Research Report: ETH Lead-Lag Strategy on Kalshi KXBTC15M

Short Disclaimer

This report presents historical simulation results only. Past performance does not guarantee future outcomes. All strategies lost money in backtesting. No live trading recommendations are made.

Intro / Thesis

Bitcoin price action on short-term binary markets often appears reactive rather than anticipatory. The working hypothesis here is straightforward: when Ethereum shows a clean, confirmed upward impulse on both 5-minute and 15-minute windows but Bitcoin has not yet fully moved, it may signal an imminent BTC catch-up move. The reverse applies for downward ETH impulses. This report tests whether acting on ETH lead-lag signals — buying YES when ETH rips and BTC is still tepid, buying NO when ETH dumps and BTC hasn’t followed — produces results that differ meaningfully from simple BTC-continuation logic.

This is a pure research exercise. One hundred variants of the same core idea were systematically explored by adjusting risk parameters, loop cadence, position sizing, and momentum sensitivity thresholds.

Variant and Strategy Explanation

Every variant shares the same foundation:

• Venue: Kalshi, on the KXBTC15M market (15-minute Bitcoin binary).
• Data Sources: Coinbase ETH-USD and BTC-USD, pulling 5-minute change, 15-minute change, and 1-minute velocity refreshed every 10 seconds.
• Entry Logic:
  • Buy YES: ETH change_5m > +0.10%, ETH change_15m > +0.15%, ETH velocity_1m > 0, BTC change_5m between −0.05% and +0.10%, BTC velocity_1m ≥ 0, price between $0.15 and $0.75, time to expiry ≥ 4 minutes.
  • Buy NO: Mirror conditions with negative ETH thresholds and neutral/subdued BTC.
• Exit Logic:
  • Profit gate: unrealized PnL ≥ +$5.00 triggers full exit.
  • Loss gate: unrealized PnL ≤ −$8.00 triggers full exit.
  • Time gate: exit all positions when ≤ 2 minutes remain to expiry.
• Risk Parameters: Max position $15, price floor $0.10, price ceiling $0.90, max loss $8.00.
• Base Position Size: 5 contracts per entry in the root strategy.

The 100 variants then permute these levers: tighter or wider ETH thresholds (change_5m from ±0.05% to ±0.50%, change_15m from ±0.10% to ±0.75%), different BTC lag bands, different position sizes (1 to 10 contracts), varied profit/loss exit levels, modified loop intervals (10s to 30s), and adjustments to the entry price ranges.

Each successful variant — meaning it executed without runtime errors — is saved as a runnable Turbine strategy under a unique slug.

Top Results

The highest-ranking variants by ROI (lowest negative ROI, since all were losers) came from a cluster that shared the same structural outcome. Here are the top performers:

Observations:

• The “best” strategies still lost roughly 2.4x to 2.5x the capital deployed. This is a severe negative return profile.
• Win rates hover around 10–10.5%, meaning roughly nine out of ten trades lost money. The profit gate of +$5 rarely triggered; the loss floor or time exit caught most positions.
• The top cluster (variants 017–020) produced identical metrics, suggesting those permutations converged on the same effective behavior — likely tighter ETH thresholds that reduced entry frequency but didn’t improve signal quality.
• Sharpe ratios are slightly negative (−0.01), which confirms the edge is not just absent but mildly destructive relative to random noise.
• Max drawdown exceeded 65% in every top case, reflecting a steady bleed rather than a single catastrophic event.

No variant in the top tier showed a glimmer of consistent positive expectancy. The best-case scenario was simply losing money more slowly.

Bottom Results

The worst performers amplified the same pattern of failure. The tail of the distribution:

Observations:

• ROI cratered to −876% in the worst group, meaning nearly 9x the deployed capital was lost over the simulation period.
• Interestingly, the worst ROI variants had smaller absolute PnL losses (−$43–44 vs −$60–63 for the top tier) but much higher ROI because they ran with smaller position sizing or different capital bases. The percentage destruction was far greater.
• Win rates remained stubbornly at 10.4–10.5%, indistinguishable from the top cluster. The difference in outcomes came from magnitude of losses on the losing trades, not from selecting better entries.
• Sharpe ratios in this group flipped nominally positive (+0.01), which is misleading — that’s an artifact of smaller position variance, not genuine profitability. A Sharpe of +0.01 is effectively zero.
• Max drawdown was actually lower in absolute terms (−46%) for the worst ROI variants, again reflecting lower absolute exposure while delivering abysmal percentage returns.

The bottom line: whether you lost 65% of a larger stake or 46% of a smaller one, no version of this strategy came close to breakeven.

Conclusion

The ETH lead-lag thesis does not hold up under systematic backtesting on Kalshi KXBTC15M. One hundred variants tested across a wide spectrum of sensitivity settings, position sizes, and risk gates, and every single one lost money. Win rates clustered around 10–10.5% regardless of configuration. The profit exit at +$5 almost never triggered; positions either bled to the −$8 loss floor or expired worthless near the close.

The core problem appears to be that short-dated Bitcoin binaries on Kalshi incorporate ETH information rapidly — fast enough that by the time the strategy detected an ETH impulse and a lagging BTC, the window for profitable entry had already closed. The 15-minute binary structure also imposes severe time decay, which works against any strategy holding positions into the final minutes.

This does not mean lead-lag relationships are fiction in crypto markets. It does suggest that on a 15-minute binary with Kalshi’s specific pricing dynamics, the signal is too noisy and the execution too slow to capture a reliable edge. Further research might explore faster data refresh, sub-5-minute ETH thresholds, or completely different market structures — but within the confines of this study, the strategy is unviable.

Each variant is preserved as a runnable Turbine strategy for anyone who wishes to inspect the parameters or run their own robustness checks.

Long Disclaimer

This report is a historical simulation analysis conducted for internal research purposes only. All performance figures are backtest results derived from historical market data and are not indicative of future returns. Backtesting carries inherent limitations: it does not account for real-world slippage, liquidity constraints, exchange downtime, or the psychological pressures of live trading. The strategies described herein all produced negative returns over the tested period.

No part of this report constitutes investment advice, a recommendation to trade, or a solicitation of any kind. Trading binary options and cryptocurrencies involves substantial risk of loss and is not suitable for all individuals. Past performance — simulated or real — does not guarantee future results.

Turbine makes no representation that any strategy described or saved on its platform will be profitable. Users should conduct their own due diligence and consult with a qualified financial advisor before engaging in any trading activity. The authors and Turbine assume no liability for any losses incurred through the use of these strategies or any derivative thereof.