Skip to content

New to Omni?·Get 20% off your first purchase with code WELCOME20 at checkout.See plans

CFA® Level II: what really drives active portfolio risk?

A multifactor model splits tracking error into factor and security-specific sources. Work through a risk budget and learn what the split reveals about a manager.

ConceptOmni Curriculum Team7 min read

Grouped and scattered wooden blocks on a navy desk illustrate common and security-specific portfolio positions.
On this page

A multifactor model separates active risk into risk from factor exposures that differ from the benchmark and risk from positions that the factors do not explain. The first is active factor risk; the second is active specific risk. Their variances, rather than their standard deviations, add to total active variance under the model's assumptions, so a candidate should calculate the components before describing a portfolio as a stock-picking or factor-tilt strategy.

Start with the benchmark, then ask what moved

Active return is the portfolio return less the benchmark return. Active risk, often called tracking error, describes variation in that difference. It is possible for a portfolio to be volatile in absolute terms yet close to its benchmark, and for a seemingly calm portfolio to be quite different from a low-volatility benchmark. The benchmark therefore belongs in the question before any factor calculation begins.

The CFA Institute multifactor reading says models can help identify active decisions and attribute risk and return. It defines active risk as the standard deviation of active returns and distinguishes factor from specific sources. This is a Level II portfolio-management question about the mechanism of risk, not a claim that any particular portfolio will outperform. For study context, see the CFA programme page and how it works.

Imagine a benchmark with broad exposures to market, sectors and investment styles. A portfolio manager may own more of one sector and less of another, or prefer companies with one common characteristic. Those positions create a different response to common drivers. A model calls that active factor exposure. The manager may also choose individual companies within each sector. The part of those company returns not captured by the model's common factors is the specific component. Specific does not mean harmless: a concentrated company position can still dominate a portfolio's active outcome.

Why the two risk components do not add directly

Risk is measured as a standard deviation, but the clean decomposition works in variance units. Squaring a standard deviation gives variance. Under the usual model separation, the residual component is uncorrelated with the factors; that assumption allows their variance contributions to add. Adding the two reported percentage standard deviations directly would overstate total tracking error. The square root returns the sum of variances to the familiar standard-deviation scale.

σA2=σF,A2+σS,A2\sigma_A^2=\sigma_{F,A}^2+\sigma_{S,A}^2
Total active variance equals active factor variance plus active specific variance in the model decomposition.

Use invented, illustrative figures to see the point. Suppose a model reports active factor risk of 3 percentage points and active specific risk of 4 percentage points. Squaring produces variance contributions of 9 and 16 in percentage-point-squared units. Their sum is 25, so total active risk is 5 percentage points, not 7. Factor risk contributes 9 divided by 25, or 36 per cent of the modelled active variance; specific risk contributes 16 divided by 25, or 64 per cent. The percentages describe shares of variance, not shares of return or probabilities of success.

Now suppose another portfolio has the same 5 percentage points of total tracking error but factor risk of 4 and specific risk of 3. Its total looks identical in a single headline figure, yet its variance split is reversed: 64 per cent factor and 36 per cent specific. A risk budget or manager review that reads only total tracking error would miss that difference. The first portfolio's risk is more dependent on its individual positions after common exposures are accounted for. The second is more exposed to systematic tilts relative to the benchmark.

The relationship also works backwards. If total active risk is 5 and factor risk is 3, specific risk is the square root of 25 less 9, which is 4. Subtracting 3 from 5 to obtain 2 would be a tempting but wrong shortcut. Keep the component labels visible while calculating, so you do not accidentally attribute the 64 per cent share to the wrong source.

Multiple factors can interact within the factor component

The factor side is not necessarily a collection of independent boxes. If the portfolio differs from the benchmark on two factors, both the size of each exposure and the covariance between factor returns affect their combined risk. Two positive exposures to factors that tend to move together can create more risk than a simple sum of isolated contributions suggests. Exposures that offset each other may produce a smaller combined result. A factor-risk number from a model has already made choices about how to estimate that covariance.

For exam interpretation, distinguish an exposure from a realised return. A portfolio can have a large active factor tilt that does not help its realised performance during a particular period. It can also earn a positive active return from an individual position while the ex ante risk estimate remains large. Attribution of what happened and forecast risk for what might happen are related but different exercises. A model's risk decomposition tells you where uncertainty is expected to come from; it does not by itself identify skill.

Model definition matters as well. A narrow factor set leaves more movement in the residual bucket. Adding a relevant factor may reclassify some apparent security-specific movement as systematic exposure. That does not change the actual holdings or returns; it changes the model's explanation of them. Similarly, a change in benchmark can change active exposures without any trades. Candidates should state the benchmark and the model basis before treating the resulting split as a physical property of the portfolio.

Use the decomposition to ask a better management question

Suppose an investment mandate says the manager will select individual companies while avoiding large common tilts. A high factor share of active variance would prompt a question about unintended sector or style positions. Conversely, a strategy deliberately expressing a factor view might reasonably have a high factor share. The split is useful because it tests the claimed source of active decisions against the measured exposures. It is not a universal ranking in which one component is always better.

Tracking error should also be kept separate from the information ratio. CFA Institute defines the latter as mean active return divided by active risk. A portfolio with the same total active risk as another may have a different information ratio if its mean active return differs. A favourable historical ratio is a description of the sample, not a guarantee that the risk source will pay off. A small denominator can also make the ratio sensitive to estimation choices.

CFA Institute's discussion of tracking-error estimates notes that estimates produced from recent conditions can change over time. That is an important qualifier when using a model output as a risk budget. A manager may appear within a numerical limit while the nature of exposures changes; conversely, a shifting covariance estimate may alter reported risk even with little change in holdings. The right next step is to inspect exposures and assumptions, then use the total figure as a summary.

The practical exam check is short. Identify the benchmark, label common-factor deviations and residual positions, square the reported risk components, add variances and take the square root only at the end. When describing the result, say whether the percentage is a share of active variance or a standard deviation. Then explain what the split says about the manager's active decisions and what the chosen factor model cannot prove.

A factor example with covariance

Consider an invented portfolio with active exposure to two broad factors. If each factor exposure contributes an illustrative 2 percentage points of stand-alone risk, it is tempting to call their combined factor risk 4 percentage points. That assumes perfect positive co-movement. If the factors are uncorrelated in the chosen model, their combined standard deviation is the square root of 4 plus 4, or about 2.83 percentage points. If they move together positively, combined risk can be higher; if they offset, it can be lower. The exposure signs and estimated covariance are essential to interpreting a multifactor report. This calculation is separate from adding the completed factor bucket to the specific bucket. Within the factor bucket, correlations among factor returns affect the result. Between the completed factor bucket and the residual specific bucket, the model decomposition assumes the residual is uncorrelated with factors. Skipping the first step can misstate factor risk; skipping the second can misstate total active risk. A factor label is not a self-contained source of risk independent of every other factor. For a risk review, ask whether the chosen factors capture the drivers that matter for this benchmark and portfolio. A change in factor definitions can move risk between named factor and residual categories. A change in estimation window can change covariance while exposures are stable. These are reasons to read the model specification alongside its output.

Sources

  1. Using Multifactor Models | CFA Institute · cfainstitute.org · Retrieved
  2. Comparing Ex-Ante Tracking Error Estimates across Time | CFA Institute · rpc.cfainstitute.org · Retrieved

Explore CFA study resources

Educational content, not investment advice. This article is written to help candidates prepare for professional exams. It is not a recommendation to buy, sell or hold any security, and it does not take your personal circumstances into account.

Omni Finance Academy is an independent exam-preparation provider and is not affiliated with or endorsed by CFA Institute, GARP, CAIA Association or CFP Board. CFA®, FRM®, CAIA® and CFP® are trademarks of their respective owners. Full disclaimer.