Zero-Intelligence Trader Models

It's tempting to assume that every statistical pattern seen in real market data — clustered volatility, a particular shape to the order book, certain price-impact curves — reflects something meaningful about how traders think and react. Zero-intelligence trader models test that assumption directly by building a market simulation populated entirely with agents that submit orders completely at random: random side (buy or sell), random size, random price around the current mid, with no view, no strategy, and no reaction to what anyone else is doing.

The point of running such a deliberately dumb simulation is to see which market-level patterns show up anyway, purely as a mechanical consequence of how a limit order book and matching engine work — order arrival, execution priority, the bid-ask spread emerging from the interaction of many random quotes. If a statistical feature (say, the rough shape of the volume profile away from the mid-price) appears in both real markets and a zero-intelligence simulation with no strategic behavior whatsoever, that's evidence the feature is a structural artifact of market mechanics rather than something requiring an explanation rooted in trader psychology or strategy. Conversely, patterns that appear in real data but never emerge from random-order simulations point to genuine strategic behavior as the cause.

This makes zero-intelligence models a useful null hypothesis and baseline rather than a realistic trading environment: nobody uses one to predict prices, but they're a standard tool for separating "this pattern is just what order books do" from "this pattern requires real trader behavior to explain," which matters when deciding whether a newly observed microstructure effect is worth building a strategy around.