DNA访谈

June 04, 2012

How do you define luck? The way I think about it, luck has three features. It happens to a person or organisation; can be good or bad; and it is reasonable to believe that another outcome was possible. By this definition, if you win the lottery you are lucky, but if you are born to a wealthy family you are not lucky —because it is not reasonable to believe that any other outcome was possible — but rather fortunate.

How is randomness different from luck? I like to distinguish, too, between randomness and luck. I like to think of randomness as something that works at a system level and luck on a lower level. So, for example, if you gather a large group of people and ask them to guess the results of five coin tosses, randomness tells you that some in the group will get them all correct. But if you get them all correct, you are lucky.

And what is skill? For skill, the dictionary says the “ability to use one’s knowledge effectively and readily in execution or performance.” I think that’s a good definition. The key is that when there is little luck involved, skill can be honed through deliberate practice. When there’s an element of luck, skill is best considered as a process. You have often spoken about the paradox of skill. What is that? The paradox of skill says that as competitors in a field become more skilful, luck becomes more important in determining results. The key to this idea is what happens when skill improves in a field. There are two effects. First, the absolute level of ability rises. And second, the variance of ability declines.

Could you give us an example? In the sport of baseball, batting average is the ratio of hits to at-bats. It’s somewhat related to the same term in cricket. In 1941, a player named Ted Williams hit 406 for a season, a feat that no other player has been able to match in 70 years. The reason, it turns out, is not that no players today are as good as Williams was in his day— they are undoubtedly much better. The reason is that the variance in skill has gone down.

Because the league draws from a deeper pool of talent, including great players from around the world, and because training techniques are vastly improved and more uniform, the difference between the best players and the average players within the pro ranks has narrowed. Even if you assume that luck hasn’t changed, the variance in batting averages should have come down. And that’s exactly what we

see. The paradox of skill makes a very specific prediction. In realms where there is no luck, you should see absolute performance improve and relative performance shrink. That’s exactly what we see.

Any other example? Take Olympic marathon times as an example. Men today run the race about 26 minutes faster than they did 80 years

ago. But in 1932, the time difference between the man who won and the man who came in 20th was close to 40 minutes. Today that difference is well under ten minutes.

How does this apply to investing? In a straightforward way. As the market is filled with participants who are smart and have access to information and computing power, the variance of skill will decline. That means that stock price changes will be random — a random walk down Wall Street, as Burton Malkiel wrote — and those investors who beat the market can chalk up their success to luck. And the evidence shows that the variance in mutual fund returns has shrunk over the past 60 years, just as the paradox of skill would suggest. I want to be clear that I believe that differential skill in investing remains, and that I don’t believe that all results are from randomness. But there’s little doubt that markets are highly competitive and that the basic sketch of the paradox of skill applies.

How do you determine the success of something, be it a song, book or a business for that matter? How much of it is luck, how much of it is skill? This is a fascinating question. In some fields, including sports and facets of business, we can answer that question reasonably well when the results are independent of one another. When the results depend on what happened before, the answer is much more complex because it’s very difficult to predict how events will unfold.

Could you explain through an example? A number of years ago, there was a wonderful experiment called MusicLab. The subjects thought the experiment was about musical taste, but it was really about understanding how hits

happen. The subjects who came into the site saw 48 songs by unknown bands. They could listen to any song, rate it and download it if they wanted to. Unbeknownst to the subjects, they were funnelled into one of two conditions. Well, 20% went to the control condition, where they could listen, rate and download but had no access to what anyone else did. This provided an objective measure of the quality of songs as social interaction was absent.

What about the other 80%? The other 80% went into one of eight social worlds. Initially, the conditions were the same as the control group’s, but in these cases the subjects could see what others before them had done. So social interaction was present, and by having eight social worlds, the experiment effectively set up alternate universes. The results showed that social interaction had a huge influence on the outcomes. One song, for instance, was in the middle of the pack in the control condition, the #1 hit on one of the social worlds, and #40 in another social world. The researchers found that poorly rated songs in the control group rarely did well in the social worlds — failure was not hard to predict — but songs that were average or good had a wide range of outcomes. There was an inherent lack of predictability. I think I can make the statement ever more general:

whenever you can assess a product or service across multiple dimensions, there is no objective way to say which is “best.”

What is the takeaway for investors? The leap to investing is a small one. Investing, too, is an inherently social exercise. From time to time, investors get uniformly optimistic or pessimistic, pushing prices to extremes. Was a book like Harry Potter inevitable as has often been suggested after the success of the book?

This is very much related to our discussion earlier about hit songs. When what happens next depends on what happened before, which is often the case when social interaction is involved, predicting outcomes is inherently difficult. The MusicLab experiment, and even simpler simulations, indicates that Harry Potter’s success was not inevitable. This is very difficult to accept because now that we know that Harry Potter is wildly popular, we can conjure up many explanations for that success.

But if you re-played the tape of the world, we would see a very different list of bestsellers.

The success of Harry Potter, or Star Wars, or the Mona Lisa, can best be explained as the result of a social process similar to any fad or fashion. In fact, one way to think about it is the process of disease spreading. Most diseases don’t spread widely because of a lack of interaction or virulence. But if the network is right and the interaction and virulence are sufficient, disease will propagate. The same is true for a product that is deemed successful through a social process.

And this applies to investing as well? This applies to investing, too. Instead of considering how the popularity of Harry Potter, or an illness, spreads across a network, you can think of investment ideas. Tops in markets are put in place when most investors are infected with bullishness, and bottoms are created by uniform bearishness. The common theme is the role of social process. What about someone like Warren Buffett or, for that matter, Bill Miller? Were they just lucky, or was there a lot of skill as well? Extreme success is, almost by definition, the combination of good skill and good luck. I think that applies to Buffett and Miller, and I think each man would concede as much. The important point is that neither skill nor luck, alone, is sufficient to launch anyone to the very top if it’s a field where luck helps shape outcomes. The problem is that our minds equate success with skill so we underestimate the role of randomness. This was one of Taleb’s points in Fooled by Randomness. All of that said, it is important to recognise that results in the short term reflect a lot of randomness. Even skilful managers will slump, and unskilful managers will shine. But over the long haul, good process wins.

How do you explain the success of Facebook in lieu of the other social media sites like Orkut, Myspace, which did not survive?

Brian Arthur, an economist long affiliated with the Santa Fe Institute, likes to say, “of networks there shall be few.” His point is that there are battles for networks and standards, and predicting the winners from those battles is notoriously difficult. We saw a heated battle for search engines, including AltaVista, Yahoo and Google. But the market tends to settle on one network, and the others drop to a very distant second. I’d say Facebook’s success is a combination of good skill, good timing and

good luck. I’d say the same for almost every successful company. The question is, if we played the world over and over, would Facebook always be the obvious winner? I doubt that.

Would you say that when a CEO’s face is all over the newspapers and magazines, like is the case with the CEO of Facebook now, he has enjoyed good luck? One of the most important business books ever written is The Halo Effect by Phil Rosenzweig. The idea is that when things are going well, we attribute that success to skill — there’s a halo effect.

Conversely, when things are going poorly, we attribute it to poor skill. This is often true for the same management of the same company over time. Rosenzweig offers Cisco as a specific example. So the answer is that great success, the kind that lands you on the covers of business magazines, almost always includes a very large dose of luck. And we’re not very good at parsing the sources of success.

You have also suggested that trying to understand the stock market by tuning into so-called market experts is not the best way of understanding it. Why do you say that?

The best way to answer this is to argue that the stock market is a great example of a complex adaptive system. These systems have three features. First, they are made up of heterogeneous agents. In the stock market, these are investors with different information, analytical approaches, time horizons, etc. And these agents learn, which is why we call them adaptive. Second, the agents interact with one another, leading to a process called emergence. The interaction in the stock market is typically through an exchange. And third, we get a global system — the market itself.

So what’s the point you are trying to make?

Here’s a key point: There is no additivity in these systems. You can’t understand the whole simply by looking at the behaviours of the parts. Now this is in sharp contrast to other systems, where reductionism works. For example, an artisan could take apart my mechanical wristwatch and understand how each part contributes to the working of the watch. The same approach doesn’t work in complex adaptive systems.

Could you explain through an example? Let me give you one of my favourite examples, that of an ant colony. If you study ants on the colony level, you’ll see that it’s robust, adaptive, follows a life cycle, etc. It’s arguably an organism on the colony level. But if you ask any individual ant what’s going on with the colony, they will have no clue. They operate solely with local information and local interaction. The behaviour of the colony emerges from the interaction between the ants.

Now it’s not hard to see that the stock market is similar. No individual has much of a clue of what’s going on at the market level. But this lack of understanding smacks right against our desire to have experts tell us what’s going on. The record of market forecasters has been studied, and the jury is in: they are very bad at

it. So I recommend people listen to market experts for entertainment, not for elucidation.

How does the media influence investment decisions? The media has a natural, and understandable, desire to find people who have views that are toward the extremes. Having someone on television explaining that this could happen, but then again it may be that, does not make for exciting viewing. Better is a market boomster who says the market will skyrocket, or a market doomster who sees the market plummeting.

Any example? Phil Tetlock, a professor of psychology at the University of Pennsylvania, has done the best work I know of on expert prediction. He has found that experts are poor predictors in the realms of economic and political outcomes. But he makes two additional points worth mentioning. The first is that he found that hedgehogs, those people who tend to know one big thing, are worse predictors than foxes, those who know a little about a lot of things. So, strongly held views that are unyielding tend not to make for quality predictions in complex realms. Second, he found that the more media mentions a pundit had, the worse his or her predictions. This makes sense in the context of what the media are trying to achieve — interesting viewing. So the people you hear and see the most in the media are among the worst predictors.

Why do most people make poor investment decisions? I say that because most investors

aren’t able to earn even the market rate of return? People make poor investment decisions because they are human. We all come with mental software that tends to encourage us to buy after results have been good and to sell after results have been poor. So we are wired to buy high and sell low instead of buy low and sell high. We see this starkly in the analysis of time-weighted versus dollar-weighted returns for funds. The time-weighted return, which is what is typically reported, is simply the return for the fund over time. The dollar-weighted return calculates the return on each of the dollars invested. These two calculations can yield very different results for the same fund.

Could you explain that in some detail? Say, for example, a fund starts with $100 and goes up 20% in year 1. The next year, it loses 10%. So the $100 invested at the beginning is worth $108 after two years and the time-weighted return is 3.9%. Now let’s say we start with the same $100 and first year results of 20%.

Investors see this very good result, and pour an additional $200 into the fund. Now it is running $320 — the original $120 plus the $200 invested. The fund then goes down 10%, causing $32 of losses. So the fund will still have the same time-weighted return, 3.9%. But now the fund will be worth $288, which means that in the aggregate, investors put in $300 — the original $100 plus $200 after year one — and lost $12. So the fund has positive time-weighted returns but negative dollar-weighted returns.

The proclivity to buy high and sell low means that investors earn, on average, a dollar-weighted return that is only about 60% of the market’s return. Bad timing is very costly.

What is reversion to the mean? Reversion to the mean occurs when an extreme outcome is followed by an outcome that has an expected value closer to the average. Let’s say, you are a student whose true skill suggests you should score 80% on a test. If you are particularly lucky one day, you might score 90%. How are you expected to do for your next test? Closer to 80%. You are expected to revert to your mean, which means that your good luck is not expected to persist. This is a huge topic in investing, for precisely the reason we just discussed. Investors, rather than constantly considering reversion to the mean, tend to extrapolate. Good results are expected to lead to more good results. This is at the core of the dichotomy between time-weighted and dollar-weighted returns. I should add quickly that this phenomenon is not unique to individual investors. Institutional investors, people who are trained to think of such things, fall into the same trap. In one of your papers, you talk about a guy who manages his and his wife’s money. With his wife’s money he is very cautious and listens to what the experts have to say. With his own money, he puts it in some investments and forgets about it. As you put it, ‘threw it in the coffee can. And forgot about it’. It so turned out that the coffee can approach did better.

Can you talk us through that example? This was a case told by Robert Kirby at Capital Guardian from the 1950s. The husband managed his wife’s fund and followed closely the advice of the investment firm. The firm had a research department and did their best to preserve and build capital. It turns out that unbeknownst to anyone, the man used $5,000 of his own money to invest in the firm’s buy recommendations. He never sold anything and never traded, he just plopped the securities into a proverbial coffee can (those were the days of paper). The husband died suddenly and the wife then came back to the investment firm to combine their accounts. Everyone was surprised to see that the husband’s account was a good deal larger than his wife’s. The neglected portfolio fared much better than the tended one. It turns out that a large part of the portfolio’s success was attributable to an investment in Xerox.

So what was the lesson drawn? Kirby drew a more basic lesson from the experience.

Sometimes doing nothing is better than doing something. In most businesses, there is some relationship between activity and results. The more active you are, the better your results.

Investing is one field where this isn’t true.

Sometimes, doing nothing is the best thing. As Warren Buffett has said, “Inactivity strikes us as intelligent behaviour.” As I mentioned before, there is a lot of evidence that the decisions to buy and sell by individuals and institutions do as much, if not more, harm than good. I’m not saying you should buy and hold forever, but I am saying that buying cheap and holding for a long time tends to do better than guessing what asset class or manager is hot.