Kalshi KXBTC15M · Coinbase BTC Momentum Research Report

Short Disclaimer

This is a historical simulation research report only. Past results do not guarantee future performance. All performance figures are in-sample and may reflect overfitting.

Intro / Thesis

This research explores whether short-duration directional momentum signals from Coinbase BTC spot markets can be used to trade Kalshi’s 15-minute binary Bitcoin contracts (KXBTC15M). The core idea is straightforward: when Coinbase BTC exhibits aligned 5-minute price change and 1-minute velocity in the same direction, Kalshi’s binary pricing sometimes lags the move. By entering between 3 and 12 minutes before expiry, at price levels that still offer meaningful upside (0.40–0.82), the strategy attempts to capture that convergence before the contract settles.

We ran a full parameter sweep across 100 combinations of price floor and price ceiling to understand how sensitive this approach is to entry zone selection. Every successfully completed variant is saved as a runnable Turbine strategy.

Variant and Strategy Explanation

The base strategy is a custom Kalshi loop running every 10 seconds.

Signals:

• Buy YES when Coinbase BTC change_5m > 0.001 AND velocity_1m > 0 — indicating upward momentum.
• Buy NO when Coinbase BTC change_5m < -0.001 AND velocity_1m < 0 — indicating downward momentum.

Entry filters:

• Contract price between the configured floor and ceiling.
• Spread ≤ 0.015 (tight enough to avoid slippage erosion).
• Time remaining between 3 and 12 minutes.
• No existing position.

Each entry uses a fixed size of 25 contracts (against a max position of 50).

Exit rules (any triggered first):

• Time ≤ 2 minutes remaining → sell everything.
• Unrealized PnL < -$25.00 (max loss floor: $30) → stop out.
• Unrealized PnL > +$15.00 → take profit.

The research sweep varied two parameters across a 10×10 grid:

• risk.price_floor: 0.05, 0.09, 0.14, 0.18, 0.23, 0.27, 0.32, 0.36, 0.41, 0.45
• risk.price_ceiling: 0.55, 0.59, 0.64, 0.68, 0.73, 0.77, 0.82, 0.86, 0.91, 0.95

This means the base DSL’s 0.40–0.82 zone was both narrowed and widened in the sweep to measure sensitivity. All 100 cells completed successfully, yielding a full performance surface.

Top Results

The highest-performing variants clustered tightly around price ceilings of 0.55 and floors in the middle-to-upper range of the sweep.

Several variants tied at the same PnL level because their entry bands overlapped on the same actual trade fills during backtesting. The top cell (floor 0.36, ceiling 0.55) produced a raw Sharpe of 0.88 and a total PnL of $62.42 on a 40-trade sample.

Important: The deflated Sharpe for this sweep is 0.39, well below 0.95, and the permutation test p-value is 0.228. This means the top result is not distinguishable from the luckiest outcome of a skill-less strategy in this parameter space. The winner ran over only 7 distinct PnL days, too few for reliable daily Sharpe estimates. The reported ROI and Sharpe should be understood as in-sample optimization outcomes, not validated edge.

Each of these variants is saved and re-runnable in Turbine.

Bottom Results

The worst-performing variants all shared high price ceilings, particularly 0.91 and 0.95.

The pattern is clear: when the price ceiling was raised to 0.91, the strategy took many more trades (186–196 vs. 40–44 in the top tier) and win rates actually increased to ~59–60%. But the higher ceiling meant entering at prices where there simply wasn’t enough room for the binary to converge profitably before settlement. Max drawdowns blew out to nearly -$119, wiping out capital despite winning more often than losing.

The takeaway: a wide entry band degrades edge by allowing entries at prices too close to settlement certainty, where directional momentum advantage is minimal.

Conclusion

The momentum signal itself shows some in-sample structure — the top variants earned positive returns over 40–44 trades, with reasonable drawdown control. The sweep reveals a narrow sweet spot: price ceilings around 0.55 and floors between roughly 0.27 and 0.41 concentrated the profitable trades.

But the robustness checks are sobering. The deflated Sharpe of 0.39 and permutation p-value of 0.228 indicate these results are consistent with selection noise from a 100-cell grid search. The winner’s PnL comes from only 7 trading days — not enough to separate signal from luck across diverse market regimes. The spread between top and bottom variants is entirely driven by the price ceiling parameter; raising it above ~0.82 systematically destroyed PnL, even while raising win rates. That is a classic sign of a parameter that controls trade frequency more than edge.

No variant should be described as strong or validated. The base thesis — that Coinbase momentum leads Kalshi KXBTC15M pricing — is plausible but unproven by this data. If pursued further, any forward testing or production deployment must use conservative position sizing and recognize that the most attractive backtest results may simply be the ones that fit historical noise best.

Long Disclaimer

This report presents historical simulation research conducted in a backtesting environment. It does not constitute investment advice, a trading recommendation, or a prediction of future results. All performance metrics — including ROI, Sharpe ratio, win rate, total PnL, and maximum drawdown — are computed in-sample over the period covered by the simulation and are subject to survivorship bias, look-ahead bias, and overfitting.

The parameter sweep described covers 100 variants derived from the base strategy DSL. The top-ranked variants reflect the optimization of those parameters over historical data. Multiple robustness checks were applied, including a deflated Sharpe ratio and a permutation test. Both indicate that the top results are not statistically distinguishable from selection noise at conventional confidence levels. The permutation test randomly scrambled the edge feed while preserving market price sequences; a p-value of 0.228 means that approximately 23% of random scrambles produced results as good as or better than the observed best, which does not meet the standard threshold for rejecting a null of no edge.

Real trading involves commissions, liquidity constraints, slippage, and execution risk not fully captured in simulation. The strategy’s reliance on tight spreads (≤0.015) and short time-to-expiry windows (3–12 minutes) makes it sensitive to market microstructure conditions that can change without notice. Historical performance should not be relied upon as a guarantee of future outcomes. Any decision to deploy capital using this or any related strategy should be made only after independent evaluation and with full awareness of the risk of total loss.