Goldman Sachs Research
US Economics Analyst
Back to the Future: Forecasting Inflation When Expectations Are Evolving (Hill/Walker/Abecasis)
  • Core CPI inflation has continued to surprise to the upside on net, with core CPI beats in each of the last two months and four of the last six reports. Additionally, the magnitude of these upside surprises is beginning to rival those of the 1960s overheating and 1970s stagflation. This raises the question of whether forecasters have simply been surprised by a strikingly long chain of unforeseen supply-side problems, or whether consensus inflation forecasts are missing something important.

  • Analyzing consensus inflation surprises over the last 75 years using the Livingston survey, we find that forecasters on average failed to incorporate the signal from short-term consumer inflation expectations, business inflation expectations, and the jobs-workers gap. This suggests that today’s models might also be underestimating the significance of these factors.

  • We test these variables in our component-level inflation models, finding significance in nearly half of the categories. Our models also continue to include (1) measures of labor and product market slack; (2) indicators related to pass-through from commodity, import, and asset prices; and (3) alternative price indices available from non-government sources.

  • We are raising our year-end core PCE inflation forecasts for 2022-2024 to 4.2%, 2.5%, and 2.3% respectively (vs. 4.0%, 2.3%, and 2.2% previously, and compared to 4.7% in May). This reflects additional upward pressure in cyclical services categories like household maintenance, personal care, and food services and accommodation. However, we are leaving our 2023 and 2024 shelter inflation forecasts unchanged despite the May price spike because alternative rent measures are already slowing and the May geographic cross-section is consistent with a one-off boost from the reversal of pandemic discounts.

  • We also introduce a new bottom-up core CPI model, which we will update alongside our PCE forecasts in our Monthly Inflation Monitor. Unlike the PCE measure, we expect year-on-year core CPI inflation to reaccelerate in late summer—from 6.0% in May to 6.3% in September—before falling to 5.5% at year-end, 2.4% in December 2023, and 2.6% in December 2024 (vs. 5.1%, 2.6%, and 2.6% previously).

  • As we highlighted last fall, the CPI-PCE gap will be much larger than usual this year because of the larger weights on the shelter and durable goods categories and differences in the measurement of health care inflation. However, we continue to expect the gap to narrow beyond this fall, when durable goods prices should start to normalize and the CPI’s health insurance category should transition from very strong to very weak as new annual source data are incorporated.

Back to the Future: Forecasting Inflation When Expectations Are Evolving

Core CPI inflation has continued to surprise to the upside this year on net, with core CPI beats in each of the last two months and four of the last six reports. The magnitude of these surprises is beginning to rival those of the 1960s overheating and 1970s stagflation. In Exhibit 1, we plot the trend in core CPI against our bottom-up inflation model and against the median estimate from the semi-annual Livingston survey, which is perhaps the longest running measure available.[1]

Exhibit 1: Large and Persistent Core Inflation Surprises Reminiscent of the 1960s Overheating and 1970s Stagflation

1. Large and Persistent Core Inflation Surprises Reminiscent of the 1960s Overheating and 1970s Stagflation. Data available on request.
Source: Department of Labor, Federal Reserve, Goldman Sachs Global Investment Research
Unlike in 2021, the broad-based nature of today’s elevated inflation is difficult to fully ascribe to microchip shortages or reopening effects, though those factors continue to be relevant (see here and here). As shown by the blue dots in Exhibit 2, our Effective Lockdown Index tracked many of the inflections in core prices between the 2020 lockdowns and the return-to-office period last fall. But as shown by the red dots in the same exhibit, this positive relationship disappeared in Q4, and the average pace of monthly core inflation actually moved higher.

Exhibit 2: Reopening Contributed to Higher Prices Last Year but Can’t Explain Most of the Recent Strength

2. Reopening Contributed to Higher Prices Last Year but Can’t Explain Most of the Recent Strength. Data available on request.
Source: Department of Commerce, Goldman Sachs Global Investment Research
This raises the question of whether forecasters have simply been surprised by a strikingly long chain of unforeseen supply-side problems, or whether consensus inflation forecasts are missing something important.

What to Expect When You're Expecting High Inflation

Long-term inflation expectations do not currently argue for sustained upward pressure. The June reading of the University of Michigan survey edged higher but only towards the top of its range from 1995-2019, which was an era of low and stable core inflation. Our composite measures of long-term inflation expectations also remain relatively range bound, as shown by the dotted lines in Exhibit 3.
In contrast, short-term inflation expectations have returned to 1980s levels across both consumers and businesses, as shown by the solid lines in the same exhibit. Relatedly, the jobs-workers gap reached a record 2% of the population last quarter and will exert additional upward pressure on wages, prices, and possibly inflation expectations.

Exhibit 3: Short-Term Inflation Expectations Are Elevated, but Long-Term Expectations Have So Far Risen More Modestly

3. Short-Term Inflation Expectations Are Elevated, but Long-Term Expectations Have So Far Risen More Modestly. Data available on request.
Source: University of Michigan, Federal Reserve, Haver Analytics, Goldman Sachs Global Investment Research
While including these measures in our inflation models would improve the fit over the last two years, we also want to avoid overfitting based on spurious relationships. To address this concern while also allowing for the possibility of sustained upward pressures, we ask which of these same signals would have been useful indicators over a much longer period.
In Exhibit 4, we analyze consensus inflation surprises over the last 75 years using the Livingston survey shown earlier in Exhibit 1. We find that forecasters on average failed to incorporate the signal from University of Michigan 1-year inflation expectations, business inflation expectations (GS composite measure), trimmed mean inflation, and the jobs-workers gap. In contrast, we do not find predictive power in UMich long-term inflation expectations or in the Fed’s Index of Common Inflation Expectations (CIE).

Exhibit 4: Inflation Expectations Surveys Could Have Prevented Some of the Post-War Inflation Forecast Errors

4. Inflation Expectations Surveys Could Have Prevented Some of the Post-War Inflation Forecast Errors. Data available on request.
Source: Department of Labor, University of Michigan, Federal Reserve, Goldman Sachs Global Investment Research
This suggests a role for some of these signals in our bottom-up models—to the extent the relationships are statistically and economically significant over the full-sample period and not just in 2021-2022. To avoid overreacting to pandemic-influenced surveys, we also assume that the inflation expectations variables return to pre-pandemic levels linearly between now and the end of 2024.[2]

Exhibit 5: Inflation Expectations Surveys Are Helpful Guideposts of the Current Trend in Core Inflation

5. Inflation Expectations Surveys Are Helpful Guideposts of the Current Trend in Core Inflation. Data available on request.
Source: Department of Commerce, University of Michigan, Federal Reserve, Goldman Sachs Global Investment Research

Updating Our Bottom-Up Core PCE Model

To better understand the implications of recent developments for the medium-term inflation outlook, we revisit our core PCE category-level inflation models to make sure that the lessons noted above are fully incorporated into our inflation forecasts. We disaggregate the core PCE index into 19 components: 11 measures of core goods and 8 measures of core services. We then use three types of explanatory variables to model the components: (1) measures of labor and product market slack; (2) indicators related to pass-through of various kinds, including commodity, import, and asset prices; and (3) alternative price indices available from non-government sources. We augment our models with the inflation expectations variables in the previous exhibit as well as the jobs-workers gap, which likely influences inflation and inflation expectations at a multi-year horizon.
Exhibit 6 summarizes the inputs used in each equation.[3]

Exhibit 6: Near-Term Inflation Expectations Are Additive in Nearly Half of Our Category-Level Models

6. Near-Term Inflation Expectations Are Additive in Nearly Half of Our Category-Level Models. Data available on request.
Source: Goldman Sachs Global Investment Research
We re-estimate the model coefficients using monthly data since 1997. We exclude the second and third quarters of 2020 from the sample to avoid distorting the coefficients with the pandemic shock. Besides incorporating the upward pressure on prices from elevated inflation expectations, our updated models now place more weight on near-term lags and alternative data and less weight on import prices and longer-term lags.
Exhibit 7 shows a breakdown of our year-over-year core PCE forecasts. We now expect somewhat higher core services inflation in December 2022 (+4.7% vs. +4.5% previously) and in December 2023 (+3.9% vs. +3.8% previously), as well as higher core goods inflation in December 2023 (-1.6% vs. -2.1% previously) and in December 2024 (-0.6% vs. -1.0% previously). Taken together, we expect year-over-year core PCE inflation of +4.2% at the end of this year, +2.5% at the end of 2023 (vs. 2.3% previously), and +2.3% at the end of 2024 (vs. 2.2% previously).

Exhibit 7: We Forecast Core PCE to Remain Above 4% Throughout 2022 Before Falling to 2.5% in December 2023 and 2.3% in December 2024

7. We Forecast Core PCE to Remain Above 4% Throughout 2022 Before Falling to 2.5% in December 2023 and 2.3% in December 2024. Data available on request.
Source: Department of Commerce, Goldman Sachs Global Investment Research

A Hot Summer for Core Inflation and the CPI-PCE Gap

We also introduce a bottom-up CPI model to complement our existing bottom-up PCE model, shown in Exhibit 8. Going forward, we will publish regular updates of our new model in our Monthly Inflation Monitor. Unlike the PCE measure, we expect year-on-year core CPI inflation to reaccelerate in late summer—from 6.0% in May to 6.3% in September—before falling to 5.5% at year-end. We expect core CPI inflation to fall more sharply next year (Dec. 2023 +2.4%), reflecting a negative swing in health insurance prices and shrinking contributions from across nearly every category but especially from normalizing auto prices, which have a larger weight in the CPI measure.

Exhibit 8: We Forecast Core CPI at 5.5% in December 2022, 2.4% in December 2023, and 2.6% and in December 2024

8. We Forecast Core CPI at 5.5% in December 2022, 2.4% in December 2023, and 2.6% and in December 2024. Data available on request.
Source: Department of Labor, Goldman Sachs Global Investment Research
We conclude with a few bottom-up component considerations and their implications for the CPI-PCE gap. As we highlighted last fall, the CPI-PCE gap will be much larger than usual this year because of the larger weights on the shelter and durable goods categories and differences in the measurement of health care inflation.
As shown by the dotted line in the left panel of Exhibit 9, the large and persistent shelter inflation category spiked to a 32-year high in May (PCE housing +0.61% mom sa). This pace overshot our shelter tracker (blue line) after underperforming it in early spring. The alternative rent measures constituting our tracker—which are much more volatile than the CPI and PCE measure—slowed sequentially in both Q4 and Q1, suggesting the possibility that the official May reading overstates the near-term trend.

Exhibit 9: Monthly Shelter Inflation Overshot Alternative Measures in May—and Those Measures Were Already Decelerating

9. Monthly Shelter Inflation Overshot Alternative Measures in May—and Those Measures Were Already Decelerating. Data available on request.
Source: Department of Commerce, Zillow, CoStar, REIS, Census Bureau, Goldman Sachs Global Investment Research
What could have driven the May strength? One possibility is a temporary boost from normalizing rent levels—perhaps related to deep one-time discounts— “months free”— during the pandemic. If a larger-than-usual share of units “reset” higher in the May sample, that would produce a temporary outsized gain in monthly shelter inflation.
We also update our model of health insurance CPI to reflect the 2021 financial results of publicly traded insurance carriers, which will begin to flow through the CPI measure this fall. As shown in Exhibit 10, the sharp increase in carrier profitability during the 2020 pandemic mostly reversed in 2021, setting the stage for outright declines in the health insurance CPI in Q4 and for most of 2023 (we assume -2.0% per month).

Exhibit 10: Health Insurance CPI Tailwind to Reverse Starting This Fall

10. Health Insurance CPI Tailwind to Reverse Starting This Fall. Data available on request.
Source: Department of Labor, Department of Commerce, Company data, Goldman Sachs Global Investment Research
Exhibit 11 summarizes our updated core inflation forecasts for the CPI and PCE measures. The core CPI-PCE inflation gap was +1.3pp in May (core CPI 6.0%, core PCE 4.7%), well above the 0.25pp historical average. As we flagged last October, three drivers underpin this outsized gap: the higher shelter weight in the CPI, the larger CPI weight on durable goods categories such as new and used cars, and sharp upward pressure on the health insurance CPI (the PCE does not include this component). Unlike the PCE measure, we expect year-on-year core CPI inflation to reaccelerate in late summer—from 6.0% in May to 6.3% in September—before falling to 5.5% at year-end, 2.4% in December 2023, and 2.6% in December (vs. our previous forecasts of 5.1%, 2.6%, and 2.6%).

Exhibit 11: CPI-PCE Gap to Widen Further This Summer, Disappear in December 2023 as Health Insurance CPI Swings Negative and Auto Prices Fall

11. CPI-PCE Gap to Widen Further This Summer, Disappear in December 2023 as Health Insurance CPI Swings Negative and Auto Prices Fall. Data available on request.
Source: Department of Commerce, Department of Labor, Goldman Sachs Global Investment Research
In terms of policy implications, while core PCE is the Fed’s preferred inflation gauge, Fed officials monitor several measures, and the CPI is probably more important than usual because it represents only consumer-paid prices—which should in principle be more relevant for inflation expectations—and because it is heavily featured by the financial and popular press. The CPI reacceleration we forecast this summer therefore represents another reason to expect the Fed to continue hiking expeditiously through the September meeting. Looking further ahead, we expect slowing growth and sequentially lower inflation to persuade the Fed to conclude its tightening campaign at the end of the year, after the funds rate reaches 3.25-3.5%.

Spencer Hill

Ronnie Walker

Manuel Abecasis

The US Economic and Financial Outlook

Data available on request.
Source: Goldman Sachs Global Investment Research

Economic Releases

Data available on request.
Source: Goldman Sachs Global Investment Research
  1. 1 ^ The survey collects data only on headline inflation, and respondents generally incorporate the recent path of oil and food prices into their forecasts of near-term inflation. Accordingly, we construct a proxy for core CPI consensus using the headline CPI forecast for quarters Q+3 and Q+4—for which the recent path of commodity prices is often a secondary or minor factor from a forecasting perspective. We find broadly similar results when replicating our analysis using headline CPI consensus, except generally lower correlations with realized inflation.
  2. 2 ^ We also cannot rule out the risk that elevated realized inflation leads short-and long-term inflation expectations even higher, particularly if the labor market continues to overheat.
  3. 3 ^ The lag structure for dependent variables differs by category based on statistical and economic significance.

Investors should consider this report as only a single factor in making their investment decision. For Reg AC certification and other important disclosures, see the Disclosure Appendix, or go to www.gs.com/research/hedge.html.