by John Coates
Many of these boxes are what are called ‘execution-only’ boxes. This type of box does not look for trades, it merely mechanises their execution. At this task, boxes excel. They can take a large block of equities, for example, and sell it in pieces here and there, minimising the effect on prices. They test the waters, looking for deep pools of liquidity, a practice known as pinging, just like a sonar searching the depths. When they find large bids hidden just below the surface of existing prices they execute a block of the trade. In this way they can move enormous blocks of stock without rippling the market. At this trading exercise, boxes are more efficient than humans, faster and nimbler. They do what Martin did when he pieced out of the DuPont trade, only they do it better. Many managers have started to ask why traders spend so much time and effort executing client trades when a box could do it just as well, and never argue over its bonus.
Other boxes do more than execution: they think for themselves. Employing cutting-edge mathematical tools such as genetic algorithms, boxes can now learn. Funds running them regularly employ the best programmers, code-breakers, even linguists, so the boxes can parse news stories, download economic releases, interpret them and trade on them, all before a human can finish reading even a single line of text. Their success has led to an exponential growth in the capital backing them, and boxes already make up the majority of trading by volume on many of the largest stock exchanges; and they are now spreading into the currency and bond markets. Their growing dominance in the markets is one of the most significant changes ever to take place in the markets. I, like many others, find the markets increasingly inhuman, and when I trade now I often have difficulty catching the scent of the market’s trail.
Human traders such as Martin are therefore in a fight for their lives. Unbeknownst to outsiders, every day a battle rages up and down Wall Street between man and machine. Some informed observers believe human traders have had their day, and will meet the same fate as John Henry, the legendary nineteenth-century railway worker who challenged a steam drill to a competition and ended up rupturing his heart.
Others, however, note with optimism that human traders are more flexible than a black box, are better at learning, especially at forming long-term views on the market, and thus in many circumstances remain faster. Evidence of their greater flexibility is found when market volatility picks up after some catastrophic event, like a credit crisis. Then managers at the banks and hedge funds are forced to unplug many of their boxes, especially those engaged in medium- and long-term price prediction, as the algorithms fail to comprehend the new data and begin to lose ever increasing amounts of money. Humans quickly step into the breach.
Something much like this occurred during the credit crisis of 2007–08. Anecdotal evidence and published fund performance statistics give us something like the following scorecard: in high-frequency trading, humans and machines fought to a draw, both making historic amounts of money; in medium-term price prediction, in other words seconds to minutes, humans pulled slightly ahead of the boxes, as flow traders made record amounts of money; but in medium- to long-term price prediction, minutes to hours or days – the boxes engaged in these time horizons are known as statistical arbitrage and quantitative equity – humans outperformed the boxes, because only they understood the implications of the political decisions being made by central bankers and Treasury officials. Thus, in what may have been the first major test of human versus machine trading, humans won, but only just. And so it is that this futuristic battle ebbs and flows.
Whatever the outcome of that battle, the financial landscape on which human risk-takers conduct their searches has been changed forever by the arrival of these machines. Governments and regulators fear the changes, suspect that the speed and opacity of the algorithms could lead to uncontrollable markets, even financial meltdown.
There is, however, another perspective from which to view the advent of these new and faster machines. They can be seen as a liberation for risk-takers. They may permit us to disaggregate the activity of trading into its component pieces and farm each of these out to the person or machine which does them best – the division of labour as applied to trading. As I say, time was when a trader had to possess good judgement, have a large appetite for risk, and be physically quick. Increasingly, though, especially at hedge funds, the roles of judgement and speed have been separated. Many portfolio managers are forbidden from executing their own trades, these being handed over to an execution desk, which then often uses execution-only boxes to place the trade. Even the appetite for risk can be stripped away from the trader and put in the hands of the floor manager. These developments mean that, increasingly, all that is required of a financial decision-maker is a good call on the market. With the help of machines, people in the financial world who have good judgement but who are risk-averse or dislike the physical aspects of trade execution could be fitted with what, in effect, amounts to a prosthetic risk-taker. Technology can lift Cassandra’s curse.
Furthermore, if the physical requirements of trading were removed, perhaps the financial playing field would be levelled in such a way that it would not be the preserve of young men. On the trading floor of the future we could have a more even balance of men and women, young and old, selected for the quality of their judgement, with the grunt work of capital allocation and trade execution being done by computers. I will return to the important issue of women in finance later in the book.
A misunderstanding may arise from the picture of the future sketched above. We may be tempted by this and similar visions to believe that our bodies will come to play a less and less important role in financial risk-taking. I do not think that follows. Computers may indeed take over the job of quick execution of trades, but our bodies will continue to be crucial for success in the markets, because they provide us with perhaps the most important data informing our call on the market, and that is our gut feelings. Recent research in physiology and neuroscience has discovered that gut feelings are more than the stuff of legend, they are real physiological entities. Gut feelings emerge from a massive information-gathering exercise conducted by the body. And the body, as we will see, remains the most advanced black box ever created.
FOUR
Gut Feelings
The financial markets are replete with stories of hunches, instincts and gut feelings. These feelings consist, according to legend, of an inexplicable conviction that an investment is destined to make or lose money, a conviction often accompanied by physical symptoms. The symptoms reported by traders and investors are often quirky, like a coughing fit before the market goes down, an itchy elbow before it goes up. George Soros, founder of the hedge fund Quantum Capital, confessed that he relied a great deal on what he called animal instincts: ‘When I was actively running the fund I suffered from backache. I used the onset of acute pain as a signal that there was something wrong in my portfolio.’
How exactly do these signals work? When we use a term like ‘gut feelings’ we imply that our brain receives information, valuable information apparently, from our body. We have seen in Chapter 2 how interoceptive pathways keep our brain constantly updated on the state of our body. The signals we considered, reporting on heart rate, blood pressure, body temperature, muscle tension and so on, served mostly homeostatic needs. Yet the notion of a gut feeling implies much more than this: it implies that gut feelings guide us in even the most complex mental tasks, like figuring out the stock market. How could information about heart rate, body temperature and the state of our immune system do that? What evidence is there that signals our brain receives from the body can help with our higher decisions? Recently there has been quite a bit of evidence. The signals flowing from the body to the brain act silently, hardly breaking the surface of consciousness, giving us a diffuse and barely perceptible sense of the body, but nonetheless act powerfully, influencing our every decision. Not only that, but without their guiding hand even the cold rationality of economic man cannot make any progress. Gut feelings are not only r
eal; they are essential to rational choice.
The necessity of gut feelings becomes all the more apparent when decisions have to be made quickly, when we are online and in the flow, as Martin was this morning when given a minute or two to price the DuPont trade and then half an hour or so to buy the bonds he sold. In situations like these he does not have the leisure to gather all relevant data, consider all possible options, probability-weight their outcomes, and systematically work through a decision tree as an engineer might when given months, years even, to solve a problem. When pressed for a decision Martin therefore needs help drawing up a shortlist of options and their likely consequences. It is in this process that his gut feelings are brought in to streamline his thinking.
CAN WE TRUST OUR HUNCHES?
As we saw in the previous chapter, a great deal of our sensation, thinking and automatic reactions take place rapidly and pre-consciously. A number of scientists have studied the differences between pre-conscious and conscious thought, and have given these two sorts of thinking some memorable names. Daniel Kahneman calls them fast and slow thinking; Arie Kruglanski and colleagues, emphasising the motor element of thought, call them locomotion and assessment; others call them hot and cold decision-making. I prefer to think of them as online and offline thinking. Colin Camerer, George Loewenstein and Drazen Prelec, three of the founders of the new field of neuro-economics, have surveyed this research and summarised the differences between the two types of brain processing, labelling them automatic and controlled thought. Most of our thinking, they point out, takes place automatically, humming along behind the scenes, quietly, efficiently and rapidly.
Automatic thought Controlled thought
Involuntary Voluntary
Effortless Effortful
Proceeds in parallel; many steps carried out simultaneously Proceeds serially; one step at a time
Largely opaque to introspection; we cannot trace the mental steps we followed when reaching a conclusion Largely open to inspection; we can retrace the mental steps we followed when reaching a conclusion
A nice illustration of automatic thinking can be found in an experiment conducted by Pawel Lewicki and colleagues in which they asked people to predict the location on a computer screen of a cross that would appear at differing spots and then disappear. Unbeknownst to the participants, the location of the cross followed a rule, so it could be predicted. However, the rule was so complicated that no participant could formulate it explicitly. Yet, despite their inability to say what this rule was, people got better at predicting the location of the cross. In other words, the participants were learning the rule pre-consciously. This is a lovely experiment. It demonstrates that many of the mental processes we commonly assume are conscious in fact take place below the surface of awareness.
The intuitions of traders most likely rely on just this sort of pre-conscious processing of correlations. When making a statement such as this I have to tread carefully, for buried here is a minefield of issues. To begin with, many economists and cognitive scientists have disputed the supposed reliability of intuition and gut feelings. Can we trust judgements, they ask, that simply pop into our heads? Are gut feelings really the oracular deliverances they are often claimed to be? Behavioural economists think not. They have convincingly and in great detail shown that much of our automatic thinking comes warped by biases that frequently get us into trouble. Others, most notably the German psychologist Gird Gigerenzer, respond that many of our automatic thinking patterns are in fact efficient adaptations to real-life problems. Nonetheless, the issue remains: if gut feelings are sometimes right and sometimes wrong, then how can we know when to trust them? If we cannot know, then frankly intuitions are not of much use. We should instead, argue many economists, psychologists and philosophers, use more controlled thinking, bring in the correctives of logic and statistical analysis, to overcome the shortcomings of first impressions.
In order to answer the question of whether we can trust intuitions, we should first recognise that intuition is not an occult gift – it is a skill. An insightful answer along these lines emerged from what began as a dispute and developed into a collaboration between Daniel Kahneman and Gary Klein, a psychologist who studies naturalistic decision-making, in other words decisions made out in the field by experts. At first Kahneman doubted the reliability of intuition, while Klein believed in it. As they hashed out their disagreements, it became apparent that their different views stemmed from the types of people they were studying. Klein was working with people who had developed an expertise in fast decision-making – firefighters, paramedics, fighter pilots – and who unquestionably did possess intuitions worth trusting. Kahneman, for his part, was working with people whose predictions performed no better than chance – social scientists, political forecasters, stock pickers – people we should listen to only with a healthy dose of scepticism. So what separates these two groups? Why does one develop skill and reliable intuition, while the other does not?
Kahneman and Klein first agreed that intuition is the recognition of patterns. When we develop a skill at some game or activity we build up a memory bank of patterns we have lived through, and of which we have seen the consequences. Later, when encountering a new situation, we rapidly scroll through our files looking for a stored pattern that most closely resembles the new one. Chess grandmasters, for example, are said to store up to 10,000 board configurations which they access for clues on what to do next. Intuition is thus nothing more mysterious than recognition.
Given this point, Kahneman and Klein went on to conclude that intuitions can be relied on only if two conditions are met: first, people can develop an expertise only if they work in an environment that is regular enough to produce repeating patterns; and second, they must encounter the patterns frequently and receive feedback on their performance quickly, for only in this way can they learn. Playing chess exemplifies these conditions: chess grandmasters play game after game, the rules are fixed, and they find out quickly if their moves were right or not. Much the same can be said of paramedics, firefighters and fighter pilots. Political forecasters, on the other hand, inhabit a world that is far too fluid and complex to produce patterns, and even if something like a pattern does emerge, it does so with such a long time lag that learning it may take a lifetime. ‘Remember this rule,’ advises Kahneman: ‘ Intuition cannot be trusted in the absence of stable regularities in the environment.’
The question for us becomes, do the financial markets present stable regularities? Only if they do can traders and investors rely on their hunches. Within economics, opinions on this question have been close to unanimous: the markets do not. The strongest statement to this effect comes from the Efficient Markets Hypothesis in economics. According to the economists arguing for this hypothesis, the market moves when new information arrives, and since news by its very nature cannot be predicted, neither can the market. The legend of traders and investors heroically drawing on gut feelings is, they argue, pure mythology. No one can predict the market, nor consistently outperform it.
But is this true? The Yale economist Robert Shiller suspects it is not. He does not buy the idea that nothing – no personal trait, no training – can improve a trader’s performance. Shiller believes, on the contrary, that investing is like any other occupation, and that intelligence, education, training and hard work can indeed improve your performance. I think he is right. I also suspect that the Efficient Market Hypothesis has been a boon to many of the physicists, engineers and code-breakers employed by hedge funds, for these scientists have been able to find tradable patterns in what efficient market theorists believed was pure noise, and to build algorithms to exploit these patterns. Efficient market theory, because it has been orthodoxy for decades, may have limited the number of competitors looking for these patterns.
My experience with traders has been that they can indeed learn patterns, that they can develop an expertise in predicting the market. I and a colleague, Lionel Page, a brilliant statistician and behavio
ural economist, tested this hypothesis by looking at how consistently a group of traders made money, consistency being determined by what is known in finance as their Sharpe Ratio. The idea behind this measure is simple: it measures how much risk was taken in making a given amount of money. For example, if one trader makes $100 million in a year, and in the course of doing so never makes or loses more than $5 million in a single day, then his performance has been steady, his risk low, and his Sharpe Ratio would be high. If another trader makes $100 million but alternates between making $500 million one day and losing it the next, then his profit looks like no more than luck, even dumb luck, his risk is far too high, and his Sharpe Ratio would be low.
The differences between these traders can be compared to the driving styles of two taxi drivers. The first holds to the speed limit and gets you to the airport in 45 minutes. The second speeds at 100 mph for 15 minutes, stops for a coffee, drives back 10 miles to pick up a newspaper, then floors it at 120 mph into oncoming cars, snarls traffic for hours afterwards and narrowly avoids several head-on collisions, but by some miracle arrives at the airport in 45 minutes. ‘See, I told you I’d get you there on time,’ he says, and with that he holds out his hand for a bonus – I mean tip. Now, which of these drivers are you going to tip? Or ride with again? Sharpe Ratios allow the banks in effect to give their traders a driving test.
According to the efficient market hypothesis, traders and investors cannot make money more consistently than the stock market itself. This statement is comparable to saying you cannot drive to the airport – obeying the speed limit, that is – in under 45 minutes. But what we found in our study was that the traders could. They were like crafty cab drivers who keep figuring out shorter routes to the airport. The S&P 500 (an index of the prices of 500 large US stocks) has a long-term Sharpe Ratio of about 0.4, yet the experienced traders in our study had Sharpes higher than 1.0, the gold standard among hedge funds.