Standard performance attribution has not changed much in three decades. The Brinson-Hood-Beebower framework, introduced in 1986, breaks returns into three buckets: what the market gave you, what you got from tilting toward the right sectors, and what came from picking the right stocks within those sectors. For a generalist fund, that decomposition is often enough. For a concentrated sector specialist, it frequently is not.
Consider a healthcare fund. If the portfolio outperformed its benchmark by 200 basis points in a given quarter, sector attribution tells you almost nothing useful. The entire fund is in one sector by construction. Of course sector allocation explains nothing; the manager is not in the business of rotating between healthcare and energy.
What actually drove that 200 bps? Was it a bet on medical devices over pharmaceuticals? An overweight in specialty biotech within devices? A handful of individual names that repriced dramatically? Or was it just a lucky sizing call on a position that already existed?
Standard attribution collapses all of that into "stock selection," which is a bit like explaining a car accident as "driver error." Technically not wrong. Not particularly illuminating.
The decomposition Kiski recently built adds an industry excess layer between sector and selection. The return stack now reads:
Market contribution + Sector excess + Industry excess + Selection + Sizing
Each layer answers a distinct question. Sector excess tells you whether the fund's broad sector exposure versus benchmark added value. Industry excess isolates the contribution from positioning within sectors, across the industry groups that compose them. Selection captures pure stock-level alpha, holding industry weights constant. Sizing captures the contribution from concentration and position scaling within that stock universe.
For a concentrated manager, this matters enormously. An allocator asking "where does your alpha come from?" deserves a more specific answer than "stock picking." The industry layer makes it possible to say: roughly half our edge historically has come from industry selection within healthcare, a quarter from individual name selection, and the remainder from how we size our convictions. That is a falsifiable, auditable claim. It is also a more credible one.
The separation of selection from sizing is worth pausing on. Many attribution systems conflate the two. But knowing that a manager is good at identifying names and bad at sizing them (or vice versa) has real implications for position limit conversations, for risk budgeting, and for understanding whether a track record is replicable as AUM grows.
A parallel workstream involved Brinson attribution at the issuer level for a China equity fund, built specifically to support an investor due diligence process. This is a different, underappreciated problem.
Many large Chinese companies trade in multiple forms: A-shares on domestic exchanges, H-shares in Hong Kong, and ADRs in New York. For attribution purposes, treating these as separate securities is often incorrect. A manager holding the ADR of a company while the benchmark holds the H-share is not making an industry bet or a stock selection bet in the usual sense. The positions are economically equivalent but structurally distinct. Naively attributing performance differences to "selection" produces noise.
Resolving this requires mapping positions to their economic issuer, then decomposing performance on that basis. Alongside standard attribution, the engagement also included active share calculations and factor tilt analysis. Active share is an imperfect metric, widely debated, but allocators still ask for it routinely during due diligence. Having a clean, auditable calculation that accounts for cross-listed securities correctly is not trivial.
Attribution methodology tends to be more important during manager evaluation than during ongoing monitoring. When an allocator is deciding whether to invest, they want to understand whether the manager's stated edge is visible in the data. "We are stock pickers" is a claim. Issuer-level Brinson attribution, properly constructed, is evidence.
The industry-layer decomposition serves a similar function. A sector specialist who can show that their alpha is concentrated in industry selection and individual name picking, rather than factor tilts or leverage, is telling a more coherent story about their process. That story is easier to believe, and easier to stress-test.
There is also a practical dimension. Attribution reports prepared for investor relations tend to be assembled under time pressure, often by people who did not build the underlying models. A system that generates clean, auditable decompositions at the right level of granularity reduces that friction significantly.
Attribution is not a solved problem. Arithmetic versus geometric linking, daily versus monthly frequency, benchmark choice, handling of cash, treatment of derivatives: each choice affects the output in ways that are not always obvious. The industry layer adds another set of choices around how industries are defined, how to handle stocks that span multiple industry classifications, and how to handle benchmark constituents with no obvious industry peer group.
None of these questions have universal answers. But they need to be asked explicitly, and documented. An attribution report that cannot be traced back to its methodology assumptions is more liability than asset in a due diligence context.
The sector-level ceiling on attribution is real. Whether adding a layer meaningfully changes how allocators evaluate managers is a more open question. The directional answer seems yes, particularly for specialists. How much it matters in practice probably depends on whether allocators have the time and inclination to engage with the detail.
