For most of a year the word was everywhere. Transitory. It was in the briefings and the broadcasts and the reassurances, and it meant that the thing you were watching would pass, that the price of the weekly shop would settle, that you did not need to change your plans. Then the plans needed changing. The shop cost more, and then more again, and the raise that was meant to be a raise turned out to be a way of standing still. Somewhere in there a suspicion hardened into a question. If the people whose job is to see this coming did not see it coming, why listen to any of them? And since this is the last chapter of a site that has spent fifty-two chapters telling you what the evidence says, the question turns on the site too. Why believe this?
The way to answer is to grade the profession the way the rest of this site grades a claim: find its record, and read it. That record is not one thing. It is two things that get confused because the same people produce both, and keeping them apart is the whole of the answer. One of them is close to an unbroken record of failure. The other is the reason the first one is worth knowing about at all.
The record on the thing everyone wants
What everyone wants from an economist is a forecast: tell me what happens next, so I can act before it does. On that, the record is close to spotless, in the wrong direction. One American archive lets anyone run the test. Since 1968 the Survey of Professional Forecasters, run now by the Federal Reserve Bank of Philadelphia, has asked a panel of forecasters every quarter where they think output is going. The forecasts and the outcomes are both published, which means the survey can be marked against what actually happened. Figure 10.1 marks it.
The 2008 case is worth walking through, because it is the one people mean when they say nobody saw it coming. In the survey taken in August of that year, weeks before the collapse of Lehman Brothers, the panel’s median forecast had the economy growing at an annualized 0.71 percent in the fourth quarter, and kept it growing through the third quarter of 2009, as far ahead as the survey reached. The same survey put the probability of output actually falling in that quarter at 46.57 percent, which is to say the panel could see the risk and still, on average, forecast growth. By the November survey, some eight weeks after Lehman, the nowcast for the quarter then underway had turned to an annualized −2.94 percent. What that quarter turned out to be depends on when you ask. The first official estimate, published early in 2009, was an annualized −3.80 percent; the latest revision puts it at −8.47 percent. The gap between those two numbers is not a rounding error. It is more than four percentage points, and it is a reminder that even the record you grade the forecast against keeps moving under you, which is why the plate above fixes on a single vintage and says which one.
None of this is a peculiarly American failing, and it is not new. Prakash Loungani, working at the International Monetary Fund, put the finding in one sentence in 2001, after checking consensus forecasts against outcomes across dozens of countries: “the record of failure to predict recessions is virtually unblemished.” His plain-English gloss was blunter. “How well do forecasters predict recessions? The simple answer is: ‘Not very well.’” Of the 60 recessions in his sample, only two had been foreseen a year in advance. A larger study in 2018, with An and Jalles, ran the test again across 63 countries and the years 1992 to 2014: of 153 recessions, 148 were still unpredicted in the April of the year before they arrived.
Two things keep that from being the whole story, and both belong in the account. The first is that “predict” here has a strict meaning. It counts a recession as foreseen only if the consensus, the average across forecasters, actually turned negative, and an average of many forecasts is arithmetically slow to go negative even when individual forecasters are placing real weight on a downturn. Loungani found that forecasters did come round, but slowly: by April of the recession year itself a downturn was forecast in only about a third of the cases; by October of that year a recession was called in the vast majority of them, and in 80 percent of the cases that October forecast still understated how bad it would be. The failure is specifically about seeing it early and sizing it right, and not about total blindness. The second is the mirror image. Forecasters are not simply pessimists who cry wolf and occasionally happen to be right. In the 2018 study, at the same year-ahead vantage, they raised a false alarm, forecasting a recession that did not come, in only 8 of 1,153 calm years. They do not predict recessions that happen, and they do not invent ones that do not. What they produce, almost always, is a forecast of ordinary times.
There is a reason the specific failure is recessions, rather than the economy in general, and it is not that forecasters are lazy. Part of it is visible in the chapter on boom and bust: the recessions that come from a credit cycle turning over hinge on when confidence breaks, and that turn is a nonlinear event with no schedule. That a build-up of credit runs ahead of such crises is written in more than a century of records; the date the music stops is not written anywhere in advance. The forecasters are failing at the timing, the hard part, and holding that apart from what the record does document is the point of this chapter.
The word that made the case
The charge people felt most directly is more recent than 2008, because they lived through it, and it deserves to be quoted rather than paraphrased. Through 2021, as prices climbed, the word from the central bank was that the rise would not last. The Federal Reserve’s rate-setting committee wrote in its statement of 28 April 2021 that “inflation has risen, largely reflecting transitory factors.” It was still saying so on 3 November, when inflation was “elevated, largely reflecting factors that are expected to be transitory.” In August, at the Jackson Hole symposium, the Fed’s chair Jerome Powell had set out the reasoning at length, judging that the high readings were “likely to prove temporary” and that longer-term inflation expectations suggested households, businesses, and market participants also believed “current high inflation readings are likely to prove transitory.”
They were not. The consumer price index rose 9.1 percent over the twelve months ending June 2022, which the Bureau of Labor Statistics recorded as “the largest 12-month increase since the period ending November 1981.” And the record shows the call being withdrawn by the people who made it. The Fed’s statement of 15 December 2021 dropped the word entirely. Two weeks earlier, testifying to the Senate on 30 November, Powell had said it out loud: “I think it is probably a good time to retire that word and try to explain more clearly what we mean.” The Treasury secretary, Janet Yellen, who had made the same call, folded her own into it before the Senate the following June: “both of us probably could have used a better term than ‘transitory.’” That is what a wrong forecast looks like when it is documented properly. A dated claim, a dated outcome that contradicted it, and dated retractions, all in the issuers’ own words. It is also, precisely, a forecast: a statement about what would happen next, of exactly the kind the record says the profession is worst at.
What the same discipline is good at
If the chapter stopped here it would have earned the reader’s distrust and misled them all the same, because forecasting is not the whole of what economists do, and it is the part they are worst at. The rest of the discipline is in the business of a different and more answerable question: what causes what. Did this policy have that effect? Does this price move that quantity? Those questions can be settled by evidence in a way a forecast cannot, and there are tools built for settling them. Figure 10.2 lays out the main ones and, next to each, the thing it cannot reach.
The reason this side of the ledger holds up better is not that its practitioners are cleverer or more honest than the forecasters, who are frequently the same people. It is that the questions are of a different kind. A forecast is a claim about a future that has not happened yet; by the time it can be checked, the moment it was needed for has passed. A causal question is about a record that already exists, and the job is to find a reading of it that a critic cannot easily dismiss, knowing that the comparison the reader really wants, the same world with and without the policy, is never fully on offer. When economists do that well, the results hold up, get replicated, and quietly become the things a chapter can point to: that a modest minimum-wage rise at the levels actually tried does not empty the shops of jobs, and that a build-up of credit runs ahead of the big busts even when its timing is anyone’s guess. Knowledge of this kind is about what is rather than what is next, and it is the part of the discipline that holds.
Economists are always wrong, so you can ignore them.
Oversimplified Moderate confidence
The claim has a true kernel, which is what keeps this ruling off backwards. On the thing the public most wants from economists, forecasting, the record really is close to a clean sheet of failure. Forecasters almost never call a recession a year out; in one study they missed 148 of 153, and in 2021 the central bank spent months calling an inflation “transitory” that reached 9.1 percent before it turned. Anyone who insists the experts have a crystal ball is answered by the plate above. Two words break the claim, though. The first is “always.” The same discipline that cannot time a recession has documented the build-up of credit that runs ahead of its financial crises, and establishes cause and effect in narrower questions well enough that the results replicate; and forecasters are not even reliably wrong in a useful direction: at the 2018 study’s year-ahead vantage they raised 8 false alarms in 1,153 calm years, against 5 of 153 real recessions called. The second is “ignore.” A profession that is bad at forecasting and good at mechanisms is not one to ignore. It is one to use for the second thing and not the first, which is a more demanding instruction than dismissal. What the record supports is narrower than the claim: the experts are worth listening to about how the economy works, and worth discounting heavily about what it will do next month, and telling those two apart is the actual skill.
Sources
- The forecasting record: Federal Reserve Bank of Philadelphia, Survey of Professional Forecasters, median real GDP forecasts against realized growth, 1968–present (of 226 surveys with a four-quarter forecast, the median was negative twice; of 23 realized negative windows, 22 began with a positive forecast). Prakash Loungani, “How Accurate Are Private Sector Forecasts?” International Journal of Forecasting 17(3), 2001 — “the record of failure to predict recessions is virtually unblemished”; two of 60 recessions foreseen a year ahead. Zidong An, João Tovar Jalles and Prakash Loungani, “How Well Do Economists Forecast Recessions?” IMF Working Paper 18/39 (2018) — 148 of 153 recessions missed a year ahead (the April of the year before), and only 8 false alarms in 1,153 non-recession years at that same vantage, across 63 countries, 1992–2014.
- The 2008 case: SPF surveys of August and November 2008 (fourth-quarter forecast of +0.71 percent annualized in August, with a 46.57 percent probability of a decline; nowcast of −2.94 percent in November); the fourth quarter’s realized growth revised from −3.80 percent at first estimate to −8.47 percent latest.
- The “transitory” episode, verbatim from the issuers: Federal Open Market Committee statements of 28 April and 3 November 2021 (word present) and 15 December 2021 (word dropped); Jerome Powell, Jackson Hole, 27 August 2021; Powell, Senate testimony, 30 November 2021, “a good time to retire that word”; Janet Yellen, Senate Finance testimony, 7 June 2022, “both of us probably could have used a better term than ‘transitory.’” The realized benchmark: U.S. Bureau of Labor Statistics, CPI-U up 9.1 percent over the 12 months ending June 2022, “the largest 12-month increase since the period ending November 1981.”
- What the discipline does well is a claim about method, not a single study: the natural-experiment and administrative-data work whose results replicate, of which the minimum-wage border-county evidence in the work volume is one worked example, and the documented build-up of credit ahead of financial crises in the chapter on boom and bust is another.
- Confidence is moderate under the rubric, which scores the weaker of evidence directness and construct match. Evidence directness is high: the forecasting record is a public archive recomputed here, and the statements are quoted from their issuers. The binding, weaker leg is construct match. The claim grades a whole profession with two absolutes, “always” and “ignore,” while the evidence is specific, about a particular horizon (recessions a year out), a particular definition of a forecast (the consensus turning negative), and a particular set of causal questions. That the pieces point the same way is clear; that they add up to a verdict on the entire discipline is the part the evidence supports only in shape, which is why the confidence is not high.
How this site is built, and what that is worth
Which brings the question back to the site itself, because the whole point of stating a rule is that it applies to the one stating it. This site does not forecast the economy: not once in fifty-three chapters is there a claim about what it will do next quarter or next year, and that is a design choice, not an oversight. The entire apparatus is built to carry the second kind of knowledge and to refuse the first. Every verdict names a claim, lays out the evidence for and against it, and attaches a confidence that is scored on how directly the evidence bears and how well it matches the thing being claimed. When the evidence is a single reconstructed case, the confidence says so and stays moderate. When the construct the evidence tests is not quite the construct the claim asserts, the confidence names that gap and drops. The method is designed so that its weakest link is visible, and so that a reader who disagrees can see exactly where to push.
That speaks to the reader who started this chapter distrusting economists and wondering whether to extend the distrust here. A profession with this record can be neither believed wholesale nor dismissed wholesale, and the verdict above rules out both. What is left is a discrimination, made claim by claim: which kind of knowledge it rests on, how direct the evidence is, and how well the evidence matches the thing claimed. That discrimination is what a verdict on this site is. It is not a ruling to be taken on trust. It is a worked example of the weighing, laid out so it can be checked and, where it is wrong, corrected. Whether it earns your confidence is not something the site can decide for you. It is the one judgment the whole design hands back.
Where the five volumes have arrived
This is the last chapter of the last book, and the site has spent five volumes doing in the small what this chapter has described in the large. It began with the frameworks, the handful of ideas that recur everywhere, and asked what each one really claims. It turned to history, and watched money and markets and firms become the things they are now, none of them inevitable. It went country by country, starting with the plain difficulty of measuring whether a place is doing well at all, and found the answers turning less on natural wealth than on whether a state can do the ordinary work of taxing and enforcing and keeping a promise. It followed work, and found that what a job pays and whether the raise arrives are outcomes of arrangements built by people, not facts of nature. And it closed here, in the pathologies, among the things that break: the products, the scams, the recoveries that skip a street, the crashes, the currencies that die, the famines, the regimes, the futures sold before they arrive, and the discipline itself.
What runs through all of it is a single habit, which is the habit this chapter has finally named. Take the claim people actually make. Find the record. Say what the record supports and, just as plainly, what it does not, and how sure it is possible to be. The economy is not a thing that happens to you, and it is not a set of predictions to be trusted or mocked. It is a set of arrangements, most of them built rather than given, and this site has read them by holding the strong evidence and the weak evidence in separate hands. That is the whole of the method, and it is the whole of what the site has to offer. What you make of any particular question now, with the record in front of you, is yours to decide.