Designing a Risk-Adjusted Performance Attribution Framework for Multi-Manager Portfolios

A risk-adjusted performance attribution framework enables family offices and multi-manager investment teams to decompose portfolio returns into distinct, actionable components: strategic allocation, manager selection, factor timing, and idiosyncratic alpha. This is essential when managing portfolios with overlapping managers, where style drift and correlated risk exposures obscure true skill. The framework aligns with Ortec Finance’s PEARL methodology, which supports decision, currency, multi-asset, equity, fixed income, and factor attribution-built on fund hierarchies that mirror investment strategy and overlay structures. The output informs decisions on manager replacement, risk budget reallocation, and overlay optimization.

The attribution architecture follows a three-layer hierarchy: (1) portfolio-level strategic decisions, (2) manager-level tactical execution, and (3) overlay or hedge-level adjustments. Each layer maps to a specific benchmark: a strategic benchmark for the portfolio, a peer-group or factor-matched benchmark for each manager, and a currency or risk-hedge benchmark for overlays. This structure mirrors the investment process and ensures that attribution reflects the decision-making sequence rather than statistical correlation alone. The framework requires a consistent set of risk factors-typically macroeconomic (e.g., inflation, growth), style (e.g., value, momentum, low volatility), and asset-class-specific (e.g., duration, credit spread)-to be applied uniformly across managers. Factor loadings are estimated using rolling regressions or factor-mimicking portfolios, and updated quarterly to capture style drift. The benchmark architecture must be embedded in the attribution engine (e.g., PEARL) to support multi-asset, multi-currency, and multi-layer decomposition.

The core attribution equation decomposes excess return over the strategic benchmark as the sum of allocation, selection, interaction, and overlay effects:

where \(w_i\) and \(w_i^b\) are portfolio and benchmark weights in asset class or manager \(i\), and \(R_i\) and \(R_i^b\) are their respective returns. Interaction terms capture the joint effect of misallocation and manager underperformance. For multi-manager portfolios, the same structure is applied recursively at the fund level: each fund’s return is decomposed into factor exposure (beta), factor timing (alpha from dynamic factor bets), and security selection (idiosyncratic alpha). Factor attribution models-such as Brinson-Hood-Beebower (BHB) extended with risk-factor loadings-allow separation of true skill from systematic exposure. Overlay decisions (e.g., currency hedging, duration targeting) are attributed separately using a dedicated overlay benchmark.

Risk adjustment ensures that returns are scaled by the risk taken to generate them. Two complementary approaches are used: (1) ex-ante risk normalization via factor loadings, and (2) ex-post Sharpe or Sortino ratio adjustment. In the ex-ante approach, each manager’s factor exposure vector is regressed against the portfolio’s factor benchmark to compute a risk-adjusted weight: \(w_i^{adj} = w_i \cdot (\beta_i^{port} / \beta_i^{manager})\), where \(\beta\) is the sensitivity to a composite risk factor (e.g., equity market, credit, volatility). This corrects for style drift and overlap. In the ex-post approach, the manager’s contribution to portfolio Sharpe ratio is computed as \(\text{SR}_i = \frac{\text{Cov}(R_i, R_p)}{\sigma_p^2} \cdot \frac{\mu_i - r_f}{\sigma_i}\), isolating the marginal contribution to risk-adjusted return. The framework also incorporates conditional value-at-risk (CVaR) adjustments for tail-risk exposures, especially relevant when managers exhibit non-normal return distributions. These adjustments are applied before aggregation to avoid double-counting systemic risk.

Manager overlap is addressed by constructing a manager-level factor covariance matrix and applying a variance decomposition algorithm (e.g., principal component analysis or factor-based clustering) to identify redundant exposures. A manager is flagged for overlap if their factor loading correlation with another manager exceeds 0.7 over two consecutive quarters. Style drift is measured as the Euclidean distance between the manager’s current factor loadings and their baseline (initial or strategic) loadings, normalized by the factor benchmark’s standard deviation. A drift threshold of 1.5 standard deviations triggers a review. The framework adjusts attribution weights dynamically: when drift exceeds the threshold, the manager’s allocation is reattributed to the nearest factor bucket (e.g., reclassifying a growth manager as a value contributor if drift persists). This prevents style drift from inflating selection alpha and ensures attribution reflects the manager’s actual contribution to the portfolio’s risk profile.

Consider a $500 million equity portfolio with four active managers, each with a $125 million allocation. Manager A (large-cap growth), B (small-cap value), C (momentum), and D (low volatility) show overlapping factor loadings: A and C share a 0.65 correlation on momentum, while B and D share 0.58 on quality. Using the framework, the portfolio’s factor benchmark is constructed from Fama-French six factors plus a momentum and low-volatility proxy. Factor attribution reveals that 62% of the portfolio’s excess return over the strategic benchmark stems from factor timing (e.g., rotating into low volatility during market stress), 28% from security selection, and only 10% from pure manager selection. After applying risk adjustment, the contribution of Manager C drops from +1.4% to +0.3% because their momentum exposure was already captured by the portfolio’s factor timing decision. Style drift analysis shows Manager A’s loadings shifted toward value (drift = 1.8σ), prompting a reclassification to the value bucket and a 0.6% upward revision to the selection alpha of Manager B. The final attribution report thus distinguishes skill from exposure, guiding the decision to reduce Manager C and increase Manager D’s allocation.