Track 1 · Machinery · lesson 8
Booms, busts, and four schools who disagree about why
13 min
Track 1 · Machinery · lesson 8
13 min
Start with what is not in dispute.
Economies do not grow smoothly. They run hot for several years — output rising, unemployment falling, credit expanding, asset prices climbing — and then they stop, sometimes gently and sometimes not, and for a period output falls, people lose work, and lending contracts. Then it starts again. This has happened repeatedly, in many countries, under many different arrangements, for as long as anyone has kept records.
That is a fact. Everything after it is a model.
Four things get run together in every argument about the economy, and separating them is the single most useful habit this track can leave you with.
A fact is something measured. A model is a story about what causes what, which the facts are consistent with. An assumption is what the model needs to be true and cannot itself demonstrate. A value judgement is a claim about what would be better, which no measurement can settle.
Most arguments that feel unresolvable are two people agreeing on the facts, holding different models, disagreeing about an assumption neither has stated, and each hearing the other's value judgement as a factual error.
Here are four models of why a boom turns. Each is held by serious people. Each explains some episodes better than others. None of them is labelled with the name of a tribe, because the tribe is not the interesting part.
The demand reading. Spending is what keeps output high, and spending depends partly on confidence, which can fall for reasons that are not proportionate to anything real. When enough people cut back at once, the cutbacks validate each other: your reduced spending is someone else's reduced income. The bust is a coordination failure, and it can persist because nobody can restart alone. It assumes output is usually limited by demand rather than by capacity.
The credit reading. Booms are financed. Debt builds quietly during the good years because lending standards relax when nothing has gone wrong for a while. At some point the borrowing cannot be serviced out of income, and everyone tries to repair their balance sheet at the same time, which crushes spending. The bust is the unwinding of an accumulated stock of debt, and its depth depends on how much was accumulated. It assumes the level of debt, not just the flow of new borrowing, changes behaviour.
The misallocation reading. Cheap credit does not just produce more investment, it produces different investment. Projects that only work at a low cost of money get started, resources move into them, and when the cost of money normalises those projects turn out never to have been viable. The bust is the discovery, and the pain is the reallocation of capital and labour out of the wrong industries. It assumes the composition of investment during the boom is systematically distorted, and that the distortion is not visible at the time.
The real-shock reading. Something genuinely changes — a technology, an energy price, a supply chain, a harvest. Output falls because the economy's actual productive capability fell or had to be reorganised, and the downturn is the adjustment rather than a malfunction. It assumes the fluctuations are responses to real changes rather than to failures of coordination or finance.
Predict
Two readings
Output falls below what the economy could produce, because of a failure of demand or of the credit system.
Output falls because resources have to move out of activities that turned out not to be worth doing.
Both sides accept
Both accept that booms and busts are real and recurrent, that credit conditions influence them, and that some downturns contain both a genuine reallocation and a shortfall of demand. Neither denies the other's mechanism exists. They disagree about which one dominates in a typical episode, and therefore about what an intervention would be acting on.
Which is which? Put each one on a side.
Unemployment rises across almost every industry at once, including ones that never expanded
Job losses are heavily concentrated in one industry that had tripled in size over four years
Firms report full order books and cut staff anyway because they cannot obtain short-term credit
Models are cheap to hold and expensive to act on. Here is the same disagreement with a decision attached.
What would you do
Take any confident claim about the economy and break it into the four pieces.
What is the fact? Usually smaller and more boring than the claim.
What is the model? Name the causal story, and check whether the fact is actually evidence for it rather than merely consistent with it.
What does the model assume? There is always one load-bearing assumption, and it is usually about something unobservable — how much slack, how forward-looking people are, whether the composition of investment was distorted.
What is the value judgement? Whose cost, borne when, is acceptable. This one cannot be settled with data and does not become more true by being asserted harder.
Check
Economies cannot be run twice. There is one path through history and no control group, the interventions are chosen in response to the conditions they are meant to address, and the number of genuinely independent episodes is small — a few dozen, in different countries with different institutions, over two centuries in which almost everything else changed.
That is a hard inference problem, not a shortage of clever people. It is worth holding on to when a confident answer is offered, in either direction: the confidence is coming from somewhere other than the evidence, and it is usually worth asking where.