Can Prediction Markets Beat the Pros?

[WITH CODE] A statistical backtest of whether Kalshi can forecast economic data better than professional surveys.

Hello and welcome back to another paid post!

Today we will take a look at whether prediction markets can forecast macroeconomic data more accurately than professional surveys, and how statistical tests show Kalshi’s ability to predict inflation, unemployment, and Fed rate decisions.

Let’s dive right in.

---

The Idea

Every month, professional economists produce forecasts for three numbers that move markets: CPI inflation, the unemployment rate, and what the Federal Reserve will do with interest rates. These forecasts drive bond positioning, mortgage pricing, and portfolio strategy.

Prediction markets offer a different approach. On Kalshi, you can buy a contract that pays $1 if CPI comes in above 3% next month. The price of that contract reflects the market’s collective probability estimate, updated continuously as new information arrives. No survey lag. No wait time period for the next Bloomberg consensus.

We ran statistical tests to evaluate how good these markets actually are. We used publicly available data from Kalshi’s research repository, FRED, the Philadelphia Fed’s Survey of Professional Forecasters (SPF), and the San Francisco Fed’s Monetary Policy database.

---

How the Contracts Work

For a given CPI release, there are contracts at multiple thresholds: ‘above 2.0%’, ‘above 2.5%’, ‘above 3.0%’, and so on. The prices of these contracts together imply a full probability distribution, not just a point estimate.

If the ‘above 3.0%’ contract trades at $0.40 and the ‘above 3.5%’ contract at $0.22, the market is saying there is roughly an 18% chance CPI falls between 3.0% and 3.5%. Apply this logic across all thresholds and you have a complete, daily-updating forecast distribution.

Kalshi covers CPI (headline and core), unemployment, payrolls, GDP growth, and Federal Reserve rate decisions. For GDP and core CPI especially, Kalshi is one of the only sources of a real-time, market-based probability distribution. Everything else either provides a point estimate or requires a few weeks wait for a survey.