Charta is on Google Play: practise on the go. Get the app
Charta

Account for what wider AI adoption could mean for the way markets behave.

FRM® Part II · Current Issues in Financial Markets · Advances in Artificial Intelligence: Implications for Capital Markets Activities · FRM2.CI.1.b

The idea

Wider AI adoption could change how markets behave in both directions. In normal conditions, faster information processing may improve price discovery, liquidity and execution quality. The concern is correlation. If many firms rely on the same foundation models, vendors and datasets, their signals and trades become similar. That produces herding, procyclical selling and amplified price moves in stress. Speed adds to this, because reactions that once took hours happen in milliseconds. AI-driven market makers may widen quotes or withdraw together, so liquidity can vanish when it is needed. Other behavioural risks are algorithms that learn tacit collusion without any communication, and AI-generated false information that automated news-reading systems act on. The usual error is assuming that more sophisticated models mean more diverse behaviour. The exam point is that common models and data create correlated behaviour and procyclicality.

A practice question

From the question bank, with the reason each wrong option is wrong.

Which is the main way that widespread adoption of similar AI models could change market behaviour in stressed conditions?

  1. Elimination of volatility because models always anticipate shocksWhy not: Assumes models foresee shocks; common signals tend to amplify, not remove, stress moves.
  2. Less need for liquidity because positions would be netted automaticallyWhy not: Nothing nets positions automatically; liquidity needs actually rise when many firms trade the same way.
  3. Slower reaction times because models require manual approvalWhy not: Reverses the effect; AI typically speeds up reaction to news and prices.
  4. More correlated trading and herding, which can amplify price moves - correct

If firms use the same models, data and vendors, their signals converge. Simultaneous selling creates herding and procyclicality, which amplifies moves in stress.

A harder one

Many asset managers fine-tune the same foundation model on the same vendor data set. A shock pushes the models' risk signals below their thresholds on the same day. Which market dynamic is most likely?

  1. Diversification of trades that dampens volatility across the whole market and all the managersWhy not: Wrongly assumes diversified behaviour; common models give similar, not diverse, trades.
  2. Synchronised selling that amplifies the fall (procyclical herding) - correct
  3. Idiosyncratic price moves in each manager's holdings onlyWhy not: Common signals produce correlated, market-wide moves, not manager-specific ones.
  4. Immediate mean reversion because the models detect valueWhy not: Nothing in the case suggests buyers appear; the models trigger sells at the same time.

Shared models and data produce shared signals. Simultaneous selling pushes prices down, which triggers further selling and amplifies the move.

The hardest tier

Reinforcement learning pricing algorithms at several firms in a thin market gradually learn to keep spreads wide, although the firms never communicate and use different data. Which concern does this describe?

  1. Herding caused by a common data set shared by all of the firms involvedWhy not: The firms use different data and nothing suggests shared signals drive the outcome.
  2. Spoofing, where orders are placed with no intention to execute and are cancelled to mislead othersWhy not: Spoofing involves placing and cancelling orders to mislead; the case describes learned wide quotes.
  3. Insider trading using privileged information obtained from a corporate sourceWhy not: No privileged information is involved; the algorithms learned from market outcomes.
  4. Algorithmic tacit collusion, which rules based on proof of agreement may miss - correct

Without any communication, learning algorithms can settle on collusive-looking outcomes. Rules that rely on proof of agreement or intent may not capture this.

In the app: 6 practice questions on this outcome, spaced review that brings it back before you forget it.

Start free →

More in Advances in Artificial Intelligence: Implications for Capital Markets Activities