Paper: arXiv 2609.23969

Authors: Alexander W. Crosier

Abstract

The sooner we receive information, and the more accurate it is, the better planning decisions we can make. Every day, prediction markets let anyone bet on tomorrow’s high temperature in cities around the world, creating a market-implied forecast built on dispersed information. We use the past five years of market data from the Kalshi exchange for seven American cities to extract, hour by hour, the market-implied forecast. We use this forecast as a measuring instrument to see how much information about the temperature the market makes public before the public forecasting system does. We race it against the leading American and European weather forecasts. In six of the seven cities we study, the market beats the most accurate single public forecast, the National Blend of Models (NBM). Aggregating every city-day, at the end of the market’s first hour of trading it beats the best single public product by about 10 percent in root-mean-square error, and holds its lead through the day, overnight, and into the target day. Looking at how the forecasts move over time, we find the National Blend travels four times further toward the market between its postings than the market travels toward the NBM. The market does not react to new weather forecast updates; instead, the forecast slowly publishes information that the market had already shared publicly.

Complexity vs Empirical Score

  • Math Complexity: 4.0/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper presents a clear and empirically strong analysis of prediction markets versus traditional weather forecasts. While the mathematical complexity is moderate, the rigorous data analysis and novel application of prediction markets to a real-world forecasting challenge are commendable. The findings are well-supported and clearly communicated.

Research Flowchart

  flowchart TD
    A["Research Goal: Do Prediction Markets Beat Weather Forecasts on Tomorrow's High Temperature?"] --> B{"Methodology: Compare Prediction Market Forecasts vs. Public Weather Forecasts"};
    B --> C["Data/Inputs: 5 Years of Kalshi Market Data (7 US Cities), Leading American & European Weather Forecasts"];
    C --> D{"Computational Process: Extract Hourly Market-Implied Forecasts, Calculate Root-Mean-Square Error (RMSE)"};
    D --> E["Key Finding 1: Market Beats NBM in 6/7 Cities by ~10% RMSE at 1st Hour, Maintains Lead"];
    E --> F["Key Finding 2: NBM Moves 4x Further Toward Market Between Postings; Market Doesn't React to NBM Updates"];
    F --> G["Key Outcome: Prediction Markets Share Information About Temperature Publicly Before Public Forecasting Systems"];

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes using prediction market prices as a continuously updated, more accurate forecast for tomorrow's high temperature.

Source sentence
We use the past five years of market data from the Kalshi exchange for seven American cities to extract, hour by hour, the market-implied forecast.

Assumes

No explicit assumptions found in the text.

Data

  • Kalshi exchange market data · Kalshi market-prices commercial
    past five years · seven American cities
    Source sentence
    We use the past five years of market data from the Kalshi exchange for seven American cities to extract, hour by hour, the market-implied forecast.

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • In six of the seven cities studied, the market beats the most accurate single public forecast, the National Blend of Models (NBM). predictive-accuracy
    Source sentence
    In six of the seven cities we study, the market beats the most accurate single public forecast, the National Blend of Models (NBM).
  • At the first hour of trading, the Kalshi market has about 10 % lower RMSE than the best single public product. performance
    Source sentence
    At the first hour of trading, the Kalshi market has 10 percent lower error than the best single public product.
  • By the final NBM posting, the market retains an advantage with roughly 11 % lower RMSE. performance
    Source sentence
    By the final posting of the NBM on the target day, the market retains its advantage with 11 percent lower error.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 21 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (12)

  1. National Blend of Models: A Statistically Post-Processed Multi-Model Ensemble (2020) — link
  2. Corporate prediction markets: evidence from google, ford, and firm X (2014) — link
  3. Weather forecasting with market prices of weather futures (2013) — link
  4. The Meteorological Development Laboratory’s Aviation Weather Prediction System (2010) — link
  5. A Simple Explanation of the Forecast Combination Puzzle (2009) — link
  6. Comparing Predictive Accuracy (1995) — link
  7. The Jackknife and the Bootstrap for General Stationary Observations (1989) — link
  8. Orange Juice and Weather (1984)
and 4 more
  1. The Use of Model Output Statistics (MOS) in Objective Weather Forecasting (1972) — link
  2. Improvement of daily temperature forecasts from a prediction market (2026)
  3. Macroeconomic Derivatives: An Initial Analysis of Market-Based Macro Forecasts, Uncertainty, and Risk (2005) — link
  4. The Use of Knowledge about Society (2005)

0 of 12 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.