Representativeness Bias in CFA Level 3: Base-Rate vs Sample-Size Neglect (With a Model Answer)

Behavioural finance questions at Level 3 have a distinctive shape. You are given a short description of an investor doing something irrational, and you must name the bias, classify it, explain the consequence, and recommend a correction. Representativeness appears in that format more often than almost any other bias, partly because it has two sub-forms that examiners enjoy distinguishing between.

It is also one of the easiest biases to identify wrongly, because three other biases produce superficially similar behaviour.

What representativeness bias is

Representativeness is a cognitive error in the information-processing category. The investor classifies new information based on how closely it resembles a familiar category or stereotype, rather than on its objective statistical properties.

The classification matters more than candidates expect, because it determines the recommendation:

• Cognitive errors stem from faulty reasoning and can be substantially corrected through education, data and better process. The prescription is to moderate them.
• Emotional biases stem from feeling and impulse and are difficult to correct. The prescription is usually to adapt or accommodate.

Because representativeness is cognitive, an essay answer recommending that the portfolio be built around the client's tendency will lose marks. The answer is to correct it, and to say specifically how.

The two sub-forms

Base-rate neglect

The investor overweights the specific descriptive information in front of them and underweights the underlying probability of the category.

Classic case: an analyst learns that a company has a charismatic founder, operates in artificial intelligence, and grew revenue 200% last year. They conclude it will become a dominant platform. What has been ignored is the base rate — the proportion of high-growth start-ups that actually become dominant platforms, which is very small. The description is representative of a successful company, so the analyst assigns it the probability of a successful company.

Recognition cue in a vignette: vivid, specific detail about a single case, with no reference to how often such cases succeed.

Sample-size neglect

The investor draws firm conclusions from samples far too small to support them, treating a short run of data as a stable property. Sometimes called the law of small numbers.

Classic case: a fund manager beats the benchmark for three consecutive years and is described as skilled. Three years is a sample of three. Given the number of managers in the market, runs of that length occur constantly by chance — in a universe of a thousand managers, roughly 125 would post three straight winning years on coin flips alone. The investor sees a pattern representative of skill and infers skill.

Recognition cue in a vignette: a specific short track record, usually three to five years, used as the basis for a conclusion about ability.

This connects directly to the fundamental law of active management, which describes expected active return. Realised results over short horizons are dominated by noise, and the framework says nothing about any single year.

How it shows up in portfolios

• Performance chasing. Allocating to whichever fund or asset class performed best recently, assuming recent returns represent future returns.
• Excessive turnover. Repeatedly reclassifying holdings as new information arrives, generating transaction costs that erode returns. This is the most commonly examined consequence.
• Confusing a good company with a good investment. A well-run business with strong products is representative of a good stock, but the price may already reflect all of it.
• Over-extrapolating growth. Projecting a short growth run indefinitely into a valuation model, which is where representativeness feeds directly into inflated terminal values.
• Style drift by proxy. Selecting managers who resemble a mental template of a good manager — pedigree, presentation, conviction — rather than on evidence.

Distinguishing it from adjacent biases

Exam questions frequently offer two plausible labels, and marks turn on choosing correctly.

• Versus confirmation bias: confirmation bias filters evidence to support a view already held. Representativeness forms the view in the first place by pattern-matching to a stereotype. Sequence is the test — did the belief come before the evidence, or from it?
• Versus availability bias: availability relies on what comes to mind easily, usually because it was recent or vivid. Representativeness relies on resemblance to a category, regardless of how easily examples are recalled. If the vignette stresses a memorable recent event, it is availability.
• Versus hindsight bias: hindsight is retrospective — believing an outcome was predictable after it occurred. Representativeness is prospective.
• Versus illusion of control: that is a belief about influencing outcomes, not about categorising information.
• Versus overconfidence: overconfidence concerns the precision of one's own estimates. Representativeness can produce overconfidence, but they are classified separately — and overconfidence is often treated as having an emotional component, which changes the recommendation.

The practical test: if the investor is misjudging a probability by relying on how closely something resembles a category, it is representativeness.

How to correct it

Because it is a cognitive error, the prescription is procedural. High-scoring answers are specific rather than generic:

• Start from base rates. Establish the unconditional probability for the category before assessing the specific case, then adjust for case-specific evidence. This is Bayesian updating in practical form.
• Set minimum sample requirements in advance. Specify what constitutes evidence — for manager evaluation, multi-year records assessed against peer dispersion rather than three years of outperformance.
• Maintain a decision log. Recording the reasoning behind each decision at the time makes pattern-matching visible in review and separates process quality from outcome.
• Use an IPS with explicit quantitative criteria. Pre-committing to selection standards removes the discretion representativeness exploits.
• Rebalance on a schedule. Calendar or threshold rebalancing mechanically counteracts performance chasing without requiring the client to overcome the bias in the moment.

A model answer structure

Level 3 answers on behavioural biases are marked on structure as much as content. A complete answer has five components, and candidates routinely lose marks by omitting the third.

1. Name the bias and its sub-form.
2. Classify it as cognitive or emotional.
3. Cite the evidence from the vignette that supports the diagnosis.
4. State the consequence for the portfolio.
5. Recommend a correction matched to the classification.

Worked example. Vignette: a client instructs her adviser to move 30% of her portfolio into an emerging markets fund that has returned over 20% annually for the past three years, saying the manager has clearly demonstrated skill.

Model answer: "The client exhibits representativeness bias, specifically sample-size neglect. This is a cognitive error. The evidence is her reliance on a three-year track record as demonstrating manager skill; three annual observations are too few to distinguish skill from chance, particularly given the number of managers in the emerging markets universe. The consequence is performance chasing and a concentrated allocation made at a point of elevated valuation, with elevated turnover costs if the pattern repeats. As a cognitive error this should be moderated rather than accommodated: the adviser should present base-rate evidence on the persistence of manager outperformance, establish a minimum evaluation period and peer-relative criteria in the IPS before any manager allocation, and apply a threshold rebalancing policy to prevent allocations drifting toward recent winners."

That answer runs about 140 words and hits all five components. Skipping the evidence step is the single most common way candidates lose marks they had the knowledge to earn — the grader is looking for the link between the text and the label, so quote the vignette back. The Level 3 essay guide covers this answer discipline in depth, and the full behavioural finance guide maps the remaining biases.