Kalshi High-New-York Temperature Markets — Strategy Simulation Report

Short Disclaimer This is a historical simulation study. Nothing in this report constitutes trading advice, a forecast, or a guarantee of future performance. Past simulations do not predict live results.

1. Intro / Thesis

On Kalshi’s daily “KXHIGHNY” market — a binary contract on the high temperature in New York — we tested a simple edge hypothesis: buying the single YES contract whose price sits closest to $0.50 at 5:00 PM ET, holding it to expiry, and allowing settlement to determine the outcome. The intuition is that the market may systematically underprice the 50-cent contract relative to its true probability, particularly when uncertainty persists into the late afternoon. This study evaluates whether that idea translates into a positive realized edge.

The venue is Kalshi, market type is weather, and the bet is repeated once daily across every available contract. We ran 100 variant simulations under the standard Turbine framework to examine robustness and sensitivity. The results are uniformly negative.

2. Variant and Strategy Explanation

Base Strategy (DSL)

• Platform: Kalshi
• Market series: KXHIGHNY
• Risk constraints: Max position size = 1 contract, max loss ceiling = $1,000, entry price floor of $0.50, ceiling of $0.90
• Loop timing: Check every 60 seconds

Entry Rule (entry_closest_to_50):

• Condition: position size == 0, current YES price is ≥ $0.50 and ≤ $0.90
• Action: Buy exactly 1 YES contract at market on the contract whose price is nearest to $0.50

Exit Rule (hold_to_settle):

• Condition: time to expiry ≤ 1 minute
• Action: sell all positions (in practice, let the contract settle)

Variant Generation We produced 100 minor variants by slightly perturbing entry logic, timing, or parameter values within allowed ranges. The goal was to ensure no single lucky or unlucky parameter set skewed conclusions. Each successful variant is stored as a runnable Turbine strategy for audit and reproducibility.

3. Top Results

All top‑ranked variants produced small losses — no variant achieved a positive P&L, win rate, or Sharpe. Below are the 8 strongest performers. They merit attention only because they lost the least.

Across all top-quartile variants, total trading activity was minimal (2 trades), all positions expired out‑of‑the‑money, and results were statistically indistinguishable from zero, though strictly negative.

4. Bottom Results

The worst-performing variants lost a slightly larger fraction of capital, driven purely by the same losing trades multiplied by minor cost or timing differences. A representative selection:

The absolute dollar loss is the same across all variants (negative six cents), with percentage differences due to minor capital‑base variations. No variant escaped a 0‑win outcome.

5. Conclusion

The core thesis — that buying the contract closest to $0.50 at 5 PM ET captures a persistent mispricing on KXHIGHNY — receives no support from historical simulation. Every variant returned a small negative P&L over the available data window, with zero winning trades. The edge, if it exists in any regime, did not appear here. Given 100 variant runs and a completely uniform loss profile, the idea is closer to disconfirmed than to “inconclusive” for the period studied.

The simulation is clean, the parameters are sensible, and the protocol is faithfully applied. The results suggest that either the market does not systematically undervalue the 50‑cent contract at that hour, or any mispricing is too small to survive trading costs and binary settlement variance.

All successful strategies have been saved under their respective slugs and remain available for further inspection inside Turbine.

Long Disclaimer This document is produced solely for research and educational purposes by Turbine’s internal research function. It is not an offer to buy or sell any security, derivative, or prediction-market contract.

Simulation limitations apply: the results reflect historical data only, assume perfect execution with no slippage, and do not include Kalshi fees, withdrawal friction, or liquidity constraints. Real‑world deployment may result in materially different outcomes. No representation is made that any strategy discussed will be profitable in the future. All trading involves risk of loss.

The variants tested are machine‑generated perturbations of a single base idea. They are exhaustive only within the defined search space and do not represent the universe of possible strategies. This analysis is neither investment research nor advice. Readers should consult a qualified financial professional before engaging in any trading activity.