily Sharpe calculations.

Bottom line: the top cell is the best of 100, but the statistical safeguards say we cannot dismiss the possibility that much of its edge comes from selection bias and a short backtest window.

Bottom Results

The poorest performer (rank 100) keeps the floor at 0.05 but sets the price ceiling at 0.55 — essentially the base band from the original strategy before tightening, with no upper‑bound adjustment. It produced:

• Total PnL: $39.38
• ROI: 131.3%
• Win rate: 45.8%
• Max drawdown: -15.45
• Sharpe: 0.48
• Total trades: 71

A cluster of low-ranked variants (ranks ~90–96) all share a ceiling of 0.95, meaning they allow entries on prices all the way up to near‑certainty. These strategies produce higher trade counts (114) and a reasonable win rate (65.4%), but their max drawdown expands dramatically to -$23.25 (-23.25%), and Sharpe drops to 0.73. The extra trades dilute the edge; you are trading more setups, often at prices where the market has already absorbed most of the available price discovery.

The pattern is clear: loosening the upper bound toward 0.95 hands back a big chunk of the improvement in risk‑adjusted terms. Tightening too aggressively to 0.55, without also shifting the floor, flips the win rate below 50% and guts PnL.

Conclusion

The backtest shows that a 45–55 entry band (or its close cousins, like 5–59) paired with three momentum confirmation signals produces the highest net PnL and win rate in this sweep. The improvement over a 40–60 or 5–95 band is material in raw numbers — higher win rate, lower drawdown, much higher terminal PnL.

But we cannot call this strategy validated or robust. The deflated Sharpe of 0.99 is below what many practitioners would require for an actionable edge after a 100‑cell sweep. The neighborhood degradation is substantial, meaning the result is fragile to small parameter changes. And the low number of distinct PnL days (7) means we have too few independent observations to trust the daily Sharpe estimates fully. The permutation p‑value of 0.001 against edge‑feed scrambling suggests the signal does depend on the actual sequence of edge data — but that alone doesn't rescue the overall statistical picture.

For a researcher, this is an interesting directional finding that merits out‑of‑sample testing with a longer history before any live deployment. The base intuition — that narrowing the band to 45–55 filters noise — holds in this dataset. Whether it holds in the next one is an open question.

Long Disclaimer

This document is a simulated historical research report produced for Turbine's internal use. It is not an offer, solicitation, or recommendation to buy or sell any security, derivative, or prediction market contract. All performance figures — including PnL, ROI, Sharpe ratio, and win rate — are derived from backtesting on historical data and are subject to known limitations: survivorship bias, look‑ahead bias, data‑snooping, and the absence of real‑world frictions such as slippage, liquidity constraints, and exchange fees.

The statistical tests reported (parameter sweep, neighborhood degradation, deflated Sharpe, permutation test) are designed to detect overfitting but cannot eliminate the possibility that the observed results are due to chance. A permutation p‑value below 0.05 does not guarantee a strategy will be profitable in live trading. Low distinct‑day counts (fewer than 10 PnL days) mean that daily statistics are unreliable. The deflated Sharpe ratio adjusts for data mining but remains an estimate; it can overstate or understate true out‑of‑sample performance.

No representation is made that any strategy discussed will achieve similar results in the future. All trading involves risk of loss. The authors and Turbine may hold positions in the markets discussed. This report should not be the sole basis for any investment decision.