Prediction markets have spent the last two years growing up. What used to be a niche corner of election-night speculation is now a multi-billion-dollar venue where the same baseball game gets priced thousands of times a night. Kalshi alone cleared more than $20 billion in volume last year, with sports now the dominant category. That kind of growth invites an obvious question: if a regulated exchange is pricing the same games the sportsbooks price, are its numbers actually right — and if they’re wrong, can you do anything about it?
A handful of recent academic studies have started answering this with real data. The short version is a useful paradox. Kalshi’s prices are remarkably accurate. They are also measurably slow. And the gap between “accurate” and “fast” is exactly where a patient trader can operate.
What “efficient” actually means
When economists call a market efficient, they mean its prices fully reflect available information — you can’t reliably beat it, because the price already knows what you know. There are two ways to test that on a prediction market, and they give different answers.
The first test is calibration: over thousands of contracts, do things priced at 60% actually happen about 60% of the time? On this measure, Kalshi looks genuinely good. Studies analyzing its markets find calibration regressions with intercepts near zero and slopes near one — statistician-speak for “the prices are honest.” A contract trading at 65¢ wins close to 65% of the time. There’s some noise at the extreme ends (very cheap and very expensive contracts), but the dominant pattern is that Kalshi prices are informative, and they get more accurate as a game approaches start time. As a snapshot of probability, the closing price on Kalshi is a better estimate than most casual bettors give it credit for.
So if you’re using Kalshi purely as a signal — a read on the true probability of an event — the research is reassuring. The crowd is doing its job.
The second test is where it gets interesting.
The 0.64 problem
Calibration asks whether prices are right on average, eventually. A different and harder question is whether they move correctly and immediately when new information arrives. This is dynamic efficiency, and it’s the one that matters if you want to trade rather than just observe.
Here’s the finding that should make every Kalshi trader sit up. A 2026 study examining how Kalshi prices respond to real-time information found that when the sharp benchmark probability for a game moves by a given amount in a one-minute window, the Kalshi price moves only about 0.64 as much in that same window.
Read that again, because it’s the whole article. When the true probability of a team winning shifts — because a lineup dropped, a pitcher got scratched, the sharp money moved — the sportsbooks reprice almost instantly. Kalshi captures only about two-thirds of that move in real time. The rest of the adjustment happens later, as the slower Kalshi market gradually catches up.
That under-reaction is not a calibration failure. Given enough time, Kalshi gets to the right number — that’s why the calibration tests look so clean. But in the moment, in the minutes after information arrives, the Kalshi price is systematically lagging the sharp consensus. It’s right eventually. It’s slow right now.
Why slow and accurate aren’t a contradiction
This sounds paradoxical until you think about who’s trading where.
The sportsbooks move enormous handle. They employ teams of traders, they sharpen their lines continuously, and any stale price gets hammered by sharp money in seconds. Their number reflects new information almost the instant it exists. That’s a market with deep liquidity and ruthless participants — it has no choice but to be fast.
Kalshi, on a regular-season game, is a thinner market. A single MLB matchup might carry only a few thousand dollars of volume. There simply aren’t enough active traders watching every contract to drag the price to fair value the moment something changes. So when the benchmark moves, Kalshi’s price drifts toward it over minutes rather than snapping to it instantly. Eventually it gets there — the calibration data proves it does — but the journey takes time, and during that journey the price is stale.
The thinness that makes Kalshi slow is the same thinness that makes it accurate-but-laggy rather than accurate-and-instant. Liquidity buys speed. Kalshi doesn’t have enough of it yet on most sports markets to be fast.
Where this leaves a trader
The academic literature on prediction markets keeps circling one hard truth: the fast inefficiencies are already gone. Studies of pure arbitrage — buying YES and NO for less than a dollar combined — find that when those windows open at all, they close in seconds, snapped shut by automated bots faster than any human could react. Cross-platform arbitrage between Kalshi and Polymarket exists in the data, but the windows are measured in seconds to a few minutes and transaction costs eat most of the theoretical profit. If your plan is to out-speed an algorithm, you’ve already lost.
But the 0.64 finding points at something different, and slower, and human-sized.
The opportunity on Kalshi isn’t lightning-fast arbitrage. It’s the lag itself. When the sharp benchmark moves and Kalshi has only captured two-thirds of the adjustment, there’s a window — not 3.6 seconds, but often minutes — where Kalshi’s price is knowably behind the consensus and drifting toward it. You don’t need to be faster than a bot to trade that. You need to notice that the two numbers disagree, take the side Kalshi is underpricing, and let the slow market catch up to the fast one.
This is convergence, not arbitrage. You’re not locking in a risk-free dollar across two platforms in milliseconds. You’re identifying a price that the research says is statistically likely to move toward the sportsbook number, taking that side, and exiting once it converges — often before the game even starts. The edge isn’t speed. It’s recognizing the lag and being patient enough to harvest it.
The honest limits
None of this is free money, and the same research that reveals the opportunity also fences it in.
The lag isn’t a guarantee on any single game. The 0.64 figure is an average across many price moves. On any individual game, Kalshi might converge upward to meet the books, or the books might move down to meet Kalshi, or news might break that changes the fair value entirely. The statistical tendency is real; the certainty on one trade is not.
Liquidity cuts both ways. The thin markets that create the lag are also hard to get size into and out of. A gap you can spot is not always a gap you can trade at any meaningful scale — and trying to exit a large position in a shallow market moves the price against you.
Fees are part of the math. Kalshi’s taker fee can quietly eat a small convergence. The move that survives the fee is the only move that counts, which is why entering as a maker — with a resting limit order rather than crossing the spread — matters so much to whether the edge is real.
Holding one side is not a hedge. Until you close, you’re carrying genuine risk. Convergence trading is lower-variance than betting the outcome, but it is not the risk-free arbitrage the textbooks describe.
The takeaway
The research delivers a clean, slightly counterintuitive verdict: Kalshi’s prices are accurate but slow. They’re well-calibrated enough to trust as a probability signal, and laggy enough — capturing only about two-thirds of a real-time information move — to trade against when you can spot the gap.
The fast money is gone; the bots took it years ago and they’ll always be faster than you. What’s left is the slow money — the minutes-long window where a thin Kalshi market hasn’t yet caught up to the sharp consensus. That window is human-sized. It’s retail-sized. And it shows up far more often than a market this accurate has any right to allow, precisely because being accurate and being fast are not the same thing.
Kalshi is efficient. Just not fast. The whole game is living in the difference.
Findings referenced: Kalshi calibration (intercepts near zero, slopes near one; accuracy improving toward game start) and the ~0.64-for-one real-time response of Kalshi midpoints to benchmark probability moves come from recent academic work on Kalshi market microstructure and real-time price discovery (Bürgi et al., 2026; and related real-time prediction-market efficiency research, arXiv 2026). Arbitrage-window and cross-platform findings draw on the broader 2025–2026 prediction-market arbitrage literature. Figures are averages across large samples and do not guarantee any individual trade. Convergence trades carry real risk — prices can move against you, holding one side is not a hedge, thin markets can be hard to exit, and fees can erase small gaps. Nothing here is financial advice.