Research Report: Kalshi BTC 15-Minute Momentum Strategy

Short Disclaimer

This is a historical simulation study only. Nothing here implies future profitability, and past results do not guarantee future performance.

Intro / Thesis

The core idea tested here is simple: Coinbase BTC-USD 5-minute price momentum may predict how Kalshi's 15-minute BTC contracts reprice during the remaining life of the contract. The strategy enters YES when 5-minute momentum is positive, enters NO when it's negative, and exits when that momentum decays back to near zero or when a -$2 unrealized stop is hit. Entry is gated on spread being tight (≤ 0.03) and sufficient time remaining (> 3 minutes).

The goal was not to optimize for maximum P&L or clever sizing — it was to isolate whether the signal-vs-price-region relationship holds at all. The sweep varied only the price floor and price ceiling filters across 100 combinations, holding everything else fixed.

Variant and Strategy Explanation

Market traded:

• Asset: BTC
• Series ticker: KXBTC15M
• Platform: Kalshi
• Strategy type: custom
• Loop interval: 10 seconds

Signal source:

• Provider: Coinbase
• Symbol: BTC-USD
• Field: change_5m (5-minute percentage change)
• Refresh: 5 seconds

Entry rules:

• Buy YES: no position, change_5m > +0.00020, spread ≤ 0.03, time to expiry > 3m
• Buy NO: no position, change_5m < -0.00020, spread ≤ 0.03, time to expiry > 3m

Exit rules:

• Stop loss: unrealized PnL < -$2.00 → sell all
• Momentum fade: position exists and change_5m between -0.00005 and +0.00005 → sell all
• Near settlement: time to expiry ≤ 3m → skip (no new entries)

Risk parameters:

• Max position: 1 contract
• Price floor and ceiling were the sweep variables, ranging from 0.05–0.45 (floor) and 0.55–0.95 (ceiling)

What this isolates: The strategy does not scale position size, does not layer entries, and does not use exotic exits. It tests whether a directionally correct read on 5-minute momentum can overcome the contract's remaining-life repricing mechanics when entries are restricted to "reasonable" price zones. The sweep then asks: which price zones, if any, make this signal work?

Each successful variant from the 100-cell sweep is saved as a runnable Turbine strategy, meaning the exact parameters and rules are preserved for further inspection or paper trading.

Top Results

All 100 parameter combinations completed successfully. The top variants by net P&L:

A few patterns stand out. First, every top variant has a price ceiling of 0.55 or 0.59 — none of the high-ceiling variants cracked the top eight. Second, floors between 0.32 and 0.45 dominate the top of the list, with 0.41/0.55 being the single best cell. Third, win rates cluster in the 42–48% range, well below 50%, which means the edge is not coming from being right more often — it's coming from winning more when right than losing when wrong (or from the stop loss and momentum fade exits cutting losers before they get too expensive).

The winner — floor 0.41, ceiling 0.55 — made $18.39 net across 1,589 trades with a max drawdown of -7.7%. That's a solid absolute number for a single-contract strategy over a historical simulation, but there are important caveats below.

Bottom Results

The worst variants by net P&L:

The bottom of the sweep is dominated by high ceilings — 0.91 and 0.95 appear in every single bottom variant. This is not a subtle pattern. When the strategy is allowed to hold contracts priced above 0.90, it systematically loses money, and it loses a lot.

Equally striking: the worst variants have the highest win rates in the entire sweep (52–55%). The signal itself is not failing more often in these configurations — it's losing more per losing trade. High-ceiling contracts mean the strategy is paying too much for YES positions or too little for NO positions when the market has already largely priced in the outcome. The momentum signal cannot overcome the adverse selection of buying into extreme price regions.

Max drawdowns of -40%+ in the bottom variants also confirm that these are not just low-P&L outcomes — they are actively destructive to capital.

Conclusion

The parameter sweep reveals a clear, monotonic relationship: the price ceiling is the dominant variable, and lower ceilings (0.55–0.59) produce positive or near-positive results while higher ceilings (0.91–0.95) produce consistent, severe losses. The signal — 5-minute BTC momentum — does appear to have some predictive value when entries are restricted to mid-range contracts, but that value is entirely destroyed when the strategy is permitted to trade in extreme price regions.

The best single configuration (floor 0.41, ceiling 0.55) produced $18.39 in net P&L with a Sharpe of 0.67.

However, the robustness statistics require a sober interpretation, and they override any positive framing from raw ROI:

The deflated Sharpe is 0.44, which is below even the 0.95 threshold the robustness framework considers minimally distinguishable from selection noise. More precisely, the deflated Sharpe (0.44) sits below the expected max Sharpe (0.70) for the sweep under the assumption of no skill. In plain terms: when you run 100 parameter combinations, some will look good by chance, and the winner here does not clear the bar needed to claim it's anything more than the luckiest cell in a skill-less sweep.

No permutation test p-value is available from the robustness block — that test did not run in this simulation, so I cannot quote a p-value. But the deflated Sharpe calculation serves the same purpose: it estimates what you'd expect from the best of 100 random tries, and the observed winner falls below that expectation.

There are two additional warnings from the robustness framework:

1. Fee drag: Fees consumed 58% of the winner's gross PnL. That's a large haircut for a strategy with this trade frequency (1,589 trades in the historical window), and it means the edge is thin before costs.

2. Low deflated Sharpe: The winner is not distinguishable from the luckiest outcome of the