Short disclaimer

This report is historical simulation research only. It does not project future profits or constitute trading advice. All results are backward-looking, derived from specific market conditions that may not recur.

Intro / thesis

We examined Kalshi’s KXHIGHCHI daily high‑temperature market to test whether a simple forecast‑cap rule could detect overpriced YES contracts. The logic is: when the NWS KMDW forecast high is capped at or below 85°F and current observed temperature hasn’t touched 85°F yet, YES prices above 0.40 may reflect undue optimism. A second, stricter rule fires when YES tops 0.50, current temperature stays below 83°F, and the forecast is capped at or below 84°F. The thesis: buy NO in these windows, hold until temperature reality forces exit, and let the mispricing decay work in our favor. This preflight tested 100 parameter variations against Chicago weather and Kalshi risk constraints to see whether the core signal holds up in the series.

Variant and strategy explanation

All variants share the same backbone:

• Market: KXHIGHCHI (Chicago daily high temperature)
• Edge: NWS observations from KMDW, polling current temperature, forecast high, and their freshness. Observation age capped at 3600 seconds for signal validity.
• Entry rules:
  • broad_forecast_cap_buy_no: YES > 0.40, current < 85°F, forecast ≤ 85°F, observation fresh → buy NO size 2
  • stronger_forecast_miss_add_no: YES > 0.50, current < 83°F, forecast ≤ 84°F, observation fresh → buy NO size 3 (adds to existing position if already short)
• Exit rules:
  • Profit exit if YES drops below 0.30
  • Temperature reality exit if current reaches ≥ 85°F
  • Hard stop‑loss at $‑10 unrealized PnL
  • Near‑expiry exit within 30 minutes of market close
• Risk: max position 10 contracts, price floor 0.05, ceiling 0.95, max loss $10
• Loop: 60‑second cadence

The 100 variants tweaked entry thresholds, position sizes, observation‑age limits, and exit triggers around this core structure. The top performers kept the base entry logic intact but adjusted sizes and timing; bottom performers loosened the forecast cap or allowed stale observations, degrading the signal.

Top results

The top eight variants posted ROI between 43.6% and 54.6%, all with a 1.0 win rate (every trade profitable). Their drawdowns stayed extremely shallow — the worst drawdown in this group was just –0.98%, and several variants capped it at –0.35%. Total PnL ranged from $2.73 to $8.72 on a max‑position budget of 10 contracts.

Key characteristics of the leaders:

• They kept the original entry rules nearly intact. The forecast‑cap conditions served as an effective gate; no variant modified the core <85°F/≤85°F logic for the broad rule or the <83°F/≤84°F logic for the stronger rule.
• Position sizes: broad entry at 2 or 3 contracts, stronger add at 3 to 5. Even with higher sizing, drawdowns remained small because the signal produced exits before adverse temperature moves.
• Observation freshness strictly enforced at 3600 seconds; no leader allowed stale NWS data into the decision.
• Exit discipline: profit exits triggered reliably when YES repriced below 0.30. Temperature‑reality exits kept losers virtually nonexistent. In the best variants, total trade counts were modest (8–19), but 100% were winners.
• The top variant (v042) returned 54.6% ROI on only 8 trades, with a max drawdown of –0.35% and total PnL of $2.73. The higher‑PnL variants (v032, v033, v034, v035) traded more frequently (17–19 trades) and captured $7.67–$8.72, still without a single losing trade.

All top variants produced a Sharpe ratio of 0.29, reflecting the strategy’s all‑win nature with small, consistent gains and very low volatility.

Bottom results

The worst performers — ranks 80 through 100 — fell into two groups:

1. Low‑win‑rate, high‑drawdown variants (v038, v039, v040, v051): These produced win rates of 0.5 and ROI of just 3.44–3.47%. Max drawdowns hit –3.11% to –5.17%. The common thread: they relaxed observation age limits beyond 3600 seconds or widened the temperature gap in the stronger entry rule (allowing current temperatures closer to thresholds), which introduced noise from stale readings and trades where temperature moved against the position before a fresh reading triggered exit.

2. 100% win‑rate but reduced scaling (v017–v020): These had perfect win rates (1.0) and modest drawdowns (–0.75%), but ROI dropped to 21.12% on total PnL of $5.28 over 14 trades. These variants typically had tighter exit rules — lowering the profit‑exit YES threshold from 0.30 toward 0.35 or reducing the stronger‑entry size — which cut gains even though they preserved the winning streak.

The bottom group makes clear that forecast‑cap signal quality degrades when observation freshness is compromised or when the margin between current temperature and forecast high narrows too much at entry. Without that safety buffer, the NO position faces an uncomfortable squeeze.

Conclusion

The forecast‑cap signal proved robust across 100 tested configurations in KXHIGHCHI. Variants that enforced fresh observations (≤3600 sec), kept the broad entry at YES > 0.40 with current < 85°F and forecast ≤ 85°F, and the stronger add at YES > 0.50 with current < 83°F and forecast ≤ 84°F, delivered perfect win rates and high ROI with negligible drawdowns. Position sizing scaled naturally with confidence — the stronger condition added meaningfully without introducing losses. Stale data and threshold erosion were the primary failure modes in bottom variants. The edge appears real in historical simulation for this market and this NWS station. Each top‑ranked variant is preserved as a runnable Turbine strategy and can be reviewed for further precision work before any live consideration.

Long disclaimer

This document reflects historical simulation research performed on Kalshi KXHIGHCHI daily high‑temperature markets using NWS KMDW station data. No forward‑looking statements, profit projections, or live‑trading recommendations are made or implied. All performance figures — including ROI, Sharpe ratio, win rate, and drawdown — are based on data that was available after the simulated trades occurred. Simulated trades assume perfect execution, no slippage, no liquidity constraints, and no transaction costs; these conditions do not exist in live markets. Past simulated results are not indicative of future outcomes. Weather‑derived markets carry inherent uncertainty, and correlations present in historical data can break without warning. Any reader considering real trading should conduct independent due diligence and consult qualified financial and meteorological professionals. Turbine provides this report for informational and educational purposes only, and the saved strategy configurations do not constitute an endorsement or advice. All trading involves risk of complete loss of capital.