Mean-Reversion Strategies for Prediction Markets

Prediction markets have grown fast. Combined trading volume on Kalshi and Polymarket exceeded $44 billion in 2025, up from roughly $16-18 billion in combined 2024 volume, with monthly volume hitting nearly $24 billion by April 2026. More volume means more noise, more overreaction, and more potential edges for systematic traders. Mean reversion is one of the oldest strategies in quantitative finance. Applying it to binary-outcome contracts requires understanding one structural risk that does not exist in equities: resolution. A stock can always revert. A contract that resolves to $0 or $1 cannot.

Why Prediction Markets Overreact

The mechanism is the same as in any market. Heavy reliance on current events means participants can overreact, pricing new information more aggressively than the probability update it actually warrants. A breaking news headline about a candidate's health, an unexpected jobs print, or a viral social post can push a contract 10 to 20 cents in minutes. Much of that move is noise. Retail participants rush the same side, liquidity thins, and the spread widens. Once the dust settles and better-informed participants re-enter, prices drift back toward prior equilibrium.

Practitioner research on prediction market microstructure consistently points to one execution constraint: mean-reversion signals tend to generate alpha only under passive limit-order execution. When aggressive market orders are used, performance degrades significantly because the spread cost eats the edge. That single finding should shape your entire execution approach: this strategy lives or dies on maker fills, not taker fills.

Identifying Candidates

Not every sharp move is an overreaction. The checklist below is a starting filter, not a signal on its own.

• Price moved more than 12-15 cents within a 2-hour window with no confirmed resolution-relevant event (i.e., the event has not actually happened yet).
• Volume spike accompanied the move, but open interest did not shift proportionally. That pattern suggests reactive retail trading rather than informed positioning.
• Resolution date is at least 7-14 days out. Short-dated contracts amplify resolution risk dramatically. A contract expiring tomorrow has almost no time for a price to revert before it settles.
• The contract is on a slow-moving underlying: economic data releases, multi-week political polling averages, weather outcomes, Fed policy. Avoid contracts where a single tweet can permanently change the true probability.
• The bid-ask spread at rest (before the spike) is 1-3 cents. Wider spreads mean you need a larger reversion just to break even on fees and slippage.

The Resolution-Risk Trap

This is the part most traders underestimate, and it is specific to binary contracts. In equities, a stock that drops 15% on bad earnings can recover over weeks. A prediction market contract that moves to 10 cents because the outcome is now unlikely will settle at 0 if the outcome does not occur. That is not mean reversion. That is the market being right.

The trap looks like this: you see a contract for 'Will X happen by month-end?' drop from 55 cents to 30 cents on news. You fade it, expecting a bounce. But the news was actually correct, and the contract grinds to zero over the next two weeks. You averaged down the whole way.

Three rules that limit resolution risk:

• Only fade moves when you have an independent probability estimate for the underlying event. If your prior is 50% and the contract just moved to 30%, you have a thesis. If you have no prior and you are just pattern-matching a big candle, you do not.
• Size inversely proportional to contract age. A 90-day contract has time to revert. A 3-day contract does not. Position sizing should reflect that asymmetry.
• Set a hard stop at a price that would imply the event is nearly certain to fail or succeed, not at an arbitrary dollar loss. A 15-cent stop on a contract you entered at 30 cents is mechanical. A stop at 12 cents, reasoning that a contract below 10 cents is near-terminal and will not revert meaningfully, is probabilistic.

Where to Trade: US Regulatory Status

US persons have two legal options. Kalshi received its CFTC Designated Contract Market (DCM) designation in November 2020, covering economics and politics initially; sports contracts were added in January 2025. Polymarket re-entered the US market through a multi-step process: it acquired QCEX, a CFTC-licensed derivatives exchange and clearinghouse, for $112 million in July 2025. The CFTC then issued an Amended Order of Designation in November 2025, and Polymarket US launched on December 2, 2025, with full KYC requirements. The international polymarket.com site blocked US users from 2022 through December 2025 following a CFTC enforcement action and $1.4 million civil penalty for operating an unregistered derivatives facility. Using a VPN to access the international site violates Polymarket's terms of service and carries no regulatory protection. For most US-based algo traders, Kalshi is still the more established venue with a longer compliance track record. Both platforms offer CLOB (central limit order book) structures, which matter for limit-order-centric strategies.

A Paper-First Implementation with Banger

The worst way to discover flaws in a mean-reversion strategy is with real money. Paper trading against the live order book is the right starting point, especially for a strategy whose profitability is so sensitive to fill quality. Banger (bangertrades.com) is a Python strategy runtime for Kalshi and Polymarket that lets you run a strategy in paper mode with one flag before ever connecting venue keys.

pip install bangertrades
banger run strategy.py --paper

A minimal mean-reversion strategy in Banger looks like this:

import banger

class MeanReversion(banger.Strategy):
    # Risk envelope: declared once, enforced by the runtime
    max_position_size = 50        # dollars per contract
    daily_loss_stop   = 200       # kill trading for the day
    max_open_positions = 4        # no more than 4 contracts at once

    def on_price_update(self, market, price, prev_price):
        move = price - prev_price

        # Only act on sharp drops with enough time to resolution
        if move < -0.12 and market.days_to_resolution > 7:
            if self.my_prior(market) > price + 0.10:  # your edge
                self.submit_limit_order(
                    market=market,
                    side="yes",
                    price=price + 0.01,  # passive, not taker
                    size=self.max_position_size,
                )

    def my_prior(self, market):
        # Replace with your own probability model
        return market.last_week_avg_price

The risk envelope (position cap, daily loss stop, max open positions) is declared at the class level and enforced by the runtime. You cannot accidentally override it in on_price_update. That matters for a strategy that can generate many signals quickly during a volatile news cycle. Banger never custodies funds; you bring your own Kalshi or Polymarket API keys.

What to Watch in Paper Mode

Run paper mode for at least two to four weeks across different market conditions before going live. The key metrics to track are not just P&L.

• Fill rate on limit orders. If fewer than 40% of your limit orders fill, your entry price assumptions are wrong and you are measuring alpha you cannot actually capture.
• Average holding period to reversion. If most positions take 4+ days to revert, you are accumulating significant overnight resolution risk.
• Loss distribution by contract age. Separate P&L for contracts with under 7 days to resolution versus over 14 days. The short-dated bucket will almost certainly drag down returns.
• False-signal rate. Count how many faded moves continued directionally rather than reverting. A rate above 40% suggests you are chasing noise rather than genuine overreaction.

The Bottom Line

Mean reversion in prediction markets is a real edge, but it requires a probability anchor, not just a price signal. Without a view on where the true probability sits, a sharp price drop is indistinguishable from the market correctly pricing in bad news. The binary resolution structure creates an asymmetric loss profile that equities traders are not used to: your upside is bounded by reversion to fair value, and your downside extends to zero. Build the resolution-risk rules first, validate fill quality in paper mode, and only then think about scaling.