New way of Brinson Style Performance Attribution

Brinson Attribution – why we split "active decisions" into components

When comparing a portfolio to its benchmark, the standard approach to explaining active return (i.e. portfolio over- or underperformance versus the benchmark) is Brinson attribution. It decomposes relative (%) performance into:

Asset Allocation Effects: did we overweight/underweight (Wp) the “right” asset classes (j) vs. the benchmark (Wb)?
Security Selection Effects: did we pick better/worse instruments (Rp) within each asset class (j) vs. the benchmark (Rb)?
an additional Interaction Effects: allocation and selection reinforcing each other

In other words, Brinson is designed to measure and explain manager-controlled decisions versus a benchmark. The key practical question then becomes: Which performance metric best reflects those decisions, especially when reallocations create real cash flows between sub-portfolios?

Measuring Returns of active investment decisions

On a portfolio level the TWR (time weighted return) is taken as the relative (%) performance figure because neutral in- and outflows should not distort the performance, since they are not under the influence of the portfolio manager.

However, if the portfolio manager reallocates assets between asset classes – e.g. deliberately overweighting US equities (Wp,j) versus the benchmark (Wb,j) and funding that by underweighting another asset class – those flows are clearly the result of active investment decisions. In that case, you want the performance measure to reflect the impact of those allocation moves.

To capture these effects, MWR (money-weighted return) is more appropriate at sub-portfolio (asset-class) levels (Rp,j). A pure TWR calculation will not reflect the impact of such decisions, because the asset-class inflows and outflows remain performance-neutral within the TWR framework.

Further advantages

If the average invested capital (instead of start or end of period weights) is taken as the weight of the asset classes you don’t need to manipulate the figures with algorithms like Carino, Menchero or the like. The contributions – and thus also the resulting attribution effects – add up cleanly across all aggregation levels.

If there are no neutral in- and outflows, TWR equals MWR on a portfolio level and therefore you will end up with the TWR return in sum.

How to handle neutral in- and outflows?

If there are performance-neutral inflows and outflows, I suggest taking the difference between TWR and MWR at portfolio level and labeling it the "client timing effect" (or a similar term).