Building defensible performance attribution

Why attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → exists

A benchmark is a reference portfolio you measure against, for example the S&P 500 or a blended index for a balanced fund. Excess return (also called active return) is simply your portfolio's return minus the benchmark's return.

Excess return is a single number. It hides everything interesting. A manager could be great at picking stocks but terrible at deciding how much to hold in each sector, and the two effects might cancel out.

AttributionAttributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → splits that single number into its sources so you can tell skill from noise.

The Brinson-Fachler model

The most widely used framework is Brinson-Fachler attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition →, published by Gary Brinson and Nimrod Fachler in the 1980s. It answers two questions for every segment (usually a sector or asset class):

• Allocation effect: Did you make money by overweighting or underweighting segmentssegmentsDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.View full definition → relative to the benchmark?
• Selection effect: Within each segment, did the specific securities you chose beat the benchmark's securities in that segment?

There is also an interaction effect, a small cross-term that captures the overlap between the two. Many shops fold it into selection to keep reports clean.

The formulas, in plain terms

For each segment *i*, define four inputs:

• wp = portfolio weight in the segment
• wb = benchmark weight in the segment
• Rp = portfolio return in the segment
• Rb = benchmark return in the segment
• Rt = total benchmark return

The Brinson-Fachler effects for segment *i*:

Sum across all segmentssegmentsDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.View full definition → and the three totals reconcile to your excess return.

Note the allocation term. It compares each segment's benchmark return to the *total* benchmark return (Rb_i - Rt). This is the Fachler refinement. It rewards you for overweighting segmentssegmentsDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.View full definition → that beat the overall market, not just segmentssegmentsDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.View full definition → with positive returns. That distinction matters and is the main reason Brinson-Fachler is preferred over the older Brinson-Hood-Beebower variant.

A worked example

Take a simple balanced fund with three segmentssegmentsDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.View full definition →: Equities, Bonds, and Cash. Here is one quarter (illustrative numbers, not real market data).

| Segment | wp | wb | Rp | Rb |

|---|---|---|---|---|

| Equities | 60% | 50% | 8.0% | 7.0% |

| Bonds | 35% | 45% | 2.0% | 1.5% |

| Cash | 5% | 5% | 1.0% | 1.0% |

First, total benchmark return:

Rt = 0.50*7.0 + 0.45*1.5 + 0.05*1.0 = 4.225%

Portfolio return:

Rp = 0.60*8.0 + 0.35*2.0 + 0.05*1.0 = 5.55%

Excess return: 5.55 - 4.225 = 1.325%. Now decompose it.

Allocation (using Rb_i - Rt):

• Equities: (0.60 - 0.50) * (7.0 - 4.225) = 0.10 * 2.775 = 0.2775%
• Bonds: (0.35 - 0.45) * (1.5 - 4.225) = -0.10 * -2.725 = 0.2725%
• Cash: (0.05 - 0.05) * (1.0 - 4.225) = 0
• Total allocation: 0.55%

Selection (wb_i * (Rp_i - Rb_i)):

• Equities: 0.50 * (8.0 - 7.0) = 0.50%
• Bonds: 0.45 * (2.0 - 1.5) = 0.225%
• Cash: 0
• Total selection: 0.725%

Interaction:

• Equities: 0.10 * 1.0 = 0.10%
• Bonds: -0.10 * 0.5 = -0.05%
• Total interaction: 0.05%

Check: 0.55 + 0.725 + 0.05 = 1.325%. It reconciles.

The story: the manager added value both ways. She overweighted equities (which beat the total benchmark) and underweighted bonds (which lagged), earning 0.55% from allocation. She also picked better securities inside each segment, earning 0.725% from selection. Selection was the bigger driver here.

That is a defensible narrative. Now let us see how the data underneath can wreck it.

Where the numbers quietly break

AttributionAttributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → math is trivial. The data feeding it is where reputations are won and lost.

1. Benchmark misalignment

If your fund holds small-cap value stocks but you attribute against a large-cap blend index, the selection effect will look enormous. You are not measuring skill. You are measuring the gap between two different universes. Always confirm the benchmark actually represents the mandate. Check the index methodology document, not just the ticker.

2. Stale or wrong holdings

AttributionAttributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → needs holdings and weights *as they actually were* over the period, ideally daily. If your system uses month-end snapshots and the manager traded heavily mid-month, the model attributes returns to positions that were not held. This is a classic source of unexplained residual, the leftover that will not reconcile.

3. Corporate actions

Stock splits, spin-offs, mergers, and dividends change share counts and prices. If your holdings data does not adjust for a split, a position's return can look like a 50% crash overnight. Providers like Bloomberg and Refinitiv handle these, but reconciliation errors still slip through.

4. Classification drift

Which sector does a diversified conglomerate belong to? Different index vendors use different schemes (for example GICS versus ICB). If your portfolio and benchmark are classified under different systems, allocation and selection effects get scrambled at the segment boundary. Use one classification consistently for both sides.

5. Currency

For multi-currency portfolios, return depends on the FX rate used and its timing. A poorly specified currency effect leaks into allocation and selection. Global mandates should use a model that isolates currency as its own effect.

The uncomfortable truth: a clean-looking attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → report can be entirely spurious if any of these inputs is off. Clients trust the decimals. The decimals are only as good as the data pipelinedata pipelineETL (Extract, Transform, Load) is a data integration process that pulls data from sources, reshapes it into a consistent format, and writes it into a target system.View full definition → behind them.

🎬 [VIDEO: "Performance AttributionAttributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → Explained (Brinson Model)" — youtube.com — a clear visual walkthrough of allocation versus selection effects with a worked example]

Making attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → defensible

"Defensible" means an auditor, a consultant, or a skeptical client can trace every number back to source data and reproduce it. Practical standards:

Reconcile to the official return first. Before you decompose anything, confirm the total portfolio and benchmark returns match your accounting system and the index provider. If the top-line number is wrong, the decomposition is theater.

Use daily data and geometric linking for multi-period reports. Single-period Brinson effects sum cleanly. Across multiple periods, returns compound, so effects do not simply add up. Standard linking methods (for example Cariño or Menchero smoothing) reconcile the multi-period total. Know which one your system uses and why.

Keep a residual line and watch it. A small residual is normal. A growing or erratic residual is a data alarm. Never suppress it to make the report look tidy.

Align with GIPS where relevant. The Global Investment Performance Standards from the CFA Institute govern how firms present performance to prospective clients. AttributionAttributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → is not strictly required under GIPS, but the same data discipline (accurate holdings, consistent valuation, documented methodology) applies directly.

Document the methodology. State the model (Brinson-Fachler), the classification scheme, the linking method, the currency treatment, and the data sources. A report without this cannot be defended when questioned.

A quick sanity checklist

Before you send an attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → report, confirm:

1. Total returns reconcile to the accounting system.

2. Portfolio and benchmark use the same classification and the same period.

3. Holdings reflect actual daily positions, adjusted for corporate actions.

4. The residual is small and explained.

5. The methodology is written down and matches what the system actually did.

Most attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → disputes trace back to a failure in one of these five, not to the math.

Key Takeaways

• Attribution turns a single excess-return number into a story of decisions. Brinson-Fachler splits it into allocation (weighting choices) and selection (security choices), which reconcile back to total excess return.
• The Fachler refinement matters: allocation is scored against the total benchmark return, so you are rewarded for overweighting segmentssegmentsDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.View full definition → that beat the whole market, not just segmentssegmentsDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.View full definition → that were merely positive.
• The math is easy; the data is hard. Benchmark misalignment, stale holdings, unadjusted corporate actions, and inconsistent sector classification silently corrupt results.