The importance of asset allocation versus security selection has generally been assessed in two ways. The first way is to consider only what is feasible from a theoretical point of view and thus not assume any limitations or constraints. The second way is to consider what is actually done and take into account the constraints asset managers encounter in their portfolio selection process, such as benchmarking that limits the portfolio selection. The authors’ approach is from a feasibility perspective. By increasing the complexity of interaction, they show that asset allocation explains the main part of the portfolio’s return variation. They stress that correlation provides information whereas risk does not because risk is scalable. For example, because of (de)leveraging, risks can be rescaled, whereas the interaction between assets is unchangeable.

The authors explain the interaction between risks as a vector in a plane: The size of the vector can be altered (leverage), but the angle at which the vectors are related to each other (correlation) cannot. They then introduce principal component analysis (PCA) to assess the effects of asset allocation and security selection without considering whether managers actually have the skill to use this strategy. Previous research has either been based on constrained asset allocation or has led to the conclusion that any comparison is not possible. The authors do not impose any constraints and instead show how different levels of variety at which assets interact still lead to the conclusion that asset selection is the main driver of return volatility.

The results could be interesting for a wide variety of professionals: portfolio managers who want to understand what drives return volatility, risk managers who want to monitor return volatility, and benchmark investors who want to know how many securities are required to track a benchmark.

The authors introduce their concept by comparing a portfolio return variation with a road network: When determining a route between two locations, a traveler will find that some road types explain more of the transit time than other road types. Similarly in portfolio management, some factors explain more of the return variation than other factors. When assessing the importance of factors, the authors do not impose any constraints on asset allocation or stock selection. Also, they make no reference to any benchmarks, which might be applicable in real life.

The authors use PCA to estimate the impact of each of the factors. Uncorrelated factors are reflected in a correlation matrix, and PCA then determines which of these factors is most explanatory and provides the most information.

They define 20 equity markets and 20 bond markets that each include 100 securities. Correlations exist between all of these investments. The authors then define four cases in which the complexity of correlations is increased: assets or cash, asset classes (equity or bonds), markets (investment location), and security selection. They show that asset allocation becomes more relevant when correlations decrease as the benefits from diversification increase. Also, asset allocation is more relevant when the number of asset classes increases.

In the most advanced correlation assessment, security selection explains one-third of portfolio return variance whereas the asset allocation decision (consisting of assets or cash investment, asset class selection, and market selection) explains two-thirds of the return variance. Even when the authors replace correlation with the measure of covariance (which can be scaled) and perform a PCA, 50% of the portfolio variance is still explained by asset selection.

The authors provide a very concise and interesting article at a time when the topic of risk has become more central. The introduction of PCA via a street map approach makes it easy to understand the basics behind this method. The article builds nicely, starting from a plain portfolio without any diversification to more advanced and complicated situations. Also, the assessment of risk and the importance of correlation is a good reminder for practitioners.