Turbine Strategy Research Report

Market: Kalshi KXETH15M (Ethereum 15-minute binary contracts)
Thesis: BTC velocity shocks as leading signals for ETH contract repricing

1. Short Disclaimer

This report presents historical simulation results. Past performance does not guarantee future outcomes. All strategies involve risk of loss. This is not investment advice.

2. Intro / Thesis

The core idea is simple: when Bitcoin makes a sharp, sudden move over a one-minute window, Ethereum markets on Kalshi don't always reprice ETH's 15-minute binary contracts fast enough. If that velocity shock is directionally aligned with Bitcoin's own one-hour VWAP (volume-weighted average price), it may signal that the real move has momentum behind it — not just noise.

We tested whether entering ETH YES or NO contracts immediately after a BTC velocity spike — filtering for VWAP alignment, sufficient time to expiry, and tight spreads — could capture a directional edge before ETH catches up. The exit triggers are ETH's own 5-minute change hitting a catch-up threshold, hard stops, or time forced exits near contract expiry.

This research runs 100 variants of that core logic, varying entry thresholds, exit catch-up levels, position sizing, and whether the VWAP guard is active. Long and short trades were tracked independently.

3. Variant and Strategy Explanation

Every variant starts from a shared base architecture:

• Venue: Kalshi
• Contract: KXETH15M — 15-minute ETH binary (YES/NO) markets
• Data sources: Coinbase BTC (price, 1-minute velocity, 1-hour VWAP) and Coinbase ETH (price, 5-minute change), refreshed every 5 seconds
• Loop interval: 10 seconds between evaluation cycles
• Risk guardrails: maximum position of 20 contracts per side, price floor 0.05, price ceiling 0.95, hard stop at -$5 unrealized PnL

Entry logic (two mirrored signals):

The base threshold is ±4.5 for velocity (measured in whatever units Coinbase velocity_1m outputs — effectively a rate-of-change metric). Variants sweep this threshold higher and lower.

Exit logic (three layers):

1. ETH catch-up: If ETH's 5-minute change exceeds +0.30 (for longs) or drops below -0.30 (for shorts), the position closes. The idea: ETH has now repriced and the edge is exhausted.
2. Time forced exit: At 120 seconds remaining to expiry, close everything — decay accelerates sharply in binary contracts near settlement.
3. Hard stop: If unrealized PnL hits -$5 on any position, cut it.

What the variants changed:

The 100 variants explored parameter sweeps including:

• BTC velocity entry thresholds (higher and lower than base ±4.5)
• ETH catch-up exit thresholds (tighter and wider than ±0.30)
• Position size adjustments
• VWAP guard on vs. off (key ablation: does VWAP alignment matter?)

Every successfully completed variant is saved as a runnable Turbine strategy, accessible by its strategy ID.

4. Top Results

The top eight variants (ranked 1 through 8) all produced identical outcomes:

What this tells us:

Every top-ranked variant lost money on every single trade. Six trades placed, six losers. The "best" strategies simply lost less total capital ($1.11) than the worst performers. The ranking is driven by lower absolute drawdown, not by any variant finding a profitable configuration.

These top variants share a common profile: they traded relatively infrequently (6 trades across the simulation period) and burned through a small amount of capital slowly. That keeps the drawdown low relative to variants that traded more aggressively, but the edge is nonexistent — zero winning trades across all top-ranked strategies.

The VWAP guard did not change the outcome for these variants. All top results include the VWAP alignment filter, yet they still failed to produce a single winning trade.

5. Bottom Results

The bottom eight variants (ranked 81 through 88) show a different pattern:

What this tells us:

Surprisingly, the worst-ranked variants lost less total money than the top-ranked ones — only $0.56 versus $1.11. But they rank lower because their ROI is worse (-11.2% vs -4.44%) on a smaller base of trades. With only 4 trades and all losers, the return on whatever capital was deployed was more negative.

The ranking penalty here comes from lower trade count combined with a worse percentage loss. These variants took fewer signals and still lost on every single one. The edge doesn't exist at any threshold tested in this group.

Again, win rate is zero across the board. These variants likely had slightly different entry thresholds that made them even more selective, resulting in fewer entries but no improvement in directional accuracy.

6. Conclusion

This thesis does not hold up in historical simulation.

Across 100 variants — covering different velocity thresholds, exit triggers, position sizes, and with or without the VWAP guard — no variant produced a positive win rate or positive total PnL. The best results merely lost money more slowly. The worst results lost a smaller absolute amount but on even fewer trades.

Key takeaways:

1. Zero winning trades anywhere. Not a single variant across all 100 simulations had a win rate above 0%. This isn't a parameter tuning problem — the core signal does not produce a directional edge in the tested data.
2. BTC velocity shocks don't predict ETH Kalshi contract direction. Whatever repricing lag exists, it's either too small, too noisy, or swamped by Kalshi market spreads and friction. The ETH catch-up exit triggers rarely if ever fired profitably.
3. VWAP filtering adds nothing here. The ablation between VWAP-on and VWAP-off variants showed no difference — both configurations lost consistently. Aligning BTC velocity with the one-hour trend didn't filter out bad signals because all signals were bad.
4. Time-to-expiry guards aren't the problem. The 7-minute minimum and 120-second forced exit are sensible protections, but they can't salvage a signal with no edge.
5. The market is efficient enough at 15-minute horizons. Kalshi market makers and participants appear to reprice ETH adequately during BTC volatility spikes, leaving no systematic lag to exploit with this approach.

Recommendation: Do not deploy this strategy family. The negative Sharpe ratio (-0.53 across all variants) and zero win rate indicate that the premise is flawed, not just the parameterization.

For researchers interested in related angles: shorter timeframes (sub-5-minute ETH markets, if they existed on Kalshi) might capture a repricing lag that 15-minute contracts smooth out. But on KXETH15M specifically, the data says no.

Each variant tested here has been preserved as a runnable Turbine strategy for reference. The full set can be accessed via their strategy slugs and IDs listed in the detailed results.

7. Long Disclaimer

This report is a historical simulation research document produced by Turbine's research analysis systems. It describes the results of automated strategy backtesting using historical market data from Kalshi and Coinbase.

No forward-looking claims are made. The performance figures shown — including PnL, ROI, Sharpe ratios, and win rates — are simulated historical results only. They do not predict or guarantee future performance, profitability, or outcomes in live trading.

All trading involves risk of loss. Binary options and prediction market contracts can lose their entire value. The strategies described here carry embedded risks including but not limited to: execution slippage, data feed latency, market impact, liquidity withdrawal, contract settlement mechanics, and platform-specific operational risks. These factors are not fully captured in simulation.

The strategies in this report are not recommendations. They are research artifacts from a systematic exploration of a specific thesis. None of the variants tested produced positive returns. Readers should not interpret any strategy discussed here as investment advice, trading advice, or a solicitation to trade any instrument on any venue.

Data limitations apply. Historical simulations use cleaned, backfilled data that may not reflect real-time conditions, including gaps, restatements, or timing mismatches between the BTC/ETH feeds and Kalshi order books. Simulation assumes fills at observed prices, which may not be achievable in live markets, particularly during volatility events.

This research was conducted by Turbine. Turbine is a prediction market research and strategy platform. This document is intended for informational and analytical purposes only. Consult qualified financial professionals before making any trading decisions.