US Economics Analyst
Nowcasting the Soft Landing: A Look Back at Our 2023 Forecasts (Hill/Abecasis)
In our final Analyst of the year, we review the economic data surprises of 2023, our ability to forecast them, and lessons learned along the way.
GDP surprised to the upside in 2023, validating our out-of-consensus view that the US economy would avoid recession. While instability in the banking system led us to lower our GDP growth forecast in March, bank deposits subsequently stabilized, the credit drag proved manageable, and activity data continued to surprise to the upside on net. A second key macro theme of 2023 was the progress on disinflation that occurred despite strong growth and auto supply issues related to China’s covid wave. Our 2023 inflation forecast led consensus higher during the spring auto supply disruptions, then led it lower in the back half of the year as services-led disinflation took hold.
In terms of economic indicators, our forecast hit rates averaged 68% directionally correct vs. consensus across top-tier indicators—up from 62% in 2022—with particularly strong performance for the GDP report (100%, 3 out of 3), core PCE (91%), core CPI (82%), average hourly earnings (75%), and the employment cost index (67%). Additionally, our 67% hit rate for the unemployment rate benefitted from our team’s soft landing call. We underperformed consensus for 2 of these 13 indicators: Philly Fed (42%) and ADP (44%). On the latter, ADP’s negative correlation with nonfarm payrolls suggests that forecasts of ADP should perhaps orient around predicting the noise as opposed to the signal.
Our hit rate for nonfarm payrolls was 55%. Our forecasts generally benefitted from Big Data indicating low layoffs and continued job growth, however we were whipsawed by seasonal labor supply constraints in May and June.
Retail sales was a second indicator for which performance was solid but nonetheless lagged prior years (57% in 2023 vs. 92% in 2022). Reviewing our track record, we find that our forecasts were more predictive of the eventual revised values and that the initial vintages suffered from an upward bias worth -0.15pp on average. These sizeable and fairly consistent downward revisions underscore the importance of monitoring alternative measures of economic activity as a crosscheck on the official data.
Our 2023 hit rate averaged 60% across second-tier indicators, with particularly strong performance for Auto sales (88%), industrial production (86%), and jobless claims (76% initial, 60% continuing), the latter benefitting from our views on residual seasonality and our soft landing call. Our key underperformer was foreign trade (13%), where we were wrong-footed repeatedly by excessive optimism around China’s reopening and the related rebound in global trade.
Markets reacted strongly to data surprises in 2023, in particular to inflation surprises, for which sensitivity was 3.5x normal for stocks and 3.0x normal for bonds (10 minutes before and after the report, GS estimates). Treasury market sensitivity to growth data was also elevated—we estimate at 40% above normal—reflecting the intensity of the “recession vs. soft landing” debate and the Fed’s data-dependent policy. We also find that the equity market has re-embraced the “bad is good” narrative, with stocks once again rallying in response to weak growth data.
Nowcasting the Soft Landing: A Look Back at Our 2023 Forecasts
In our final Analyst of the year, we review the economic data surprises of 2023, our ability to forecast them, and lessons learned along the way.
GDP growth surprised to the upside in 2023, validating our out-of-consensus view that the US economy would avoid recession. Unlike consensus, we saw the risks to the economy tilted towards reacceleration, not recession. We expected the large growth drags from tighter monetary and fiscal policy to fade in 2023, and saw strong personal income growth as a key driver of healthy consumption growth. As shown in Exhibit 1, activity data subsequently surprised to the upside, and we boosted our GDP growth forecast another 0.4pp above consensus between January and early March.
Exhibit 1: Growth Exceeded Even Our Optimistic Expectations in 2023
Source: Bloomberg, Goldman Sachs Global Investment Research
In the spring, instability in the banking system led us to lower our GDP growth forecasts in the expectation that rapid deposit outflows would lead banks to cut back on credit. But after a short-lived decline, bank deposits remained broadly stable in the second half of the year, and the slowdown in bank lending was in line with the historical relationship between credit and financial conditions.[1] Activity data also continued to surprise to the upside on net, with real consumer spending on pace to rise nearly 3% annualized in the second half of the year despite the resumption of student loan payments.
Turning to our quarterly GDP tracking estimates, we were correctly above consensus for nearly the entire year (with the caveat that Q4 has not been officially reported), as shown in Exhibit 2. For Q1, our forecast led consensus higher during the quarter itself, however GDP ultimately disappointed consensus’s (and our) upwardly revised projections, due to a drag from the difficult-to-forecast inventory component. Our quarterly forecasts were consistently above consensus for the rest of the year, and GDP growth has so far indeed surprised to the upside.[2]
Exhibit 2: GS Quarterly GDP Tracking, Versus Consensus and Atlanta Fed GDPNow
Source: Bloomberg, Federal Reserve, Department of Commerce, Goldman Sachs Global Investment Research
One surprising macro theme of 2023 was that strong GDP and payroll gains did not prevent inflation from declining significantly, to 3.2% in November compared to 5-5½% in 2022 (core PCE, year-on-year). As shown in Exhibit 3, our inflation forecast generally led consensus higher during the spring and early summer, when China’s covid wave and a lull in winter microchip supply caused another jump in used car prices. Our inflation forecast then generally led consensus lower in the back half of the year as services-led disinflation took hold.[3]
Exhibit 3: Inflation Surprised to the Downside in H2 After Upside Surprises Earlier in the Year
Source: Bloomberg, Goldman Sachs Global Investment Research
While our current 2023 Q4/Q4 core PCE forecast of 3.1% is roughly what we forecasted a year ago, the sequential pace of inflation slowed more quickly than we expected in the second half, with core PCE inflation of just +2.0% annualized in Q3 (and so far in Q4). With year-over-year core PCE inflation now likely to reach +2.5% by March, we anticipate that the FOMC will cut the funds rate five times next year.
Indicator Forecast Performance
In Exhibit 4, we review our 2023 economic indicator forecasts based on our hit rate—the share of forecasts that were directionally correct versus consensus.[4] Our hit rate averaged 68% across 13 top-tier, market-moving indicators—up from 62% in 2022 and compared to 62% on average since 2017 (when we began systematically incorporating Big Data into our forecasts). Our 2023 hit rates were above 50% for 11 out of these 13 indicators, with particularly strong performance for GDP (100% or 3 of 3 directionally correct vs. consensus, advance reading), core PCE (91% directionally correct vs. consensus), core CPI (82%), core capex orders (80%), and average hourly earnings (75%). Performance was also strong—67% hit rates—for the employment cost index, ISM manufacturing, and the unemployment rate.
Across these and other indicators, forecast performance benefited from our application of Big Data, our suite of proprietary indicators, and the insights of our equity analysts. As in previous years, our monthly core PCE inflation forecasts benefited from the detailed CPI, PPI, and import price source data that we incorporate into our estimates.[5] And our forecasts for core capital goods orders, ISM, and industrial production (86% hit rate) benefitted from our tracking of foreign industrial indicators and US freight data.
Exhibit 4: What Worked in 2023? GS Forecast Accuracy for Key Market-Moving Indicators
Source: Goldman Sachs Global Investment Research
For the unemployment rate, our forecast performance benefitted from our high-conviction view on a soft landing, as well as on Big Data that continued to indicate this outcome—specifically that employment growth generally remained at or above potential. But while our 67% hit rate in 2023 compares favorably to our longer-term record, we were nonetheless wrong-footed in May and October by the dramatic underperformance of the household survey. This underperformance itself in part reflected the shift away from self-employment and the trend toward multiple jobholders—neither of which our nowcasting inputs are well-suited to assess.
We underperformed consensus with our Philly Fed (42% directionally correct) and ADP (44%) forecasts. On the former, our Philly Fed forecasts were overly optimistic early in the year with respect to the China reopening and the timing and magnitude of the East Asian manufacturing rebound. On the latter, ADP employment growth has exhibited a negative correlation with nonfarm payrolls since the most recent methodology changeover (-0.26 since August 2022, first-reported basis). The lesson we glean is that optimal forecasts of ADP should perhaps orient around predicting the noise rather than the signal—at least in periods of relative stability in the labor market.
We achieved a 55% hit rate on our nonfarm payrolls forecasts for 2023 (6 of 11 forecasts directionally correct vs. consensus)—an improvement vs. 2022 (33%) but below our 61% average since 2017. In 2023, our Q1 forecasts benefitted from Big Data indicating low layoffs and from our expectation that the tight labor market would cause firms to pull forward hiring earlier in the year. However, we were whipsawed at the start of summer by labor supply constraints, which we assumed would weigh on May job growth but evidently proved binding in June.
Retail sales was a second key indicator for which performance was solid (57% hit rate) but nonetheless lagged prior years (92% in 2022 and 73% since 2017). Reviewing this middling performance, we find that our forecasts were more predictive of the eventual revised values and that the initial vintages suffered from an upward bias worth -0.15pp on average. [6] As shown by Exhibit 5 comparing the first-reported data (light blue line) to the revised vintages (dark blue line), the pace of monthly retail sales growth has been revised down by 0.15pp on average so far this year—adversely affecting our performance in months including March and October.[7]
Exhibit 5: Retail Sales Data Were Consistently Revised Lower In 2023
Source: Census Bureau, Bloomberg, Goldman Sachs Global Investment Research
One possible explanation is nonresponse bias, for example if underperforming brick-and-mortar firms were less likely to submit revenue data on a timely basis. Regardless of the driver, these revisions underscore the importance of monitoring alternative measures of economic activity, which in some instances may be equally or more accurate than the official data.[8]
Turning to our successful retail sales forecasts—including the large spending beats for January, April, and July—performance benefitted from our application of Big Data (credit card spending, Adobe ecommerce panel) and our views on post-pandemic seasonality (July 2023 upside driven by Amazon Prime Day, January 2023 upside driven by residual seasonality).
Forecast Accuracy Across Second-Tier Data
In Exhibit 6, we perform the same hit-rate analysis for 17 second-tier indicators, which on average generate smaller or negligible market reactions. Our 2023 hit rate averaged 60% across these indicators—below the 70% rate in 2022 but close to the 61% average since 2017. Hit rates were above 50% for 12 of the 17 indicators, with particularly strong performance for auto sales (88% hit rate in 2023), industrial production (86%, discussed previously), and jobless claims (76% initial, 60% continuing). Our housing indicator forecasts also performed well, including for pending homes sales (75%), construction spending (71%), new homes sales (67%), and existing homes sales (64%).
Exhibit 6: Second-Tier Indicator Forecast Accuracy
Source: Goldman Sachs Global Investment Research
Our application of Big Data drove our strong forecasting performance for auto sales and the various housing releases. We believe our strong performance for initial jobless claims (76% hit rate, up from the prior 3-year average of 54%) reflected our detailed analysis of residual seasonality as well as our broader expectation of a soft-landing: consensus on average forecasted a 3.6k weekly rise in jobless claims which proved far too pessimistic (+0.3k as reported so far in 2023). Our UMich inflation expectations forecasts (67% hit rate) once again benefitted from our monitoring of gas prices and of the inflation news heard by consumers.
In terms of underperformers, our hit rate for consumer confidence once again lagged (33%), as we underappreciated the late-summer decline in sentiment catalyzed by higher interest rates and lower stock markets. Our very poor performance for the goods trade balance—just 13% directionally correct—in part reflected excessive optimism with respect to China’s reopening and the rebound in global trade that was deferred until the second half of the year.
Market Reactions to Economic Data
2023 was also a strong year for data sensitivity. As shown in Exhibit 7, Treasury market reactions to growth surprises jumped from negligible levels in 2021 to 40% above normal in 2023, on our estimates, with a 1.4bp reaction in the 10-year yield per standardized growth surprise. This elevated sensitivity in large part reflected the intensity of the “recession vs. soft landing” debate, as well as the high level of the funds rate and the Fed’s data-dependent policy.
Exhibit 7: Treasury Yields Remain Highly Sensitive to US Growth Data, Even in Q4
Source: Goldman Sachs Global Investment Research
Despite the easing of recession fears, sensitivity to growth surprises remained high in Q4, suggesting that fixed income investors believe that activity and labor market performance—and not solely inflation—will influence the timing and magnitude of Fed easing in 2024. This is consistent with Powell’s renewed concern with the labor market that he expressed at the December FOMC meeting.[9]
Turning to equities, stock market sensitivity to growth data was low by historical standards in 2023 (see Exhibit 8). However, these averages conceal a more nuanced picture, namely, an oscillation within the year from a “bad news is good news” interpretation, to a “good news is good news” regime, and back again. These sensitivity regimes broadly tracked the dominant market dynamic of each period, from “rate hike fears” in the winter, to “recession fears” during the SVB shock and the associated “growth relief” that followed. This dynamic then gave way to “inflation relief” and “rate cuts” in the second half of the year.
Exhibit 8: “Bad News Is Good News” in the Stock Market
Source: Goldman Sachs Global Investment Research
In previous research, we found that the market’s implied probability of a hard landing moved decisively lower between May and July of this year. And as shown by the blue dot, equity reactions to better growth news had turned negative again by Q4, with equities once again tending to rally on weak growth data. We believe this reflects the likelihood that weak-but-positive growth reduced the risk of additional rate hikes in 2023 and will increase the number of rate cuts in 2024. Looking ahead, we expect these “perverse” equity reactions to continue next year, as long as recession fears remain on the back burner and the Fed’s easing plans continue to depend on the pace of growth as well as on inflation.
Exhibit 9: Rates Remain Sensitive to Inflation Data Even As Overheating and Recession Concerns Have Diminished
Source: Goldman Sachs Global Investment Research
We view the market’s continued intense focus on inflation data as consistent with our forecast of five 25bp cuts in the Fed funds rate next year—two more than shown by the FOMC median projection. We forecast core PCE inflation to approach the target next year, falling to 2.5% in March and 2.2% in December. This would undershoot the Fed’s expectation of 2.4% (December SEP median, Q4 average), paving the way for more easing than the Committee currently envisions.
Spencer Hill
Manuel Abecasis
The US Economic and Financial Outlook
Source: Goldman Sachs Global Investment Research
- 1 ^ Nonbank lenders also filled some of the financing gap left by banks, further dampening the growth drag from tighter bank lending standards.
- 2 ^ GDP was particularly strong in Q3, at +4.9% annualized. The Atlanta Fed’s GDPNow tool was correctly well above both our and consensus expectations for most of that quarter, and as in Q1, the volatile inventories component explained all of the difference between our forecast and the official print (and most of the difference in our forecast relative to GDPNow). For Q4, we currently estimate GDP rose +1.4% annualized, above consensus of +1.2% currently and well above its projected decline as of late-summer. With a Q4 government shutdown averted and a limited impact on auto production from the autoworkers strike, the growth headwinds we assumed now appear milder.
- 3 ^ In the first half of the year, we raised our core inflation forecast several times as progress on core goods prices and shelter inflation proved slower than we expected. But since then, core services prices have decelerated by more than we anticipated, driven in part by the continued rebalancing of the labor market. As a result, we now expect core PCE inflation to end the year at +3.1% in Q4 and 2.9% in December, roughly in line with our forecasts as of one year ago.
- 4 ^ Excluding exactly-in-line readings and based on first-reported actuals on the Bloomberg ECO screen. We ignore observations that are exactly in line. In other words, the hit rate is the number of directionally-correct forecasts divided by the number of correct and incorrect forecasts. This construction also allows the interpretation of >50% hit rates as “net favorable” and <50% hit rates as “net unfavorable.”
- 5 ^ Because we forecast monthly core CPI and PCE on an unrounded basis (e.g. 0.23%), comparing our forecast to rounded median consensus (e.g. 0.2% or 0.3%) is not a perfect benchmark. When evaluated on a rounded basis—a smaller sample—our core CPI hit rate rises from 82% to 100% (2 out of 2 correct calls) and our core PCE hit rate rises from 91% to 100% (4 out of 4 correct calls).
- 6 ^ Control measure (ex-auto, gas, building materials, mom sa).
- 7 ^ Our hit rate would rise 10pp to 67% if evaluated using revised data, even if November remains an incorrect forecast on this basis.
- 8 ^ More generally, this serves as a reminder that contemporaneous seasonal adjustment tends to absorb some of the data surprise, which in turn argues for attenuating our forecasts of the initial vintages (particularly when the forecasting models themselves are estimated on revised data).
- 9 ^ “We’re getting now back to the point where both mandates are important—they’re more in balance.”--Chairman Powell, December FOMC press conference.
- 10 ^ The limited number of inflation releases limits the reliability of these parameter estimates over shorter windows, as they can be influenced by outliers such as the outsized -11bp reaction to the -0.07pp surprise in the October core CPI (+0.23% mom sa vs. consensus of +0.3%). Accordingly, we believe inflation sensitivity in the Treasury market is currently somewhere in between, perhaps around 3bps per standardized inflation surprise).
- 11 ^ A key example is the euphoric reaction to the October CPI report on November 14, which generated a 1.0% rally in S&P 500 futures in the 10 minutes around the release.
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.