style="font-size:15px;"> Another popular approach, called Bayesian, treats all risk like bets on a sporting event. This is more accurate than the frequentist approach because it can handle events that only happen once, with some unknown potential outcomes and only guesses about probability. But it is still a limited model that does not capture all important aspects of risk. Risk managers draw on a broad spectrum of risk models, frequentist and Bayesian, plus models drawn from evolution, statistical thermodynamics, behavioural studies and game theory. And they know that even with all the different analytic approaches, important aspects of risk are missed.
Consider a teacup. You know that teacups can shatter into shards and dust, and also that shards and dust never spontaneously recombine into a teacup. Why? Because of all the possible arrangements of the atoms that make up a teacup, only a negligible fraction actually are a teacup. That’s all you have to know to predict that a teacup is fragile. It can shatter, but it can’t self-construct. Any sufficiently large change in conditions – impact, temperature or others – will destroy it. If I have a china shop, I know that it won’t last forever; I don’t need a bull to destroy it. Risk and time are enough.
Some things in the universe do come into being spontaneously – stars, for example, and people and crystals. In many cases these things gain from disorder and change. They can be destroyed, but they can also recreate without outside help.
The same thing is true of human plans and institutions. Some are fragile. Disorder and change only hurt them. Such plans will fail, however solid they seem. Perversely, people often respond to risk by building in more fragility, making the teacup heavier and stronger but no less exposed to risk and time. Risk managers don’t ask how strong your teacup is, they ask how it will respond to the unexpected events that the future will bring. Will it gain or lose? That’s what really matters, because although the events are individually unexpected, you can be certain that unexpected events will occur.
I don’t talk much about measuring risk. For the most part, risk that can be measured can be insured, avoided, hedged or diversified away. Generally I insist that line risk takers do all the measurement and mitigation they can before I take over the job of managing the residual risk.
Of course, there’s room for risk measurement in risk management but less than outsiders tend to think. In addition, it’s definitely true that bad risk measurements, as well as inappropriate attempts by inexperienced risk managers to measure non-measurable risks, do a lot more harm in risk management than good risk measurements do good. (I talk about the various components of risk in Chapter 6.)
To see what I mean, consider the graph in Figure 1-1, which shows the distribution of daily returns for the S&P 500 index over the last 50 years.
© John Wiley & Sons, Inc.
Figure 1-1: Daily returns on the Standard & Poor 500 stock index from 1965 to 2015.
You have various ways to measure the spread illustrated by this graph. You can compute a standard deviation, a mean absolute deviation, an interquartile range or something else. For that matter, you can just reproduce the graph. However, there’s something misleading about representing the data this way: You cannot see the essential risk on this graph, and the risk you think you see is largely irrelevant.
Now the 150 per cent increase from the 99 per cent of normal days isn’t insignificant. However, most of the action, especially to a risk manager, happens in the 1 per cent of extreme days, which are nearly invisible. This percentage isn’t true just of stock market returns, but also true of many important things in the world.
Consider the risk going forward, which of course is what matters. Suppose that you’re considering an investment in stocks with a 1,000-day horizon – about four years of trading days. You expect to get 990 normal days in which the market moves less than 3.5 per cent. You may get 996 or 987 or even 1,000 such days; but you won’t get much different from 990. Also, getting a few days more or less won’t matter much because the average return on these days is 0.04 per cent, and no day can make a difference of more than 3.5 per cent. With 990 or so events and limited range, you’re highly likely to get something quite close to the expected outcome. Moreover, you have lots and lots of historical data on what happens on normal days, so you’re reasonably confident you know what the expected outcome is. There just isn’t a lot of risk in 99 per cent of the days, and what risk does exist can be easily handled by front-line risk takers. After all, if they couldn’t handle the stuff that happens 99 days out of 100, you’d have noticed long ago.
You also expect to get about five days when the market loses more than 3.5 per cent, plus about five days when the market gains more than 3.5 per cent. However, there’s a lot of potential variability around those numbers. You might get 2 or 8 or even 0 or 10 or more of either one. Each one of these days is significant as they average about a 5 per cent move, and may be as large as -28 per cent or +18 per cent. With only a few events, you can get outcomes far away from the mean. Moreover, you have little historical data, you don’t really know how big these days can get; and you can’t be confident that your front-line risk takers are prepared for them unless you check.
Another problem is that the big days in the market can seldom be tied to observable economic events. On normal days, some fraction of stock price movements occurs in discrete jumps after clear news events such as central bank actions or corporate earnings announcements. A lot of unexplainable noise (price movements that cannot be easily explained) is evident too (which doesn’t stop commentators from jumping in with explanations after the fact), but it’s possible to imagine that prices are changing in response to economic news. On many of the biggest days, no news turns up at all, and on others, the extent and timing of the price move is inconsistent with the news the market is supposed to be reacting to.
If that weren’t enough, not all the days the stock market makes big moves are abnormal; some are just normal big moves. On the other hand, on some abnormal days, the market behaves strangely but prices don’t move a lot by the end of the day, such as the Flash Crash of May 2010 or the Quant
