I was recently sent a note about the issue of calculating attribution on balanced portfolios. There seems to be a lot of confusion about something that really isn't that difficult.
A balanced portfolio consists of two or more asset classes. For our example we'll take the simple case of a portfolio with equities and fixed income. And so, how do we calculate attribution? We should begin by thinking of what questions we want to answer.
At the highest level, we want to know how the allocation decision worked; that is, how the allocation of the portfolio's assets across the two asset classes performed. We would expect to have a strategy or benchmark with which we will compare our performance. And so, if for example we have a strategy which calls for 60% equities and 40% bonds, but if we decided to underweight equities (with 50%) and overweight bonds (with 50%), then we want to determine if these decisions were good or not. And so, how do we accomplish this? We can easily employ one of the "Brinson" models (either the Brinson, Hood, Beebower (BHB) or Brinson, Fachler (BF)). The allocation effect tells us whether these weight adjustments were good or not. If the managers of the equity and fixed income portions were selected by us (e.g., in a fund-of-funds situation), then we will learn whether or not our selections were good ones.
We now move down to the asset class levels themselves, where we would calculate the attribution here using models approprirate for the respective asset classes (e.g., for equities a Brinson-type model will be fine; for fixed income, we would use a fixed income model).
I think that a lot of people think that attribution for balanced portfolios is somehow more complex; perhaps because they think that one must asses everything simultaneously, using a single model. But that isn't the case as it would only generate misleading information. One must always be sensitive to what the questions are and what asset classes we're dealing with.