How did you become interested in the impact of events on financial markets?
I went to graduate school at the University of Chicago, and I had a lot of friends who were traders on the exchanges in Chicago. When I would describe the research that academics were doing, they would respond by saying things like, “You should come down and see how the markets really work.” So I went down to the trading floor for an unemployment announcement on a Friday morning. I get down there and everybody’s just sitting around, and things are dead quiet; prices are not moving at all. Then all of a sudden, the number came across on the TV screens and the place just exploded. There was a massive move in both bond and equity prices. I quickly realized that this sort of thing is not in our models.
Traditional models in finance don’t deal with events, such as a Federal Reserve announcement, an unemployment announcement, an inflation announcement, or for an individual firm, an earnings announcement. Much of academic finance focuses on understanding the movements of prices over lower frequencies, over weeks or months or even years. In these models, movements over a given time period are assumed to be approximately normally distributed. Price move up, prices move down, but in a “normal” fashion. However, if you take these models seriously, it’s clear that they are going miss events, because these events are precisely the opposite of typical “normal” movements: they are large and rapid — not normal in both the intuitive and the statistical sense.
It’s important to recognize that there are two types of events. Certain events, like a stock market crash, are unpredictable in both their timing and size. Other events, like macro announcements or earnings announcements, have a known timing but an unknown market impact — you don’t know until the news is released whether prices will increase or decrease.
Is it possible to quantify the importance of these types of events?
In my dissertation, I focused on building models with events in interest rate markets and developing statistical tools to quantify their importance. Basically, what I found is that these events are extremely important and they swamp the importance of the nonannouncement days. Intuitively, the vast majority of the variance in bond prices occurs on announcement days, even though there are only a couple of important announcements each month.
Most of academic finance is focused on understanding how risks are priced, and the important question quickly becomes, how should one compensate investors for bearing these sorts of risks? For example, if you were going to sell a put option on the S&P — where if the S&P falls very quickly and by a large amount you’re going to lose a lot of money — what sort of compensation would you demand for taking those types of risks? As another example, a moment before a macroeconomic announcement comes out, if you were to bet on the announcement, what sort of compensation would you want for being exposed to this massive risk, which occurs only over a moment?
Although my dissertation was largely statistical, over the last two or three years I’ve been trying to quantify what a reasonable premium would be for taking these sorts of event risks. You build a statistical model of how stock prices behave and then compare the model prices to derivative prices, and from those you can get a sense of the compensation that investors are receiving for taking these sorts of risks. Essentially, you focus on derivatives whose payouts are closely related to events and ask, Is this risk appropriately priced?
How do you determine whether the risks are priced appropriately?
On the one hand, you have a theoretical framework for how risks should be compensated; on the other hand, you see what happens with market prices and you also have a model. What you can look at, for instance, is what are the returns on various strategies that expose people to risks? This is best understood in the context of options on indices like the S&P 500, where there are trading strategies that appear to be very profitable. Let’s say you write put options that are out of the money — what’s commonly called crash insurance. If there’s no crash, you make a lot of money. Over periods of time, it looks like you’ve made a lot of money, and every now and then a crash comes along and you’re wiped out. In that case, it appears that the options are sort of appropriately priced.
The procedure of research is largely iterative: you realize that an existing model is wrong and you improve it, like in the case of incorporating event risk. Given the new model, strategies that looked profitable in the context of this other model all of a sudden don’t necessarily look that profitable, as the high returns were in fact just compensation for a missing risk factor. In the case of the S&P 500 put options, you realize that every 5 or 10 years there’s going to be a big down movement. While you may make 5 percent per month normally by writing puts, you’re going to lose 75 percent when this one event happens — and it will happen sooner or later.
The model allows you to contrast what should theoretically be happening with what is currently happening and what has happened in the past. Sometimes you write down what appears to be a reasonable model that captures the important sorts of risk, and even with this model, you still can’t justify current prices. In such a case, it appears that there may be trading opportunities.
Michael Johannes is associate professor of finance and economics at Columbia Business School.
