Methodology — Multi-period (linked) Brinson attribution

The PORT Attribution tab's default. This page explains what the linked attribution computes and why it is trustworthy. The exact period-construction parameters, exclusion thresholds and gating rules are proprietary calibration and deliberately not enumerated.

Why linked, not single-period

A single Brinson–Fachler computation run over a multi-year window with today's weights has two distortions: look-ahead (today's book and today's benchmark weights did not exist at window start) and a compounding gap (sector effects computed on multi-year total returns cannot reconcile to a compounded active return except by accident). Industry practice — including Bloomberg PORT — is to run single-period attribution per month and link the monthly effects across the window. That is what we do.

The method

1. Monthly grid. The analysis window is broken into calendar months; sector effects are computed per month with that month's beginning weights and that month's returns (Brinson–Fachler (1985) allocation / selection / interaction — see the single-period Brinson methodology).
2. Weights per period. Without a transaction record, per-month weights are derived from a buy-and-hold counterfactual consistent with each holding's own realized returns — disclosed in the tab as "assumes the current book was held, unrebalanced, through the window." The benchmark side applies the same drift arithmetic to cap-weighted sector proxies, which removes the benchmark-side look-ahead. Holdings whose price history does not adequately cover the window are excluded from the linked computation, with the excluded weight disclosed rather than silently absorbed.
3. Carino linking. Monthly effects are combined with Carino logarithmic linking coefficients (Carino, 1999) — the same standard linking used for contribution-to-return — so that the linked sector effects sum exactly to the true compounded active return over the window. This reconciliation is test-pinned; it is the defining invariant of the method.

Transaction-aware mode (saved portfolios)

When the analysis references a saved portfolio with a dated transaction ledger that fully covers the window, the counterfactual is replaced by actual positions: beginning-of-month weights come from the ledger's dated quantities at as-of prices, and each sector sleeve's monthly return is a Modified Dietz return with intra-month trades as time-weighted external flows (see the Modified Dietz methodology) — so trades move capital without masquerading as performance. The mode is all-or-nothing: partial ledger coverage never blends conventions. When the ledger can't support the computation the tab falls back to the counterfactual mode and says why — never silently.

Honest limits (disclosed in the tab)

• The counterfactual mode assumes no trades over the window — like every static-book attribution, it attributes a hypothetical held book.
• The benchmark composition is today's universe drifted back, not point-in-time index membership — index adds/drops are invisible.
• ETF look-through uses today's constituents.
• The chain covers the window's jointly-priced months; months priced on only one side are trimmed with a caveat, and a gap in the middle abstains rather than splicing compounding around it.
• Mid-month entries join the weights at the next month start (their entry-month P&L is not attributed), and positions opened and closed within a single month never appear in the weights; both conventions are disclosed in the tab when they bind.