Performance Attribution: How the Numbers Actually Reconcile

A portfolio returns 11.7% against a benchmark of 10%. Everyone in the room agrees on the 1.7 points. Nobody agrees, yet, on where they came from.

That is the actual question performance attribution answers, and it is why it shows up in almost every technical round: not "how much did you beat the benchmark by" but "which of two different skills earned that number." One is choosing the right sectors. The other is choosing the right names inside them. A manager can be excellent at one and mediocre at the other, and a client paying an active fee has every right to know which.

3

pieces active return splits into

allocation, selection, interaction

1

number they must sum to

the total active return, exactly

2

decompositions

one for equities, a different one for bonds

What attribution answers

Two different skills, paid for differently

An active manager earns a fee for one of two things, and usually a blend of both: reading the macro picture well enough to be in the right parts of the market, or reading individual companies well enough to pick the right ones inside a part of the market everyone can see. These are not the same talent.

A macro-minded allocator can be right about which sectors will lead and indifferent at picking within them. A stock-picker can be brilliant at finding mispriced names and mediocre at deciding how much of the portfolio should sit in that sector in the first place.

Performance attribution exists because a single number, "beat the benchmark by 1.7 points," cannot tell you which of those two people you are looking at. Decomposing it can, and a candidate who can do that decomposition out loud is showing something a memorized definition cannot: that they understand what the number is made of.

The three terms, defined before any formula

Before any arithmetic, the three pieces in plain words:

• Allocation is the return earned purely from weighting a sector differently than the benchmark does. Being overweight a sector that outperforms adds to it; being overweight one that lags subtracts from it, regardless of which stocks you held there.
• Selection is the return earned from picking better securities inside a sector, measured as if the sector were held at the benchmark's own weight. It isolates picking skill from sizing skill by holding the sizing constant.
• Interaction is what is left over: the extra return that only shows up when a sector bet and a good pick inside that sector happen at the same time. It is a cross term, not a separate decision, and it is usually the smallest of the three.

Why interaction gets its own name

Interaction exists because allocation and selection are each calculated as if the other decision had not happened, and real portfolios do not work that cleanly. If a manager is ten points overweight a sector and also picks the best-performing names inside it, some of the extra return belongs to neither the sizing decision alone nor the picking decision alone. It belongs to both happening together, and the arithmetic needs somewhere to put that.

A worked example: four sectors, one active return

The formulas are easier to trust once they have run on real numbers, so here is a small portfolio, four sectors, built to reconcile exactly.

The setup

The portfolio is overweight Technology and underweight Financials by five points each; Energy and Healthcare sit at the benchmark's own weight. Every sector's own return differs from its benchmark counterpart, some better and some worse.

Two totals, one number

The benchmark's blended return is 10.0%: an equal quarter of 16%, 8%, 10% and 6%. The portfolio's blended return is 11.7%, adding each sector at its own portfolio weight. The active return is the gap between them, 1.7 percentage points, and the whole exercise below is showing where those 1.7 points actually came from.

The decomposition, sector by sector

The formula for each sector runs the same way every time. Allocation compares the sector's own benchmark return against the benchmark's total, weighted by how far off-benchmark the portfolio's sizing is. Selection compares the portfolio's return in that sector against the benchmark's, weighted at the benchmark's own size so sizing cannot inflate it. Interaction multiplies the same two gaps, the sizing gap and the return gap, against each other.

Take Technology on its own. The portfolio is five points overweight, and the sector beat the total benchmark by six points, so allocation contributes 0.05 times 6, or +0.30%. Selection compares the portfolio's 20% against the sector benchmark's 16%, at the benchmark's 25% weight: 0.25 times 4, or +1.00%. Interaction is the sizing gap times the return gap: 0.05 times 4, or +0.20%.

Run the same three calculations across all four sectors and the totals are:

The check that makes the exercise worth doing

Allocation effect

+0.40%

Selection effect

+1.00%

Interaction effect

+0.30%

Total active return

+1.70%

0.40 plus 1.00 plus 0.30 reconciles exactly to the 1.70% of active return computed directly from the portfolio and benchmark totals above. Nothing here is estimated; an interviewer can check every line on paper.

A second worked example: right result, wrong reasons

The first example shows the mechanic. This one shows why anyone bothers running it: it can catch a portfolio that looks skilful and is not.

The setup

Three different sectors this time, so the two examples never blur together: Consumer, Industrials and Utilities.

The portfolio is forty points overweight Consumer, the best-performing sector, and correspondingly underweight the other two. Inside every sector, though, the portfolio's own return trails its sector's own benchmark return.

The total looks like skill

The benchmark's blended return is 8.0%. The portfolio's is 12.4%. Active return: plus 4.4 points, a genuinely strong number that would read well alone on a fact sheet.

The decomposition says otherwise

Run the same formulas and the story flips. Allocation contributes +6.00%, entirely from being heavily sized into Consumer before it ran. Selection contributes -1.20%, negative in all three sectors: the manager's own picks lagged the sector benchmark everywhere they were tested. Interaction contributes -0.40%.

A client reading only the headline number would call this manager skilled. One who asked for the decomposition would ask a sharper question: what happens next quarter, once the sector rotation reverses.

One period is not evidence

None of this proves the manager is bad, any more than the first example proved the first one was good. One quarter of negative selection could easily be noise rather than a real weakness; a genuine skill judgment needs the same decomposition run across many periods, watching whether selection stays reliably negative or was simply unlucky once.

Attribution is arithmetic, not a verdict on talent. It tells you where a number came from; only repetition tells you whether that source is repeatable.

Why the interaction term exists, and why some firms fold it away

The cross term, in one sentence

Interaction is the return that belongs to the sizing decision and the picking decision acting together, not to either one alone, and it only appears where both bets land in the same sector.

The case for keeping it, and the case for folding it in

A firm that reports interaction separately is telling a client something specific: how much of the outperformance came from a sector bet and a good pick happening to coincide, rather than from either skill on its own. Some firms consider that genuinely informative, especially when it is large, because a big interaction term often means the manager concentrated conviction rather than spreading it thin.

Other firms fold interaction into the selection line instead, and that is a defensible choice rather than a shortcut. If a stock-picker is skilled enough to put their best ideas where the portfolio is already most overweight, that concentration is arguably part of the selection skill itself, not a separate accident. Both conventions are used across the industry; neither is the "correct" one.

Why an interviewer asks about it anyway

Allocation and selection can both be recited from a definition without anyone knowing whether the candidate has built the arithmetic. Interaction cannot, quite as easily, because getting the sign and the size right requires understanding that it is a product of two gaps rather than a definition to memorize. It is the one part of this topic that is hard to fake.

What a negative interaction term is telling you

Interaction does not have to be positive, and a negative one is not a mistake. It means the sizing bet and the stock-picking edge worked against each other: an overweight sector where the picks underperformed, or an underweight sector where they outperformed anyway.

The second worked example above shows both versions inside one net number. Consumer was overweight with negative selection there, producing a -0.80% interaction on its own. Industrials and Utilities were both underweight with negative selection too, and underweighting a sector where selection is weak actually produces a small positive interaction, +0.20% apiece. The three net to -0.40% overall, and a candidate reporting only that net figure has missed that it hides two different stories, not one.

This is exactly why it tends to come up as a follow-up question rather than a first one: a candidate who can walk a negative interaction term apart sector by sector, rather than just accepting the sign, is showing they understand it as a product of two gaps and not a mood the portfolio was in.

Why the arithmetic matters once real money is involved

A pension fund or an endowment paying an active fee is implicitly buying a story about where the extra return comes from, and attribution is how that story gets checked rather than just told. A manager whose whole edge is one lucky sector overweight is a different product than one whose edge is genuinely picking better names, and each deserves a different fee.

This is also the conversation after a bad quarter. A client seeing a negative active return wants to know whether it was a sector call or stock selection that went wrong, because the fix and the forgiveness differ. Attribution turns "we underperformed" from a verdict into a diagnosis.

How it reaches the client

None of this stays on an analyst's spreadsheet. Once checked and signed off, the decomposition becomes a page inside the quarterly client report, active return at the top, allocation and selection underneath, with a short paragraph on what drove each one. A relationship manager preparing for a client call reads that page first, because it is where the hardest question in the meeting usually starts.

That downstream use is why the numbers cannot be approximate: a report whose pieces do not sum to the total is a visible, checkable error on a document a paying client reads closely, and that is a large part of why this exact arithmetic gets asked on a whiteboard rather than assumed.

The ninety percent statistic everyone gets backwards

What the 1986 study measured

A number circulates constantly in this part of the industry: that asset allocation explains roughly ninety percent of a portfolio's returns, and stock selection barely matters at all. It comes from a real, well-known 1986 study of large pension funds, and the number itself, 93.6%, is accurate. What it measured is not what it gets used to argue.

The study looked at how much a single fund's own quarterly returns moved around over time, and found that the fund's policy allocation explained the overwhelming majority of that movement. That is a different question from "what separates a good manager's returns from a bad one's," which compares different funds against each other rather than looking at one fund's returns across time.

A follow-up review of citations to the original study checked how often it got cited for that second, harder question rather than the first one it actually answered. The large majority of the citations made exactly that substitution.

What it is not saying

Multi-period attribution, and why the numbers don't just add up

The compounding problem

Everything above reconciles cleanly because it covers one period. Extend the same portfolio across four quarters and the neat addition stops working, because portfolio and benchmark returns both compound. The sum of four quarters' worth of allocation, selection and interaction numbers, added the simple way, will not equal the gap between the portfolio's compounded annual return and the benchmark's compounded annual return. The single quarters are each correct on their own; stacking them with plain addition is not.

Two names worth knowing

The fix is a linking method that spreads a compounding adjustment across each period's effects, so the whole year reconciles exactly the same way the sector effects reconcile within one quarter. Two names come up constantly in this corner of the field, Cariño and Menchero, each behind a widely used linking method, alongside other contributors who have proposed variations on the same idea.

None of them change what allocation, selection and interaction mean. They change how a full year's worth of single-period numbers gets stitched together without leaving anything unexplained.

Fixed income runs a different decomposition entirely

Curve, spread and carry

Everything so far assumes a sector, a stock, and a return that can be compared to a peer inside that sector. A bond does not have a sector in the same sense, and its return does not come apart the same way. Fixed income attribution instead splits a bond portfolio's return into:

• Curve — the return earned or lost from how the yield curve itself shifted, twisted or bent, and from where the portfolio's duration was positioned along it.
• Spread — the return from credit spreads widening or tightening, isolated from the curve move itself, which is the closest fixed income equivalent to a stock-picking call.
• Carry — the income a bond earns simply from being held, plus the return from rolling down the curve as it ages, which accrues whether or not anything else in the market moves.

Where this decomposition came from

The equity-style sector decomposition does not map cleanly onto a bond portfolio, so fixed income attribution was built as its own framework rather than borrowed. Duration risk replaced the market-return term that equity attribution uses as its anchor, which is the reason the two decompositions look almost nothing alike on paper despite answering the same underlying question: where did the extra return actually come from.

Currency attribution: the third axis global mandates can't ignore

Everything above assumes one currency. A global mandate never gets that luxury, and most explanations of attribution skip straight past it.

Local return and currency return are not the same decision

A US-based investor holding a German bond earns two things at once: whatever the bond returned in euros, and whatever the euro did against the dollar over the same period. Those are two separate decisions, often made by two different people, and collapsing them into one number hides which one actually drove the result.

Currency attribution pulls them apart. The local return gets corrected for the currency move so it reflects what the bond or stock did in its own market, and the currency return is reported separately, apart from anything the local manager chose. A currency-hedged benchmark needs a third piece again: the hedge itself, priced through forward contracts, adding or subtracting its own return independent of both the local market and the spot move.

Why it belongs in the same conversation as allocation and selection

The logic is identical to sector allocation and stock selection, just applied to a different axis. Being overweight a currency that strengthens works the same way as being overweight a sector that outperforms. A genuinely global mandate is judged on three decisions at once, not two, and a candidate who can only decompose the first two has half the picture for anything that is not a single-currency book.

Who runs this, and why it's a route into the front office

The manager does not usually build this table themselves. A dedicated performance team does: pulling holdings and returns from the accounting system, running the calculation, checking it reconciles, and packaging it for whoever explains it to a client next.

A real, credentialed specialism

This is enough of a distinct skill to carry its own qualification, the Certificate in Investment Performance Measurement, run by CFA Institute for performance analysts, investment consultants and portfolio managers who need this arithmetic fluent from the other side of the desk. Calculating a return correctly, then decomposing it without quietly introducing an error, is a specialized competence, not an afterthought bolted onto an accounting job.

Why it's a genuine way into a front-office seat

The route from operations into a front-office seat treats performance measurement as one of the strongest paths across that line, and this article is the technical reason why.

Someone who has spent two years reconciling attribution reports has built the exact fluency this piece has been asking for, having seen hundreds of real decompositions and learned to catch a reconciliation error before a client does. That is not adjacent experience for a research or portfolio seat; it is the same arithmetic, already proven under pressure.

What gets asked, and how a good answer sounds

Reciting the formula versus understanding it

Reciting the formula

• States the definitions of allocation, selection and interaction
• Cannot walk a real number through the calculation
• Treats interaction as an afterthought or skips it
• Cannot explain why the three numbers must sum to the total

Understanding it

• Picks up a pen and works a small example unprompted
• Gets the sign right on interaction before checking it
• Explains why some firms fold interaction into selection
• Connects the split back to what a client is paying for

The mistake that ends the interview

A round built around this topic tends to ask for one of a few things: a plain-language definition of the three terms, a worked decomposition on a small example, an explanation of why interaction exists, or a question about what a client would actually want to know from the split. None of these require memorizing a textbook page. All of them reward having built the arithmetic at least once.

How to get fluent at this

1. Build one small example from scratch, by hand, three or four sectors, and check that your own numbers reconcile before trusting them.
2. Say the definitions out loud without the formula first. If a plain-English version does not come easily, the formula is memorized rather than understood.
3. Deliberately construct a case with a large interaction term and explain, out loud, why it is large and what that implies about how the portfolio was built.
4. Learn the fixed income version as a separate framework, not a variant, so the two never get conflated in the room.
5. Practice the client-facing version of the answer, not just the arithmetic: why does this split matter to whoever is paying the fee.

The one number nobody can fake

Performance attribution asks one real question inside a much bigger one: when a portfolio beats its benchmark, was that a sector call or a stock call, and can the manager prove it rather than just describe it. Allocation, selection and interaction are the three-way answer, and they reconcile exactly to the total or something in the arithmetic is wrong.

Learn to build the example on paper, know why interaction is the one nobody can fake, and know that fixed income throws the whole framework out and starts again with curve, spread and carry. Walk in able to do all three and the rest of the round takes care of itself.