Brinson attribution answers the question every portfolio review eventually reaches: did we win because we overweighted the right segments, or because we picked the right securities inside them? The framework comes from Brinson, Hood and Beebower's 1986 paper "Determinants of Portfolio Performance," and forty years later it is still the standard first decomposition of active return against a benchmark.
The Setup
Take a portfolio and its benchmark, and cut both into the same segments — sectors are the classic choice, but countries, industries or asset classes work identically. For each segment i you need four numbers:
- wp,i — the portfolio's weight in segment i
- wb,i — the benchmark's weight in segment i
- rp,i — the portfolio's return inside segment i
- rb,i — the benchmark's return inside segment i
The whole game is explaining the active return — the portfolio's total return minus the benchmark's — as a sum of per-segment effects. Brinson attribution splits it into three.
The Three Effects
Allocation
Allocationi = (wp,i − wb,i) × (rb,i − rb)
The reward for overweighting segments that beat the overall benchmark and underweighting segments that lagged it — judged entirely on benchmark returns, so it isolates the top-down bet. Note the second term compares the segment to the total benchmark return rb: overweighting a segment that merely matched the benchmark earns nothing, which is exactly right.
Selection
Selectioni = wb,i × (rp,i − rb,i)
The reward for holding better securities inside a segment than the benchmark held — judged at benchmark weight, so it isolates the bottom-up skill from the sizing decision.
Interaction
Interactioni = (wp,i − wb,i) × (rp,i − rb,i)
The cross term: the extra return from having overweighted a segment where your picks also beat the segment. It is neither purely allocation nor purely selection, and it is the piece practitioners argue about.
The original 1986 model (BHB) reports interaction as its own line. The Brinson-Fachler variant folds interaction into selection by computing selection at portfolio weight — most commercial attribution systems do this, because a standalone interaction line is hard to assign to anyone's decision. When two reports disagree on selection, check which convention each used before blaming the data.
A Worked Example
Two segments, one period. The benchmark holds 50/50 Tech and Utilities; Tech returns 10%, Utilities 2%, so the benchmark earns 6.0%. The portfolio overweights Tech 70/30 and its picks return 12% in Tech and 1% in Utilities, earning 8.7% — an active return of +2.7 points to explain.
| Segment | wp | wb | rp | rb | Allocation | Selection | Interaction |
|---|---|---|---|---|---|---|---|
| Tech | 70% | 50% | 12% | 10% | (0.20)(10−6) = +0.80 | (0.50)(12−10) = +1.00 | (0.20)(2) = +0.40 |
| Utilities | 30% | 50% | 1% | 2% | (−0.20)(2−6) = +0.80 | (0.50)(1−2) = −0.50 | (−0.20)(−1) = +0.20 |
| Total | 8.7% | 6.0% | +1.60 | +0.50 | +0.60 |
The three effects sum to +2.7 points exactly — the decomposition is complete by construction, no residual. And it already tells a story the headline number hides: this manager's edge was mostly the allocation call (overweighting Tech and underweighting Utilities were each worth +0.8), while stock selection contributed only +0.5 net, because the good Tech picks were half undone by weak Utility picks.
The Multi-Period Problem
Brinson attribution is exact for a single period. Across many periods the effects do not compound cleanly: monthly allocation effects, multiplied through time, no longer sum to the multi-period active return, and a residual appears. Commercial systems close the gap with smoothing or linking algorithms (Cariño and Menchero are the common ones). The practical advice: treat linked multi-period effects as a well-motivated approximation, and be suspicious of any report that shows years of linked attribution with no note on the linking method.
What Brinson Cannot See
The framework is segment-blind by design: it credits "selection" for anything that beats the segment, without asking how. A portfolio that beats its sector benchmarks by systematically tilting toward, say, low-volatility or high-profitability names will show glowing selection skill — when the true driver is a factor exposure anyone could have bought. That is not a flaw in the arithmetic; it is a boundary of the question the model asks. The modern complement is factor-based attribution, which regresses the same active return onto factor exposures and asks how much survives as genuinely idiosyncratic.
- FM103's Post-Mortem applies the Brinson framework to factor strategies — decomposing a backtest's return into per-factor contributions rather than per-sector ones.
- The Specific Return panel isolates what no factor explains — the residual closest to true selection skill.
- Every attribution runs on simulated, point-in-time books: the same discipline the research shelf uses.
Start with Brinson when the question is organizational — who added value, the allocator or the picker. Move to factor attribution when the question is scientific — whether the "picking" was skill or a factor in disguise. The two answers disagree more often than either report admits, and the gap between them is usually the most interesting number in the review.