The idea unconscious biases influence decision-making in financial markets is nothing new. But behavioural finance has taken on new relevance in the age of Big Data and artificial intelligence.
The combination of digital tools and behavioural expertise has positive implications
Life used to be nasty, brutish and short. As hunter-gatherers out on the savannah 200,000 years ago, we were surrounded by mortal threats: ravenous beasts, raging heat, torrential storms. Safeguarding resources was key to survival. Spearing a gazelle for dinner meant quickly hauling it back to the cave before the local sabre-toothed cat took an interest.
In such a harsh environment, caution was an evolutionary advantage. The fear of loss became hard-wired into the brain. Other evolutionary quirks mean we experience the same stress responses when sitting in traffic and fretting about finances as we did when being chased by predators in prehistoric times: useful as a one-off, but corrosive when turned on chronically. What once helped us survive now threatens our health.
The idea primal emotions influence human choice and behaviour is nothing new. In the 1970s, pioneering psychologists Daniel Kahneman and Amos Tversky proved our decisions are often swayed by the neurological biases we inherited from our earliest ancestors. We get tired, stressed and distracted. We often lack self control and have a sense of fun, making our behaviour hard to pigeonhole into neat categories. By way of example, a shocking study in 2011 revealed judges dish out harsher sentences just before lunch.
Despite this, we are now in a far better position to do something about our failings as powerful data-driven technologies enable us to identify and correct for these errors. Informed by the latest neuroscience and empowered by new digital tools, economists are assembling vast empirical datasets and identifying irrational behaviours on a societal scale.
Drawing on these insights, policymakers are rolling out digital platforms to encourage citizens to improve their health and save for the future. In the finance industry, advisers are firing up high-tech risk-profiling tools to guide their clients through investment strategies, while asset managers are deploying computer programmes to ‘de-bias’ their portfolios and boost returns.
None of these methods is completely infallible, and an overreliance on computer models brings risks of its own. But used in the right way, digital technologies can spark us into awareness of the unexamined tendencies that shape behaviour – and allow us to transcend our all-too-human flaws.
Nudge, nudge; think, think
Start with the economics. Taking their cue from Adam Smith’s The Wealth of Nations (1776), neoclassical economists long assumed people act according to a rational calculation of self-interest. While this assumption might make for useful explanatory models, it comes with a significant problem: it doesn’t tally with how people actually behave. And, somewhat ironically, Smith was acutely aware of our human foibles; his earlier Theory of Moral Sentiments (1759) had roots in social psychology. Early 20th century heavyweight economists such as Thorstein Veblen and John Maynard Keynes also built behavioural aspects into the core of their thinking.
More recently, Kahneman’s research has helped guide economics back to these early insights – showing that we think according to two different systems. ‘Fast’ thinking (also known as system one) is typically automatic, unconscious and swayed by physical or emotional responses, while ‘slow’ thought (‘system two’) is more logical and calculating. Our system two brains might kick in when we are working out a mathematical problem or reading poetry, but most of the time we act according to system one.1
"Much of what we do every day we do in automatic mode,” explains Hannah Lewis, founder of Behave London, a consulting firm that applies psychological insights in business and finance. “Running on automatic isn’t necessarily always bad – most of the time it’s perfectly helpful. Unfortunately, we’re often running ‘sub-optimal’ behaviour. It can be difficult to change our automatic mode unless we experience some kind of disruption.”
Kahneman’s two-speed conception of human thought has its roots in the neurological structures of the brain, which have remained more or less constant since early Homo sapiens vied for supremacy with the Neanderthals. Because human beings’ system one brains are much alike, we are governed by a common set of cognitive biases – or mental heuristics, in the psychological parlance – that lead us to behave in similar ways in similar situations. It follows that while human behaviour might be irrational, it is systematic and even predictable.
This insight forms the basis of behavioural economics, a discipline that has risen to prominence in recent years thanks largely to the work of Richard Thaler, a professor at the University of Chicago Booth School of Business. Because it relies on empirical observations about how people behave – rather than theoretical assumptions on how they ought to behave – behavioural economics opens up a whole range of practical applications that neoclassical models cannot reach.
Thaler has written several amiably witty books outlining these applications, but his key contribution can be summed up in a single word: ‘nudge’.2 He argues the framework within which individuals make decisions can be optimised to nudge them to recognise their biases and counteract them. Thaler has argued nudging will only become more effective as “powerful statistical tools and datasets” enable us to track people’s behaviour with more precision.3
The digital nudge
Digital nudging techniques are being used to ‘de-bias’ financial behaviours
So it has proved. One of the most successful examples of Thaler’s nudging principle is the automatic enrolment of employees into pension schemes in the US and the UK. Behavioural science shows people tend to favour short-term gains over longer-term prosperity. This bias may have worked for our hunter gatherer, who needed to concentrate on where his next meal was coming from. But it’s a problem for a modern 30-something who urgently needs to start building their retirement pot.
By making it more difficult to opt out of pension schemes than it is to enrol on them, governments have been able to nudge savers into taking their future needs more seriously. In the US, automatic enrolment is estimated to have boosted annual savings rates by $7.4 billion.4 Nudging techniques have also been used to encourage people to add loft insulation to their homes to reduce their energy bills, remind them to pay their tax bills on time and even donate their organs after they die.
But nudging is now being applied on a bigger scale. Alex Pentland, a professor at the Massachusetts Institute of Technology (MIT), has conducted cutting-edge research into how digital tools and psychological insights can be used to gain a picture of how people behave en masse. His work shows nudges and small incentives can be deployed at the level of whole communities – and even populations – to predict and optimise social behaviour. This is known as ‘social physics’.
“The biggest limitation of [traditional] behavioural economics is that it is not systematic and holistic: it can't predict what will happen, only tell you some possibilities for how things can run off the rails,” Pentland says. “That means you need a data-driven approach to understanding behaviour – for example, social physics – to keep track of what is actually happening versus what you suspect might be happening.”
There is a dark side to the use of data to manipulate social behaviour – witness the Russian government’s deployment of cyber-propaganda techniques to disseminate political messages on social media. However, the combination of data-crunching tools and behavioural expertise has many positive implications for public life; from the provision of healthcare to the management of transport infrastructure.
For example, the San Francisco Mass Transit Authority successfully eased transport congestion on its network by using large-scale GPS data to track and predict people’s movements. By designing smartphone apps to nudge commuters using mobile games and monetary incentives, the authority was able to make the entire system more efficient.5
Even as behavioural economics entered the mainstream and began to influence government policy, one industry was curiously slow to recognise its significance: finance. This was partly due to a long-held assumption financial markets provide sufficient incentives for investors to behave rationally, making emotion and bias irrelevant. As Thaler once quipped, if some financial professionals do something stupid, there are plenty of others willing to take their money.
This conviction underpinned the efficient market hypothesis (EMH); the idea market pricing always incorporates the relevant available information, rendering the hunt for undervalued stocks futile. The global financial crisis of 2008-’09 was a salutary reminder of how irrational behaviour can drive market swings.
The ensuing turmoil exposed financial institutions’ overreliance on quantitative risk models based on mathematical probability, including Value-at-Risk (VaR), which plots a bell-shaped range of possible outcomes. Such metrics failed to account for the ways investors behave under pressure, says George Lagarias, senior economist at Mazars Wealth Management.
“There were two axioms by which we used to work prior to the financial crisis. One was that investors are inherently rational. The other was that, if you have enough data, markets will tend to show a ‘normal’ distribution of outcomes. Both of these axioms were tested and promptly rejected during the financial crisis. And if the models are failing, this shows you need to look at investor behaviour.”
More specifically, the fallout from the crisis focused attention on the biases that drive financial decisions. More than 170 individual cognitive biases have been identified. ‘Herding’ denotes the tendency for investors to copy each other’s strategies even as asset bubbles form, while ‘loss aversion’ leads them to do whatever it takes to avoid the psychological pain of losing, including taking big risks to dig themselves out of negative positions.
New data-driven tools can be used to identify the operation of bias at a macro level, which could help prevent future crises. Conducting experiments on an online trading platform, Pentland’s team at MIT was able to identify incipient herd behaviour as it developed; by tweaking the flow of information investors received, he led them to adjust their strategies and prevented bubbles from forming.6 Regulators are already beginning to create supervisory technology (or ‘SupTech’), which uses machine learning to monitor financial stability in a similar way.7
The fear of loss can also play out at an organisational level. Research from the International Monetary Fund (IMF) shows institutional investors such as sovereign wealth funds, pension schemes and insurers acted pro-cyclically during the financial crisis.8 Some investors were able to successfully exploit their risk tolerance, but many were driven by loss aversion to reduce risk at the wrong time, despite their long investment horizons.
If irrational behaviour can explain hasty decisions, it may also account for damaging institutional lethargy. For example, pension fund trustees have in some cases shown a reluctance to hedge a greater proportion of scheme liabilities, even if that would be the most appropriate course of action, because doing so would mean capitulating on previous market calls, and accepting they were wrong.
“If you have lived with a view that yields are too low and expressed that view by not fully hedging, it is awfully hard to change that view now,” says John Dewey, head of investment strategy in Aviva Investors’ solutions function. “It can also be more comfortable for pension trustees to stick with an existing strategy rather than undertake a more appropriate strategy that may in retrospect perform worse and be attributed to those individuals. Loss aversion and individual career risk play a part, as does a reluctance for large changes (anchoring and familiarity bias), and irrelevant comparisons to peers (herding).”
Similarities abound in the insurance market, as Iain Forrester, head of insurance investment strategy at Aviva Investors, explains. “Behavioural issues have contributed to the apparent reluctance of insurers to diversify into new assets,” he says. “Status quo and familiarity biases in their asset allocations, exacerbated by regulatory considerations, are leading to conservativism, resulting in insurers achieving sub-optimal outcomes with their capital.”
Nudging for advisers
At a micro level, digital tools can be used to make investors aware of their biases before they buy or sell securities, nudging them to focus on how their desired outcomes might be achieved. The financial advice industry is leading the way in this area. Advisory firms have devised innovative ‘risk profiling’ platforms that can map individual investors’ personalities, quantify their risk capacity and highlight their unconscious biases.
“There are three aspects to this: digital, data and design,” explains Greg Davies, head of behavioural finance at Oxford Risk, a company specialising in behavioural software to help people make better financial decisions. “Think of it as a Venn diagram: you have digital platforms as a mechanism for delivery of information; data that can enable you to personalise what you put in front of clients through the digital channel; and a design that makes the platform comfortable and easy to use. Then you have behavioural science at the centre to pull all of these elements together.”
These platforms are not infallible, and no one has yet devised a model that would correct all behavioural errors. Nevertheless, the enormous power of these digital tools raises a potential problem with nudging: its inherent paternalism. Thaler prefers the phrase ‘libertarian paternalism’ – which emphasises the scope for freedom of choice – but in practice nudging requires no active engagement from the nudgee. The risk is that non-professional investors are nudged by algorithms into making decisions whose implications they only dimly understand.
Davies argues the true potential of computer-based financial platforms lies in giving people the information they need to make informed and conscious choices, not in the kind of automated machine operations that eliminate the role of flesh-and-blood human beings. Properly designed, digital tools can incorporate nudging methods alongside other protocols to ensure clients are engaged, educated and aware of the true complexity of financial decisions.
These methods include ‘gamification’ techniques, or game-like elements that encourage people to engage and learn comfortably in complex environments. As Jeremy Leadsom, head of UK wholesale at Aviva Investors, puts it: “The great thing here is that technology is being used as an enabler and enhancer to the advisory process and experience. Real-time data can be combined with behaviourally-informed risk profiles to offer up effective prompts at key points in clients’ lives. What’s more, the frequency and tone of communications can be tailored to best meet a client‘s disposition and preferences.”
De-biasing in asset management
Nudging is also on the rise among professional investors, with the asset management industry waking up to the potential of behavioural science. As with individual investors, the most effective methods do not use machines to take matters out of human hands; rather, they nudge investors into awareness of their biases so they avoid costly mistakes.
Gulnur Muradoglu, professor of finance at Queen Mary University of London and director of the Behavioural Finance Working Group, says the key to successful nudging in asset management is to identify and target particular irrational behaviours to prevent them from reoccurring.
“There are some asset management companies that do this sort of training with their fund managers, but it has to be specific about your actions [to work],” she says. “Fund managers can be encouraged to check their previous forecasts on a timely basis, so they are not under the illusion they are right every time. That will help them calibrate themselves better, and have more realistic expectations about the future.”
Digital nudging techniques are being used to ‘de-bias’ specific behaviours in this way. In a recent research paper, consulting firm McKinsey & Co. described an approach known as ‘de-biasing’, which it estimates could lead to improvements of fund performance of between 100 and 300 basis points per year among asset managers.9
The process begins with the deployment of data analytics software to undertake performance decomposition, which indicates how certain types of decisions (stock selection, the timing of asset purchases or sales) have contributed to or detracted from historic returns. The results are combined with findings gathered from detailed psychometric questionnaires that pick up on the emotional and environmental factors that influence fund managers’ decisions.
The nature of the nudge will depend on the bias identified through these methods. ‘Visual nudging’ uses fund managers’ software to automatically present them with alternative metrics about the structural environment – such as analysts’ upgrades or price performance relative to other stocks in the sector – they might not have considered. Visual nudging has been found to be particularly effective in addressing an error known as ‘anchoring’, a tendency to base or ‘anchor’ decisions on illogical reference points.
Giles Parkinson, global equities fund manager at Aviva Investors, uses a visual nudge known as ‘clean-sheet redesign’ to help him avoid this pitfall. “One common tendency among investors is to anchor their thinking to the price they paid for a stock. They might have paid $100; if it falls to $80, they might hold on to it for too long in the hope that it will rise back to the original level. This is totally irrational,” he says.
“By disabling the book-cost display on the reporting software, which would otherwise provide a constant reminder of the amount I paid for a particular stock, I ensure I am continually reappraising the portfolio with fresh eyes. It’s a way of asking myself: ‘If I was starting from scratch, would I own this stock?’ It’s good mental discipline,” Parkinson adds.
So what does the future hold for behavioural finance? With the rise of Big Data and machine-learning algorithms, some investment firms have spotted a new alpha opportunity. The quantitative hedge fund industry has developed sophisticated computerised investment models that can ruthlessly zero in on mispricing or arbitrage opportunities human traders are too slow to spot.
Even the most sophisticated AI-led investment tools have built-in limitations, however. Machine algorithms risk what is known as ‘overfitting’, a tendency to make conclusions on the basis of random correlations, mistaking noise for signals.10 A deeper problem is that, while algorithms tend to be good at exploiting a particular inefficiency, they are less good at adapting when the environment shifts.
“Most quant models are designed to exploit a particular factor, whether that is value, momentum or size,” says Parkinson. “There will be times those factor strategies do not work, either because the factor is cyclical, or because the structure of the market has changed and the opportunity has disappeared permanently. What do you do as a quant investor in that situation? Do you leave the model running in the hope it will work again, or do you decide it is the model that is flawed?”
Due to their difficulty in adapting to contextual changes, autonomous machines risk becoming trapped in feedback loops, repeating the same trades without registering their distortive effects on the wider market. The Financial Stability Board warned of this danger in a recent paper on AI.11
Then there is the risk the computers could simply malfunction, as was the case with high-frequency trading firm Knight Capital. The company collapsed in 2012 after it sustained losses of $440 million due to a flaw in its algorithms, inflicting disruption on other investors. In a note on the incident, analysts at research provider Gavekal observed: “Sometimes all computers do is replace human stupidity with machine stupidity, [which] can devour markets far faster than any human panic can achieve”.12
Beyond the cave
For all our flaws – the unconscious biases, the proneness to stress and anxiety – humans still have key advantages over machines: the ability to adjust to the uncertainty of a rapidly-changing environment, the capacity to appreciate ambiguity and nuance. There are advantages to be gained from incorporating computing power into investment decision-making, but only when it complements human judgement. Bringing together design, data and psychology, the digital nudge offers a way to combine the best traits of humans and machines.
Take chess. Despite massive leaps forward in AI, the best players aren’t algorithms but so-called ‘centaurs’; human players given the freedom to consult computers that can alert them to potential pitfalls before they make their own decisions as to the next move. In the same way, data-led tools can help both professional and non-professional investors become aware of their biases, and consciously work to counteract them.
While it is no magic bullet, the combination of digital platforms and psychological insights is already having a transformative impact across economics, government policy and financial decision-making. Behavioural science teaches us that part of our brains will always remain in that dimly-lit prehistoric world, in which we relied on instinct to survive. By using the power of data, we are beginning to emerge from the cave, one carefully-constructed nudge at a time.
1 Kahneman, Thinking, fast and slow, 2011
2 Thaler and Sunstein, Nudge: Improving decisions about health,wealth, and happiness, 2008
3 Thaler, ‘Behavioural economics: Past, present and future,’ American Economic Review, May 2016
5 ‘Incentives shift BART riders out of the morning rush,’ Bay Area Rapid Transit System website, August 2017
6 Pentland, Social Physics: The new science of information flow, 2014
7 ‘Artificial intelligence and machine learning in financial services: Market developments and financial stability implications,‘ FSB, November 2017
8 ‘Procyclical behaviour of institutional investors during the recent financial crisis: causes, impacts, and challenges,’ International Monetary Fund, September 2013
9 ’An analytics approach to debiasing asset management decisions,‘ McKinsey & Co., December 2017
10 ‘Fintech: Search for a super-algo,’ Financial Times, January 2016
11 ‘Artificial intelligence and machine learning in financial services,’ FSB
12 Op cit.
13 Thaler, Misbehaving: The making of behavioural economics, 2015
14 ‘Procyclical behavior of institutional investors during the recent financial crisis,’ IMF, September 2013