Historical

“…Democrats appeared to have done particularly poorly among Hispanic men in 2022. In the VoteCast data, Democrats carried this group by a scant 6 points (51-45). Compared to 2018, Democratic Hispanic working-class support is down 20 margin points, more than double the decline among Hispanic college voters (9 points). AP VoteCast estimates the decline in Democrats’ advantage among the nonwhite working class as 14 points between 2020 and 2022, 23 points between 2018 and 2022, and (splicing in some Catalist data, which overlap pretty well with VoteCast data where they overlap) an astonishing 33 point drop between 2012 and 2022….”

Ruy Teixeira, "Democrats’ Hispanic Problem: The Sequel,"The Liberal Patriot, November 17, 2022, https://theliberalpatriot.substack.com/p/democrats-hispanic-problem-the-sequel

Democrats’ Hispanic Problem: The Sequel

Democrats are feeling pretty good about themselves these days after a 2022 election that went way, way better than they (and most observers) thought it would. Those on the Democratic left, in particular, are congratulating themselves that they were right about…well, everything. Ezra Levin, who co-founded the activist progressive group, Indivisible said: "The great thing about having your strategy being proven correct is that you don’t have to rethink your strategy…We would have, if the red wave materialized. But it didn’t have the potency that we thought."

God forbid anyone should do any rethinking. That would never do.

The fact remains, however, that while they held the Senate (and may possibly pick up a seat), they lost the House, as well as the nationwide popular vote by 3-4 points. That’s a swing of 7-8 points toward the Republicans compared to Biden’s 4 point national advantage in 2020. The Senate map for the Democrats in 2024 looks absolutely terrifying: Democrats will have to hold seats in a wide range of red and purple states—Arizona, Maine, Michigan, Minnesota, Montana, Nevada, Ohio, Pennsylvania, Virginia, Wisconsin and West Virginia. The Republican seats that will be up are all in solid GOP states, with the possible exceptions of Texas and Florida (and we saw what just happened there). Plus there’s no guarantee that the 2024 Democratic nominee—likely Biden—will get to run again against the Democrats’ preferred opponent, Trump. It could be someone much tougher to take down.

In light of this, it’s worth considering the possibility that Democrats did not, in fact, fix all their problems in 2022 and that some of these may be lurking beneath the surface to undermine their chances—perhaps fatally—in 2024. One such problem is the Democrats’ Hispanic voter problem. In 2020, Democrats’ advantage among Hispanic voters declined nationwide by 16 points relative to 2016. Democrats had hoped to stop the bleeding in 2022. Did they?

It does not appear so. Prior to the election, the AEI demographics tracker, which averages poll subgroup results, found the Democratic Congressional margin among Hispanic voters consistently 7-9 points below its 2020 level and 17-19 points below its 2018 level. Results from AP/NORC VoteCast indicate that the drop in the 2022 election was actually larger than that foreshadowed by the pre-election data. These data show Democrats carrying Hispanics nationwide by just 56-39 in 2022, a 12 point decline in margin relative to 2020 (18 points relative to 2018). For what it’s worth, the less-reliable network exit polls, show an identical decline in Hispanic support between 2020 and 2022.

Digging more into the available data, some other troubling signs can be discerned beneath this broad national trend (all data from VoteCast except election returns).

1. Democrats appeared to have done particularly poorly among Hispanic men in 2022. In the VoteCast data, Democrats carried this group by a scant 6 points (51-45).

2. Democrats did worse among Hispanic working class (noncollege) voters than college voters. They carried Hispanic working class voters by just 15 points, compared to 21 points among their college-educated counterparts. Compared to 2018, Democratic Hispanic working class support is down 20 margin points, more than double the decline among Hispanic college voters (9 points).

3. With a few exceptions like Nevada, Democratic decline in Hispanic support can be seen across states. The most striking example of this is in Florida where DeSantis carried the Hispanic vote by 13 points, a 22 point swing from Biden’s 9 point margin among this group in 2020. He carried heavily Hispanic Miami-Dade county, historically the Democrats’ firewall, by 11 points. He carried Osceola County by almost 7 points—a county where Puerto Ricans, among the most Democratic of Hispanic subgroups, loom large. Shockingly, statewide DeSantis actually split the Puerto Rican vote 50-50 with Democrat Charlie Crist.

4. Democrat Val Demings also lost the Hispanic vote in Florida to Marco Rubio by 11 points. And statewide Democrats lost the Hispanic House vote by 7 points to the Republicans.

5. In Texas, Republicans got 43 percent of the statewide House vote among Hispanics, losing it by only 10 points to the Democrats. Beto O’Rourke did somewhat better than Biden in the Rio Grande Valley but still carried Hispanics statewide by just 14 points.

6. The poor showing of Democrats in the Hispanic Texas House vote was driven by working class Hispanics. They gave Democrats a mere 6 point margin statewide (51-45) compared to 24 points among college Hispanics (61-37). But working class Hispanics are three-quarters of the Texas Hispanic vote.

7. In Arizona, the GOP got 41 percent of the statewide House vote among Hispanics, cutting their deficit with this demographic to 13 points. Democrats’ House margin was down 6 points from Biden’s margin among this group in 2020. Katie Hobbs also ran behind Biden in her Hispanic support while Mark Kelley did a little better. But Kelley’s healthy margin of victory (5 points) is clearly attributable to a sharp improvement in support among white college graduates relative to Biden.

8. Even in California, Democratic House candidates in 2022 ran 11 points behind Biden’s margin in the state in 2020. This decline can be illustrated by looking as some heavily Latino CDs in Los Angeles County (vote counts ongoing): CA-31, Biden +31, 2022 House +16; CA-35, Biden +28, 2022 House +12; CA-38, Biden +30, 2022 House +13; CA-42, Biden +46, 2022 House +34.

These data suggest Democrats are far from out of the woods in terms of their Hispanic voter support. In fact, they indicate the problem is getting worse. More broadly, Democrats would be well-advised to look at these results in the context of their ongoing decline in working class support among nonwhites. AP VoteCast estimates the decline in Democrats’ advantage among the nonwhite working class as 14 points between 2020 and 2022, 23 points between 2018 and 2022 and (splicing in some Catalist data, which are consistent with VoteCast data where they overlap) an astonishing 33 point drop between 2012 and 2022.

I’d say that qualifies as a problem—and one that’s very, very far from being fixed.

The Verdad team shows that from 2001-22, firms with the highest analyst growth forecasts actually delivered the lowest total returns, as EBITDA multiple contraction more than offset high EBITDA growth that fell short of initial analyst forecasts.

“…The above chart shows the total returns of each quintile of growth prediction broken out into categories. The data above suggests that while firms with high forecast growth indeed grew at impressive rates, they also experienced the greatest contractions in their multiples. In fact, the same three factors that predicted growth—analyst forecast EBITDA growth, improvement in the TEV/EBITDA multiple, and trailing EBITDA growth—were the exact same factors that predicted multiple contraction. As a result, higher growth net of capital consumption for these firms was more than offset by higher multiple contraction. This is the cruel result of a lack of persistence in growth. Valuation multiples contract because growth does not persist….”

Greg Obenshain and Brian Chingono, "Predicting Growth,"Verdad, October 2022, https://mailchi.mp/verdadcap/predicting-growth
Predicting Growth

Two weeks ago, we extended the 2001 findings in the paper The Level and Persistence of Growth Rates through June 2022 and confirmed that, indeed, there is no persistence in long-term earnings growth beyond chance. The chart below shows this graphically. The straight black line at the top is what we’d expect if growth perfectly persisted, meaning that companies that were above the median growth rate in year 1 were again above that rate in years 2, 3, 4 and 5. The gray line is what we would expect if growth does not persist at all, meaning that a company that was above median growth in year 1 had a coin flip of a chance at being above median growth in year 2, so that that probability of being above median growth for two years in a row is 25% and just 3.215% for five years running. The blue line is Chan et al.’s original finding from that paper and the dotted black line is ours.

This result led us to quip that analysts might as well plug in the same long-term growth assumptions for a SaaS company as a coal mine. And we stand by this. The data shows that we need to be very humble when plugging a terminal growth assumption into a discounted cash flow model. A good assumption is the GDP growth rate.

But the observation that growth rates do not persist is not the same as saying they are not predictable in the short run. And last week we showed that, while analysts are wildly over-optimistic in their growth estimates, they do, in fact, sort companies correctly into the high and low growers. While it may indeed be true that growth does not persist, surely we can predict with some confidence that the IT services firm with a growing customer backlog is likely to grow faster next year than the auto parts supplier who just lost their largest contract.

So can we make some reasonable predictions about growth? One way to do this is just to select those variables that have predicted growth in the past. When we run regressions on EBITDA growth, the three variables that are most significant and have the highest magnitude of impact are, in order, analyst expected EBITDA growth, followed closely by both the trailing improvement in the TEV/EBITDA multiple and trailing EBITDA growth.

We can build a simple growth factor that equal weights these three variables. And this factor works very well to predict the relative magnitude of short-term growth. Below we show the same persistence chart, except this time we show the persistence of growth for the top quintile of our growth factor and the bottom quintile of our growth factor. The black line is again random chance.

But, alas, there is a problem: predicting growth isn’t the same as predicting returns. And it turns out that our growth prediction is actually negatively correlated with total returns. Below we show the realized EBITDA growth by predicted growth quintile and the realized total return.

The key to understanding this apparent paradox is to decompose total returns into the contributions of growth, changes in multiple, leverage, and distributions. The below chart shows the total returns of each quintile of our growth prediction broken out into these categories.

This is the cruel result of a lack of persistence in growth. Valuation multiples contract because growth does not persist. In fact, the lack of dispersion of returns between the high growers and the low growers points to the efficiency of markets. Growth is more or less correctly priced. And that’s what we’d expect. As every good analyst learns, it’s not the forecast that matters, but the forecast relative to what is already priced in.

Stocks that price high growth are particularly susceptible to multiple contraction. So an analyst building a DCF model should remind themselves that the higher the growth rate, the lower the exit multiple relative to the entry multiple should be. Analysts should assume that growth mean reverts to the market within five years. This is the argument we have made from the beginning and an eminently reasonable approach, even for the fastest growers, as we showed in Figure 2. To give a sense of what this should look like, we provide a cheat sheet below. For each projected growth quintile, we show the starting annualized two-year growth projection and the subsequent realized median annual growth rates.

We’ve spent three weeks now looking at growth, only to come to the conclusion that predicting growth well doesn’t lead to being able to forecast total returns any better. Next week we plan to look at what fair multiples should be for a terminal value assumption and why some companies deserve higher valuations, independent of their growth rates.

From 1997 to 2021, US securities analysts systematically overestimated 2-year cumulative revenue growth forecasts by 4.3% and 3-year forecasts by 4.6%, but their forecast errors were even larger for EBT: 28.4% high for 2-year and 36.8% for 3-year forecasts

Using a database of all US stocks from 1997 to 2021, we measured the median analyst estimate for growth over the next two to three years across a range of earnings metrics. We then compared these forecasts against the actual median outcomes over the next two to three years. While the estimated and actual growth rates in the table below are annualized for simplicity, the error is shown on a cumulative basis to reflect the magnitude of the mistakes at the end of each horizon. Analysts’ forecast errors seem to be bigger precisely in the areas that should matter most for stock prices: earnings further down the income statement that go to equity investors, and earnings further out in time.

Brian Chingono and Greg Obenshain, "The Sages of Wall Street," Verdad, October 2022, https://mailchi.mp/verdadcap/the-sages-of-wall-street

The Sages of Wall Street

Last week, we showed that past earnings growth does not predict future earnings growth. When earnings growth is measured over long horizons, there is little to no evidence of persistence beyond chance, even among the high-fliers that have grown the fastest in the past.

But of course, investors don’t simply look in the rearview mirror when projecting future earnings growth. They also evaluate additional information, including interviews with management, studies of firms’ competitive landscapes, and forensic reviews of any clues that are hidden in the footnotes of firms’ financial statements.

But how good are security analysts at predicting growth? In the paper we referenced last week by Chan et al. (2001), the authors state that “it is commonly suggested that one group of informed participants, security analysts, may have some ability to predict growth.... Over long horizons, however, there is little forecastability in earnings, and analysts’ estimates tend to be overly optimistic.”

We tested this finding for ourselves and came to a similar conclusion, as shown in the table below. Using a database of all US stocks from 1997 to 2021, we measured the median analyst estimate for growth over the next two to three years across a range of earnings metrics. We then compared these forecasts against the actual median outcomes over the next two to three years. While the estimated and actual growth rates in the table below are annualized for simplicity, the error is shown on a cumulative basis to reflect the magnitude of the mistakes at the end of each horizon.

While analyst estimates of future revenue are relatively close to actual outcomes, their accuracy deteriorates considerably as you move further down the income statement toward measures like Earnings Before Tax (EBT) that better reflect the earnings that are available to equity investors. And in each case, the cumulative error widens as you move further out in time from two years to three years in the future.

What if we replaced the analyst forecasts with a naïve estimate that corresponds with long-term GDP growth? Since 1947, US GDP growth has averaged 6.3% per year on a nominal basis (including inflation of around 3.5% per year). In the table below, we show the cumulative errors from forecasting 6.3% annualized growth for every company, compared to the cumulative errors from analyst forecasts.

This finding confirms Chan et al.’s observation that “it is difficult to see how the profitability of the business sector over the long term can grow much faster than overall gross domestic product.”

Fair enough, but fundamental investors might object that they don’t invest in average companies. What we want to understand is the distribution of forecasts versus actual results. Are analysts effective at separating high- and low-growth firms more broadly?

And here, we find some evidence of analyst forecasting skill. Below we show a table where we sorted analysts’ estimates of revenue and EBITDA growth into quintiles, and actual growth outcomes into quintiles. We then evaluate how accurate analysts are at selecting which growth quintile a company belongs to.

An accurate prediction would be a company where analysts expect top quintile growth, and actual growth over the next three years is indeed in the top quintile. These accurate predictions are tagged with a one in our data, and all other outcomes are tagged with a zero. The average of this binary column gives us the sorting accuracy of analyst forecasts. As shown in the table below, analysts are up to 50% accurate in the highest-expectation quintile, whereas they are around 30-40% accurate in other forecast quintiles, a significant improvement over random chance (20%).

But there’s a wrinkle to this analysis.

If a stock is richly priced because analysts expect 20% annualized earnings growth over the next three years, investors in that firm generally won’t earn a significant return premium unless its realized earnings growth exceeds 20% per year. On the other hand, if realized growth comes in at 10% annualized, the price decline could be catastrophic, as investors have seen when the pandemic-era “stay-at-home” trades like video conferencing and home exercise reversed dramatically in 2022. So we really care about the errors relative to the forecasts.

In the table below, we show median three-year forecasts and outcomes within each quintile of analyst expectations. The results indicate that analysts are generally directionally correct (the highest-expectation stocks grow faster than the lowest-expectation stocks), but the errors are significantly bigger in the higher-expectation quintiles than the lower-expectation quintiles, and errors are also significantly bigger the further down we look on the income statement.

Perhaps these results simply reflect human bias toward overconfidence, and more accurate predictions could be made by algorithmically weighting metrics from the financial statements? After all, the superiority of mechanical (statistical) predictions over clinical forecasts has been well documented across a range of disciplines in numerous studies. In the next installment of our research series, we’ll explore whether better growth forecasts can be made using quantitative measures.

Brad Setser, “Brad Sester On Current Accounts Surpluses And FX Reserves,”@Brad_Setser, October 10, 2022, https://twitter.com/Brad_Setser/status/1579659816372666368

“As @BobEUnlimited notes, one big global shift has come from the change in the U.S. petrol balance. I think there has been another: a shift from a world where big current account surpluses show up in fx reserves to one where they don’t. Prior to the global financial crisis — and in the few years after it – reserve growth in China, Saudi Arabia and Russia more or less tracked to their combined current account surplus. That isn’t the case now…”

“… This matters because the global current account surplus today is almost entirely in China (on track for at least $400b), Russia (on track for $250b or so), and the Gulf (on track for $400b or more), and none are showing in any reserve growth. The reasons differ (rate differentials and exporters hoarding dollars for China, sanctions for Russia, the Saudi shift away from SAMA toward a sovereign fund) but the net effect is the same — none are likely to add to their formal reserves in 2022. This matters for financial flows (less demand for bonds, probably less demand for euros too — for strange reasons), but it also simply makes the global flow of funds hard to track. It also has led to a broad misunderstanding about what is driving the big reported fall in global reserves. That fall is mostly from valuation changes – combined with real selling by EM Asian countries other than China. The big surplus countries (China, Russia and Saudi Arabia, which is representative of the Gulf) aren’t selling (unlike in 15), but they also aren’t adding. Their missing accumulation thus is helping to drive the large reported fall in global reserves. Last thought: Russia’s reported current account surplus in the first part of 2022 is almost equal to that of China. That reflects high oil and low Russian oil imports. But it also reflects the big gap between China’s customs surplus and its BoP surplus…”

“…Bottom line: there hasn’t been this kind of surge in the aggregate surplus of the big authoritarian global powers that hasn’t been primarily intermediated through their relatively conservative central banks in recent financial history.”

Inventors in East Asia now file significantly more international patents than their North American counterparts to protect their intellectual property around the world

“…The figure above shows the evolution in the number of foreign patent applications filed by different regions of the world in foreign jurisdictions. By 2003, East Asia (mainland China, Hong Kong, Japan, Macao, Mongolia, South Korea and Taiwan) had surpassed North America.According to this measure of foreign patents, Japan, accounting for 61% of the region’s foreign patent applications from 2010 to 2019. South Korea, China and Taiwan followed, filing 18%, 12% and 7%, respectively. East Asia has risen as the main destination where inventors around the world seek patent protection. Of all the patents filed abroad by applicants in North America during the 2010s, 41% went to East Asia, while just 13% stayed in the region, and 11% went to Western Europe…”

Ana Maria Santacreu and Jesse LaBelle, "East Asia Outpaces Other Regions in International Patenting," Federal Reserve Bank Of St. Louis, October 3, 2022, https://www.stlouisfed.org/on-the-economy/2022/oct/east-asia-outpaces-other-regions-international-patenting

East Asia Outpaces Other Regions in International Patenting

The global need for patent protection has increased substantially over the last couple of decades. Because patents are territorial (they offer protection only in the jurisdiction where they are granted), innovators typically seek patent protection in multiple countries where they are interested in doing business to prevent the theft of ideas or to keep technological competition at bay. That was the case of the United States in Asia during the decades following the formation of the World Trade Organization, as we documented in a December 2021 essay in Economic Synopses.

In this blog post, we document changes in the patterns of patenting across regions around the world. We use data on patent applications filed by 213 jurisdictions, grouped into 14 regions.

The figure below shows the evolution in the number of foreign patent applications filed by different regions of the world in foreign jurisdictions. By 2003, East Asia (mainland China, Hong Kong, Japan, Macao, Mongolia, South Korea and Taiwan) had surpassed North America (Canada, Greenland and the United States) in terms of filing patents in other jurisdictions.

East Asia Rises as a Patent Application Destination

Exploring patenting activity in East Asia further, we found that the main innovator, according to this measure of foreign patents, was Japan, accounting for 61% of the region’s foreign patent applications from 2010 to 2019. South Korea, China and Taiwan followed, filing 18%, 12% and 7%, respectively, of patent applications abroad.

East Asia has risen not only as an innovation hub (i.e., filing patents abroad) but also as the main destination where inventors around the world seek patent protection.

Of all the patents filed abroad by applicants in North America during the 2010s, 41% went to East Asia, while just 13% stayed in the region, and 11% went to Western Europe. Those in Western Europe showed a slight preference for East Asia, with 37% of their patents filed abroad going to the region, compared with 34% filed in North America.

Even East Asian countries have a preference for their own region. Of all the patent applications filed by East Asian countries to other jurisdictions, 43% went to North America, while 36% stayed within the region, and 11% went to Western Europe.

These patterns of international patenting indicate a bias for seeking patent protection in East Asia, not only by countries outside the region but also by intraregional countries. The shift toward Asia has not always been the case. Instead, in the 1980s and early 1990s, the other most innovative regions were filing for patents with one another, while Asia was behind in both the origin and the destination of patent applications.

High-Tech Competition Builds in East Asia

What explains these new trends? There could be several reasons why East Asia has become the main destination where countries around the world seek patent protection.

First, Asia is an important destination of technology-intensive goods imports and foreign direct investment from North America and Western Europe. Hence, inventors in these developed regions may want to protect their exported technologies from theft and misappropriation by competitors in Asian countries.

Second, Asian economies are becoming innovation powerhouses, increasing competition for high-tech goods within the region. For instance, the Japanese and South Korean technology sectors have grown substantially, whereas China and Taiwan have been dominating the manufacturing of semiconductors. This regionally strong growth may have led to a “race to patent,” as each country attempts to become the dominant innovative force in that region and ward off competition.

In summary, foreign patenting has been increasing substantially in recent decades. While the U.S. and Western Europe have historically been seen as innovation powerhouses, East Asia has quickly established itself as both the main origin of innovations and the main destination for patent protection.

Note

This region comprises the Aland Islands, the Channel Islands, Denmark, Estonia, the Faroe Islands, Finland, Iceland, Ireland, the Isle of Man, Latvia, Lithuania, Norway, the Svalbard and Jan Mayen islands, Sweden and the U.K.

Rising assets values reduced labor force participation for 51-65 year olds by 0.7mm people as of Sept. ‘21, 15% of overall LFP reduction since Jan. ‘20. It’s unclear whether recent declines in asset prices will reverse this trend. “…As of September 2021, the labor force participation rate was 61.6%, down from a pre-pandemic peak of 63.4% in January 2020. I try to quantify how much of the 1.8-percentage-point drop in LFP during 2020 and 2021 may have been related to rising asset values during that same period. I focus on households with members between 51 and 65 years of age who are thus more likely to retire. The computed change in the value of wealth leads to an estimated drop in LFP of 0.92 percentage points for this age group, which corresponds to 0.7 million people—about 15% of the total drop.Nor is it clear whether these labor supply effects are symmetric: Would people increase their LFP by the same amount if their wealth fell…”

Miguel Faria e Castro, "Asset Returns and Labor Force Participation During COVID-19,"Federal Reserve Bank of St. Louis, January 6, 2022, https://research.stlouisfed.org/publications/economic-synopses/2022/01/06/asset-returns-and-labor-force-participation-during-covid-19

Asset Returns and Labor Force Participation During COVID-19

As of September 2021, the labor force participation (LFP) rate1 was 61.6%, down from a pre-pandemic peak of 63.4% in January 2020. This 1.8-percentage-­point decline in the LFP rate corresponds roughly to a net decrease of 4.7 million workers in the U.S. economy, compared with the pre-pandemic period.2

In a previous essay, I discussed the large increase in excess retirements during the pandemic period relative to previous trends.3 An increase in retirements or a decrease in the number of retired people returning to work are among the reasons the LFP may have decreased and remained low. In that essay, I explained that there are many reasons older people may have chosen to accelerate their retirement plans, including greater risk of death conditional on infection and the need to care for family members. Another factor that may have enabled these retirements was rising asset values during the pandemic period. The figure shows that some of the most common asset classes, such as stocks or housing, posted historically high cumulative real returns during 2020 and 2021.4 Thus, these years were associated with significant increases in wealth for households exposed to these assets.

Second, I impute the realized returns on several asset classes during the 2020-2021 period, which allows me to estimate the evolution in the value of net worth for households in the sample (assuming the composition of their portfolio remained fixed).

In the final step, I use standard estimates for the elasticity of LFP rates to changes in wealth; that is, I estimate how much the LFP rate is predicted to change for each household in the sample, given the estimated change in net worth. Standard theories of household behavior predict that people tend to work less in response to increases in wealth, and many studies have tried to empirically measure the magnitude of these effects. I combine the results of one such study5 with my estimates for the evolution of net worth to obtain estimates for the change in LFP of individual households, which can then be aggregated into an estimated change of the LFP rate.

I focus on households with members between 51 and 65 years of age who are thus more likely to retire. The computed change in the value of wealth leads to an estimated drop in LFP of 0.92 percentage points for this age group, which corresponds to 0.7 million people—about 15% of the total drop.

It is important to emphasize that rising asset valuations may not have necessarily caused people to exit the labor force (or remain out of it). These resulting increases in wealth may have, however, compounded other reasons to stop working, as mentioned in the second paragraph of this essay (i.e., risk of death from infection and increased domestic responsibilities).6 Finally, it is not clear whether these reductions in LFP are permanent, as people could eventually return from retirement, which is particularly likely for younger households. Nor is it clear whether these labor supply effects are symmetric: Would people increase their LFP by the same amount if their wealth fell? These questions are left for further research.

Notes

1 The labor force participation rate is the percentage of the civilian noninstitutional population 16 years and older that is working or actively looking for work.

2 This calculation uses as a reference point the civilian noninstitutional population as of September 2021, which was equal to 261.8 million people.

3 Faria-e-Castro, M. "The COVID Retirement Boom." Federal Reserve Bank of St. Louis Economic Synopses, No. 25, 2021;https://doi.org/10.20955/es.2021.25.

4 For reference, the cumulative real returns on stocks and housing for a period of the same length ending in December 2019 were 14.8% and 2.4%, respectively.

5 Benson, D. and French, E. "How Do Sudden Large Losses in Wealth Affect Labor Force Participation?" Federal Reserve Bank of Chicago Letter No. 282, January 2011;https://www.chicagofed.org/publications/chicago-fed-letter/2011/january-282.

6 See, for example, the following NPR story: "These Older Workers Hadn't Planned To Retire So Soon. The Pandemic Sped Things Up";https://www.npr.org/2021/08/23/1028993124/these-older-workers-hadnt-planned-to-retire-so-soon-the-pandemic-sped-things-up.

“…This paper studies the impacts of Covid-19 illnesses on the labor supply of U.S. workers using longitudinal data from the Current Population Survey (CPS). We first document that the rate of week-long health-related absences has been substantially elevated during the pandemic relative to pre-pandemic seasonal patterns, and that these excess absences appear to reflect Covid-19 illnesses. In a typical pandemic week, about ten workers per thousand missed an entire week of work due to their own health problems, as compared to six health-related absences per thousand workers in an average week from 2010 to 2019.Excess absences covary with reported Covid-19 cases both in the national time-series and a state panel, and are higher among occupations with higher Covid-19 exposure. Second, we show that health-related absences generate persistent reductions in labor supply. Using an event study, we find that workers who miss an entire week due to probable Covid-19 illnesses are approximately 7 percentage points less likely to be in the labor force one year after their absence compared to otherwise-similar workers who do not miss work for health reasons. These estimates appear representative for Covid-19 illnesses specifically, not only as an average over myriad health issues, and they are about the same as the average effects of pre-pandemic health-related absences. One reason why Covid-19 illnesses reduce labor supply is that they push older workers into retirement. Our results are thus best interpreted as identifying not “long Covid” in isolation but rather the overall labor-supply adjustment, across mechanisms, induced by Covid- 19 illnesses. We next use our results to assess the aggregate impacts of Covid-19 illnesses on the U.S. labor force participation rate. Combining our event-study results with the excess rate of health-related absences, we estimate that Covid-19 illnesses have reduced the labor force participation rate by 0.2 percentage points, or by approximately 500,000 workers, through June 2022. This point-in- time impact is near our estimates of the steady-state impact of Covid-19 illnesses at the average rate of health-related absences in 2021. These losses fundamentally reflect an overall increase in the health-related absence rate with no change in the average long-run impact of an absence. We also discuss additional considerations that would increase or reduce the total effect of Covid-19 illnesses on labor supply relative to our preferred estimate. Overall, the loss in U.S. labor supply from Covid-19 illnesses appears substantial, comparable to an additional year of U.S. population aging at its current pace.Finally, we calculate the average cost of a Covid-19 absence in forgone labor earnings through fourteen months after the absence, including both extensive and intensive margins of adjustment. The average Covid-19 absence results in forgone earnings of at least $9,000, about 90 percent of which is due to persistent labor-supply reduction beyond the week-long absence itself. This economic cost can be viewed as a revealed-preference measure of the medium-term health impact of Covid-19 illnesses. In aggregate, we calculate that the per-year value of the lost labor supply is approximately $62 billion, which is about half of estimated losses from cancer or diabetes….”

Evidence of Covid Impact on LFP

“…We find that Covid-19 illnesses have likely become, over the last two years, an important contributor to the net change in the participation rate. By June 2022, Covid-19 illnesses reduced the participation rate by 0.18 percentage points, with a range of 0.13 to 0.22 percentage points, as we show in Appendix Figure A21. These reductions in the participation rate imply that approximately 500,000 adults are neither working nor actively looking for work due to the persistent effects of Covid-19 illnesses, with a range of 340,000 to 590,000 adults. This point-in-time participation-rate loss is near the steady-state loss associated with the 2021-average rate of health-related absences. That is, if the health-related absence rate remains near its 2021 level, and if the impacts of Covid- 19 absences are unchanged, then our estimates suggest the participation rate will be persistently reduced by approximately 0.2 percentage points….”

Gopi Shah Goda and Evan Soltas, "The Impacts of Covid-19 Illnesses on Workers,"National Bureau Of Economic Research, September 2022, https://www.nber.org/papers/w30435

“…Then again, there is another composition effect at work in these statistics that I haven’t discussed. While women’s labor-force participation rates rose from the 1950s through 1990s, men’s fell. So while the current labor-force participation rate for all Americans in their late 50s and early 60s is near an all-time high, for men it’s well below the rates that prevailed before the mid-1970sThe puzzle of why labor-force participation fell so much among US men of all ages was a major topic of economic discussion in the 2010s that’s probably due for a revival. For those in their late 50s and early 60s the decline wasn’t all bad news, as generous early retirement packages surely drove some of it. Those are less common than they used to be, whileboosting male 55-64 labor force participation back to, say, the August 1970 rate of 82.6% would add 2 million would-be workers to the US job market. Bringing the participation rate for so-called prime-age men (25-54) back to 1970 levels would add another 4.5 million.Reversing changes that happened half a century ago is surely a lot harder than reversing those of the past couple of years. But maybe it’s the next frontier….”

Justin Fox, "The 55-and-Older Labor Market Exodus Is a Statistical Mirage,"Bloomberg, September 9, 2022, https://www.bloomberg.com/opinion/articles/2022-09-09/americans-older-than-55-haven-t-given-up-working

The 55-and-Older Labor Market Exodus Is a Statistical Mirage

I’m in my late 50s and I’m on the job. So are more than 70% of Americans aged 55 through 59, according to the Bureau of Labor Statistics. In fact, my age group’s labor-force participation rate was approaching record levels earlier this year before undergoing its usual summer decline.

Americans in their early 60s are pretty close to record labor-force participation, too. The big declines are all among those 65 and older. But — and here’s where things get really weird — those declines are still smaller in percentage-point terms than those for the 55-and-older group overall. This strange data quirk is the result of so-called composition effects, caused by the shifting age distribution within the age group, which I’ll discuss a little later. First, though, let’s get around the strange quirk by looking at labor market changes for smaller age groups.

The reason the 55-and-older numbers get so much attention is that it’s the one of the few age groups for which the BLS provides seasonally adjusted labor market data. The unadjusted numbers come with big seasonal swings and, especially when you’re looking at narrow age groups such as 55 through 59, a fair amount of month-to-month statistical noise. Still, there are ways to deal with both, such as comparing three-month averages from this summer with those from before the pandemic in 2019.

As for the two under-60 age groups that haven’t returned to or surpassed their summer 2019 participation and employment rates, those in their early 20s may be staying in college longer to make up for pandemic gap years and other delays, or still struggling to recover from the many disruptions to the job market and society in general in 2020. For those in their late 30s child-care complications could be a culprit, although curiously, women in that age group are back to pre-pandemic levels while men are not.

For workers 60 and older, who in the decades before the pandemic had been experiencing steady gains in labor-force participation and employment, it makes sense that a disease that was far deadlier for the elderly drove many out of the labor force — because of either fear of catching Covid-19 or struggles with Long-Covid complications after catching it. The remarkable performance of stocks and other assets in 2020 and 2021 may also have led some affluent older workers to retire earlier than planned.

But again, members of the 60-to-64 age group left only briefly. Their labor-force participation rate hit an all-time high in March 2020, when low response rates amid lockdowns may have skewed the data, and came close to that again last November. (“All-time” for these numbers only goes back to June 1976, but statistics available back to 1948 for the broader 55-to-64 age group indicate that recent rates are the highest over that period too.)

Those 65 and older have in fact seen a clear drop in labor-force participation after decades of gains. But 65-plussers made up only 6.6% of the labor force in August, compared with the 16.4% share accounted for by those 55 through 64. Their 1.1 percentage point labor-force participation decline since summer 2019 works out to 626,835 missing would-be workers — not nothing, but also not that huge a deal in a labor force of almost 165 million.

The labor-force participation decline for the broad 55-and-older category is 1.7 percentage points whether measured in seasonally adjusted data from February 2020 or in unadjusted summer-2019-to-summer-2022 comparisons, as I did with the other age groups. How can this be bigger than the decline for those 65 and older, not to mention those in their late 50s and early 60s? Well, the youngest baby boomers are turning 58 this year, meaning that the younger end of the 55-and-older group is now being replenished with the less-numerous members of Generation X.

Because of this, the average age of Americans in the 55 and older category rose even through the ravages of Covid-19 and will continue doing so. 1 Meanwhile, labor-force participation tends to peak around age 40, and decline as people age beyond that. So with each passing year, the 55-plus labor-force participation rate will drop even if the rates for every year of age within the group remain the same. The reverse transpired as the baby boomers began to enter the age group two decades ago, as will also be the case for the 75-plus age group that they started joining last year.

Then again, there is another composition effect at work in these statistics that I haven’t discussed. While women’s labor-force participation rates rose from the 1950s through 1990s, men’s fell. So while the current labor-force participation rate for all Americans in their late 50s and early 60s is near an all-time high, for men it’s well below the rates that prevailed before the mid-1970s