The Sign of the Feedback

Diamond and Dybvig (1983) explained bank runs as an equilibrium held together by belief: if I expect you to withdraw, I withdraw, and a bank that would have been solvent on Tuesday fails on Monday. The model also shows how to remove it: a credible guarantee makes waiting the sensible choice. Read the later history against the paper and something uncomfortable appears. Where deposits are imperfectly insured, runs of this kind remain possible, as Northern Rock in 2007 and Silicon Valley Bank in 2023 demonstrated. The theory's best vindication is an absence it helped bring about. Physics has no equivalent achievement. A field whose successes can delete their own evidence is doing something strange, and the strangeness is the best clue to what economics is.

Whether it counts as a discipline depends on what we take a discipline to be. Ownership of subject matter will not do the work. Poverty belongs to sociologists as much as to us, regulation to lawyers, choice to psychologists, and the border disputes are territorial rather than intellectual. Technique fails as a criterion too: one department holds archival historians, randomised trials, axiomatic proof and work that is ethnography in all but name. What holds a field together is a durable habit of asking particular questions, agreeing what counts as an answer, and quarrelling about why an answer failed. Economics has that habit, organised around constraint, incentive and interdependence. It also has a shared embarrassment: the people it studies read what it writes.

MacKenzie's (2006) account of Black-Scholes-Merton says more than the loose thought that ideas influence behaviour. Sermons influence behaviour. What happened in Chicago after 1973 was infrastructural. The formula arrived on printed sheets that market makers carried onto the floor, then in software, then in the shared sense of what counted as a hedged book, helping arbitrage away prices that strayed from it. Observed prices came closer to the model's values because the machinery producing those prices had the model inside it. Value at Risk repeated the pattern in a more official register: after the 1996 Market Risk Amendment, banks could use their own VaR models to set the capital they held against market risk, giving a modelling convention the force of a constraint. When everyone measures risk the same way, the measured risk begins to move with the measuring (Danielsson and Shin, 2003). Once a model is wired in that deeply, a gap between prediction and data admits two readings: the model is wrong, or not yet fully installed. Falsification turns into a question about plumbing.

This should unsettle us more than it does, because the same reflexivity runs backwards just as easily. Lucas (1976) showed that estimated relations shift when policy changes the rules under which expectations form. The Phillips curve was a lever that snapped when gripped. Publicity ruined it. Set that beside Black-Scholes and two celebrated arguments about the same phenomenon give opposite answers to one question: what happens to a regularity when everybody knows it?

Reflexivity has a sign, and the sign comes from the structure of the regularity rather than the quality of the model. Some regularities run on shared understanding. Expectations are part of the mechanism, so telling people about it supplies the belief the mechanism needs. An auction allocates because bidders know the rules they are bidding under (Milgrom, 2004); an inflation target works only if wage setters expect it to hold (Kydland and Prescott, 1977); where sovereign debt markets admit a good and a bad equilibrium, a promise can select between them. De Grauwe (2012) made that case about the eurozone; Italian yields fell after Draghi's three words in July 2012, before a bond was bought. Other regularities run on the opposite fuel. They need the people inside them not to see the pattern, because seeing it is enough to make acting on it worthwhile: an exploitable trade-off between inflation and unemployment, an arbitrage, a waiting-time target that rewards whoever learns to sort the queue. Publicity feeds the first kind and dissolves the second. Goodhart's Law (1975) is the second case compressed into a maxim; it reads less like a discovery than the Lucas critique because managers had known it for years. Before publishing, ask which fuel your regularity burns. No chemist faces that question.

Hold both limbs at once and something appears that neither Lucas nor Goodhart quite says. Inflation targeting is the deliberate sale of one sign to buy the other. A central bank able to surprise the public owns an ignorance-dependent asset, valuable in proportion to how few people anticipate its use and worthless once they do. Kydland and Prescott's advice amounts to giving that asset away and taking a coordination-dependent one in exchange, an expectation that strengthens as it spreads. Forty years of published forecasts, minutes and forward guidance can be read as a portfolio decision about which kind of reflexivity a central bank wants to hold. The sign is something institutions can be built around.

Hayek (1945) makes this harder. If knowledge is dispersed and local, the economist must guess which description of the situation people will act under, and models are one of the places descriptions come from. They teach us what to notice. Portfolio insurance, grown from the same options theory that had bent prices into line, handed many institutions the same instruction to sell as prices fell; the Brady Report (1988) named that selling as one element in the crash of October 1987, and MacKenzie reads the episode as counterperformative. One theory, both signs, fourteen years apart. The sign is not a property of the model. It depends on how many people hold it and what they can see while holding it, which means it can invert quietly, with no equation amended. League tables begin as measurements of universities and end as descriptions of what ambition feels like to a seventeen-year-old.

The obvious objection is that almost no regularity is pure. Most have a coordination limb and an exploitable one, and the sign depends on which margin is pressed. That sharpens the demand rather than dissolving it, since nobody is asked to call the sign correctly in advance. The requirement is to say what would flip it, and that can be checked afterwards against what happened. Much of economics is not reflexive at all. Harvests answer to rainfall, budget constraints bind whether or not they are theorised, and a badly specified model can fail before anyone reads it. Being self-confirming is not the same as being false: a crude pricing convention that lets strangers trade is true enough for the work it does. Medicine blinds its trials because patients respond to what they are told, and yet the pancreas does not read the leaflet. We can borrow randomisation. We cannot borrow an object that stays still when unobserved.

What economics is asked for keeps the problem in front of it. Historians reconstruct paths from traces that will not recur; economists are asked to build things, and the things are rules. Deaton and Cartwright (2018) put the scale-up problem carefully, and it is sharper than external validity. A pilot measures responses to a scheme nobody has learned to anticipate; a national rollout is a scheme everybody has learned to anticipate. The estimate does not decay with distance. It decays with publicity.

Economics already knows how to handle this in one corner of itself. A mechanism is incentive compatible when its rules leave agents with an incentive to behave as intended even after they understand the mechanism, a stability condition under publicity written directly into the design. Mechanism design must satisfy that condition because its objects are rules, and rules get read. My proposal, and it is a proposal rather than a report of current practice, is that the condition should be generalised from mechanisms to claims. A complete economic claim would state who must know it for it to hold, what they must believe, which behaviour depends on that belief, and what becomes of the mechanism when the belief becomes general. A claim offered without this is incomplete in the way a prescription without a dose is incomplete. Not humility as good manners; a component of the claim, whose absence is a specification error.

Is economics its own discipline? Yes, though not because it has fenced off a method. Its logic is borrowed, its tools shared, its subject matter contested by anyone who wants it. What is unique is the demand its object makes. Economists work on a system that reads their conclusions and rearranges itself around them, and no other field is so regularly obliged to say in advance how its own influence will alter the evidence for it. That obligation, met or dodged, is the methodology. It makes the economist an observer, an engineer and a moving part, sometimes within a single paper. A physicist's theory has only to survive being read. Ours have to survive being used, and the honest ones say in advance how they expect to die.

Bibliography

• Danielsson, J. and Shin, H. S. (2003) 'Endogenous risk', in Field, P. (ed.) Modern Risk Management: A History. London: Risk Books, pp. 297–316.
• Deaton, A. and Cartwright, N. (2018) 'Understanding and misunderstanding randomized controlled trials', Social Science & Medicine, 210, pp. 2–21.
• De Grauwe, P. (2012) 'The governance of a fragile eurozone', Australian Economic Review, 45(3), pp. 255–268.
• Diamond, D. W. and Dybvig, P. H. (1983) 'Bank runs, deposit insurance, and liquidity', Journal of Political Economy, 91(3), pp. 401–419.
• Goodhart, C. A. E. (1975) 'Problems of monetary management: The UK experience', Papers in Monetary Economics, Vol. 1. Sydney: Reserve Bank of Australia.
• Hayek, F. A. (1945) 'The use of knowledge in society', American Economic Review, 35(4), pp. 519–530.
• Kydland, F. E. and Prescott, E. C. (1977) 'Rules rather than discretion: The inconsistency of optimal plans', Journal of Political Economy, 85(3), pp. 473–491.
• Lucas, R. E. (1976) 'Econometric policy evaluation: A critique', Carnegie-Rochester Conference Series on Public Policy, 1, pp. 19–46.
• MacKenzie, D. (2006) An Engine, Not a Camera: How Financial Models Shape Markets. Cambridge, MA: MIT Press.
• Milgrom, P. (2004) Putting Auction Theory to Work. Cambridge: Cambridge University Press.
• Presidential Task Force on Market Mechanisms (1988) Report of the Presidential Task Force on Market Mechanisms. Washington, DC: U.S. Government Printing Office.