Economists have long had an awkward hole in their account of why economies go through booms and slumps. In the standard model, what a country produces depends on what it puts in: workers, machines, buildings. Divide output by those inputs and you get productivity.
The problem is that measured productivity swings about a great deal over the business cycle. It rises in booms and falls in recessions – and the conventional story is that these are swings in technology. But technology does not plausibly get worse in a recession. Nobody ‘forgot’ how to build a car in 2009.
In a new study – which I co-authored with Yan Bai (University of Rochester) and Kjetil Storesletten (University of Minnesota) – we offer a different explanation (Bai et al, 2026). It starts with something seemingly mundane: goods have to be bought before they count as having been produced.
The restaurant problem
Consider a restaurant. In the standard theoretical view, its output is a function of its kitchen, its tables and its staff. But a restaurant with empty tables produces nothing. What it actually produces depends on how many diners walk in.
The same is true far more widely: dentists need patients; car dealers need shoppers; and consultants need clients. Production requires a match between someone who wants to sell and someone who wants to buy – and that match takes effort on both sides.
Once you take this idea seriously, demand stops being merely a claim on output and becomes an input into it. When customers search harder – via more shopping trips, more browsing, more willingness to go out – more of the economy’s existing productive capacity gets used. Output rises with no change in technology, machines or hours worked. And because the statistician measures productivity as output divided by conventional inputs, this shows up in the data as a rise in productivity.
Do people really shop more during booms?
The short answer is: yes, they do. Using the American Time Use Survey, we find that the hours that households spend shopping move closely with the business cycle: the correlation with GDP is 0.56, and with measured productivity 0.52. When the economy is strong, people spend more time searching for things to buy. When it is weak, they stop looking.
That is a suggestive fact, not a theory. To find out whether it matters quantitatively, we build a model of the goods market in which firms post prices and customers weigh price against congestion – the risk of turning up and finding nothing that they want. Firms in markets where sales are less likely charge more, compensating for the empty tables.
We then drop this friction into an otherwise entirely standard model of the economy – no sticky prices, no market power, nothing exotic – and estimate it on US data using Bayesian methods.
How much of the business cycle is explained by this?
In our baseline estimates, shocks to the desire to search account for 38% of the variance of GDP, 39% of the variance of measured productivity and 28% of the variation in consumption. Across the specifications we try, the range is 39-60% of the movement in output and measured productivity, and up to half the movement in consumption.
When we do not force the model to match observed shopping time – taking a broader view of what search effort means – the figure rises to around 60%.
These findings survive the obvious objections. They hold when we let firms vary how intensively they use their capital, and when we let households and firms hold inventories and durable goods.
Why these findings matter
Three things follow from our analysis. First, a substantial part of what economists have been recording as technology shocks may just be ‘demand in disguise’. That is not a small bookkeeping matter: entire bodies of macroeconomic research rest on the size and timing of technology shocks, and policy conclusions are drawn from them.
Second, the result rescues an old Keynesian idea – that consumer demand has real effects – without any of the machinery usually thought necessary to deliver it. There are no sticky prices in our model. Prices are perfectly flexible, markets clear and the outcome is efficient. Demand matters anyway, because matching takes effort.
Third, it changes how economists should read a productivity slowdown. If measured productivity falls partly because customers have stopped looking, then the remedy is not the same as it would be if the economy had genuinely forgotten how to produce. Policies that support demand may raise measured productivity, rather than merely raising output at productivity’s expense.
The mechanism is also more general than shopping. When an economy is slack, buyers can be choosy and get exactly the specification they want. When it is tight, they compromise. Some of what looks like firms becoming more productive in a boom is really households accepting ‘less good’ matches – and doing more of the work of finding them.
