No single risk model performs best in every circumstance—which is precisely why Qontigo provides clients with multiple views of risk that differ by horizon (short- and medium-horizon) and by factor structure (fundamental and statistical) .
While investors may prefer one of the various versions of our risk models as a primary choice, we encourage clients to monitor their portfolio risk with all risk models, especially in times of market stress. Fundamental risk models are more easily interpretable by portfolio managers as the fundamental factors are well defined and commonly used in the investment process. Statistical models may capture transient factors that drive returns in the market. Longer horizon models might be preferred by investment managers with a more long-term view and less frequent rebalancing timeframes. The risk estimate of shorter horizon models react more quickly to changes in the volatility environment providing managers with early indications as to how their portfolio might be impacted over a longer horizon.
Different investment strategies and the risk models used to analyze risk may lead to different levels of responsiveness in risk measurement during periods of rapid change, as we have seen in the current crisis. For illustration, we used a value index in the following analysis. Any strategy or index would lead to similar observations and the conclusion is: we benefit by monitoring our risk with multiple risk models – particularly in periods of crisis.
In periods of crisis where volatility rapidly increases, all risk models will respond to the increased risk environment, with shorter horizon models capturing the changes more accurately than their fundamental counterparts. The following chart shows four predicted risk estimates: Fundamental model in blue, Statistical model in orange, Mid Horizon in “full line”, and Short horizon in “dashed line”. The dotted grey line shows ex-post/realized risk on a rolling 60-days basis.
In more stable periods (YTD up to Feb 20, 2020), all four risk estimates track closely together, giving us confidence in our risk monitoring. In a crisis period where volatility increases significantly and quickly, short horizon models give an estimate of risk closer to recently realized risk. While the statistical model sometimes captures emerging factors in certain periods, that has not been the case so far during the coronavirus crisis, where the statistical models lagged more than the fundamental models. During this intense crisis, Axioma’s fundamental factors have captured a great deal of the cross-sectional variation in returns, mainly because the crisis is fundamental in nature, with clear expected outcomes for industries and company characteristics, such as leverage, market sensitivity, profitability, size, etc. As we have shown over the past few weeks in numerous posts, the underlying details, factor returns, factor correlations, and changes in specific risk, all give depth and understanding of the drivers of these changes.
Looking at the active risk of the value index vs the broader index, we observe a stable level of active risk until February 20th, with all four risk models giving close estimates of risk. With the onset of the coronavirus crisis, however, active risk estimates immediately begin to diverge—again, the short horizon estimates being closer to realized risk. This time, the statistical risk model seems to be giving a slightly improved active risk estimate. This probably varies depending on the active strategy being monitored, but clients in this environment have reported to us active risk levels that have increased in a similar manner, which required more frequent rebalancing of their portfolios to bring active risk within tolerable boundaries. Portfolio managers monitoring their risk with only one risk model in a crisis period like the current one may possibly be managing their risk with incomplete information. Regardless, monitoring active risk with multiple risk models provides more information and enhanced insights, enabling managers to obtain a deeper understanding of their portfolio risk and, more importantly, to act accordingly.
Using custom risk models for improved accuracy
Recognizing that any risk model will, to some extent, lag in a rapidly increasing volatility environment, some clients utilized Qontigo’s Risk Model Machine (RMM) software to quickly build even shorter horizon risk models to monitor the risk of their portfolios during the coronavirus crisis. RMM also allows for embedding a manager’s alpha signals into custom risk models for even more accurate risk estimates. The combination of embedded alpha signals with more short-term parameters allows RMM’s custom risk models to deliver the most accurate estimates of risk during highly volatile periods.
Using more than one risk model in portfolio construction during a crisis
We showed the benefit of using multiple risk models in monitoring the risk of any portfolio. What about portfolio construction? The Axioma Portfolio OptimizerTM has a unique feature that allows users to easily include more than one risk model in either the objective function and/or in risk constraints.
For example, if a manager targets 3% active risk using the fundamental mid-horizon model, one could potentially also constrain a little bit more loosely the short-horizon active-risk estimate at 3.5%. The optimizer would construct an optimal portfolio that respects both constraints on both risk models at the same time. In fact, one can use two or even more risk models in the same optimization—a unique feature of the Axioma Portfolio Optimizer.
Multiple views of risk deliver more information and insights to investment managers. Qontigo provides four different risk models to give investment managers multiple views of their portfolio risk. The benefits of monitoring risk with multiple risk models in stable periods are less pronounced but become acute in periods of crisis. Even in the absence of a clear crisis, however, using multiple risk models can help managers obtain additional insights when there is unusual activity in the market.
Qontigo also provides the tool to build custom risk models which investment managers can calibrate closely to their investment process. These leads to even more accurate estimate of their portfolios risk – particularly better quantification of the risk that they take on the factors driving their active bets.
Finally, more than one risk model can be used in the same optimization, to make sure that portfolio risk is properly controlled in periods of rapidly increasing volatility, as in the current coronavirus crisis.
 See “Using Multiple Risk Models for Superior Portfolio Management – A Practice Not Just For Quants” by Melissa Brown and Chris Canova, Axioma Research Paper No. 032, August 2011.