Public Research Report: KXBTC15M Slow-Climb Reversion Strategy

Short Disclaimer

This is a simulated historical research report. Nothing here is financial advice or a prediction of future returns.

Intro / Thesis

The core idea under test was simple: in 15-minute Kalshi BTC markets (KXBTC15M), a small positive 5-minute BTC change on Coinbase might signal a slow climb that fails to complete. If that were true, buying NO when change_5m is between 0 and 0.005 could capture reversion before settle.

The base DSL encodes exactly that. It enters NO on small positive 5-minute moves with at least 5 minutes left, exits on a NO price above 0.80 or near expiry, and includes a stop-loss at -2.0 unrealized PnL.

The honest headline after running 100 parameter variants: no edge found. The thesis does not hold up in this simulation.

Variant and Strategy Explanation

The strategy family tested is a custom Kalshi strategy on the KXBTC15M series, looping every 10 seconds. It uses Coinbase BTC-USD change_5m as the only edge field, refreshed every 5 seconds.

The 100 completed variants come from a two-axis parameter sweep:

• risk.price_floor: 0.05 to 0.45 in 10 steps
• risk.price_ceiling: 0.55 to 0.95 in 10 steps

Everything else in the base DSL was held fixed: entry rule, exit rule, stop-loss, position size (2 contracts per entry), and max position (10 contracts).

Each successful variant is saved as a runnable Turbine strategy, so any cell in the sweep can be re-run or inspected as a standalone strategy by its slug.

Top Results

The top of the sweep is not a story of success. It is a story of flat zero PnL with very few trades.

The best-ranked variant is:

• Label: BTC strategy · floor 0.05 / ceil 0.59
• Rank: 1 of 100
• Total PnL: 0.00
• ROI: 0%
• Total trades: 4
• Win rate: 0%
• Max drawdown: -1.12
• Sharpe: -0.41

Ranks 2 through 8 are effectively identical in outcome: 0 PnL, 4 trades, 0% win rate, -1.12 max drawdown, -0.41 Sharpe. The only thing changing across the top block is the price ceiling parameter, from 0.64 up to 0.91.

There is no variation in outcomes because the strategy barely trades. Four resolved trades over the entire simulated period, none of them winners. The top results are not "good" — they are flat and statistically empty.

The robustness statistics also flag this clearly:

• Winner has 4 resolved trades (< 30): metrics are statistically unreliable
• Winner has 1 distinct PnL day (< 10): daily Sharpe rests on too few observations
• 50% of winner fills are at prices <0.10 or >0.90, where fill assumptions are least trustworthy

The permutation test result is unambiguous. The p-value is 1.000, both upper and lower. In plain terms: the strategy's edge-feed timing beat 0.0% of 1,000 re-sweeps against time-scrambled versions of the same feed. The real best result is indistinguishable from noise.

Per the analysis rules, I must state this plainly: the top results are consistent with selection noise and overfitting. No variant is strong, validated, or promising.

Bottom Results

The worst-ranked variant is:

• Label: BTC strategy · floor 0.05 / ceil 0.55
• Rank: 100 of 100
• Total PnL: -0.09
• ROI: -0.92%
• Total trades: 5
• Win rate: 0%
• Max drawdown: -1.22
• Sharpe: -0.39

The only meaningful difference between the bottom variant and the top variants is one extra trade and a small negative PnL. The ceiling of 0.55 appears to slightly alter trade frequency, but the outcome is still zero wins and a modest loss.

The bottom block in the JSON then repeats the top variants, which is a data artifact of how the results were compiled. The substantive bottom result is the -0.09 PnL at rank 100.

The spread between best and worst is tiny: 0.00 versus -0.09. That is not a robust strategy family. That is a flat line with noise.

Conclusion

The slow-climb reversion thesis did not survive contact with simulation.

Across 100 parameter variants, every single variant produced either zero or slightly negative PnL. Win rate is uniformly 0%. Trade counts are too low to draw any statistical inference. The permutation p-value is 1.000, meaning the results are indistinguishable from random timing.

The top result being 0 PnL is not a near-miss. It is the absence of signal. The strategy barely trades, and when it does, it does not win.

This is a clean negative result. The slow-climb NO edge on KXBTC15M, as specified in the base DSL, does not exist in this historical simulation.

Long Disclaimer

This report is a historical simulation research artifact produced for internal analysis purposes. It is not an offer, solicitation, or recommendation to buy or sell any security, contract, or financial instrument.

All results are derived from backtested or simulated data over a specific historical window. Backtested performance is not indicative of future performance. Simulated fills, especially at extreme prices below 0.10 or above 0.90, may not reflect actual market liquidity or executable prices on Kalshi or any other venue.

The market described is a Kalshi crypto market on the KXBTC15M series. Prediction markets and crypto assets involve substantial risk, including rapid loss of capital. Kalshi contracts are subject to Kalshi's rules and regulatory framework.

The strategy DSL and variant parameters are provided as technical documentation of what was tested. They are not a recommendation to deploy any strategy live.

Key limitations of this study:

• Small sample size: The best variant produced only 4 resolved trades and 1 distinct PnL day. No reliable statistical inference can be drawn from this sample.
• Permutation test p-value = 1.000: The edge-feed timing performed no better than randomly scrambled versions of the same feed. The results are consistent with pure noise.
• Uniform zero win rate: Every variant in the top and bottom blocks had a 0% win rate. There is no evidence of any directional edge.
• Extreme price fills: Half of the winner fills occurred at prices below 0.10 or above 0.90, where realistic fill assumptions are weakest.
• Historical simulation only: No live trading occurred. No fees, slippage, or venue-specific execution constraints were modeled beyond the DSL assumptions present in the simulation.

Each successful variant is saved as a runnable Turbine strategy for audit and reproducibility purposes. That does not imply endorsement or live deployment suitability.

If you are considering trading Kalshi crypto markets, you should independently evaluate the risks, your own financial situation, and the relevant regulatory and tax implications.