key: cord-0980164-opyuzpve
authors: Furman, Jason
title: Why Did (Almost) No One See the Inflation Coming?
date: 2022-04-08
journal: Intereconomics
DOI: 10.1007/s10272-022-1034-9
sha: d74148475f1d7811f25b2eb3db4ad153937fcb2c
doc_id: 980164
cord_uid: opyuzpve

To understand the possible trajectory of inflation in 2022 and beyond, it is helpful to understand why the United States and Europe had so much inflation in 2021.

The United States and Europe are both experiencing the fastest infl ation in a generation. The infl ation in both economies was not foreseen by the standard economic models used by offi cial and private sector forecasters. This failure should lead to some reassessment of the models and should increase uncertainty and concern about the trajectory of infl ation going forward. In particular, policymakers should not rely on statistical relationships that held in the decades before the pandemic when making predictions in today's very diff erent environment. This situation calls for both rethinking the underlying economics -for example, infl ation will play a more salient role in setting wages and prices at its faster pace so wage-price passthrough could be higher as well -and widening confi dence intervals to refl ect the greater uncertainty.

Specifi cally, the linear Phillips curve with anchored expectations failed to predict the infl ation of 2021 because, by construction, that Phillips curve can essentially never predict high infl ation. Even with a massive fi scal stimulus that cut the unemployment rate to the likely impossibly low level of 1%, the infl ation rate would still be predicted to remain below 3%. An alternative model, in which fi scal stimulus predicts nominal (not real) demand, real output can rise but not above its short-run potential, and infl ation is the diff erence between the two, does a much better job of making sense of the extraordinary infl ation in 2021 by dispensing with the labor market intermediation.

Despite the shared underestimation of infl ation, the specifi c situations diff er on the two sides of the Atlantic with infl ation running considerably higher in the United States than in Europe and the GDP recovery conversely further behind in Europe. Policy in Europe should avoid the trap of being too driven by developments and news in the United States.

Forum such judgmental adjustments, however, the models still would not have predicted much infl ation -because no amount of unemployment rate reduction can generate much infl ation from these models.

nomic Advisers (2009) . In the case of the normal multipliers, this results in the economically absurd forecast of a 1.1% unemployment rate in the fi rst quarter of 2021, a sign that something is wrong with this methodology -a topic I will return to. 2 Even the implausibly low unemployment rate would not have been expected to translate into much infl ation using a conventional approach. Ball et al. (2021) , for example, estimate that the Phillips curve has a slope of -0.17 -i.e. each one percentage point reduction in the unemployment rate boosts the infl ation rate by 0.17 percentage points. Figure 3 shows predicted infl ation based on this Phillips curve. (Note, a Phillips curve approach cannot explain the U.S.-euro area infl ation diff erential and in fact would predict higher infl ation in Europe because employment was higher there.)

In summary, forecasters using major models should have been nervous that absent any judgmental adjustments, their models were forecasting GDP well above potential and implausibly low unemployment rates. Regardless of 2 In reality, no forecasters predicted an unemployment rate like this and in fact virtually all of them expected the unemployment rate to still be above the pre-COVID-19 rate by the end of 2021. Most forecasters, however, did expect GDP to be above its pre-pandemic trend by the end of 2021. The diff erence was bridged by implicit or explicit forecasts of a temporarily very large increase in productivity growth -the residual between GDP growth and employment growth. IHS Markit's forecast for 2021Q4 GDP, for example, increased by 5.4% between December 2019 and June 2021. 

The net eff ect is ambiguous but evidence from earlier in the pandemic suggests this was more plausibly negative than positive because, fi rst, the initial wave of COVID-19 in 2020 reduced infl ation. Second, the timing of the infl ation generally followed the reopening of the economy, rising fi rst in the United States when its economy was reopening and then later in the euro area when its economy reopened somewhat later. This suggests that the net eff ect of the coronavirus pandemic is to suppress infl ation and that infl ation would have been even higher without the Delta and Omicron variants. Even if not correct, it is unlikely that the Delta and Omicron variants had a large positive eff ect.

The shift from services to goods. Another candidate for the error term is what could be described as a taste shock: For example, people felt unsafe in the gym so, instead of paying gym memberships, they bought exercise bicycles. To the degree that the supply of goods is more inelastic than the supply of services, this would increase infl ation. There are two issues with this theory, however. First, the increase in goods spending seems more a consequence of the overall level of demand than a taste shift -goods spending was considerably higher in the spring of 2021, as COVID-19 case numbers were low and falling, than it was in the winter of 2020-21, when case numbers were high and rising. Moreover, both goods and services spending was higher in the United States than in Europe (although service spending in the two economies had largely converged by the end of 2021). This suggests that it was the economic impact payments and other fi scal support that drove goods spending not a taste shift. Second, while it is plausible that the supply curve for goods is more inelastic than the supply curve for services, there still would have been some additional services infl ation if there had been less of a shift from goods to services. As a result, the goods-services shift is at most part of the story of the error term.

There is no doubt that supply chain disruptions explain some of the increase in infl ation, most notably in microchips and the dynamics of rental fl eet purchases and sales of used vehicles. These drove the spectacular increase in motor vehicle and parts prices that contributed 1.1 percentage points to core personal consumption expenditure (PCE) infl ation in 2021. But a lot of the so-called supply chain issues are really large increases in demand coming up against supply that was relatively inelastic. The result was a combination of higher prices and higher quantities. U.S. ports, for example, were not disrupted and, in fact, were processing 18% more than in 2019. This, however, was not enough to keep up with demand -as a result, both prices and quantities increased.

The "great resignation" as labor force participation remains low. This is a U.S.-specifi c factor of low labor force par-Where did all the infl ation come from if not from a linear Phillips curve?

The general Phillips curve is:

Infl ation = expected infl ation -θ*(unemployment -natural rate of unemployment) + error term

The discussion of the increase in infl ation is organized around the diff erent terms of this equation.

A positive error term: Supply shocks and COVID-19 taste changes One possibility is that the infl ation we have seen refl ects the error term -unforeseeable events that happened essentially outside the economic model and did not transmit to infl ation through aggregate demand or the labor market. It is likely that is part of the story but probably only part of the story. Some of the main candidates for the error term include:

The emergence of the Delta and Omicron variants of . Slowing the reopening of the economy is commonly cited as a reason that infl ation was higher than expected in the second half of 2021. But the rapid reopening of the economy as people were vaccinated in the fi rst half of the year was also cited as a reason for rapid infl ation then. While it is possible both arguments were true, it seems unlikely. The resurgence of COVID-19 likely raised durable goods prices but lowered service and gasoline prices. Dec. 2020 IHS baseline Actual Low multipliers due to NPIs Normal multipliers slack. But in 2021, job openings and the quits rate both soared, suggesting a much tighter labor market than indicated by the unemployment rate, especially earlier in the year. Some evidence suggests that the ratio of unemployed to job openings and quits are both better predictors of infl ation than the unemployment rate; using them would add more to the infl ation prediction (Furman and Powell, 2021; Domash and Summers, 2022) . 3

The fi nal possible explanation for the recent infl ation, sticking with the linear Phillips curve model described at the start of this section, is that expected infl ation increased. This is also not a satisfying explanation because any increases in expected infl ation mostly followed the price and wage increases instead of preceding them and the increases in expectations were relatively small. At the end of 2020, fi nancial market expectations for infl ation over the next fi ve years were very low, they rose sharply starting in December 2020 but settled only modestly above their normal level (see Figure 4) . Consumers did increase their near-term infl ation expectations sharply starting in January 2021. And, as shown in Table 1 , forecasters were well behind infl ation all year. Moreover, most forecasting models incorporate long-run, not short-run, infl ation expectations and those 3 In the prediction models reported by Furman and Powell (2021) , they fi nd the adjusted R 2 in predicting core CPI is 0.47 for quits as an explanatory variable, 0.45 for the ratio of unemployed to job openings, 0.35 for the unemployment rate and 0.22 for the prime-age (25-54) employment rate.

ticipation. It may have played a role in increasing infl ation by reducing supply. But it also decreased demand so the net eff ect on infl ation is unclear and unlikely to be very large. Also, the eff ects of the great resignation on infl ation depend on interactions with other fi scal support. It is possible that this could have had more of a role in the early part of 2021 since many people who were not working could get unemployment insurance suffi cient to maintain their consumption until September 2021. Now, however, people returning to work are likely to increase both supply and demand.

A steeper Phillips curve or tighter labor markets Part of the disconnect between predicted and actual infl ation may be related to the slack term in the Phillips curve. Some plausible modifi cations could add at most about one percentage point to infl ation, bridging part of the gap, but not all of it.

A steeper Phillips curve. It is possible that low unemployment translates into a larger increase in infl ation than the 0.17 percentage point assumed above based on Ball et al. (2021) . There are a number of diffi culties in estimating the slope of the Phillips curve and more eff ective monetary policy and anchored infl ation expectations can create the statistical illusion that the Phillips curve is fl atter than it seems. Hazell et al. (2022) get around these issues by using state-level data on the relationship between unemployment and infl ation fi nding a coeffi cient that is closer to 0.3. Even this, however, cannot generate much infl ation, certainly nothing like the roughly 5% infl ation the United States has been experiencing.

A temporarily higher natural rate or a speed limit. Even if the natural rate of unemployment was 3.5% in the run-up to the pandemic, it was likely higher during 2021, especially the fi rst half of the year as it takes time for people to connect to jobs, hysteresis temporarily raises the natural rate, the pandemic itself temporarily disrupted people from taking jobs, and unemployment insurance reduced the willingness of people to take jobs. Alternatively, it is possible that there is a "speed limit" of how fast employment can improve without triggering infl ation (which could be modelled as a temporarily higher natural rate that only falls slowly; Turner, 1995) . These changes, however, could not add much to infl ation because they are limited by the relative fl atness of the Phillips curve itself -even a 5% natural rate combined with a 0.3 slope of the Phillips curve would generate less than an additional 0.5 percentage point of infl ation.

Alternative measures of slack showed a tighter labor market. The unemployment rate is the standard measure of Forum terms -asking questions like, "If you give people $100, how much do they increase their nominal spending?" (e.g. Parker et al., 2013; Sahm et al., 2012) . This suggests a simple three-step framework for thinking about infl ation in 2021:

1. Use multipliers to predict nominal GDP.

2. Use the productive capacity of the economy adjusted down for the eff ects of the pandemic to predict real GDP. That is, assume that there is a limit on the amount that fi scal support can increase real production.

3. Price increases are the residual.

What this approach means for U.S. infl ation in 2021 can be discerned from looking at the prediction the standard multiplier models had for output relative to potential (see Figure 5 ), which is just another way of showing the results of the multiplier exercise reported in Figure 1 . This shows that output was projected to be about 1% to 4% above the pre-pandemic projection of potential in 2021Q4, depending on the multipliers. Even hitting the pre-pandemic projection for potential would have been hard given that the population was smaller due to reduced immigration and excess deaths, the capital stock was smaller due to foregone investment, the COVID-19 pandemic was still disrupting production, and U.S. income support policies like unemployment insurance and stimulus checks caused sustained reductions in labor supply. On the other remained well anchored all year (Reifschneider and Wilcox, 2022) .

Looking at reasonable ranges for the parameters of the linear Phillips curve above, it is possible to make changes that generate some additional infl ation but they do not plausibly account for all of the infl ation in 2021. Moreover, many of these changes are ad hoc and may not actually be right. And none of them provide a particularly satisfying explanation.

To understand the limits of the linear Phillips curve for this situation, consider a much more extreme policy. Imagine that households were each given $100,000 in 2019. An economist using a linear Phillips curve would not predict very much infl ation because the policy could not lower the unemployment rate below 0% and so the tight labor market (the only way the Phillips curve incorporates demand) would not add much more than one percentage point to the infl ation rate. But clearly a forecast that this policy would lead to only 3% infl ation is absurd.

The better way to think about this thought experimentand the less extreme actual policy carried out in 2020 and 2021 -is through a highly nonlinear model. There is some evidence that the Phillips curve itself is nonlinear (e.g. Nalewaik, 2016; Fair, 2021; Forbes et al., 2021) . But a lot of other research has found that any nonlinearities in the Phillips curve are not robust or are unstable and that it is better to work with a linear one (e.g. Marcellino, 2008) . Moreover, even a standard nonlinear Phillips curve would struggle with the fact that we have never before seen core PCE infl ation jump in this way despite the unemployment rate being well within its normal range. Additionally, it is not a particularly satisfying way to generate ex ante predictions -the nonlinear Phillips curve would not be a good way to predict the diff erences in infl ation that would result from increasing the hypothetical helicopter drop from $100,000 to $1 million per household.

A better model is to dispense with the additional step of modelling the impact of demand on the labor market and the labor market on infl ation and instead just go straight from nominal demand to infl ation.

Most of the microeconomic research that has been used to develop fi scal multipliers actually looks at parameters like the marginal propensity to consume in nominal Normal multipliers Low multipliers due to NPIs 2 0 2 1 Q 2 2 0 2 1 Q 3 2 0 2 1 Q 4 2 0 2 2 Q 2 2 0 2 2 Q 3 2 0 2 2 Q 4

In the context of the linear Phillips curve, low levels of labor market slack suggest more infl ation in 2022 than in 2021 but the error term (e.g. supply shocks), which likely added to infl ation could move to zero or even negative in 2022. This would cause infl ation to fall relative to 2021.

The trickiest issue to assess is the infl ation expectations term. Reifschneider and Wilcox (2022) model this term as largely based on professional forecasts of infl ation over the next ten years, which have been stable. But in the current context, short-run infl ation expectations may be more relevant and appear to be becoming embedded in wage and price setting (Furman, 2022) .

All in, core infl ation is likely to be lower in 2022 than it was in 2021. However, with several forces pushing infl ation higher, it may still end up in the 3.5% to 4.5% range, depending on the measure used. Moreover, it is plausible that infl ation in 2023 will exceed infl ation in 2022 if, for example, there is an unusually large one-time decline in goods prices in 2022 due to a glut in the supply of cars.

The European situation is somewhat diff erent from the U.S. one because GDP growth has been weaker and infl ation has not increased as much, with the apparently smaller European fi scal response likely at least partially responsible for the diff erence.

hand, it is plausible that pre-pandemic expectations for potential were about one percentage point too low given that they assumed the natural rate of unemployment was 4.4% while 3.5% was completely plausible (CBO, 2020b). Accounting for these off setting eff ects, assuming the economy was capable of producing at the level of the prepandemic projection of potential was a plausible but likely upper bound assumption -leaving about one to four percentage points of additional infl ation above the baseline expectation for infl ation. 4

What does this mean for the outlook for infl ation?

Modelling nominal demand and real supply is less useful for forecasting infl ation in 2022 and 2023. It is almost certainly the case that reduced fi scal and monetary support will slow nominal demand growth. But it is also almost certainly the case that with the economy close to its potential real GDP, growth will slow too.

4 Two caveats are in order. Baseline expectations for infl ation might have been below 2% absent the two rounds of fi scal support in December 2020 and March 2021. IHS Markit, for example, was expecting core PCE to be 1.8% for 2021 Q4/Q4 in its December 2020 baseline. On the other hand, it is possible people had enough excess savings from the transfers in 2020 and the reduced consumption in that year to fi nance an above-normal level of spending when the pandemic receded and that giving them additional money in this context would have a very low marginal propensity to consume. In this case, the multiplier might be towards the low end but the underlying baseline infl ation it would be adding to could be towards the high end.

Real gross domestic product Forum close to its previous trend even amidst a massive economic contraction, but that was nothing compared to the huge increase in disposable personal income in the United States, as shown in Figure 7 . Other aspects of stimulus may also have been larger in the United States, which, for example, had a considerably larger and less targeted grant program for small and medium-sized businesses, called the Paycheck Protection Program, than anything in the major European economies. The result of the increase in disposable personal income in the United States is that U.S. consumption, particularly of goods, greatly outpaced European consumption.

The fl ip side of the larger fi scal support and faster pace of U.S. GDP recovery has been higher infl ation in the United States, as shown in Figure 8 . The United States and Europe have been hit by diff erent supply shocks. The increase in the price of used cars is a bigger deal in the United States, where they are a larger part of the consumption bundle, but Europe has been hit by much larger increases in spot natural gas prices. Europe also had a lower infl ation rate in 2020, in part because of the way that temporary value added tax reductions fed into the infl ation rate, and experienced larger base eff ects as its economy moved towards normalizing in 2021. Overall, U.S. core infl ation is well above its 2% target trend, whereas the core harmonized index of consumer prices (HICP) in Both the United States and the euro area suff ered rapid reductions in GDP when the COVID-19 pandemic hit, followed by partial bouncebacks as restrictions eased. The United States, however, has had a stronger GDP recovery, both in absolute terms and relative to its prepandemic trend, as shown in Figure 6 . This stronger U.S. GDP recovery has materialized despite U.S. employment lagging employment in Europe as the United States has experienced a large withdrawal and only partial return to the workforce. The gap between these is made up for by the increase in average work hours and a temporary boost in productivity in the United States relative to Europe.

It is diffi cult to make a meaningful comparison of the size of the U.S. and European fi scal responses because of differences in how fi scal stimulus is described and measured. Ex post defi cits and debt are also of limited use, especially when, for example, Germany's defi cit numbers appear to refl ect macroeconomically unmeaningful charges that increase the defi cit and debt in 2021 to make it easier to satisfy the debt brake in future years. The degree to which the U.S. fi scal support was considerably larger than European fi scal support can be seen by comparing the trajectory of disposable personal income in the major economies. Germany and France successfully protected disposable personal income, keeping it

Disposable personal income relative to trend Forum the European Central Bank to be more patient in tightening monetary policy, giving the economy more room to recover and more cushion against spillover from the Russian invasion.

While getting infl ation under control and keeping expectations anchored is critical in both economies, central bankers also need to be thinking about changing the infl ation target itself. Given the decline in equilibrium interest rates, a higher target, like 3%, would give more room for policymakers than the current 2% one. It is possible that the current moment will turn into an opportunity to achieve this new target. But even keeping infl ation to 3%, especially in the United States, will be a challenge.

the euro area still falls slightly short of 2% annual growth since the start of the pandemic.

Predicting infl ation is hard, understanding what to do about it is even harder. The Federal Open Market Committee's expectations for its own interest rate path are much more moderate than even a very dovish version of a Taylor-type rule would imply, as shown in Figure 9 . This may well be the appropriate expectation for policy given the many uncertainties in the real economy and fi nancial market reactions, the rapidly diminishing fi scal support for the economy and the desire to avoid risking a recession. But it is very far from the way policy has ever been conducted before.

Europe is closer to its infl ation target and further away from its output target. Moreover, Europe faces a potentially much more serious economic impact from the Russian invasion of Ukraine. As a result, it makes sense for

Federal funds rate and policy rules in %

Notes: The fi gure shows three versions of a Taylor-style rule. The "Taylor rule" uses the Taylor (1993) rule's weight of 0.5 on infl ation and 0.5 on the output gap, a natural real federal funds rate of 0.5% and a natural rate of unemployment of 4.0%. The "Dovish balanced rule" raises the weight on the output gap to 1.0, lowers the natural real federal funds rate to 0.0% and the natural rate of unemployment to 3.5%. The "Inertial dovish balanced rule" is the same as the "Dovish balanced rule", only it places 20% weight on this formula and 80% weight on the value of the federal funds rate in the last period. 

Measuring U.S. Core Infl ation: The Stress Test of COVID-19, NBER Working Paper, 29609. Congressional Budget Offi ce (2020a), Key Methods That CBO Used to Estimate the Eff ects of Pandemic-Related Legislation on Output

Estimates of Job Creation from the American Recovery and Reinvestment Act of 2009, US Government Printing Offi ce

The Economic Impact of the American Recovery and Reinvestment Act Five Years Later

How Tight are U.S. Labor Markets?

What Do Price Equations Say About Future Infl ation?

Low Infl ation Bends the Phillips Curve around the World, NBER Working Paper

Will Anchored Infl ation Expectations Actually Anchor Infl ation?, RealTime Economic Issues Watch, Peterson Institute for International Economics

What is the Best Measure of Labor Market Tightness?, RealTime Economic Issues Watch, Peterson Institute for International Economics

The Slope of the Phillips Curve: Evidence from U.S. States, NBER Working Paper

A Linear Benchmark for Forecasting GDP Growth and Infl ation?

Non-Linear Phillips Curves with Infl ation Regime Switching

Consumer Spending and the Economic Stimulus Payments of

The Case for a Cautiously Optimistic Outlook for US Infl ation

Check in the Mail or More in the Paycheck: Does the Eff ectiveness of Fiscal Stimulus Depend on How It Is Delivered

Discretion versus Policy Rules in Practice

Speed Limit and Asymmetric Infl ation Eff ects from the Output Gap in the Major Seven Economies