Skip to content

FRM Exam Part II · The Financial Stability Implications of Artificial Intelligence

AI Herding, Market Correlation and Volatility

Updated 11 October 2026 · Fact-checked

AI herding is when many firms use similar models, data and strategies, so they trade the same way at the same time. This raises correlation, amplifies price moves and makes markets procyclical. To answer exam questions, identify the common driver, the feedback loop and the systemic outcome.

Understand Market Correlations and Herding Behavior

Start with diversification. Markets stay stable partly because participants hold different views, use different data and trade at different times. When one seller exits, another buyer steps in.

AI can weaken this. If many firms buy the same vendor model, train on the same data, or use similar machine learning methods, their signals look alike. They then buy and sell together. This is herding: many participants take the same action because of a common input or a common method, not because they copied each other on purpose.

Herding raises correlation. Assets and strategies that used to move independently begin to move together, especially in stress. Diversification benefits that a risk model assumed in calm periods can disappear when they are needed most.

Procyclicality is a different idea. It means the system amplifies the economic or market cycle. Models that cut risk when volatility rises force selling in falling markets and buying in rising markets. Many similar AI models reacting to the same volatility signal make this stronger. Speed matters too: automated trading can turn a small shock into a fast price spiral, as in a flash crash, where prices drop sharply and recover within a short time as liquidity vanishes.

The result is higher volatility, thinner liquidity and greater systemic risk. Supporting factors are model uniformity, data concentration (a few data or model providers), limited explainability, and automated execution that leaves little time for human intervention.

Key formulas to remember

Portfolio variance of two assets
σp² = w1²σ1² + w2²σ2² + 2·w1·w2·ρ·σ1·σ2
Higher correlation ρ raises portfolio risk. Herding pushes ρ up, so diversification benefit falls.
Perfect correlation limit
If ρ = 1, σp = w1σ1 + w2σ2
No diversification benefit. Used to show the worst case when everyone trades alike.
Herding causal chain
Common models/data → similar signals → synchronized trades → higher correlation → price impact → volatility and liquidity strain → feedback
Use this chain to structure any conceptual answer.

How to solve Market Correlations and Herding Behavior questions

Use this method for any scenario question on AI, correlation and herding.

  1. 1Read the scenario and find the common element: shared vendor model, shared data, similar training method or similar objective.
  2. 2Decide the mechanism: herding (same action), procyclicality (amplifying the cycle) or both.
  3. 3Trace the feedback loop: signal, trade, price move, new signal, more trades.
  4. 4Name the market effect: higher correlation, higher volatility, lower liquidity or a flash crash.
  5. 5Link to the risk measure: correlation rises, so portfolio VaR rises and diversification falls. Calm-period estimates understate risk.
  6. 6Pick the mitigation that matches the cause: model and data diversity, circuit breakers, human oversight, stress tests, monitoring of concentration.
  7. 7Check each option for absolute words like always or eliminates, and choose the most precise answer.

Quickest way: Common driver, feedback, effect

When to use it: Use for conceptual multiple-choice questions with four plausible options.

  1. Ask: what do the firms share? Models, data or providers point to herding.
  2. Ask: does the reaction amplify the cycle? If yes, it is procyclicality.
  3. Pick the option that links the shared input to correlated trading and volatility.
  4. Reject options claiming AI always diversifies or always reduces risk.

Common mistakes in Market Correlations and Herding Behavior

  • Treating herding and procyclicality as the same thing.

    Both amplify market moves, so they seem identical.

    Fix: Herding is many participants acting alike. Procyclicality is amplification of the cycle, often through rules or models that cut risk in downturns. They reinforce each other but are distinct.

  • Assuming herding needs participants to communicate or copy each other.

    The everyday meaning of herding implies imitation.

    Fix: In AI contexts herding often arises from independent use of the same models or data, with no coordination.

  • Saying correlation rises so individual asset volatility must fall.

    Mixing up portfolio risk and single-asset risk.

    Fix: Higher correlation raises portfolio risk by cutting diversification. Volatility of individual assets can also rise from synchronized trading.

  • Claiming AI always increases systemic risk.

    Overgeneralizing from flash crash stories.

    Fix: AI can improve risk management and liquidity provision too. The risk comes from uniformity, concentration, speed and weak oversight.

  • Choosing a mitigation that does not address the cause.

    Memorizing a generic list of controls.

    Fix: Match the control: diversity of models and data for herding, circuit breakers for speed-driven spirals, stress testing for correlation breakdown.

Worked examples

Example 1

A fund holds two equally weighted assets, each with volatility 20%. With correlation 0.2, compute portfolio volatility. Then compute it if AI-driven herding pushes correlation to 0.8. Comment.

Show the solution
  1. Weights are 0.5 each, so σp² = 0.25×0.04 + 0.25×0.04 + 2×0.25×ρ×0.04.
  2. That is 0.01 + 0.01 + 0.02ρ = 0.02 + 0.02ρ.
  3. For ρ = 0.2: σp² = 0.02 + 0.004 = 0.024, so σp = √0.024 ≈ 15.49%.
  4. For ρ = 0.8: σp² = 0.02 + 0.016 = 0.036, so σp = √0.036 ≈ 18.97%.
  5. Portfolio volatility rises by about 3.5 percentage points, so VaR scales up by roughly 22%.

Answer: Portfolio volatility rises from about 15.5% to about 19.0%. Herding-driven correlation erodes diversification, so a VaR model calibrated on the low-correlation period understates risk.

Example 2

Many asset managers license the same AI volatility-targeting model. A volatility spike triggers all of them to cut equity exposure at once. Which description fits best: herding only, procyclicality only, or both? Explain the market effect.

Show the solution
  1. The common element is the shared model, so firms act alike on the same signal. This is herding.
  2. The signal is rising volatility, and the action is selling in a falling market. This amplifies the downturn, which is procyclicality.
  3. Selling pushes prices down and volatility up, which triggers further cuts. This is a feedback loop.
  4. Liquidity thins as many sellers face few buyers, raising correlation across holdings.

Answer: Both. Shared model use causes herding, and volatility-based de-risking is procyclical. The result is a self-reinforcing sell-off with higher volatility, higher correlation and weaker liquidity.

Exam tips

  • Look for the shared input in the question stem. It usually signals the correct answer.
  • Keep herding, procyclicality and flash crash separate in your head. Options often swap them.
  • Be wary of absolute words such as always, eliminates or guarantees.
  • For numeric items, recompute portfolio risk with a higher ρ and state that diversification falls.
  • For mitigation questions, choose the control that targets the stated cause.

Practice questions from The Financial Stability Implications of Artificial Intelligence

Market Correlations and Herding Behavior in other exams

The same ground in other exams, if you are preparing for more than one or want another angle on it.

Market Correlations and Herding Behavior: frequently asked questions

How can AI models increase market correlation and volatility?

When firms use similar models, data or vendors, their signals and trades align. Synchronized buying and selling moves prices together, raises correlation and can amplify volatility, especially when liquidity is thin.

What is the difference between herding and procyclicality?

Herding is many participants taking the same action, often because of shared models or data. Procyclicality is the amplification of the market or economic cycle, for example by de-risking in downturns. They often occur together.

Can AI trading cause a flash crash?

It can contribute. Fast automated trading on similar signals can drain liquidity and push prices sharply in minutes. Circuit breakers, human oversight and monitoring are the typical safeguards.

Does AI always raise systemic risk?

No. AI can improve risk measurement and liquidity provision. Systemic risk rises mainly when models, data and providers are concentrated and oversight is weak.