"Bonding Bots" in Prediction Markets
If you buy a dollar for 98 cents often enough, eventually you will buy something that is not a dollar.
In prediction-market circles, a “bond” is a contract trading very close to $1 because the outcome looks decided:
The game is in the 9th and one team is up eight runs
The election has been called
The movie has already premiered, whatever.
A “bonding bot” systematically buys these near-certain contracts and collects the last few cents at settlement. Institutional bond traders typically call it pull to par, as you’re buying something at a discount to face value and clipping the yield as it matures.
The math on it is honestly seductive, at first:
Buy YES at $0.98.
Contract settles at $1.00.
Collect $0.02, roughly 2% on capital, often within minutes.
Recycle the capital into the next “decided” event.
Repeat across every market on the board, every night.
Run that loop across a full slate of outcomes and the equity curve looks gorgeous. Straight line up, win rate north of 99%, the kind of thing that screenshots well.
The problem, however, is the skew. When you buy at $0.98, you’re risking 98 cents to make 2, so one miss wipes out roughly 49 perfect trades.
The payoff profile (chart below) is a long staircase of two-cent steps up, and a trapdoor that gives back all of them at once:
It’s the same profile as selling deep out-of-the-money puts: a long stream of small wins funding one occasional catastrophic loss.
The Run That Didn’t Count
On July 24, that trapdoor opened at Oracle Park, a little after 1 a.m. Eastern.
We’re not a sports publication, so to keep things high-level:
In the final inning of a tie game, Team A made a play that visually won it by multiple runs
Data feeds posted the multi-run final and bots rushed the “win by 2+” contracts up to $1
The official scoring was subsequently corrected under an MLB rule
The “bond-like” contracts closed the market at $0, with aggregate bonder losses in the seven figures.
This was the last game of the night, so a good chunk of that carnage likely happened while the owners of those bots were asleep. Kalshi appears to have paid traders on both sides afterward, though there’s no official postmortem on the mechanics yet:
Now, believe it or not, even after a face-plant like that, running one of these can be genuinely profitable with a bit of sophistication.
Well, Maybe a Bit of Free Money
A naive bonding program essentially buys anything with a high price:
while markets_open:
for market in board:
if feed_status == "final" and ask < $1.00:
buy(everything)
In contrast, a more sophisticated operator knows which two cents are actually free, and it generally doesn’t even take that much.
Keep reading with a 7-day free trial
Subscribe to Alphanume Research to keep reading this post and get 7 days of free access to the full post archives.