Want More Accurate Economic Forecasts? Consumer Surveys May Hold the Key.

When forming inflation expectations, respondents were most likely to rely on their own shopping experiences, or on what their friends and family were telling them about their shopping experiences, the researchers learned. National data sources, such as traditional media outlets that publish reports about inflation, were far less influential.

When inflation spiked in spring 2023, D’Acunto, Karlsruhe Institute of Technology’s Daniel Hoang, the Bank of Finland’s Maritta Paloviita, and Weber reran a similar, smaller survey in Finland and obtained similar results. They also found some patterns in what consumers said. For example, cognitive ability may affect expectations. Among the male-only survey participants (who were members of the Finnish Defense Forces, which conscripts nearly all men in Finland), there were widespread forecasting errors about the future path of inflation, but the magnitude of missing the mark was smallest among men with the highest IQ, the results demonstrate.

There may also be a gender-based nuance to inflation expectations, according to separate research by D’Acunto, Malmendier, and Weber, and it points back to people forming opinions on the basis of their lived experience. From surveying both male and female heads of the same US household, the researchers find that women tend to have persistently higher inflation expectations than men. The gap in their data was about 40 percent higher in households where men never did the grocery shopping, but it disappeared in households where men at least occasionally made grocery runs. This suggests it’s not necessarily financial literacy or business-media appetite shaping opinions but, rather, personal experience.

To measure inflation, the US Bureau of Labor Statistics compiles prices for what it describes as a “representative basket of consumer goods and services” that reflect “inflation as experienced by consumers in their day-to-day-living expenses.” In reality, there’s variation in what people buy, and that variation may be meaningful. In that same survey of heads of households, when respondents were asked what shaped their expectations, people who went grocery shopping mentioned milk, bread, and eggs frequently. But men were substantially more likely than women to refer to the price of gasoline. The most-used measure of inflation, the core Consumer Price Index, excludes both grocery and gas prices because of their volatility. Thus, this broad CPI measure could be less helpful than essentially a household-level version. If the CPI says the overall inflation rate is 3 percent but many consumers are experiencing something closer to 7 percent, any forecast that uses the CPI will be off the mark.

Complicating things further, people may also react most to the prices of goods they buy frequently, suggests the research by D’Acunto, Malmendier, Ospina, and Weber—even where they’re shopping will affect their views about inflation, says Weber. “The price trends you see at Whole Foods versus Trader Joe’s can be quite different,” he says.

Expectations can also be affected by faulty personal memories, D’Acunto and Weber find in a field experiment. They asked nearly 40,000 US residents enrolled in the NielsenIQ Consumer Panel where they thought inflation would land in the next 12 months, also asking people what they paid for milk recently and what they thought it cost a year prior.

The researchers compared the recollections with the earlier amount the survey participants actually paid and find that respondents typically said the year-ago price was lower than it was—and felt that the price had risen more sharply than it had. They also reported higher expectations for what milk would cost in a year’s time and expected higher overall inflation in turn.

Economists tend to reject consumer survey data as being full of noise. Many respondents give answers that are so different from others, and so far from rational after-the-fact realizations, that they don’t seem meaningful. But another explanation is that people are submitting survey responses that are shaped by their own abilities, circumstances, and sometimes-flawed memories. What’s more, they’re also shaped by behavioral biases: People tend to pay more attention to price increases than to decreases, find D’Acunto, Malmendier, Ospina, and Weber. When milk prices rise from $2.49 a gallon to $2.99, and then fall back to $2.49, the increase weighs more heavily on people when they form expectations.

When you take all this into consideration, D’Acunto and Weber argue, survey data look less random and actually make some sense.

Taking the new approach for a spin

Could survey expectations, in context, help forecasters predict real-world behavior? Various research projects have been tying people’s expectations about a recession, or housing prices, or indeed inflation, to their decisions about investment, spending, and employment. For one example, in a study of German consumers reacting to a 2005 announcement that the government planned to increase its value-added tax (similar to a US sales tax), D’Acunto, KIT’s Hoang, and Weber tie expectations of rising prices to a higher willingness to purchase big-ticket items such as furniture and electronics. (Read more in the Spring 2021 story “How Central Bankers Misjudge Forward Guidance.”) With a back-of-the-envelope calculation, the researchers estimate that the announcement caused consumption of those big-ticket items to grow by about 10 percent in the year after the policy change was announced.

University of Texas’s Olivier Coibion, University of California at Berkeley’s Yuriy Gorodnichenko, and Weber find similar evidence for US consumers in the NielsenIQ Consumer Panel Data. When people are guided by survey prompts to raise their inflation expectations, they say they will increase their subsequent spending—and then do.

Someone’s willingness to purchase or borrow is still a step from the actual action they take. But D’Acunto, Weber, and University College London’s Xiao Yin more recently conducted a field experiment that connected the dots between what consumers predict will happen and real-life borrowing decisions.

The team collaborated with a leading Chinese bank between 2020 and 2023 to survey more than 10,000 clients periodically about their projections for income and borrowing in the coming six months. Access to the participants’ bank-account activity enabled the researchers to analyze how an unexpected income change (from a promotion or a job loss, say) affected a consumer’s outlook. They were able to track credit card spending and other debt obligations through the bank accounts, and they incorporated consumer borrowing data from a national database that tracks all individual borrowing. If a person suddenly earned more or less than anticipated, the researchers could see not only how that affected their spending and borrowing but how they responded to the next survey.

Consumers seem to be overly optimistic, according to the research. Both average and median income expectations for six months ahead exceeded the income participants actually had at the end of that time.

The researchers estimate that for every $1 unexpected positive change in income, the average participant overestimated their income for the next six months by 40 cents. (The researchers converted all currency into US dollars.) Participants whose income fell also miscalculated, expecting their income six months out to be lower than it ended up being. The bigger the shock, the more participants overshot their expectations for what would happen.

Overconfidence is a common behavioral bias, but it isn’t typically part of classic forecasting models. These results suggest it should be, argue the researchers. Overestimating future income affected borrowing: On average, for every dollar of income that was expected but didn’t materialize within six months, unsecured debt increased by 7 cents. Given that people borrowed on the basis of an income boost that didn’t materialize, it isn’t surprising that the researchers estimate a significant increase in the probability of default on those loans.

These patterns could inform macro models of consumption patterns, write D’Acunto, Weber, and Yin. They built a model incorporating their newfound variable of misguided income expectations, finding that the resulting patterns for spending, borrowing, and defaults tracked those in the actual banking, credit, and survey data. A positive income shift raised expectations for more income growth in the future, which led to increased borrowing. When the researchers ran the same model without the subjective expectations data—relying instead on assumptions grounded in classical, rational economic theory—those patterns didn’t emerge.

They also compared their model to a well-established narrative that describes how financial crises happen: Overreaction to good news causes consumers to increase borrowing too much. Similarly, people overreact to bad news by tamping down on borrowing more than necessary. D’Acunto, Weber, and Yin find that their model generated a similar cycle. Income expectations, borrowing amounts, and default probabilities all increased when the model considered consumers’ subjective income expectations—but didn’t when the model substituted in “rational” economic expectations.

How might this sort of approach work in the real world? As a proof-of-concept exercise, the researchers applied their theory to the 2008–09 financial crisis. They find that income jolts spurred overconfident expectations and, in turn, a big jump in borrowing in the years leading up to 2008. They were able to closely model this borrowing spike, but when they manufactured a parallel universe in which there was no role for subjective income expectations to affect consumer borrowing choice (as in rational economic theory), debt levels and defaults didn’t skyrocket. Rather, they stayed in line with what they were in 2004.

Given the postcrisis interest in better understanding and anticipating drivers of consumer debt cycles, plus more attention being paid now than was a decade ago to all things inflation-related, this outcome suggests a potential role for consumer expectation surveys. And central banks seem receptive. “Surveys have repeatedly proven their usefulness over the past five years,” the ECB’s de Guindos told attendees of this past fall’s conference on expectations surveys.

But all still rely on theory when accounting for consumer expectations—likely because even though mounting research suggests theoretical models are imperfect, “the academic literature hasn’t fully converged on a view of what you should replace rational expectations with,” says Weber. Until there’s some consensus on that point, central banks may keep asking consumers what they think and producing forecasts that don’t incorporate the answers.