Live System

Weather Edge

A quantitative trading system that finds mispriced daily high temperature brackets on Kalshi prediction markets by fusing 194 AI and physics-based weather model forecasts against real-time market prices.

Win Rate

100%

2 of 2 trades profitable

Total PnL

+$2.17

Realized across all closed positions

Avg Confidence

92

Only trades scoring 90+ are executed

Cities Covered

5

NYC, CHI, DEN, MIA, LAX

Ensemble Models

194

Members across 5 model families

Uptime

24/7

Cron-scheduled scans 5x daily

How It Works

Every day, Kalshi lists binary contracts on daily high temperatures for 5 US cities. The market is inefficient because most participants rely on a single NWS point forecast. Weather Edge exploits this by building a full probability distribution from 194 ensemble forecast members, applying physics-based corrections, and finding brackets where the true probability significantly exceeds the market price.

1

Ensemble Ingestion

Pull 194 forecast members from 5 model families via Open-Meteo API

2

KDE Probability Engine

Gaussian kernel density estimation smooths discrete members into a continuous PDF

3

Physics Corrections

Apply wind mixing, wet bulb depression, and rounding adjustments

4

Market Comparison

Compare model probability against Kalshi bid/ask to find mispriced brackets

5

Confidence Scoring

All 5 checks must pass (ensemble spread, model agreement, NWS alignment, observations)

6

Automated Execution

Limit orders placed at bid+1¢ for maker fee (0%). Position monitor handles exits.

Alpha Strategies

A

Midnight High

Post-frontal cold advection sets the daily high at midnight before cold air arrives. The system detects when overnight temps exceed afternoon forecasts.

B

Wind Mixing Penalty

Strong winds prevent super-adiabatic surface heating. Gusts above 15 mph mechanically cap temperatures 1-2°F below clear-sky forecasts.

C

Rounding Arbitrage

NWS rounds to the nearest whole degree. A physics model suggesting 34.4°F means the reported high lands in a different bracket than 34.5°F.

D

Wet Bulb Depression

Daytime precipitation probability above 40% caps the high below the dry-bulb forecast through evaporative cooling.

E

NWS vs Ensemble Divergence

When the NWS point forecast diverges more than 2°F from the 194-member ensemble mean, the ensemble captures newer data the forecaster may have missed.

Risk Management

Half-Kelly Sizing

Max 10% of balance per trade. Half-Kelly criterion balances growth with drawdown protection.

Automated Exits

Freeroll at 2x, efficiency exit at 90¢, trailing stop. Position monitor runs every 5 minutes.

7 Pre-Trade Guards

Kill switch, daily trade count, circuit breaker, intraday drawdown, correlated exposure, bot window, duplicate order guard.

Tech Stack

Python 3asyncioKalshi API (RSA-PSS)Open-Meteo Ensemble APINWS APIKDE (scipy)Claude AIDiscord WebhookscronECMWF AIFSECMWF IFSGFSICONGEM

Trade Log

Real trades executed by the system. Every position is logged with full transparency — entry price, exit strategy, confidence score, and P&L.

This is a personal research project. Not financial advice. Prediction markets carry risk of total loss. Past performance does not guarantee future results. The system trades with a small account to validate the quantitative approach.