Using Performance Analysis to Inform Ex-Ante Risk

Sean Carr and Dean McIntyre contributed their time and insights to this article.

About as many people are against consistency between an investment company’s analytics and attribution platforms as are against motherhood and apple pie. However, the advantages are not limited to the consistent measurement of tilts and outcomes in the investment process. This article suggests what I believes is a novel, but important, application of performance analysis to ex-ante risk analysis.

While most active investment managers expect to earn their alpha through security selection, they also tilt their beta exposures to generate alpha. If there are no beta tilts associated with the security selection, the overall returns distribution of the portfolio does not change. In contrast, beta tilts and how they are applied over time can change the portfolio’s return distribution. This is important, as a key assumption in ex-ante risk analysis is that the portfolio’s composition remains unchanged over the risk horizon. However, with active management of market factors, the portfolio composition, and thus the beta exposures, change over the risk horizon, and change the expected ex-ante risk distribution as well.

Let’s look at a simple example from the perspective of a fixed income portfolio manager, who is actively managing exposure to the two-, five-, and 10-year portions of the yield curve. The performance analyst could mine the attribution history to measure the change in relative duration in the next period given a change in yield in the previous period. If the analyst has too much time on their hands, the study, for each yield point, could extend to something like:

• If yields rise, then what is the probability that relative duration rises?
  • In this up/up state, what is change in duration?
• If yields rise, then what is the probability that relative duration falls?
  • In this up/down state, what is change in duration?
• If yields fall, then what is the probability that relative duration rises?
  • In this down/up state, what is change in duration?
• If yields fall, then what is the probability that relative duration falls?
  • In this down/down state, what is change in duration? 

To incorporate this into risk analysis, the risk analyst must be using a Monte Carlo VaR framework. As the Monte Carlo samples yield changes, extra random numbers must be sampled to determine the appropriate duration move given the change in yields (please note that this analysis cannot be incorporated in a linear risk model approach).

This performance study was applied to a sanitized portfolio composite of duration and attribution, extracted from actual manager data. Using a simple Monte Carlo VaR-based model of yield curve risk, I calculated the ex-ante risk of this active portfolio with and without the beta adjustments over time. In this particular example, the VaR dropped by 10% on average. Over time, as the duration mismatches change, the relative positioning between the ex-ante risk measures change as well.

There are several implications of this addition to ex-ante risk management. First, the risk of a portfolio is not limited to the market exposures, but includes important effects resulting from the manager’s response to market conditions. The inclusion of the beta-adjusted response thus gives the performance analyst more information about the proposed and realized risk/return trade-offs in their strategy. These effects can be asymmetric and unexpected. The decomposition of the beta adjustment is an opportunity to fine-tune the specifics of the strategy. Finally, the beta decomposition and application to the ex-ante risk is only possible when the same platform is used for both performance attribution and ex-ante risk measurement.

In conclusion, this is a novel application of performance attribution analysis to ex-ante risk measurement, which in turn, could lead to more stable outperformance and lower overall risk.