Performance Attribution

Performance Attribution does not exist as a formal concept. The portfolio returned 12% last year, and when the board asks 'why?', the PM says 'we picked good stocks.' There is no decomposition of returns, no factor analysis, no distinction between allocation decisions and selection skill. Good performance is attributed to genius; bad performance is attributed to bad markets. Luck and skill are indistinguishable.

None — AI cannot perform Performance Attribution because no attribution records exist. Distinguishing alpha from beta, skill from luck, or allocation effect from selection effect is impossible without formal attribution data.

Performance Attribution is calculated quarterly in a spreadsheet by a junior analyst. Returns are decomposed into allocation and selection effects using a basic Brinson single-period model. But the calculation is manual, the benchmark mapping is maintained in a separate file, and sector classifications are assigned by the analyst's judgment. Two analysts running the same attribution get different results because they classify securities into sectors differently.

AI could read the quarterly attribution spreadsheet, but cannot reproduce the calculation because methodology, benchmark mapping, and classification rules are not formally documented. Trend analysis across quarters is unreliable due to inconsistent methodology application.

Performance Attribution records are generated monthly using a documented Brinson methodology with standardized fields — total return, benchmark return, allocation effect, selection effect, interaction effect, and currency effect by sector. The attribution system uses an official benchmark mapping and GICS sector classifications. Results are consistent across runs. But attribution is single-factor (sector) only — there is no factor model decomposition, no risk-adjusted analysis, and no linking of attribution results to the investment decisions that drove them.

AI can generate time-series analysis of Brinson attribution results, identify persistent allocation or selection skill by sector, and flag sectors where the PM consistently destroys value. Cannot explain returns through factor exposures or connect attribution results to specific investment decisions.

Performance Attribution records include both Brinson sector attribution and multi-factor model decomposition. Returns are explained through factor exposures (market beta, size, value, momentum, quality, volatility) alongside traditional allocation and selection effects. Risk-adjusted metrics (information ratio, tracking error decomposition) are calculated at the portfolio, sector, and security level. A CIO can query 'how much of the Global Equity fund's outperformance came from factor tilts versus pure stock picking?' and get a rigorous, decomposed answer.

AI can perform comprehensive performance diagnostics — identifying factor crowding, detecting style drift, evaluating PM skill net of factor exposure, and comparing attribution across managers and time periods. Cannot yet link attribution outcomes to the specific Research Signals and Trade Orders that produced them.

Performance Attribution records are schema-driven entities with formal relationships to Trade Orders, Research Signals, and portfolio positions. Each attribution component links to the investment decisions that created it — selection effect in Technology traces to specific security overweights, which link to the Research Signals that motivated them and the Trade Orders that implemented them. An AI agent can ask 'which Research Signals from the fundamental team generated positive selection effect exceeding 25 basis points after accounting for execution costs from TCA?' and get a precise, fully attributed answer.

AI can perform end-to-end investment process evaluation — from signal generation through execution quality to realized performance, attributing every basis point of return to its originating decision. Full autonomous investment process optimization recommendations.

Performance Attribution is a living, real-time entity. Returns decompose continuously as market prices tick, positions change, and factor exposures shift. Brinson effects, factor contributions, and decision-level attribution update in real-time. The attribution is not calculated at period end — it is generated continuously from the streaming event history of every investment decision, execution, and market movement. The CIO sees a real-time dashboard where every basis point of return is traced to its source: factor exposure, sector allocation, security selection, or execution quality.

Fully autonomous Performance Attribution intelligence. AI generates, decomposes, and explains portfolio returns in real-time. Attribution is a continuous signal, not a periodic report — enabling real-time investment process optimization.

Ceiling of the CMC framework for this dimension.