Kalshi BTC 15-Minute Binary: EMA/SMA Crossover Strategy Research Report

Short Disclaimer

This report presents historical simulation results only. Past performance does not indicate future results. These findings are for research purposes and do not constitute investment advice.

Intro / Thesis

We tested whether momentum persistence after moving-average crossovers on Coinbase BTC spot data could generate small, consistent profits in Kalshi's KXBTC15M binary market. The core idea: when a 5-minute EMA-26 crosses above or below a 5-minute SMA-50, short-term directional bias often carries into the next several minutes. By filtering for tight Kalshi spreads (≤$0.03) and using tight exits (+/-$1.00 unrealized PnL), we aimed to capture that follow-through before 15-minute settlement, with position stacking allowed to compound signals.

The hypothesis was modest in scope—not home runs, but repeated base hits. We ran 100 completed variants to see if any parameter combinations made this edge tradable.

Variant and Strategy Explanation

Each variant used the same base structure with permutations across the parameter space. The strategy monitors Coinbase BTC-USD for EMA-26/SMA-50 crossovers on 5-minute candles. When bullish (EMA > SMA) with Kalshi spread ≤$0.03, it buys YES contracts. When bearish (EMA < SMA) under the same spread condition, it buys NO contracts. Positions stack if signals repeat before exit. Every position closes when unrealized PnL hits +$1.00 (take profit) or -$1.00 (stop loss).

Variants differed in their specific parameter combinations—things like position sizing increments, loop timing adjustments, and risk threshold calibrations. The base DSL specifies a max position of 50 contracts, price floor/ceiling of $0.05/$0.95 to avoid extreme binaries, and a $10.00 max loss ceiling per trade sequence.

Each successful variant is saved as a runnable Turbine strategy with its unique slug identifier.

Top Results

The top tier clustered tightly, suggesting a stable parameter region. Four variants tied for first place with identical metrics:

The #1-4 group traded more frequently (545 vs. 410 trades) and achieved higher absolute PnL, but with deeper drawdowns (-$139.57 vs. -$111.78). The #5-8 group was more selective with slightly better win rates. All shared a low Sharpe of 0.07—positive but indicating noisy, uneven returns. The ~40% win rate is notable: these strategies were profitable not because they won often, but because wins and losses were capped asymmetrically by the tight exit rules, and position stacking let winners run briefly while losers were cut quickly.

Bottom Results

The bottom quartile tells its own story. Here the same structural approach produced near-total capital erosion:

The worst performers actually had similar or slightly better win rates than the top tier (~41-42% vs. ~40-41%). The difference was catastrophic drawdown management. Variants 041-044 limited max drawdown to -$31.11 but still lost 98% of capital—suggesting many small losses grinding down equity. Variants 045-048 allowed deeper drawdowns (-$62.35) with similar win rates, also ending at -96% ROI.

This is critical: the bottom variants weren't wrong more often. They were wrong at the wrong times, or their parameter combinations let losses compound before the -$1.00 stop could trigger effectively. The spread filter and exit rules that helped top variants apparently misfired under certain parameter configurations.

Conclusion

The EMA-26/SMA-50 crossover approach on KXBTC15M showed conditional historical viability. A tight cluster of variants (017-020) produced positive returns with aggressive trade frequency, while a slightly more conservative group (013-016) achieved decent ROI with lower drawdowns. The edge is fragile: win rates around 40% require disciplined risk mechanics, and the 0.07 Sharpe across all top performers indicates substantial return volatility.

The bottom performers are equally instructive. Near-identical win rates produced near-total losses, demonstrating that in binary markets with tight exits, parameter tuning around execution timing, position sizing, and drawdown tolerance matters enormously. The spread filter (≤$0.03) was likely binding—when it worked, it worked; when parameter drift weakened its effectiveness, the same structural signals became loss generators.

We would characterize this as a narrowly positive historical result with high implementation risk. The strategy is not robust across all parameter combinations. Traders considering deployment should note the deep drawdowns (-$111 to -$140) required to achieve modest absolute profits ($12-17), and the very low Sharpe suggesting significant equity swings. Each top variant is available as a runnable Turbine strategy for further paper testing.

Long Disclaimer

This research report is published by Turbine for informational and educational purposes only. It does not constitute investment advice, financial advice, trading advice, or any other type of advice. Nothing contained herein should be construed as a recommendation to buy, sell, or hold any financial instrument or to engage in any specific trading strategy.

Historical Simulation Only: All results presented are derived from historical backtesting and simulation. These results are hypothetical and do not represent actual trading. They do not reflect the impact of market liquidity, slippage, execution delays, or other real-world frictions. Past performance, whether actual or simulated, is not indicative of future results. The strategy may perform differently in live market conditions.

No Future Profit Implication: The report explicitly does not imply or guarantee future profits. The top-performing variants identified in this study may produce losses in live trading. Market conditions change, and edges that appeared in historical data frequently disappear or reverse.

Risk of Loss: Trading in prediction markets, binary options, and cryptocurrency-related derivatives involves substantial risk of loss. The maximum drawdown figures shown (-$31.11 to -$139.57 in simulation) do not represent the maximum possible loss. Traders can lose more than their initial investment in certain market conditions or with certain position structures.

Strategy Variability: The 100 variants tested represent a sample of parameter combinations. Other untested combinations may produce superior or inferior results. The clustering of top and bottom performers suggests sensitivity to specific parameters, but does not establish causal relationships.

Data and Model Limitations: The study relies on Coinbase BTC-USD price data and Kalshi market data. Data quality, availability, and accuracy are not guaranteed. The EMA and SMA calculations, spread measurements, and unrealized PnL computations are model-based and may differ from actual platform calculations.

Regulatory and Compliance: Kalshi markets are regulated by the Commodity Futures Trading Commission (CFTC) and are available only to eligible participants in permitted jurisdictions. Cryptocurrency trading and prediction market participation may be restricted or