# Top 10 Risk Analytics

February 28, 2011 2 Comments

Comments are Welcome. Please click LEAVE A COMMENT right above this text…

I’m often asked for my “top 10” risk measures, and it happened again just the other day. The trouble is that a top 10 list is often interpreted as “No. 10 does matter all that much” and “I really only care about No. 1.” We love top 10 lists because they provide a linear answer, and our brains *love* to think linearly. We really like it when someone tells us what the single most important answer is to whatever we’re trying to solve. One of the tenets of this blog is that risk analysis, market behavior and economics are *not* linear systems and that thinking about them linearly is inappropriate and dangerous.

I’ll give examples from two very different non-linear systems: sports and piloting an airplane. What is the *single most important* factor in having a winning baseball team? *Surely* it’s pitching, because without an A team in the bullpen, good hitters on the opposing team will dominate. But what about having your own set of sluggers? *Surely* the answer is having quality hitters at the plate so you score more runs. Wait – what about fielding and hustle and smart base running? Baseball, like all sports, is a complex dynamic system and there is no answer to “what is *the* most important attribute.” Of course, that doesn’t stop the sports announcers from attempting to present it that way. The amount of debate among experts about the most important factor is in itself a pretty good sign that you’re dealing with a complex system. Similarly, what is the most important thing for a pilot to monitor while flying? Should she most closely monitor Altitude? Air speed? Weather conditions? Flaps?? Fuel?? The answer is obvious: all of the above, plus some! I, for one, would never get on a plane whose pilot was only concerned with a “the top 10 things to keep track of.”

So I hope you agree that I’m not dodging the question when I write that the very idea of a “top 10” list of risk analytics is the wrong thing to look for. Still, some people press the issue and ask me “**if you could only have one risk analytic, what would it be?**” My answer, which is not tongue in cheek nor glib nor snarky, is simply “**if I could only have one risk analytic, I would not manage the portfolio.**” Just like I hope the pilot would answer “**If I could only have one instrument on my airplane, I would not fly it.**”

In other words, there are several risk analytics that are necessary, and none of them are sufficient. So, instead of giving a top 10 list, I will provide a “Must Have List” in no particular order.

- Tail Risk Analytic
**:**this can be a Monte Carlo or an Extreme Value Theory approach to understanding how bad things can get. Using the Normal distribution for any confidence interval larger than about 95% is not advisable. Warning: if your Monte Carlo uses the Normal distribution and doesn’t revalue the non-linear instruments such as options, then you’ve not really gained much. Any analysis that looks at confidence intervals of 98%, 99% (or larger) needs to use a so-called ‘fat tail’ distribution. - Volatility Stability Analytic: this can be based on VaR, expected loss or any other such analytic. The important thing is to monitor how the risk analytic changes over time. This tells you whether you’re taking more or less risk than you were over various periods of the funds’ life.
- Goodness of Fit Analytic
**:**regardless of which measures of risk you choose to use, it’s very important to investigate how well the risk models fit the portfolio. This isn’t one of those small technical matters than can be reasonably glossed over, despite the fact that many professional portfolio managers do exactly that. I tried to come up with several pithy analogies for this one, but nothing really worked. Why would you ever use a model that doesn’t fit? Why would you ever not examine if a model fits? All models will give numbers – and if you don’t examine their quality, how do you know if it’s any good? - Risk Attribution: determining how the different parts of the portfolio contribute to your risk. Ideally, you’d also have a return attribution analysis that breaks things out the same way so that you can identify how much each part of the portfolio contributes on a risk-adjusted return basis.
- Consistency: this isn’t a risk analytic in and of itself, but rather a statement that you should not be switching your suite of risk analytics every time a new measure comes on the block. There is value to having a set of measures that you get to know really well and that you understand intuitively. That’s not to say that you shouldn’t ever add or take away from this list. Adding a new analytic should be done slowly and deliberately – add the number to the reports but take the appropriate time to get to know how it behaves and what it really tells you before you make too many changes because of your new toy.
- Short-term vs. long-term ratio: this can also be based on a number of different individual analytics (like VaR, volatility or other measures of loss likelihood), and is the ratio of the same analytic computed using only short-term data as opposed to the exact same analytic computed using long-term data. For example, a VaR measure that uses a 50-trading day historical window versus a 2-year historical window. This tells you whether the markets you are invested in is more or less risky now than it has been in recent times.
- Sensitivities: this could be delta/gamma, or duration/convexity or beta or vega or theta or rho. It’s the set of simple and often linear (yes, linear) measures that describe the responses of individual securities to movements in their underlying driver of returns and risk.
- Realistic Stresses: understanding how the portfolio is likely to respond to large changes in the markets. Many people think that volatility measures risk, but that’s hardly the case. Volatility measures what happens in usual circumstances – the ‘normal’ variations that take place, well, normally. Stress tests measure what happens in abnormal circumstances, and they’re really important for just that reason. Examples of realistic stresses include simultaneous movements in the largest drivers of risk and correlations going to extremes like +1 and 0.

It’s important to remember that none of these is sufficient in and of itself. Can you have a solid risk management program if you only look at 6 of these 8? Sure. If you only use 4 of these 8? Probably not. That’s a lot more useful, I think, than a “top 10 list.”

Damian, thank you for providing such thought-provoking analogies. I believe there is a dilemma in our Risk minds between the complexities of Risk and the need of simplicity to exaplin Risk. Your contributions to Riskology and the readers’ comments strive to find the right balance between these two contrasting forces. While the complexity is inevitable, simplicity can be reached with a compelling visualization of data and a narrative of these data that pulls everything together. Well crafted reports help with the visualization. I belive that only experience and discussions among peers can help with the effective narrative.

David,

Thanks for the comment and compliment. I especially like the visualization part — it helps tell the right story.