Skip to content

Adverse Selection

Definition

Adverse selection in financial markets refers to the risk that a market maker (or any liquidity provider) faces when trading against a counterparty who possesses superior information. The market maker posts a bid and ask; if the incoming order is from an informed trader who knows the true value of the asset is outside the current spread, the market maker will consistently lose. This is the cost of providing liquidity.

Historical context

Formalised in the Glosten-Milgrom (1985) model of the spread: the equilibrium bid-ask spread is set wide enough that the market maker breaks even across both informed and uninformed order flow. Kyle (1985) modelled the informed trader's optimal strategy as a linear demand schedule, giving the famous \(\lambda\) price impact coefficient.

How it works

In a market with mixed informed and uninformed traders: - Uninformed (noise) traders arrive randomly and provide profits to the market maker. - Informed traders arrive when they have private information; they always trade on the right side of the spread. - The market maker cannot distinguish between them in real time. - The spread must compensate: losses to informed traders are subsidised by profits from noise traders.

The permanent price impact of a trade is driven by the probability that it comes from an informed source. High adverse selection → wide spread, lower depth.

Empirical validation in this wiki

The flash crash analysis in explainable-crypto-microstructure provides a compelling natural experiment: - Taker strategy (aggressive orders): collapses during the flash crash — the taker is picking off stale quotes but also getting run over by informed sellers. - Maker strategy (passive orders): survives — consistent with makers withdrawing or resting orders that benefit from the post-crash mean reversion.

This supports classic adverse selection theory: during extreme events, the composition of order flow shifts heavily towards informed (or at least directional) flow, and liquidity provision becomes dangerous.

order-flow-filtration operationalises this: filters targeting the parent orders of executed trades isolate informed flow, strengthening the OBI-return signal.

Implications

  • Bid-ask spread is partially an adverse selection premium.
  • Market depth decreases when adverse selection risk is high (informed flow suspected).
  • Quote withdrawal (market makers pulling liquidity) is a rational response to detected informed flow.
  • order-flow-imbalance can be used to detect adverse selection risk in real time.

Open questions

  • How can market makers distinguish informed from uninformed flow in real time?
  • Do structural LOB filters (as in order-flow-filtration) reduce adverse selection costs?
  • Is adverse selection risk quantifiably higher in crypto markets than equity markets?

Connections