There's a nice juxtaposition between two of the biggest stories of recent days: retail investors piling into Game Stop; and the reluctance of ethnic minorities to take Covid vaccines and the slowness of the European Medicines Agency to approve them.

The thing is that these are all stories about extreme attitudes to risk. Small investors buying a dodgy company looks like risk-loving behaviour, whilst refusing to approve or take the vaccine looks like high risk aversion.

What explains such a big contrast?

Conventional neoclassical economics has no answer. It sees them as mere differences in tastes, which it regards as exogenous.

This won't do. It's just shrugging your shoulders. Not that this is the only problem they have. As Thaler and Rabin have shown, the standard approach to attitudes to risk predicts that we reject even good bets. And in fact, it struggles to explain such a basic fact as why people both gamble and buy insurance.

There is, however, a much better way of thinking about people's attitudes to risk, inspired by a paper (pdf) by Michael Woodford and colleagues.

When we think about a risk, such as a 50-50 chance of winning or losing £100, it's natural to ask: how would I feel if I lost or won £100? We try to translate expected monetary values into mental states. But this translation, says Woodford, is prone to "noisy coding" and yields answers that are only "approximate". Faced with the same gambles, therefore, willingness to bet can vary from person to person not because of differences in the marginal utility of wealth, but because of differences in coding. And in fact, the same person's willingness to bet will vary from context to context because the coding varies.

This might seem trivial, but it explains a lot. For example:

- Why we gamble and buy insurance. The questions "how would I feel if my house burnt down?" and "how would I feel if I lost £1000 in the casino?" have answers which are coded very differently. It's not that we have different marginal utilities over wealth as Friedman and Savage claimed (pdf). It's that these are utterly different questions.

- Why we pay more to cut risk from (say) 10% to zero than from (say) 30% to 20%. It's difficult to translate a 20% chance into mental states, so we still face uncertainty. A zero probability, however, eliminates this uncertainty. And we're willing to pay for that.

- Why some people back longshots or speculative stocks despite their poor prospects. In terms of expected value, a 1% chance of a £1000 win is worse than a 15% chance of a £100 win. Some people, though, code it differently. They figure that £1000 will do more than ten times as much good for them as £100 - for example by allowing them to break even after past losses or to pay off their debts. (This is of course compatible with prospect theory).

- Why some are reluctant to take the vaccine. Mere talk of the issue raises the subject of death and so colours the coding: "the jab could kill me". This is especially true for people who distrust authority.

- Why our attitudes to risk aren't always accurate as a guide to future utility. Just before the 2008 crash, Christoph Merkle asked UK equity investors how they'd feel if they lost a lot. He then surveyed them after they had actually lost money. And he found they were less miserable than they expected to be. This is consistent with noisy coding. In early 2008, it was difficult to translate a 20% loss into a mental state. By the end of the year this difficulty had disappeared. (I suspect people in early 2008 under-estimated the extent to which they'd be comforted by the knowledge that everybody else was in the same boat.)

- Why bubbles sometimes form. One driver of them is a shift in coding. Investors change from "this stock could lose me money" to "if I don't buy I'll miss out." And - in Game Stop's case - to "my buying will sock it to hedge funds." In this way, investors anaesthetize themselves against the possibility of loss.

- Why people more capable of cognitive reflection (pdf) tend to take more risks. It's because they are better at coding expected monetary values into mental states, and so face less uncertainty.

What we have here, though, is not merely a story about attitudes to risk. It's also a story about how some of the presumptions of neoclassical economics are doubtful. It presumes that our preferences can be described by a simple equation describing marginal utility. But this needn't be so. Our psychology is more interesting than that. It also presumes that preferences are just given. But in fact we can intelligently ask how they are formed. For example, past losses can make us more risk-seeking to get even, and good stories can encourage us to take risk. (A general failing of neoclassical economics is its incuriosity about genealogy). It also presumes that preferences are a guide to our future happiness. But as Merkle has shown, this is not the case: where there is noisy (or plain bad) coding, our preferences need not fulfil our interests.

What looks like an arcane area of economics is therefore central not only to how we behave in important respects but also to some of the shortcomings of neoclassical economics.