Forgot your password?
Close
typodupeerror
Stats

Kalshi Prediction Markets Match or Beat Traditional Forecasting Tools For Macro Indicators, NBER Study Finds (nber.org) 25

Posted by msmash from the wisdom-of-degenerate-crowds dept.
A new NBER working paper from researchers at the Federal Reserve, Northwestern's Kellogg School and Johns Hopkins finds that Kalshi -- the largest federally regulated prediction market in the U.S., overseen by the CFTC -- produces macroeconomic forecasts that match or beat those of professional forecasters and traditional financial instruments like fed funds futures.

The study compared Kalshi-implied forecasts for the federal funds rate, CPI inflation and unemployment against the New York Fed's Survey of Market Expectations and Bloomberg consensus. Kalshi's modal forecast correctly predicted the federal funds rate on the day before every FOMC meeting since 2022, something neither the survey nor fed funds futures achieved. For headline CPI, Kalshi's median and mode produced a statistically significant improvement over Bloomberg consensus.

Kalshi also fills a gap no other financial market covers: real-time probability distributions for GDP growth, core CPI, unemployment, and payrolls. The paper documented how these distributions shift in response to macro news -- positive CPI surprises moved the mean of the fed funds rate distribution four times more than negative ones. Trading volumes on the platform have grown to nearly 100 million contracts for a single FOMC meeting, supported by liquidity from Susquehanna, Citadel, and Two Sigma.
This discussion has been archived. No new comments can be posted.

Kalshi Prediction Markets Match or Beat Traditional Forecasting Tools For Macro Indicators, NBER Study Finds More

Kalshi Prediction Markets Match or Beat Traditional Forecasting Tools For Macro Indicators, NBER Study Finds

Comments Filter:
  • Disaster futures are now a thing :o

  • Time of prediction (Score:5, Insightful)

    by larryjoe ( 135075 ) on Monday February 09, 2026 @10:18PM (#65979112)

    The New York Fed's Survey of Market Expectations and Bloomberg consensus are surveys filled out 1-2 weeks before the rate announcement. It's not a surprise for a daily Kalshi guess to be more accurate the day before the announcement. It would be more impressive if the Kalshi market were as accurate 1-2 weeks before the rate announcement, as that would be a more apples-to-apples comparison.

  • Insider trading (Score:3)

    by Wolfling1 ( 1808594 ) on Monday February 09, 2026 @10:58PM (#65979154) Journal
    Its funny how insider trading predictions are so insanely accurate, hmm?

  • Insider trading? (Score:5, Insightful)

    by sinij ( 911942 ) on Monday February 09, 2026 @11:05PM (#65979162)
    There is a large number of people with first-hand privileged knowledge of upcoming events that you could probably make a case that increased accuracy is due to insider trading.

  • Not just in the headline of tfa.

    Here's a neat little gem:

    ...real-time probability distributions for GDP growth, core CPI, unemployment, and payrolls.

    How in the fuck would such a thing be meaningful? It's not like there's an ensemble of parallel universe economies out there you could interrogate with your handy dandy interdimensional portal to validate the quality of this probability distribution. There's exactly one economy and exactly one right now and exactly one actual growth rate, cpi, etc.

    There *is* such a thing as an error distribution of an estimator for any particular statistic. But this

    • You're not a Bayesian, are you? ;-)

    • ... one occurrence.

      If you know the answer, you're not dealing with probabilities. An error distribution, as you noted, works backward from a known answer.

      A probably distribution for an estimate, is a guess upon a guess. However, the factors that change an estimate, are far less random than the estimate they produce. In short, there's a limit to how much the estimate can change. Calculating that is the point of a distribution and it's it why we can say, this known result is not random, given this confidence/error level.

  • Hey, if somebody really understands what calculation they're doing to measure "accuracy," I think a brief explanation would be very welcome. I'll admit I gave up when their top-line, explicit example involved a pair of contracts selling for .67 and .57- I'd say spending $1.24 for $1 is the very definition of an inaccurate market.

  • Metrification decreases accuracy (Score:5, Insightful)

    by gurps_npc ( 621217 ) on Tuesday February 10, 2026 @12:26AM (#65979230) Homepage

    The second a measuring method becomes officially used, it drops in accuracy.

    When something is not official, no one tries to game the system. You get more honest answers and nobody is spending millions/billions of dollars to get the 'right' answer.

  • These are all basically following the model of the Delphi Polls from his novel, "The Shockwave Rider". The only step left is for the government to start prioritizing funding for projects based on what people think is likely to happen and we'll be all the way there.

  • keep it to themselves, as that's how they make money. Thus, comparing this to public models doesn't mean a whole lot, only that they have a bigger than average mouth.

  • I thought it was cereal (Score:3)

    by fortunatus ( 445210 ) on Tuesday February 10, 2026 @07:14AM (#65979668)
    Oh wait, that's Kashi...
  • I remembered reading this a while ago and looked in my Goodreads history. There it was, it was first published in 2004, and I read it in 2013.

    The Wisdom of Crowds, by James Surowieki. He reported on the surprising accuracy of large, monetised, prediction markets.
  • Somebody placed $500,000 dollars on a prediction that Lady Gaga would appear in the Super Bowl halftime show. INSIDER TRADING at its finest. This is an ongoing story and will not be the last, as the ability to essentially place bets on all aspects of life breeds means and opportunity to rig the system.

Slashdot Top Deals

System checkpoint complete.

Close