Nobel Laureate Robert Engle on High-Frequency Trading and Portfolio Management
In part one of my interview with Nobel laureate Robert Engle, we discussed the development of the ARCH model, the global financial crisis, systemic risk, and forecasting liquidity with ARCH models. In this part, we will cover the application of ARCH models in high-frequency trading and how he thinks risk models should be applied in portfolio management.
CFA Institute: Let’s switch gears and talk about some of the most practical ways traders and portfolio managers leverage risk modeling. How about we start with high-frequency trading?
Robert Engle: There is no reason why you can’t calculate value at risk at a millisecond interval, except that we probably need a better volatility model, a volatility model that is specific to the time frame. One way people have done this is sort of what I call the brute force approach, which is you take thousands of observations every day, then you take thousands of observations the next day, put them all together to make them a million observations long, and fit it into the GARCH model. And it turns out that doesn’t work very well because it spends most of its effort predicting time of day shapes. So the decay rates of volatility are like an hour long. It predicts two days ahead. The volatility is not affected by what we know today. And we know that is not true. We need a more complex model.
How should we go about doing that?
An idea I liked (and published) is to basically treat the daily part and the time and date effect as three components. What [the model] is really trying to do is to predict the volatility in all these buckets. That’s the way you can build high-frequency volatility models.
It starts with an ordinary daily model. We use a lot of historical data to predict the volatility the next day. The next step is to [ask], what is the typical shape of each day? If you take daily returns and divide them by their predicted variance from the first model, and then you look at the volatility in the first minute of this process, then the volatility of second minute of the process, that would give a standard deviation pattern that is much higher at the beginning of the day and gets quieter in the middle of the day. If you take each one minute return and divide it by this predicted variance, which now has two pieces, the daily and the intradaily pattern, then you string together what’s left over to get these million observations to build the GARCH model. It doesn’t have to predict the time of day shapes. It doesn’t have to predict the long-run volatility; it only has to predict the short run. Just by simply working with the data and using the traditional GARCH methods, you can get a high-frequency volatility model.
My MBA students get to see this. It is also in a paper in the Journal of Financial Econometrics. I’ve had a variety of people tell me confidentially that it’s done a lot of good for them.
Let’s get to some of the other real-life applications. How about portfolio construction? What role should risk modeling play in portfolio construction?
In portfolio construction, volatility and correlations are central inputs. Once you have identified your alpha, then the question is how you construct the portfolio. It’s really all about what assets you put in there so I think that GARCH models have an important role to play. Until fairly recently, there was a concern that these methods didn’t work very well. In fact, we now have new ways of estimating covariance matrices, changing over time for very large size covariance matrices. I think that these are potentially very useful for both risk management and portfolio construction. All these new ideas in asset allocation, risk budgeting and risk parity models and so forth, are really tied to volatility processes.
It is interesting that people have spent a lot of time and energy researching expected returns. But for expected volatility, people seem happy to use historical volatility. Is this one area that volatility forecasting mechanisms can potentially add a lot of value?
I like to focus attention on exactly what you said: [the different approaches institutions take to] expected return forecasting and risk measurement. In almost all financial institutions, they are done by different people. The expected returns are done by portfolio managers who generate the alphas and don’t pay much attention to the risk. And then you have a risk manager who is the policeman supposed to say what you can do. We know from theory that risks and returns are tied together. It’s all one problem; it’s not two separate problems.
I think the reason is that it is very hard to identify expected returns. People typically use low-frequency data. Then the risk manager come in, they get their risk so much faster than the alpha signals come out. (That’s not so true for the high-frequency guys because they get high-frequency alpha. That’s a different problem for asset allocation.) It’s natural that they come out of separate departments but really from a portfolio point of view they should be together. There are a lot of ways you could do that.
I agree. I see tremendous value in combining both. You see that actually gets us to the next point, fundamental investing versus quantitative investing. What are some of the quantitative methods you think fundamental managers can benefit from the most?
My sense is that fundamental managers often don’t do the job of managing the risks, so they identify the alpha but don’t make the most of the alpha that they have. They may or may not do a good job of controlling execution cost.
I think quants typically are working out of a different speed from the fundamental people, just as what we are saying. So if you are following a quant model, you can’t be bothered looking at balance sheets. They come out so infrequently. You really have to just look at what’s in the data.
And maybe you look at news feeds; some of the quants nowadays are easily parsing the news as it comes in and translating that into signals. (They call it computational linguistics?) That sounds like a good name.
I don’t think there is one right way to do investing. The markets accommodate all these different points of views, and actually there is money to be made [using many of these approaches] although it’s getting harder all the time. People have the same skill sets, and they are all pretty well tagged up.
Thank you for an absolutely fascinating discussion.
If you enjoyed this post, you should subscribe to The Enterprising Investor
Please note that the content of this site should not be construed as investment advice, nor do the opinions expressed necessarily reflect the views of CFA Institute.
Photo credit: ©iStockphoto.com/enot-poloskun