Research Report: KXETH15M 15-Minute ETH Momentum Strategy

Short Disclaimer

This is historical simulation research only. Nothing here implies future profits or live trading performance. All results come from backtests of a parameter sweep on historical data.

Intro / Thesis

The idea under test was whether short-horizon ETH momentum, confirmed by BTC's 1-minute velocity, could identify near-term directional edges in Kalshi's 15-minute ETH binary markets (KXETH15M). The base logic required three signals to agree:

• ETH 5-minute change
• ETH 1-minute velocity
• BTC 1-minute velocity

When all three pointed the same direction, the strategy would buy that side between 45–55¢ with a tight spread, expecting short-duration continuation toward fair value. A stop-loss at -$4.50 and forced exit at 5 seconds to expiry capped downside.

We ran a 10×10 parameter sweep across price floors and ceilings — 100 variants total, all of which completed. The goal was not to find a "best" setting, but to test whether the edge survived variation in entry bands, exit bands, and data-refresh behavior at the edge level.

Variant and Strategy Explanation

The strategy is a custom Kalshi loop running every 10 seconds on the KXETH15M series. It fetches Coinbase ETH and BTC spot data, refreshes every 10 seconds, and applies the following flow:

1. Entry: If ETH 5-minute change, ETH 1-minute velocity, and BTC 1-minute velocity all point the same direction, and the best ask on the chosen side is between the configured floor and ceiling, and the spread is ≤ 3¢, the strategy buys $10 on that side.

2. Timing constraint: Entries are only allowed when time to expiry is between 5 seconds and 5 minutes, excluding the final 5 seconds.

3. Risk controls: One entry per market, max position of 100 contracts, price bounds enforced, stop-loss at -$4.50, and mandatory exit at 5 seconds before expiry.

The parameter sweep varied two risk bounds:

• price_floor: 0.05 to 0.45 in ten steps
• price_ceiling: 0.55 to 0.95 in ten steps

Each combination is saved as a runnable Turbine strategy, so any variant can be loaded and re-run against fresh data if desired.

Top Results

The best raw performer by net P&L was the combination of floor 0.41 / ceiling 0.73, with these figures:

• Net P&L: $120.12
• ROI: 120.12%
• Trades: 178
• Win rate: 68.4%
• Max drawdown: -$14.58
• Sharpe: 0.45

A cluster of near-identical results appeared at floors from 0.09 to 0.45 with ceiling fixed at 0.73. Those variants produced $118.21 to $120.12 net P&L, 67.1% to 68.4% win rates, and Sharpe ratios around 0.42–0.45. The floor parameter mattered less than the ceiling once the lower bound was above roughly 0.09.

The ceiling of 0.73 produced the strongest marginal results, with a mean P&L near $118.5 across all floors in that bucket. In other words, capping entries at 73¢ consistently outperformed wider entry bands in this backtest.

However, the raw numbers must be viewed through the robustness statistics. The deflated Sharpe is 0.23, far below the 0.95 threshold that would suggest the top result is distinguishable from the luckiest skill-less trial. The expected maximum Sharpe from the sweep, assuming no underlying edge, was 0.60. The observed 0.45 Sharpe on the top variant is below that noise ceiling.

Bottom Results

The worst-ranked variants were concentrated at the ceiling 0.95 end of the sweep, regardless of floor. Representative examples:

• floor 0.05 / ceiling 0.95: Net P&L $24.74, max drawdown -$67.30, Sharpe -0.03, 297 trades, 66.2% win rate
• floor 0.05 / ceiling 0.91: Net P&L $24.97, max drawdown -$72.25, Sharpe -0.03, 265 trades, 63.0% win rate
• floor 0.32 / ceiling 0.95: Net P&L $25.69, max drawdown -$66.35, Sharpe -0.02, 295 trades, 66.7% win rate

The pattern is clear: allowing entries all the way up to 95¢ destroyed the strategy's edge. These variants still had respectable win rates — around 63–67% — but the losses on the losing side were far larger relative to the winners. Max drawdowns ballooned to roughly $67–72, more than four times the drawdown of the best variant.

Wider ceilings meant the strategy was paying too much for binary contracts near the extremes, where the directional momentum signal offered little marginal value. The degradation was not gradual; it accelerated sharply once the ceiling moved above roughly 0.82.

Conclusion

The thesis had surface-level support: the raw performance gradient across the ceiling parameter was clean and directionally consistent. Ceiling 0.73 represented a real local optimum, and the strategy's edge degraded meaningfully as entries were allowed at higher prices. Win rates held up in the high-60s even in poor variants, suggesting the three-signal confirmation did filter some low-quality trades.

But the robustness test tells a different story. The deflated Sharpe of 0.23 is below the expected maximum Sharpe from a skill-less sweep (0.60). That means the top results are consistent with selection noise and overfitting across the 100-cell parameter grid. I cannot describe the floor 0.41 / ceiling 0.73 variant as strong, validated, or promising based on this evidence.

The most honest summary is: the sweep produced an orderly-looking result that falls short of statistical robustness under permutation testing. A promising ceiling band exists in the raw data, but it is not yet distinguishable from luck over this many variants.

Long Disclaimer

This report is a historical simulation research document produced for Turbine's research workflow. It does not constitute investment advice, a recommendation to trade, or a prediction of future performance.

All performance figures — including net P&L, ROI, Sharpe ratio, win rate, and drawdown — are derived from backtests on historical market data. Backtests are subject to numerous limitations, including but not limited to:

• Survivorship and selection bias in the strategy development process
• Parameter overfitting when many variants are tested on the same dataset
• Differences between historical fills and real-time execution
• Slippage, fees, and liquidity constraints not fully reflected in simulation
• Regime shifts in crypto markets that may invalidate past relationships

The deflated Sharpe statistic used in this research is a deliberate attempt to penalize for multiple testing across the parameter sweep. A low deflated Sharpe does not prove the strategy will lose money live. It indicates that the observed top result is not statistically distinguishable from the best result one would expect from random noise given the number of variants tested.

Each variant's configuration is saved as a runnable Turbine strategy for research continuity. Running a saved strategy against future data does not guarantee that historical patterns will persist.

Turbine Research does not warrant the accuracy, completeness, or usefulness of this analysis for any trading or investment purpose. Past performance, whether real or simulated, is not indicative of future results. You should consult multiple sources and consider your own risk tolerance before making any trading decision.

Trading binary options on crypto assets involves substantial risk of loss. Positions can expire worthless. You should never trade with capital you cannot afford to lose.