Alternative risk premia and the advantage of cluster analysis

The return patterns for alternative risk premia strategies are not the same. The fact that there are risk premia differences is not out of the ordinary and should not be surprising; however, we can learn a lot by grouping the different risk premia to find what strategies can be substitutes and complements. 

Using information from the paper "A Framework for Risk Premia Investing: Anywhere to Hide" by Kari Vatanen and Antti Suhonen, we can find different patterns with the correlations across different risk premia strategies. In the first table, 28 different strategies are bundled into eight style groups. For each of the composites, there is presented an average correlation. The style groups are based on an assessment of style and not any statistical analysis. 

There are interesting commonalities as well as differences across style groups. For example, there is a wide difference between equity trend and the rest of the time series momentum group. The skew for equity trend is negative and the kurtosis is high even though the correlation between constituents is similar. The raw Sharpe ratios are very high for short volatility carry although there is significant negative skew. Clearly, there are significant difference both intra-style and across ARP styles with a lot of potential diversification benefit. 

Nonetheless, a very different perspective is gained from conducting a cluster analysis of the different styles. Cluster approach is sensitive to the data used and the methods employed, but this analysis shows that the groupings of style is not the same as what occurs when it is done through some naming convention. For example, credit carry and curve are more closely tied with equity size than with bond styles. FX carry is more closely related with short volatility carry. This grouping makes sense because FX carry is known to be volatility sensitive, but it is not something that will jump out at the investor who just looks at correlation across style names. 

The dendrogram developed in this paper also links some strategies together that make sense, albeit the similarity may not be immediately apparent. For example, trend is tied to equity defensive strategies and then to commodities and bonds. Credit is related to carry and then to value. 

There are two major groupings that could be called defensive risk-off and offensive risk-on strategies. Offensive will be focused on credit, carry, and value while defensive strategies are focused on trend, rates, commodities, and equity defensive. The clusters provide a nice simple framework for strategy classification. 

Given this deep knowledge on strategy differences, portfolios can be constructed that have a tilt to either defensive or offensive risk profiles. If there is a desire to minimize the strategies employed, cluster thinking can help with reducing duplication across strategies. Diversification analysis can be more nuanced than just looking at a correlation table. The value with ARP strategies is that focused mix and matching can be done in ways that are not available through hedge funds.