On January 14, a political market on Polymarket dropped from $0.55 to $0.32 in 8 minutes. A single trader panic-sold $40K of YES shares after a misleading tweet from an unverified account went viral. Within 2 hours, the price was back at $0.51.

That 19-cent round trip was not an anomaly. Prediction markets overreact to news constantly, and it's one of the most consistent sources of alpha we've found.

1Why prediction markets overreact

Stock markets have circuit breakers, market makers with deep pockets, and institutional investors who buy dips rationally. Prediction markets have none of that.

Thin order books amplify moves

A $20K market order on a $100K market can move the price 15-20%. The same order on the S&P 500 would move the price 0.000001%. Prediction markets are structurally prone to outsized moves.

Retail traders panic

When a price drops sharply, other holders see their position losing value and sell too. This cascade creates a spike far larger than the underlying information warrants.

No automatic stabilizers

There are no market makers obligated to provide liquidity. No circuit breakers to halt trading. No algorithmic stabilization. When the order book is thin, it stays thin.

2Detecting the spike

We define an overreaction as a price move of 15% or more within 30 minutes that isn't supported by a confirmed, material news event. The system monitors every market's price history in rolling 30-minute windows and flags any that cross the threshold.

When a spike is detected, the LLM checks whether there's a legitimate reason for the move. It scans news APIs, Twitter/X, and official sources. If the move is supported by confirmed news, we don't fade it. It's a real repricing, not a panic. If the move is driven by unverified rumors, a single large trader, or cascading sells with no new information, it's a candidate for fading.

15%+Spike threshold

30 minDetection window

AI-verifiedNews check

3Fading the move

"Fading" means trading against the spike. If the price dropped, we buy. If it spiked up, we sell. The thesis is simple: the price will revert toward the pre-spike level because the spike was driven by panic, not information.

We don't bet on a full reversal. Our target is a 50-70% retracement of the spike. If a market dropped from $0.55 to $0.32 (a 23-cent move), we target a recovery to roughly $0.44-0.48. This conservative target means we capture the easy part of the reversion without holding through the noise at the extremes.

Entry and exit mechanics

We enter with a limit order at the spike price (not at the current price after partial recovery). If the spike is still happening, we get filled at maximum dislocation. The exit is a limit order at our target retracement level, with a 24-hour time stop. If the price hasn't reverted in 24 hours, the spike may have been real and we cut the position.

4When NOT to fade

The most dangerous mistake in this strategy is fading a move that's actually correct. A 20% drop because a major news event just happened isn't an overreaction. It's the market doing its job.

We never fade these

Confirmed news from verified sources: If AP, Reuters, or an official government source confirms the information, the price move is justified.
Markets near resolution: In the final days before resolution, prices should be moving toward $0 or $1. Large moves are expected, not overreactions.
Correlated moves across multiple markets: If five related markets all move in the same direction, it's probably real information, not a single-market panic.

5The edge in numbers

Overreaction fading produces fewer trades than our other strategies, typically 2-5 per week. But the edge on each trade is consistent: 4-8% returns with a high win rate because we're only fading moves that the AI has confirmed are not supported by real news.

4-8%Average edge

2-5/weekTrade frequency

50-70%Target retracement

24hrMax hold time

When everyone else is panicking, the bot stays calm. That emotional discipline, or rather, complete lack of emotion, is the real edge. The math just tells it where to trade.