Kalshi KXHIGHNY Forecast-Cap Strategy Research Report

Short Disclaimer

This report presents historical simulation research conducted by Turbine. It does not constitute trading advice, and past simulated performance does not guarantee future results. All figures are hypothetical and derived from backtesting.

Intro / Thesis

We tested the idea that expensive YES contracts in Kalshi’s KXHIGHNY daily high-temperature market might be worth fading when the National Weather Service forecast for LaGuardia Airport (KLGA) suggests the 72°F bucket floor is unlikely to be reached.

The core thesis is straightforward: if the NWS forecast high is capped near or below 72°F, and the observed current temperature hasn’t yet approached that level, then an expensive YES contract (implying high market confidence the temperature will hit 72°F) may be overpriced. Selling NO against that optimism could capture a mean-reversion edge.

This research reran the base concept with broader entry conditions than a prior iteration that produced zero trades. We removed the restrictive pre-expiry entry gate found in earlier work and deliberately widened the conditions to see whether the forecast-cap signal survives contact with more permissive entry logic, varied price thresholds, different position sizing, and loop-cadence tweaks.

Variant and Strategy Explanation

All 100 variants descend from a base DSL strategy targeting the KXHIGHNY series on Kalshi. The base strategy uses two entry rules, two profit-taking exits, two risk-management exits, and NWS KLGA weather data refreshed every 60 seconds.

Entry rules (base):

• broad_forecast_cap_buy_no – triggers when YES price is above $0.40, current KLGA temperature is below 72°F, and the NWS forecast high is at or below 72°F. Opens a NO position of size 2 contracts.
• stronger_forecast_miss_add_no – triggers at an even higher conviction level: YES above $0.50, current temp below 70°F, and forecast high at or below 71°F. Adds 3 contracts to the NO side.

Exit rules (base):

• exit_profit_yes_repriced_lower – sells all NO if the market reprices YES below $0.30.
• exit_temperature_reaches_bucket_floor – sells all NO if current temperature reaches 72°F, indicating the thesis is wrong.
• exit_stop_loss – sells all NO if unrealized PnL hits -$10.00.
• exit_near_expiry – sells all NO within 30 minutes of market expiration to avoid settlement pin risk.

The 100 variants explored parameter sweeps across:

• Entry price thresholds (varied above and below the base $0.40 and $0.50 levels)
• Current temperature thresholds for entry
• Forecast high thresholds
• Position sizing per entry
• Max position limits
• Loop refresh intervals
• Price floors and ceilings
• Exit profit-take levels
• Stop-loss trip levels
• Time-to-expiry gate values

Each successful variant is saved as a runnable Turbine strategy with a unique slug, meaning any of these configurations can be deployed or re-tested directly on the Turbine platform without manual reconstruction.

Top Results

The top-ranked variants were nearly identical in performance, with only negligible differences separating the first eight slots. Here are the leaders:

The best variant achieved the least-bad result in a field where no variant turned a profit. Four total trades were executed, none won, and the maximum drawdown was contained to -$0.50 across all top performers. Sharpe ratios were effectively zero (negative single digits annualized), reflecting flat-to-slightly-negative risk-adjusted returns with no meaningful positive edge detected.

A win rate of zero across four trades in the top cohort suggests the entry signals fired rarely and never resolved favorably during the backtest window.

Bottom Results

The worst-performing variants traded more frequently and lost more money, compounding the lack of edge with greater exposure:

These bottom variants traded 14 times — more than triple the top cohort — and sustained drawdowns exceeding $2.60. Again, win rates were zero across the board. The increased trade count came from more permissive entry thresholds in these parameter sweeps, which triggered signals on a wider range of forecast and temperature conditions. Rather than finding new profitable regimes, those looser gates simply opened the door to more losing trades.

The pattern is clear: variants that traded more lost more, and no parameter combination in the tested space produced a profitable tilt.

Conclusion

This 100-variant sweep of the KXHIGHNY forecast-cap fade strategy produced no profitable configurations. The thesis — that expensive YES contracts can be faded when the NWS forecast high is capped near or below the 72°F bucket floor — did not generate positive expectancy in historical simulation.

The top variants lost modestly while trading infrequently. The bottom variants lost more by trading more often. Zero wins across all variants that triggered trades indicates the core signal did not separate favorable from unfavorable outcomes during the backtest period.

Widening the entry conditions from a prior zero-trade iteration succeeded in generating activity but failed to surface an edge. The forecast-cap signal, at least as parameterized here, did not survive broader entry logic or varied thresholds.

We are not concluding that temperature-forecast signals on Kalshi are useless — only that this specific formulation, in this market, over this historical window, did not work. The forecast-high information may still have value in combination with other features (humidity, time of day, rate of temperature change, cloud cover, seasonal factors), but the simple current-temp-plus-forecast-cap rule was insufficient as a standalone entry trigger.

All 100 variant configurations remain saved as runnable Turbine strategies for reference, auditing, or adaptation into future research.

Long Disclaimer

This report is produced by Turbine for internal research purposes and is made publicly available in unlisted form as a historical simulation artifact. It is not investment advice, a solicitation, or a recommendation to trade any financial instrument.

All performance figures are hypothetical and derived from backtesting on historical data. Simulated results do not represent actual trading and may not account for real-world factors including, but not limited to: market impact, slippage, liquidity constraints, exchange fees, counterparty risk, execution delays, or the inability to execute at quoted prices. Backtested strategies may perform differently — including substantially worse — in live markets.

The maximum loss figures shown assume all exit conditions executed perfectly without slippage or gap risk. Real trading losses could exceed those shown. The win rates, Sharpe ratios, and ROI figures are computed from a finite set of simulated trades and may not be representative of future performance if market conditions change or if the strategy is deployed on different time periods.

Turbine does not guarantee the accuracy or completeness of the NWS weather data used in this research. Weather observations and forecasts are subject to revision, and any discrepancies between research data and live data could materially affect strategy outcomes.

Past simulation results, whether positive or negative, do not predict future results. Strategies that lose money in backtests may sometimes perform well in new market regimes, and strategies that show profits in simulation may fail in live conditions. Readers should conduct their own diligence before relying on any research output.

Each variant described in this report is preserved as a Turbine strategy object. These objects are research