Estimating Multi-Asset Fund Exposures with Return Regression
Summary
The document asks how to infer the asset-class weights of a multi-asset fund from its daily returns when the fund’s actual holdings are unknown. The proposed approach regresses fund returns on broad asset-class returns, potentially using rolling windows to track changing exposures. The question suggests omitting an intercept and treating cash as the constant exposure, while using residuals to look for asset classes missing from the model.
The response endorses multiple regression and points to return-based style analysis. It also suggests principal component analysis when candidate asset-class return series are highly correlated. Correlated or cointegrated benchmarks can make individual coefficients unstable, and the exchange offers no formal method for detecting omitted exposures or validating estimated weights. It does not provide data or results, so the suggestions are a starting point for analysis rather than evidence about any particular fund.
Key ideas
- Regressing fund returns on asset-class returns can estimate return-based exposures.
- Rolling regressions can be used to examine how estimated exposures vary over time.
- Residuals may indicate that the selected benchmarks do not explain the fund’s returns fully.
- Highly correlated asset-class series can make regression coefficients unreliable.
- Return-based style analysis and principal component analysis are suggested approaches.
Tags
Full text
# Existing research on fund performance contribution analysis? # Existing research on fund performance contribution analysis? Consider the following problem: - You would like to understand how a particular Multi Asset Class fund is invested (ascertain the weights attributed to each asset class) - You have at your disposal a series of daily returns for that fund - You also have daily returns for broad asset classes (Gov Bond index, local equities, foreign equities, High Yield, ...) - Crucially, you don't know in which of those asset classes the fund is invested in (could be only a subset of these or worse, could be in an asset class outside the set) How would I go about determining how the fund has been invested: - On average over the entire period - On a rolling window basis And how might I determine if the fund is invested in assets not in the above set? My thoughts on the matter: - I could use rolling regressions on the daily returns, excluding an intercept term (the CASH asset class would act like my constant anyway). - From that regression, the coefficients would be my weights and the error term would provide clues as to whether the model specification is lacking any asset classes - An issue with this approach might be the unreliability of the coefficients if two or more asset classes are highly cointegrated (eg: Small and large cap equities or AAA and AA rated bonds) I've looked far and wide, but I haven't been able to find a single source that looks at this particular problem. If someone can point me in the right direction, I would be very appreciative. Thank you for your help and insight! ## Answer by R. Steigmeier (score 1) https://quant.stackexchange.com/a/54312 You're on the right track. Use Multiple Regression, see also return based style analysis If you are concerned with time series being to correlated you could also use Principal Component Analysis (PCA).
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.