If you only look at the quoted spread, Polymarket and Kalshi look like the same market. On MLB moneylines they both quote a median 1¢ bid-ask spread — tight, identical, unremarkable. But the spread is the cover of the book, not the book. Behind that same 1¢ quote, one venue can absorb roughly fifteen times more size before the price moves.
We can show this because we capture the full Level-2 depth live, tick by tick, on both venues — the data neither exchange publishes historically and nobody can reconstruct after the fact. What follows is a first, deliberately small look: ~3 days of Polymarket depth and 1 day of overlapping Kalshi depth on MLB game moneylines (June 21–23, 2026), off-season. Treat every number as illustrative, not seasonal truth.
The quoted spread is a decoy
Across the window, the top-of-book spread is a median 1.0¢ on both venues — and not just on average: the 25th and 75th percentiles are also 1¢. If you were shopping on spread alone, you'd flip a coin between the two books — which is why our prediction market platform comparison weighs fees, geography, and API access alongside liquidity.
Now ask the question that actually matters for anyone trading size: how much USD do you have to spend to walk the ask up by 1¢?
- Kalshi: a median of roughly $297,000 to move the ask 1¢.
- Polymarket: roughly $15,000–$17,000 to do the same.
Same quote, ~15–20× the depth on Kalshi. And it isn't one whale propping up a single market — the depth is broad-based: 12+ distinct Kalshi game tickers each show $400K–$580K of median depth to move 1¢. The gap holds across every price bucket, not just near 50/50.
Polymarket depth varies wildly by sport
The same measurement across Polymarket's sports shows how thin the long tail gets (median USD to move the ask +1¢):
| Market | ~USD to move ask +1¢ |
|---|---|
| FIFA World Cup | ~$52,000 |
| MLB | ~$15,000–$17,000 |
| League of Legends | ~$4,300 |
| WNBA | ~$2,000 |
| Tennis | ~$1,200 |
| CS2 | ~$650 |
| NBA (off-season) | ~$25 |
If you're backtesting an esports or tennis strategy and assuming you can get filled at the quote, this is the table that should worry you. A 1¢ "edge" means nothing in a book that moves 1¢ on a few hundred dollars.
The part that kills a popular strategy: imbalance has no signal
Order-book imbalance — the ratio of bid size to ask size — is one of the most-cited microstructure signals in equities. The folklore: when bids stack up, price drifts up. We tested it directly here.
At the ~30-second horizon we can observe, imbalance predicts essentially nothing:
- correlation(imbalance, next-snapshot mid move) = +0.007 (MLB), +0.010 (FIFA World Cup), −0.005 (Tennis)
- R² ≈ 0.00005
- ~89% of forward moves are exactly zero to begin with
That null is the honest headline. And unlike a positive finding, it's robust to small samples — autocorrelation can inflate a fake signal, but it cannot manufacture a zero. If you're building an imbalance-based bot on prediction markets, the burden of proof is on the signal, and at this horizon it isn't there.
What we are not claiming
Honesty matters more than a clean story, so the caveats are load-bearing:
- This is ~3 days of Polymarket depth and 1 day of Kalshi depth, off-season, preceded by a known capture gap (Apr 22–Jun 14). Off-season books are likely thinner and less representative than peak season.
- The exact Kalshi-vs-Polymarket multiple is day-sensitive — it rests on a single overlapping day of Kalshi depth.
- The book is captured at a ~20-level cap, so any "+5¢ depth" figure is a truncated lower bound.
- These are resting-book spreads, not the effective taker spread. The real cost of crossing adds ~3–5¢ of platform friction on top.
No tradeable edge is claimed. The point is the opposite: the quoted spread is not liquidity, and the most popular liquidity signal doesn't work here.
See the data
We publish a free, score-synced cross-venue matched-book sample (Polymarket × Kalshi on the same games, one timeline) on Hugging Face and Zenodo (DOI 10.5281/zenodo.20816908). Load it in DuckDB and reproduce the depth comparison yourself.
If you need the full historical depth archive — the part that can't be backfilled — start with how to get historical Polymarket order-book data and the data archives, or see our vendor-neutral comparison of where to get historical Polymarket and Kalshi order-book data, which covers competing providers too.