Short Disclaimer

This report simulates historical strategy research only. No future performance is implied or guaranteed. All PnL figures are simulated and do not represent actual trading results.

Intro / Thesis

The common intuition on crypto prediction markets goes something like this: ETH leads, BTC follows. When Ethereum rips higher on Coinbase, traders rush to buy YES on Kalshi's 15-minute BTC contracts, expecting Bitcoin to play catch-up. The same logic applies on the downside—ETH dumps first, so bet YES on BTC dumping too.

Our research tested whether that assumption might be exactly backwards for KXBTC15M. Specifically, the thesis was: when ETH shows a clean impulse but BTC velocity is not confirming the move, Kalshi traders may systematically overprice the probability of BTC catching up. Fading that expectation—buying NO on ETH up/BTC not confirming, and buying YES on ETH down/BTC not confirming—could potentially capture a mispricing in how the market interprets the ETH-BTC lead relationship in very short time windows.

The base strategy implemented this by using Coinbase ETH and BTC data (5-minute change, 15-minute change, 1-minute velocity), layered with price filters and time-to-expiry guards on Kalshi contracts. Exit conditions included PnL gates and a near-expiry close.

Variant and Strategy Explanation

The base strategy (shown in full YAML above) operates as follows:

• Market: Kalshi KXBTC15M (15-minute Bitcoin binary contracts)
• Data sources: Coinbase ETH-USD and BTC-USD, refreshed every 10 seconds
• Entry logic (contrarian):
  • ETH up, BTC not confirming → buy NO: Requires ETH change_5m > 0.1%, ETH change_15m > 0.15%, ETH velocity_1m positive, but BTC change_5m between -0.1% and +0.05% AND BTC velocity_1m ≤ 0. Price must be between $0.25 and $0.85, with at least 4 minutes to expiry. Position size: 5 contracts.
  • ETH down, BTC not confirming → buy YES: Mirror image. ETH change_5m < -0.1%, ETH change_15m < -0.15%, ETH velocity_1m negative, but BTC change_5m between -0.05% and +0.1% AND BTC velocity_1m ≥ 0. Price between $0.15 and $0.75, 4+ minutes to expiry. Position size: 5 contracts.
• Exit logic:
  • Take profit at unrealized PnL ≥ $5.00
  • Stop out at unrealized PnL ≤ -$8.00
  • Close all positions with ≤ 2 minutes to expiry
• Risk limits: Max position 15 contracts, max loss $8.00, price floor/ceiling enforced

From this base, 100 variants were systematically tested. Parameters explored across the variant grid included:

• Tightening and widening the ETH impulse thresholds (change_5m and change_15m)
• Adjusting BTC non-confirmation bands (what counts as "not confirming")
• Shifting entry price windows
• Varying position sizes (contract counts)
• Modifying PnL exit thresholds
• Changing the time-to-expiry minimums
• Altering the loop cadence

This was not an optimization exercise. It was an exploration of whether the contrarian framing held water under any reasonable parameter set. All 100 variants completed simulation.

Each successful variant explored is saved as a runnable Turbine strategy, meaning the exact YAML configuration, parameters, and logic are preserved for audit or further testing.

Top Results

The top-ranked variants all produced nearly identical results—a recurring pattern across the best-performing configurations. Here is what the top tier looks like:

Kalshi variant 017 (Rank #1):

• Total PnL: -$11.52
• ROI: -46.08%
• Max drawdown: -$11.52
• Sharpe: -0.03
• Win rate: 3.03%
• Total trades: 71

Variants ranked #1 through #8 all share these exact metrics: -$11.52 total PnL, -46.08% ROI, -$11.52 max drawdown, -0.03 Sharpe, win rates hovering around 3%. The consistency here is telling—no parameter set in the upper tier meaningfully changed the outcome. The best the contrarian approach could muster was still deeply negative, with ruin-level drawdowns and a win rate indistinguishable from random chance (if not worse).

The top variants appear to have tilted parameters toward slightly wider entry price bands and minimally adjusted non-confirmation thresholds, but none of this moved the needle. The strategy lost money methodically regardless of how the dials were tuned.

Bottom Results

The worst-performing variants paint an even bleaker picture, though the spread between "top" and "bottom" isn't as wide as one might expect—suggesting the core idea itself was the problem, not parameter selection.

Kalshi variant 001 (Rank #81):

• Total PnL: -$9.42
• ROI: -188.4%
• Max drawdown: -$9.42
• Sharpe: -0.04
• Win rate: 3.03%
• Total trades: 66

Variants #81 through #88 all show -$9.42 total PnL and -188.4% ROI, with the same ~3% win rate. While the absolute dollar loss appears smaller than the top variants ($9.42 vs $11.52), the ROI is dramatically worse because these variants likely traded with smaller average position sizes or had different sizing logic that made each dollar lost represent a larger percentage of deployed capital.

More importantly, the bottom tier confirms that tightening parameters (earlier versions of the variant sweep tended toward stricter entry conditions) did nothing to rescue the thesis. In fact, tighter filters created fewer trades (66 vs 71) but proportionally worse outcomes, suggesting that when the strategy did fire under conservative conditions, it was even more reliably wrong.

No variant anywhere in the 100-test run showed positive expectancy. Not one.

Conclusion

The ETH-leads-BTC contrarian thesis fails decisively in this test. Across 100 variant simulations, the strategy consistently destroyed capital with win rates that make a coin flip look sophisticated. A few specific takeaways:

1. The core signal is unreliable: ETH impulse without BTC confirmation does not reliably predict BTC non-movement over the next several minutes in KXBTC15M contracts. If anything, the market's expectation of catch-up may be correct on average, or the noise in 5-minute granular crypto data overwhelms any edge the contrarian framing might possess.

2. Parameter sensitivity is negligible: The clustering of results—top variants at -46% ROI, bottom variants at -188% ROI, no positive outliers—indicates the idea itself is the problem, not the specific thresholds. No reasonable parameter set salvaged the approach.

3. Transaction costs and spread not included: These simulations do not account for Kalshi's bid-ask spread, slippage, or any fees. Real-world implementation would likely perform even worse than these already-negative figures suggest.

4. Lead-lag relationships are noisy at this frequency: ETH and BTC are highly correlated assets. Divergences at the 5-minute level may reflect microstructure noise, liquidity differences between Coinbase markets, or random timing variance rather than a predictable pattern that can be faded.

The base strategy YAML serves as a clean, well-structured template for testing alternative theses on KXBTC15M. The ETH-BTC contrarian thesis, however, should be shelved. The data does not support it.

Long Disclaimer

This document is a historical simulation research report produced for internal analysis purposes. It does not constitute investment advice, a trading recommendation, or a solicitation to buy or sell any financial instrument.

All performance figures, including total PnL, ROI, Sharpe ratios, win rates, and drawdowns, are simulated results derived from historical market data. Simulated performance does not represent actual trading and is subject to inherent limitations, including but not limited to: the absence of live market impact, the assumption of perfect order execution at modeled prices, the exclusion of commission and fee structures, and the impossibility of perfectly replicating real-time data latency and order book dynamics.

Past simulated performance does not guarantee future results. The strategy variants described in this report produced negative returns in historical simulation. There is no basis in this research to expect positive returns in live trading. Any trader considering similar approaches should assume a high probability of loss.

Price data sourced from Coinbase via the provider interfaces specified in the strategy YAML. Kalshi market specifications and contract behavior reflect available documentation at the time of research. Market rules, contract terms, and liquidity conditions are subject to change.

Turbine is a strategy research and automation platform. Strategies saved on the platform are runnable configurations that execute logic defined by the user. Turbine does not endorse any specific strategy, guarantee performance, or provide financial advice. All trading decisions remain the sole responsibility of the user.