Our Calculation Methodology
A transparent breakdown of how we estimate probabilities and compare data.
1. Transparent Heuristic Temperature Model
Instead of claiming proprietary "AI prediction" capabilities, WeatherMarketWatch uses a transparent, rules-based heuristic temperature-distribution model to estimate outcome probabilities.
We model the high temperature on the target calendar day as a Normal Distribution \(N(\mu, \sigma^2)\):
- Mean (\(\mu\)): Centered at the official National Weather Service (NWS) point forecast high temperature.
- Standard Deviation (\(\sigma\)): Scales dynamically based on the forecast horizon (days remaining until settlement) and weather risk flags:
- Same day (nowcast mode): \(\sigma = 1.0^\circ\text{F}\)
- 1 day out: \(\sigma = 1.5^\circ\text{F}\)
- 2 days out: \(\sigma = 2.0^\circ\text{F}\)
- 3+ days out: \(\sigma = 2.5^\circ\text{F}\)
- Thunderstorm risk flag: Add \(+0.5^\circ\text{F}\) to \(\sigma\)
- Precipitation / Rain flag: Add \(+0.3^\circ\text{F}\) to \(\sigma\)
- Overcast / Heavy clouds flag: Add \(+0.2^\circ\text{F}\) to \(\sigma\)
2. Outcome Probability Integrals
Because daily high temperatures are reported in whole degrees, we apply half-degree offsets to represent outcomes as continuous brackets. We integrate the Normal Cumulative Distribution Function (CDF):
- Exact value outcomes (e.g. 85°F): We calculate the probability that the high falls in the range \([84.5^\circ\text{F}, 85.5^\circ\text{F})\):
Probability = Φ(85.5, μ, σ) - Φ(84.5, μ, σ) - Range brackets (e.g. 84-85°F): We integrate from the lower bound minus 0.5 to the upper bound plus 0.5:
Probability = Φ(85.5, μ, σ) - Φ(83.5, μ, σ) - Open-ended upper bounds (e.g. 86°F or higher):
Probability = 1.0 - Φ(85.5, μ, σ)
3. Confidence & Model Maturity Rules
To prevent overconfidence, we apply the following safety caps:
- Database maturity cap: All confidence outputs are strictly capped at Medium until the database has compiled at least 50 resolved comparable markets.
- Boundary flag: If the forecast mean (\(\mu\)) lies within \(0.5^\circ\text{F}\) of any outcome threshold, the market is flagged, and confidence is reduced due to boundary volatility.
4. Simulated Paper Trading Positions
Paper signal simulations apply conservative execution assumptions:
- Order sizes: Fixed at \(10.00\) virtual dollars per eligible signal.
- Exchange fees: We apply a \(1.0\%\) flat transaction fee penalty on entry.
- Slippage: We apply a \(0.5\%\) slippage penalty on entry.
- Triggers: Signals are only simulated if the spread is narrow (\(< 0.10\)), liquidity is sufficient (\(> \$100.00\)), and the forecast-vs-market difference exceeds \(5.0\%\).