Research Report: KXBTC15M Late-Expensive Momentum Strategy

1. Short Disclaimer

This report is historical simulation research. It does not predict or guarantee future performance. All results are based on backtesting across past KXBTC15M contract windows. Real-money trading involves execution, latency, and fill risks not fully captured here.

2. Intro / Thesis

We tested a short-term BTC momentum strategy on Kalshi's KXBTC15M contracts. The core idea: in the final two minutes before expiry, when the market is already trading at an extreme ($0.90 or above on either the YES or NO side), the alignment of the underlying BTC price with its 1-hour VWAP and 5-minute Coinbase momentum acts as a reliable directional signal. The thesis is that this confluence of factors — late-stage positioning, agreement between a medium-term volume anchor and short-term spot momentum — tends to persist through settlement rather than mean-revert.

We used a parameter sweep across 100 combinations of price floors (minimum allowed trade price) and price ceilings (maximum allowed trade price) to understand how sensitive the idea is to the entry band. This report covers what worked, what didn't, and whether the best results are distinguishable from luck.

3. Variant and Strategy Explanation

All variants share the same rule logic, derived from the base DSL. The strategy operates on a 10-second loop and applies two entry rules, only when flat (no open position) and within 120 seconds of expiry:

• Bullish entry: Triggered when the YES side is at or above $0.90, BTC's 5-minute price change on Coinbase is positive, and BTC spot is above its 1-hour VWAP. The system buys YES.
• Bearish entry: Triggered when the NO side is at or above $0.90, BTC's 5-minute change is negative, and BTC spot is below its 1-hour VWAP. The system buys NO.

Both entries use a fixed $10 position size per trade. A hard stop-loss exits all positions if unrealized P&L reaches -$4.00.

The 100 variants differ only in risk.price_floor and risk.price_ceiling, which act as guardrails that block entries outside specified price bands. The sweep tested floors from $0.05 to $0.45 and ceilings from $0.55 to $0.95, creating a 10×10 grid. Every variant was simulated over the same historical KXBTC15M series. Each successful variant has been saved as a runnable Turbine strategy for further review.

The asset is BTC, the venue is Kalshi, the series ticker is KXBTC15M, and all BTC reference data comes from Coinbase (spot price, 5-minute change, 1-hour VWAP).

4. Top Results

The best-performing variant used a $0.05 floor and $0.95 ceiling — essentially the widest allowable band — producing a net P&L of $40.03 across 62 trades, with a win rate of 67.7%. ROI measured 400.3% against the $10 position size, and max drawdown was limited to -$2.19. This variant is rank 1 in the sweep.

The second-place variant, with a floor of $0.05 and ceiling of $0.91, generated $29.34 over 34 trades with a 35.3% win rate. While its total P&L is lower, the Sharpe ratio (0.32 vs. 0.45 for the top variant) and substantially lower win rate suggest a more volatile, less consistent edge.

Several variants tied for ranks 3–8 with $27.60 in total P&L across 30 trades and a 26.7% win rate. These all used the $0.05 floor and ceilings ranging from $0.55 to $0.77. The clustering of identical results at these lower ceilings reflects ceiling constraints limiting trade frequency, but not meaningfully altering the payoff profile on the trades that did occur.

The marginal analysis from the parameter sweep confirms a clear relationship: the $0.05 floor consistently outperformed higher floors, and the $0.95 ceiling produced the single best outcome.

However, three critical caveats apply:

1. Fill assumption risk: 97% of the top variant's fills occurred at prices below $0.10 or above $0.90. These are the least trustworthy price levels for simulated fills — real-world execution at extremes often differs from theoretical fills due to spread, liquidity, and queue position.

2. Deflated Sharpe ratio: The deflated Sharpe for the winner is 0.14, substantially below the expected maximum Sharpe of 0.58 for a sweep of this size. The neighborhood degradation metric of 0.30 also indicates that small parameter changes produce meaningfully different outcomes. This is a textbook sign that the top result is consistent with selection noise — we tested 100 combinations and found the luckiest one.

3. Permutation test: We ran a Monte Carlo permutation test (415 re-sweeps against time-scrambled versions of the BTC feed) to assess whether the strategy's timing edge is real. The headline result (p = 0.002) suggests the edge-feed timing beat 99.8% of scrambled alternatives. This indicates some genuine predictive structure in the BTC data timing. However, this test does not scramble Kalshi market prices — only the external feed — so it cannot rule out that the signal is simply a weak but real timing effect amplified by favorable market fill assumptions.

Taken together, the top variant is the best in the sweep, but it is not robust by standard quantitative standards. The result is not strong enough to describe as validated or promising.

5. Bottom Results

The bottom cohort — ranks 37 through 44 — consists entirely of variants with a $0.14 floor paired with ceilings from $0.55 to $0.86. Every one of these produced zero trades and zero P&L.

The reason is structural: a $0.14 floor prevents any entry unless both sides of the market are $0.14 or above. In the KXBTC15M series, late-stage market pricing typically pushes one side toward $0.90+ while the other side falls well below $0.14. The floor condition therefore never satisfied alongside the other entry requirements, resulting in a dead strategy.

This is a useful negative result. It confirms that the strategy's edge (to the extent one exists) depends entirely on allowing entries at very low contract prices. Tightening the floor even slightly from $0.05 to $0.09 dropped average P&L from $29 to $16 across the sweep marginals. Moving to $0.14 removed all activity. The idea is fragile to entry-band restrictions.

6. Conclusion

The late-expensive-momentum concept shows surface-level appeal — a high win rate and attractive ROI in the top variant — but the simulation evidence does not support confidence in the edge. The deflated Sharpe ratio of 0.14 is weak, the neighborhood degradation is high, and the winner's performance depends heavily on fills at extreme prices where real-world execution is least reliable.

The permutation test provides modest evidence that BTC feed timing carries some predictive value, but it cannot rescue the overall robustness picture. A strategy that only works at the widest price band and degrades quickly with minor parameter changes is more likely a product of overfitting than a durable market inefficiency.

We do not recommend this strategy for live deployment without further out-of-sample testing and careful attention to real-money fill quality at the price extremes. The saved Turbine strategies remain available for monitoring, but the burden of proof has not been met.

7. Long Disclaimer

This research was conducted through historical simulation of Kalshi's KXBTC15M prediction market contracts. All performance figures — including P&L, ROI, Sharpe ratios, win rates, and drawdowns — are derived from backtesting and do not represent actual trading results.

Simulated fills assume instantaneous execution at the prevailing market price at the time of signal. In live markets, order queue position, bid-ask spread, partial fills, and exchange latency can produce materially different outcomes — particularly at price extremes ($0.05 or below, $0.90 or above) where liquidity is thinnest and fill assumptions are most fragile. The warning that 97% of winner fills occurred at these extremes should be taken seriously.

The parameter sweep tested 100 combinations of two risk guardrails. The deflated Sharpe ratio of 0.14 and neighborhood degradation of 0.30 are quantitative indicators that the best result is consistent with selection bias from testing many variants. The permutation test produced p = 0.002 against time-scrambled BTC feeds, which suggests some timing structure in the external data, but this test does not validate the full strategy including market-price-based entry conditions.

No forward-looking claims are made. Past simulation performance, even when statistically suggestive, does not imply future profitability. This report is provided for research and analytical purposes only and does not constitute trading advice or a recommendation to buy or sell any financial instrument.