Research Report: Kalshi 15-Minute BTC Binary Late-Drift Strategy

Short Disclaimer

This report is historical simulation research only. Past simulated performance does not guarantee future results. All trades and metrics are hypothetical and do not represent live trading.

Intro / Thesis

We investigated whether a rules-based strategy could capture late-settlement drift in Kalshi’s 15-minute BTC binary contracts. The core idea is straightforward: when one side is already strongly favored (YES above 0.80 or NO above 0.80, meaning YES below 0.20), and we see confirming momentum from the underlying Bitcoin market, the remaining minutes before settlement often see continued drift toward the winning side. Entry timing, spread control, and tight stop-losses are the key operational guardrails.

We ran 100 systematic variants, all derived from the same base logic, and this report covers what worked best, what failed entirely, and what patterns we observed across the full set.

Variant and Strategy Explanation

Every variant tested was a customization of a single base DSL strategy that operates as follows:

Market and instrument

• Venue: Kalshi
• Series: KXBTC15M (15-minute BTC binary, YES/NO)
• Contract risk limits: max position 8 contracts, price floor 0.10, price ceiling 0.90

Edge data

• Bitcoin spot price and 1-hour change from Coinbase, refreshed every 10 seconds

Core entry rules (base logic)

• YES momentum entry: price >= 0.80, BTC 1h change >= +0.5%, spread <= 0.03, time-to-expiry between 4m and 5m, and current position under 8 contracts → buy 5 YES contracts
• NO momentum entry: price <= 0.20 (meaning NO is heavily favored), BTC 1h change <= -0.5%, same spread, time, and position constraints → buy 5 NO contracts

Risk and exit rules

• Stop-loss: sell entire position if unrealized PnL drops to -$3.00 or worse (per-contract stop discipline)
• Exit at settlement: close everything when 30 seconds remain
• Re-entry: allowed after a stop-out if conditions are met again before the time window closes

Across the 100 variants, we explored tweaks to entry thresholds, momentum strength, spread tolerance, timing windows, and stop distance. Each successful variant is saved as a fully runnable Turbine strategy under its own strategy slug for future deployment if desired.

Top Results

Eight variants produced identical top-tier performance with a 26.8% simulated ROI, 1.34 total PnL, a 0.33 Sharpe ratio, and a modest 6 total trades across the backtest period. Win rate settled at 66.7%, meaning two out of three bets paid off. Max drawdown was -$0.26, which is notably small.

The standout variants (ranked 1 through 8) share these common features:

Key observations from the top cohort:

• Low trade frequency, high conviction. Only 6 total signals fired, which implies the entry conditions are genuinely restrictive. That is desirable in a late-drift strategy: you want clean setups, not constant churn.
• The stop-loss did its job. The -$0.26 max drawdown across the simulation means the stop-loss rule clipped losers before they could spiral. The tight per-contract stop (-$3.00 unrealized) paired with small position sizing meant losses were contained without preventing re-entry on better setups.
• 66.7% win rate is solid for binaries. In binary markets with only two outcomes, a 67% hit rate on a strategy that takes late, high-probability entries is consistent with the thesis that settlement drift exists when both price and momentum align.
• Re-entry logic provided value. Some of these variants likely used re-entry after a stop-out to catch a second attempt within the same expiry window. The data does not break out re-entry separately, but the ability to re-enter after a whipsaw without waiting for a new contract is structurally important.
• Sharpe of 0.33 is modest but real. Given the ultra-short holding period (minutes), a positive Sharpe on such a small sample is notable. It would not impress a multi-asset portfolio manager, but for a single-market binary strategy, it indicates the edge is not purely random.

Each of these top variants is saved as a runnable Turbine strategy under its own slug — for example, custom-on-kxbtc15m-e910c9ac5608 for variant 061 — meaning they can be deployed, monitored, or further refined without rebuilding from scratch.

Bottom Results

The bottom eight variants all share the same profile: zero trades, zero PnL, zero drawdown, zero win rate. They are ranked 41 through 48 and are functionally indistinguishable from each other:

These are not "losing" strategies in the conventional sense — they simply never fired. This is almost certainly because their entry parameters were set too restrictively: likely a combination of tighter momentum thresholds, narrower time windows, or stricter spread requirements that historical conditions never satisfied during the simulation period.

The lesson: there is a cliff between "selective enough to filter noise" and "too strict to ever trade." The top variants found the balance. The bottom variants fell off the other side. In practice, a strategy that never trades is just dead code — it cannot compound, cannot prove itself, and cannot pay for its own existence. These bottom variants serve as a reminder that over-optimization without sufficient slack in the parameters produces beautiful logic with zero utility.

Conclusion

The simulation results are internally consistent with the thesis. When a 15-minute Kalshi BTC binary is already heavily skewed (0.80+ or 0.20-) with confirming 1-hour BTC momentum and a tight spread, entering in the final 5 minutes while protecting the position with a hard stop-loss produced a 26.8% simulated return on a small number of high-conviction trades. The stop-loss kept worst-case outcomes trivial, and the re-entry permission meant the strategy was not one-and-done.

The wide gap between the top and bottom cohorts also tells us something useful: parameter selection matters enormously. Slight differences in threshold stringency separate a tradeable strategy from a dormant one. Anyone looking to operationalize this approach should focus on the specific parameter combinations present in the top eight variants, all of which are captured and runnable inside Turbine.

This remains historical research. Live markets introduce slippage, fill uncertainty, and regime changes that no backtest captures. But as a structured exploration of a specific hypothesis, the data supports the view that late-settlement drift in these binary markets is real enough to build rules around.

Long Disclaimer

This document is a historical simulation research report produced for analytical purposes only. It does not constitute investment advice, a trading recommendation, or an offer to buy or sell any financial instrument. All performance figures — including ROI, PnL, Sharpe ratio, drawdown, win rate, and trade counts — are hypothetical and derived from backtesting on historical data. They do not represent actual trading results and do not account for real-world factors such as liquidity constraints, execution slippage, commission costs, market impact, or platform availability.

Past simulated performance is not indicative of future results. The strategies described carry risk of total loss. Binary options involve fixed-payout outcomes and can move adversely with little warning. Any person considering deploying these or similar strategies should understand the risks fully and, if