Following large trades on Kalshi: 14,000 tests of whether it works

When someone buys $500 or more of a single contract on Kalshi, should you follow them or bet against them?

It is a reasonable question with an unreasonable amount of folklore attached. Large trades get read as smart money by some people and as a liquidity event to be faded by others, and almost nobody has the data to say which. This lab papers both directions of every qualifying print of every qualifying print, the same trade followed and faded at the price the book was actually showing, so the question answers itself.

15,003 settled trades across ten market families.

Following beats fading, and it is not close

Following is roughly break-even before fees. Fading loses money outright, before you have paid the exchange anything. Whatever a large trade is on this venue, it is not noise to be sold into.

Where the price the whale paid changes everything

Splitting by what the whale themselves paid:

Two things fall out of that table.

Fading never works. Not in one band. Every cell is negative both before and after fees, and it gets worse as the whale pays more.

The expensive end is the only thing that clears the fee. Following someone who paid 86–97¢ is the single profitable cell in the grid, and it is profitable after the exchange takes its cut, which almost nothing else in this lab manages.

That is worth dwelling on, because it is the opposite of what I expected. The bands exist in the first place because an earlier, smaller sample suggested the high end was a trap: a whale bulldozing a market that was already priced correctly, with nothing left to capture. With far more data, the high band is the only one that pays.

It holds across markets, not just in aggregate

Following at 86–97¢, broken out by what is being traded:

Positive in most families rather than one outlier dragging an average, which is the difference between a finding and a fluke.

What this does not prove

The window is short. 7 days, 2026-07-28 to 2026-08-03. A large number of trades in a small number of days is still a small number of days, and any market can behave unusually for a week. This table gets republished as the sample grows, and the number worth trusting is the one that survives months.

Nothing here moved real money. Every trade is simulated at the ask the exchange was reporting, charged Kalshi's real fee. These are taker entries, so the rate is ceil(7 × price × (100 − price) / 10000) cents a contract; resting orders are charged nothing on all but about 130 series, which the fee study lists in full.

Most of these fills cannot be verified. The lab records order-book depth for Bitcoin and esports markets and can check entries against the size actually resting there. It does not record depth for challenger tennis, MLB, soccer, basketball or football, which is most of the whale grid, so those entries are taken at a reported price that cannot be re-checked afterwards.

And the effect is small. The best cell earns a fraction of a cent per trade. That is a real edge and a fragile one; it would not survive worse fills, and it says nothing about whether a person clicking manually could capture it.

The full board is published every week, and the per-trade record is available to members, so none of this needs taking on trust. If a figure here does not survive your arithmetic, say so.

The Seats lists every strategy this lab runs, including the whale grid's price bands and the Kalshi tickers each family covers.