We identified herding choices in two situations: first, when the participant decided to buy a share after seeing information that most of the herd (i.e. three or four out of four) had bought it too; and second, when the participant decided not to buy a share after seeing information that most of the herd had not bought it either. Contrarian anti-herding choices were identified in the opposite situations – when a participant bought the share even though most of the herd had not bought it, or when she did not buy the share even when she could see that most of the herd had bought it. We also analysed the impact of some of the participants’ individual differences, which we captured by asking them to complete some biographical questionnaires and personality tests before we brought them into the brain scanner.
Figure 4. Financial herding and anti-herding in the brain scanner: task structure and brain activations in amygdala, prefrontal cortex and anterior cingulate cortex respectively.
A strong herding tendency was identified in our participants. They were copying the herd’s decisions far more often than we would expect if they were just deciding randomly. This confirms evidence from a diverse range of sources, that humans have a strong tendency towards herding; anti-herding is much more unusual. We are copycats much more often than we are contrarians. In order to pick up what was going on in the brains of our experimental participants we focused our analysis on those brain regions commonly implicated in decision-making. One of these is the amygdala, part of the limbic system (a collection of brain areas associated with emotional processing) and thought to be involved when we are processing negative emotions, including fear. Another is the ventral striatum, an area implicated in the processing of rewards, the focus of Schultz’s reward prediction error model. Finally, we looked at activations in the anterior cingulate cortex, an area generally associated with higher cognitive functioning. There is some evidence that the anterior cingulate cortex operates something like Plato’s charioteer: it steps in to resolve neural conflicts, including in situations where System 2 reason and System 1 emotion are competing.22
To explain what neuroscientists mean by a neural conflict, we can look to other studies from social neuroscience. A classic neuroeconomic study of social conflict was conducted by American neuroscientists Alan Sanfey, Jonathan Cohen and their colleagues.23 The team brought nineteen people into their lab and asked them to play the ultimatum game, a famous experimental game widely used by behavioural economists to capture people’s social preferences – that is, people’s propensities to be selfish or generous.24 As with many variants of this experiment, Sanfey and his colleagues split their participants into two groups. They gave one player (the ‘proposer’) $10 and asked them to divide the money between themselves and a second player (the ‘responder’). If the responder accepts the proposer’s offer, then the money will be allocated accordingly. If the responder rejects it, however, then neither player gets any money. The challenge for the proposer, then, is to figure out the lowest possible offer that the responder is likely to accept. Standard economics, at least if starkly presented, predicts that an offer of $1 should do it. If both players are rational, selfish maximisers, then the responder would not reject an offer of $1 when the alternative is $0, because a rational economic decision-maker will always prefer something to nothing. In contrast to the predictions of mainstream economics, however, many experiments with the ultimatum game show that proposers are surprisingly generous, and will offer not much less than 50 per cent of the total, while responders will reject relatively large offers, even if those offers are much greater than $1. This is interpreted by many as evidence of our socialised natures. Our propensity towards generosity means that proposers are inclined towards ‘fairer’ offers – where fair is defined as something approximating a 50:50 split. When responders decide that proposers are making unfair offers, they will punish the proposers by vetoing the offer even though the veto leaves the responder with nothing too.
Sanfey and his colleagues were interested to see how their responders would behave if they decided that they were being treated unfairly. They scanned the brains of the responders, targeting three main regions of interest: the insula, parts of the prefrontal cortex, and the anterior cingulate cortex. Neuroscientists think that the insula is implicated when we feel negative emotions like disgust. Disgust can be understood not only as physical repulsion, such as the feeling when we smell a foul odour. Disgust also has a social corollary: the disgust we feel when being unfairly treated. The experimenters found significant activations in all these brain areas. When their proposers offered the responders much less than half the money, activity in the insula captured the responders’ social disgust of feeling cheated. This emotion was so strong that the responders were inclined to punish the unfair proposers by rejecting their mean offer, even when it meant getting nothing themselves. Sanfey and his colleagues also inferred that the prefrontal cortex was driving more economically sensible choices. On purely economic grounds, it is better to win a small sum than nothing at all. The anterior cingulate cortex was acting as arbiter, reconciling the conflict between the cognitive desire for more money and emotional responses such as anger and resentment that trigger retaliation when a person feels wronged.
For our experiment, we used the fMRI scanner to capture a different dimension to these social emotions – specifically the types of cognitive or emotional responses that drive us when we are herding or rebelling. With regard to the cognitive dimension, we hypothesised that our participants might be driven by a form of self-interested herding – linking to the Bayesian social learning experiments of Anderson and Holt that we introduced in chapter 1. Their findings were consistent with the idea that herding copycats are using Bayes’ rule to reconcile contradictions between private information and social information. In our experiment, the private information was the objective evidence communicated via the share price charts and the social information was conveyed via the images of the choices of the herd.
We studied the fMRI evidence and identified significant differences in brain activity between the first phase, when the participants were looking at the share price chart (their private information), and the second phase, when they were looking at images depicting the herd and their choices (their social information). When the participants were looking at the social information about the herd’s choices, areas of their ventral striatum were more strongly activated than when they were not looking at the social information. This finding is consistent with the idea that social information triggers reward learning. Non-social factors were important too, specifically the different participants’ preferences for different types of stocks. There were two broad types of participants – some who preferred stocks with high average values, and others who preferred stocks with low average values. Why would someone prefer a stock with a low average value? Post-experiment questioning revealed that some participants thought that a stock with a low value today might turn into a high value stock in the future. Either way, both groups were predicting the likely rewards from the stocks and the ventral striatum was activated more strongly when participants were buying the type of stocks they generally preferred.
The activations in other brain areas differed depending on whether our participants were herding or anti-herding. When participants were herding, they showed significant activations in the amygdala – an area, as noted above, associated with processing negative emotions such as fear. This finding is consistent with the idea that herding and fear are somehow related. Perhaps when we are feeling fearful we want to avoid risks, and are thus more likely to collect together in groups and conform with the herd. We also found significant activations in the anterior cingulate cortex of the contrarians making anti-herding choices. One possibility, similar to Sanfey and his colleagues’ interpretation of their brain scanning evidence, is that the anterior cingulate cortex is mediating a neural struggle. The ventral striatum is capturing our desire for reward, the amygdala is capturing the fear associated with the risk of disagreeing with th
e herd, and the anterior cingulate cortex is mediating this neural conflict.
As well as picking up some emotional processing, our experiments also captured how people respond to private and social information. Our evidence linked to two of herding’s facets: first, how people use social versus private information; and second, how reason and emotion interact when they are balancing these different sources of information. Our fMRI evidence was consistent with the idea that a mixture of objective, cognitive and subjective, emotional influences was driving decisions to join the herd. Our finding links with insights from neuroscientists Ramsey Raafat, Nick Chater and Chris Frith, based at University College London. They identify the transmission of thoughts and information between individuals as a key characteristic of human herding, and suggest that interactions between unconscious ‘automatic contagion’ and conscious ‘rational deliberation’ drive this facet of herding.25 Our experiments also illustrated something about the interplay between the rational, economic influences associated with economists’ theories of self-interested herding and other social scientists’ theories about the emotional drivers of collective herding. How can we link this experimental evidence with Kahneman’s dual systems model? If Bayesian explanations are true, then whether information is private or social shouldn’t matter. We process it all using higher cognitive functioning, drawing on our System 2 thinking. But then, why do we see activations in the emotional processing areas when people are thinking about what others in a herd are deciding? Paralleling the interpretation of the Engineer-Lawyer fMRI evidence about System 1 thinking from De Neys and colleagues, if Bayesian explanations are only part of the story (not necessarily false, just not the only thing going on), then we will see activations in areas usually associated with emotional, intuitive decision-making. The fact that neural areas associated with emotional processing are activated during herding suggests that it is not all about cool, calm calculation, despite what many economists might claim.
Herding heuristics
As we have seen, neuroeconomic experiments can capture interactions between emotion and cognition, connecting with Kahneman’s division of System 1 thinking and System 2 thinking. Another insight from Kahneman’s model links to the speed of decision-making. As we noted above, System 1 often dominates because it requires less cognitive effort. When we herd, is this because System 2 is lazy and we want to avoid the time and effort it takes to do the careful reasoning when there are quicker decision-making tactics available? If so, then following the herd will not be a controlled, logical choice. Instead, it may be a quick, automatic response driven by System 1. This connects with another set of insights from Daniel Kahneman about simple cognitive tools known as heuristics – quick decision-making rules. Perhaps in our herding experiments we were picking up the operation of a herding heuristic.26
How do herding heuristics work in practice? We herd because it is quicker and easier just to follow others, even if there is a chance we are simply copying their mistakes. Imagine you need to buy a new fridge, and you know that your neighbour has just spent a lot of time investigating the best brand of fridge to buy. Why would you repeat all that effort when you could just ask them for a recommendation? Your heuristic is to ask your neighbour. This will save you time and energy. But the problem with heuristics is that, whilst they are quick and convenient and often work well enough, they sometimes, though not always, lead to systematic mistakes – what behavioural economists and economic psychologists call behavioural biases. When we follow others, we may be leveraging valuable social information, or we may just be repeating their errors. Our neighbour may have bought their fridge on impulse, perhaps just because their neighbours had bought the same, without properly checking its specifications. If we follow them, then we too might end up with a second-rate fridge.
Cognitive psychologists Amos Tversky and Daniel Kahneman identified three main groups of heuristics and related biases: the availability heuristic, the representativeness heuristic and anchoring/adjustment. When we use the availability heuristic, we judge the chances of a specific event happening according to how easy it is for us to retrieve and recall relevant information. When we see a crowd in front of us, it is a clear and salient signal. Our vision of the herd in front of us is readily available and close to the top of our minds. By looking at the herd and simply copying them, we can circumvent other more costly cognitive devices which require more time and effort, for example memory or calculation. When we use the representativeness heuristic we are judging the likelihood of an outcome by comparing it with what we interpret as similar experiences and events in the past. This encourages us towards herding because we will assume that others’ decisions provide clues as to how we should judge a situation ourselves. When we use anchoring and adjustment heuristics, herding will emerge if we are using the group’s consensus as a social reference point. Again, this saves us time and effort because we don’t have to start from scratch each time we are faced with a new choice. For example, if we are buying or renting a house, we may choose an area where members of our family or friends have recently moved. We use the information they give us about prices, local amenities and transport links as a reference point and, taking account of our own preferences too, we adjust our own choices around this reference point.
The psychologists Gerd Gigerenzer and Daniel Goldstein emphasise that using these heuristics is not irrational. Using heuristics is sensible. We build them into our routines to save us the time and energy that we would need to exert if we thought deeply about everything we do. For copycats, imitation is a ‘fast and frugal’ heuristic. It is a cognitive short-cut helping us to make quicker and more efficient decisions in social situations.27 When we have social information we can be more selective. We can bypass laborious information-gathering exercises. Buying a fridge is a relatively simple choice – we want a machine that chills and freezes. For more complex purchases, we might need more help. For example, if we have a few friends and family who have recently bought new smartphones or computers, we may use a herding heuristic to guide our purchases. This might be more sensible than copying our neighbours’ fridge choices, as noted above, because most of us can easily grasp the basic functions of a fridge. In the case of phones and computers, however, the information and options available are often overwhelmingly confusing. So, it makes sense to look to other people we know, who might know more than we do. We copy their purchases because we assume that they are knowledgeable, have researched what has the best capabilities and what is the best value, and can understand all the esoteric technical information about the various options. We couldn’t do a better job ourselves in deciding which phone or computer to buy, so we rely on our herding heuristics to speed up and simplify what might otherwise be a complex, time-consuming and confusing decision-making process. Herding helps us to deal with the problem of choice and information overload – both of which might otherwise paralyse our decision-making.
All this raises the question of why and how we developed our herding tendencies in the first place. Where do our herding instincts come from, and do they suit the modern world? What connections are there with our basic instincts, developed over millions of years of evolutionary history? Evolutionary biology offers some key insights – not only about the evolutionary origins of dual-thinking systems in humans and other animals, but also the ways in which our instincts to herd might have served important purposes in primitive environments where resources were scarce. The quick, automatic and instinctive System 1 styles of thinking and deciding are older in evolutionary terms, whereas System 2 thinking, associated with conscious, deliberative, cognitive effort, has evolved more recently.
We have seen that there are number of ways in which we can apply Kahneman’s division of System 1 fast and System 2 slow thinking styles in reconciling the different conceptions of self-interested herding and collective herding explored in chapters 1 and 2. Neuroeconomics adds to these insights by providing some evidence that our herding decisions involve complex interac
tions between cognition and emotion. They cannot be categorised, in any binary way, as either rational or irrational. Our choices to follow others may reflect a mixture of logical decision-making and more unconscious and emotional influences.
Fast interconnectedness in our modern globalised world affects our daily lives like never before. Technologies such as social media connect us closely together with complete strangers, sometimes many thousands of miles away. Information, goods and services, food and addictive substances are abundant and easy to come by. We can rapidly choose to consume something new with a click of a mouse button. This strange new world has enabled our basic copying instincts to spread on a massive scale. We can follow the social information collected via online reviews on Airbnb, Uber, eBay, TripAdvisor and price comparison sites, amongst many others. Whether we can or should trust these sites as much as we would a friend illustrates some of the limitations of copying behaviours when they emerge on an epidemic scale. When all our interactions are so anonymous, information can be manipulated to encourage us to follow fake news and other spurious information about what other people choose or think. Another worrying implication is that collective herding can dominate self-interested herding far more easily because modern technology suits our fast System 1 thinking style. When all our decisions can be made so quickly, there may be no time for the slow and careful reflection associated with System 2 thinking.
Copycats and Contrarians Page 9