by Thom Hartle
Andrew Lo, Harris & Harris Group Professor of Finance at the Massachusetts Institute of Technology's (MIT) Sloan School of Management, director of MIT'sLaboratory for Financial Engineering and founder of Sloan's Track in Financial Engineering, is a radical of sorts because of his frank and open interest in technical analysis. Further, he's an academic who's perfectly willing to admit that technicians are often more open-minded about the markets than the academics who are his peers. But he'll also tell you that technicians are getting too far away from the fundamentals of technical analysis, and getting overly enamored with the modern bells and whistles. To find out what else Lo had to say, STOCKS & COMMODITIES Editor Thom Hartle spoke with him via telephone on September 22, 1997, discussing technical analysis of yore, why statistics are misapplied and how artificial intelligence may not be the great end-all that it's proclaimed to be.
Andrew, you describe yourself as a black sheep in the academic community because of your interest in technical analysis. What started you off on the subject?
I would say my interest in technical analysis began when I was in high school. I would read The Wall Street Journal and the charts would look awfully interesting, but I couldn't make heads or tails of it. That was probably my first interest in the idea of forecasting stock returns graphically, as opposed to numerically, but I really didn't do much with that back then. My interest in technical analysis, the way it is currently practiced, really came about when I was in graduate school.How so?
As a graduate student, I got interested in the random walk hypothesis. The notion of a random walk and, more precisely, the continuous time version of the random walk, or what is known as geometric Brownian motion, is absolutely fundamental to modern financial economics. I was very curious about this because there was an enormous amount of research being done using Brownian motion as a model for stock prices. However, relatively little work had been published in the recent literature on testing the random walk hypothesis. There had been some research in the 1960s and even before that, but there really hadn't been much focus on testing the random walk hypothesis using recent data.It was just accepted as fact?
Yes, it seemed as if most academics took for granted that it was true. Of course, if it is true, that would eliminate the value of technical analysis in its simplest form -- looking at past price patterns to forecast future price changes.What led you to think that technical analysis could add value to the investment process?
The work that I did in 1988 with my colleague Craig MacKinlay at the Wharton School suggested that the random walk hypothesis was not satisfied by broad stock market indices, such as a value-weighted or equal-weighted index of US equity returns in the recent past. So that was the first inkling that technical analysis might offer value.How do violations of the random walk explain the markets' support of technical analysis?
If the random walk hypothesis is false, that would seem to argue that there are predictable patterns in stock returns, even using just past prices as data. The predictability may not be simple, there may not be linear predictability, but there is definitely something there. That's what opens the door for methods such as technical analysis to be effective in providing useful information about future price movements.You used the phrase linear predictability. Can you elaborate with an example of linear predictability and its alternative?
An example of linear predictability is if, with some regularity, next week's return tends to be two-thirds this week's return, so that a 3% return this week implies a 2% return next week on average. Of course, this is a very simple linear relationship, a very artificial example. An example of nonlinear predictability is if return volatility were linearly predictable -- 3% volatility this week implying 2% volatility next week on average. Since volatility is a nonlinear function of returns, linearly predictable volatility implies a nonlinear relationship between returns this week and next.Many technicians focus on both the linear and nonlinear relationships by borrowing methods from other fields, such as engineering. Is that appropriate?
My view of technical analysis may differ from that of the typical reader of STOCKS & COMMODITIES. I take a much broader view. To me, technical analysis, at least from a historical perspective, has used a variety of tools, not simply the tools of engineering, such as moving averages and linear filters and so on, but actual geometric head-and-shoulders patterns, double-bottom reversals, support and resistance levels. Those are the kinds of classical technical tools that first caught my attention.Pattern recognition?
That's right. One of the trends I see, particularly in your magazine, is that there seems to be more and more emphasis on trying to incorporate the jargon of engineering applications into technical analysis. I think that's unfortunate.Why?
Because, in my view, technical analysis has something to contribute to the investment process, but technicians are, in some instances, somewhat insecure about their contributions. I get the impression they feel it is important to take on the mantel and the trappings of engineering applications. I think that's counterproductive, because technical analysis is not an aspect of engineering. Let me put it this way. There is a very important difference between technical analysis and a field like, say, digital signal processing. The difference is that the signal-to-noise ratio for financial markets is much lower than it is in engineering applications.What do you mean by "signal-to-noise ratio"?
The signal-to-noise ratio is a measure that engineers use to determine how much noise there is in their attempts to measure some underlying phenomenon. For example, if you're trying to transmit a TV program, you generate a certain set of electromagnetic signals. Those signals are going to be corrupted by noise from background radiation, transmission noise, and so on. So the signal-to-noise ratio measures how reliable your signal is relative to the noise that might obscure it. An alternative way to look at this is how challenging a problem it is for you to extract the signal, given all the noise.That challenge isn't too far from the technician measuring the trend of the market.
Yes, and you can view the investment process as a kind of a signal detection problem. You've got a signal in terms of the underlying stock market. You know the underlying stock's going to be moving in a certain direction -- that's the signal -- but the signal is obscured by noise because there are random fluctuations that affect the stock price. Those random fluctuations have nothing to do with the underlying value of the stock.Financial economics has focused far too long on simple linear models. The fact is, the world is not linear. Linearity is not a good approximation for many markets.
