Macro Roundup

Rising earnings inequality is dominated by rising between-firm inequality. Our analysis as well as the recent literature emphasizes that this largely reflects how firms are organizing themselves in terms of their workforce. High (low) earnings workers are more likely to work with each other (increased segregation), and high (low) earnings workers are more likely to work at high (low) firm premia firms (sorting). Our contribution is to highlight the dominant role of industry effects in accounting for this structural change of how firms organize their workforces. Most of rising between-firm inequality is accounted for by rising between-industry dispersion in earnings. The between-industry component accounts for 61.9% of total increasing earnings inequality, and 72.8% of between-firm inequality growth. This changes the narrative of the sorting and segregation contributions. High (low) earnings workers are more likely to work with each other in specific industries and high (low) earnings workers are more likely to work in high (low) average firm premia industries. Not only do industry effects dominate but it is a relatively small share of industries that account for virtually all the increasing dispersion in earnings across industries. We find that about ten percent of the 301 detailed 4-digit NAICS industries account for almost 100% of the rising between-industry dispersion, while accounting for less than 40% of employment. The ten percent of industries that account for virtually all of the increase are drawn from the top and bottom of the earnings distribution in terms of industry-level averages. For those industries at the top of the earnings distribution, their contribution is dominated by rising inter-industry earnings differentials. For industries at the bottom of the earnings distribution, their contribution is dominated by shifts in employment to these very low earnings industries. For both sets of industries at the top and the bottom of the earnings distribution, increased sorting and segregation between industries dominates but increased dispersion in between industry firm premia also plays an important supporting role. Increased sorting is relatively more important for the rising between-industry dispersion from the industries at the bottom of the earnings distribution. In contrast, increased segregation is relatively more important for the rising between industry dispersion from the industries at the top of the earnings distribution. The dominance of industry effects is closely linked to the rising importance of mega (10,000+) firms in the U.S. economy. The increasing share of employment accounted for by mega firms is concentrated in the thirty 4-digit industries that account for virtually all of rising between-industry dispersion. This rising share of employment at mega firms is accompanied by a declining size-earnings premium in the eleven low-paying industries. For mega firms in the nineteen high-paying industries in the top 30, earnings premia rise sharply relative to other industries (albeit not as rapidly as other large but not mega firms in these industries). Our findings imply that understanding rising earnings inequality during the last several decades requires understanding the restructuring of how firms organize themselves in a relatively small set of industries. Moreover, since it is the between-industry contribution that dominates, it is the common effects of re-organization across firms in the same industry that matter.Many mechanisms such as changing technology, market structure, and globalization likely underlie rising earnings inequality. The focus of future research on the impact of such changes on rising earnings inequality should be on the uneven and concentrated impact of such mechanisms across industries. The top ten percent of industries that account for virtually all of rising between-industry inequality are not randomly spread across the distribution of industries but concentrated in specific industry clusters in the tails of the earnings distribution. At the high end, dominant industries are drawn from high-tech and STEM intensive industries, finance, mining, and selected industries in health. At the low end, dominant industries are drawn from selected industries in retail and health. Notably absent are the vast majority of industries in manufacturing. The top thirty industries are in industry clusters that have exhibited structural transformations that have been the subject of independent study. Our findings imply that the role of inter-industry earnings differentials and the changing composition of employment across industries is much more important for understanding earnings inequality than suggested by the recent literature.

“….We provide detail about these thirty industries in Table 3 (the industries in Table 3 are sorted by NAICS). The largest contribution is from Restaurants and Other Eating Places (7225), which alone accounts for 16.9% of between-industry variance growth. The second-largest contribution occurs among Other General Merchandise Stores (4529), which accounts for 6.8%. While the most important two industries to increasing inequality tend to offer low-paying jobs, the other three industries that account for more than 5% of between-industry variance growth are high-paying: Software Publishers (5112), Computer Systems Design (5415), and Management of Companies (5511). What about the other 271 4-digit NAICS industries? The contributions of these industries to between-industry variance growth are summarized in Table 2. There are 145 industries that each contribute approximately 0.0% (to be precise, greater than −0.05% and less than 0.05%) to between industry variance growth. This says that almost half of all 4-digit NAICS industries contribute essentially nothing to inequality growth. There are 71 industries that contribute between 0.05% and 1.0%, accounting for 22.3% of between-industry variance growth. These industries are basically offset by another 55 industries that have a negative contribution (< −0.05%), accounting for −20.3% of between-industry variance growth. As seen in Table 4, the top thirty industries include nineteen high-paying industries that account for 54.1% of between-industry variance growth, and eleven low-paying industries that account for 44.1% of between-industry variance growth. The other 271 industries that have small contributing and offsetting contributions to increasing inequality do not occur systematically among high-paying vs. low-paying industries. 146 high-paying industries account for 1.3% of between-industry variance growth, and 125 low-paying industries account for only 0.6% of between-industry variance growth….”

“…Are there common characteristics that underlie the top ten percent of industries that contribute to between-industry variance growth? The top thirty industries reflect a small number of industry clusters that are notable for undergoing structural transformations that have been the subject of independent analysis. Eleven of the nineteen high-paying industries have been defined as high-tech industries in terms of STEM intensity by Hecker (2005) and Goldschlag and Miranda (2016).13 These innovative industries in combination account for about one-third of the between-industry increase in earnings dispersion. The transformation of the retail sector accounts for another one-third of the increase….”

Superstar Firms

“….Table 7 shows descriptive statistics of employment and earnings in mega firms and non-mega firms in our four industry groups. One immediate result in Table 7 is that employment has shifted over time to the top thirty industries. The employment share of the top thirty industries increased by 8.2 percentage points, with most of this increase (6.0 percentage points) among the eleven low-paying industries. The employment share of the other 271 industries analogously declined by 8.2 percentage points, with most of this decline (6.8 percentage points) among the 146 high-paying industries. The substantial increase in the employment share of the top thirty industries is driven by mega firms. This is evident in both Table 7 and Figure 2. The employment share of the eleven low-paying industries increased in every size class, with mega firms exhibiting the largest increase (2.5 percentage points). The nineteen high-paying industries had a smaller increase in employment, but most of this increase (1.4 percentage points of the 2.2 percentage point total) is accounted for by mega firms. Given the high average relative pay of mega firms in the high-paying industries (57.6 log points, or 77.9%) and the low average relative pay of mega firms (-49.2 log points, or -63.6%) in the low-paying industries, these shifts in employment to mega firms contributed to rising between-industry earnings inequality…”

John Haltiwanger Henry Hyatt and James Spletzer, "Industries, Mega Firms, and Increasing Inequality,"Institute of Labor Economics, March 2022, https://docs.iza.org/dp15197.pdf

“…A natural starting point is to group industries by their contributions to increasing inequality, which we explore in Table 2. There are five industries that each contribute more than 5% of between-industry variance growth, accounting for 40.7% of between-industry variance growth. These five industries have 8.8% of total employment. An additional twenty-five industries each contribute between 1% and 5% of between-industry variance growth, accounting for 57.4% of between-industry variance growth. In total, the top thirty industries - about ten percent of all 4-digit NAICS industries - account for 98.1% of between-industry variance growth and 39.3% of employment. As nearly two-thirds of the growth in U.S. earnings dispersion has occurred between industries rather than within them, these thirty industries account for most of increasing inequality….”

“…A growing number of studies attribute increases in earnings inequality to rising between-firm dispersion. 1 We confirm this pattern with comprehensive U.S. matched employer-employee data from 1996 to 2018. Our contribution is to explore and emphasize that rising between-firm dispersion mostly occurs at the industry level. Rising between-industry dispersion accounts for most of the overall increase in earnings inequality, and is driven by a relatively small number of industries. About ten percent of 4-digit NAICS industries account for virtually all of the increase in between-industry dispersion, while accounting for less than 40% of employment. These industries are in the tails of the earnings distribution including high-paying industries such as Software Publishing (5112) and low paying industries such as Restaurants and Other Eating Places(7225).3Remarkably, the remaining ninety percent of 4-digit industries individually contribute little to rising between-industry earnings inequality Importantly, it is increased sorting and segregation between industries, rather than between firms within industries, that primarily matters for rising earnings dispersion. We find differences in the roles of sorting, segregation, and pay premia based on whether the industries tend to be low-paying vs. high-paying. The top ten percent of industries that contribute to rising inequality include nineteen that are high-paying. These industries account for 54.1% of the increase in between-industry inequality. The top three of these are high-paying, high-tech service industries - Software Publishers (5112), Computer Systems Design (5415), and Other Information Services (5191) - and, in total, eleven of these nineteen high-paying industries are high-tech. As discussed in Oliner, Sichel, and Stiroh (2007) and Fernald (2014), these industries are characterized as the source of rapid technological advances. These industries play an outsized role in the tendency for high-paid workers to work both for high-paying firms (sorting) and with each other (segregation). More generally, we find a dominant role for segregation - employees with high worker effects concentrated among each other - in the contribution of these nineteen high-paying industries to increasing inequality. Eleven low-paying industries are in the top ten percent of industries that dominate rising earnings inequality. These industries in combination account for 44.1% of the increase in between-industry inequality. More than one-fourth of the increase is accounted for by just three of these eleven: Restaurants and Other Eating Places (7225), Other General Merchandise Stores (4529), and Grocery Stores (4451). These industries have gone through substantial changes in recent decades, moving away from single establishment firms to large, national chains, see Foster, Haltiwanger and Krizan (2006), Foster et al. (2016), and Autor et al. (2020). In all three of these industries, sorting provides the largest contribution to rising inequality. The dominant role of sorting holds more generally among the eleven low-paying industries that have contributed to rising inequality. A distinctive feature of the dominant ten percent of industries is that they exhibit a sharp increase in the share of employment at mega firms, which we define as firms with more than 10,000 employees. Strikingly, the remaining ninety percent of industries exhibit small declines in the share of employment at mega firms. For the low-paying dominant industries, there is a sharp decline in the earnings of mega firms relative to earnings of the average industry (averaging over all 301 industries). This sharp decline is accompanied by a decline in the size-earnings premium within these low-paying industries. For the high-paying dominant industries, the mega firms experience a substantial increase in earnings relative to both small firms in the same industry and to earnings of the average industry. Thus, we find that the rise in “superstar” firms (see, e.g., Autor et al. (2020)) is concentrated in these dominant industries with accompanying systematic changes in the size-earnings premia…”

Data and descriptive statistics

“…We use Longitudinal Employer-Household Dynamics (LEHD) linked employer-employee data, which is created by the U.S. Census Bureau as part of the Local Employment Dynamics federal-state partnership. The LEHD data are derived from state-submitted Unemployment Insurance (UI) wage records and Quarterly Census of Employment and Wages (QCEW) data. Every quarter, employers who are subject to state UI laws - approximately 98% of all private sector employers, plus state and local governments - are required to submit to the states information on their workers (the wage records, which record the quarterly earnings of every worker in the firm) and their workplaces (the QCEW, which provides information on the industry and location of each establishment). The wage records and the QCEW data submitted by the states to the U.S. Census Bureau are enhanced with census and survey microdata in order to incorporate information about worker demographics (age, gender, and education) and the firm (firm age and firm size). Abowd et al. (2009) provide a thorough description of the source data and the methodology underlying the LEHD data. A job in the quarterly LEHD data is defined as the presence of an worker-employer match, and earnings is defined as the amount earned from that job during the quarter. Because states have joined the LEHD program at different times, and have provided different amounts of historical data upon joining the LEHD program, the length of the time series of LEHD data varies by state. We use data from the 18 states that have data available from 1996:Q1 through 2018:Q4, which gives us annual data for 23 years. We restrict the LEHD data to jobs in the private sector….”

Evidence

“…The statistics in Table 1 demonstrate that 4-digit NAICS industry accounts for almost two-thirds of the growth of earnings variance. In this section, we present a descriptive analysis to learn where in the earnings distribution industry is important. We first estimate annual earnings for each of the percentiles 1 to 99 for the first (1996-2002) and the third (2012-2018) 7-year intervals, and then calculate the difference between the first and third intervals for each percentile.10 In our analytical sample, comparing the first and the third intervals, annual earnings declined by more than five log points for the first 34 percentiles, and declined for the first 61 percentiles (Figure 1(a)). However, earnings at the top increased substantially. Earnings in the top 23 percentiles increased by more than 5 log points (5.1%), and earnings in the top 13 percentiles increased by more than 10 log points (10.5%)….”

In a new paperwe document an important determinant of how much an employer pays: the industry. An employer’s industry is determined by its production processes or its products. Pay differentials across industries are large and have widened over time, leading to increases in labour earnings inequality. For example, restaurants tend to be low-paying employers. Our data indicate that in recent decades in the US, about one-tenth of the rise in labour earnings inequality can be attributed to restaurants employing more people and paying them less. We use matched employer-employee data for 18 US states to study labour earnings inequality (hereafter, simply ‘inequality’; note that this excludes self-employment and capital income). Thirty industries (out of a total of 301) can account for nearly all of the rise in inequality from the late 1990s to the late 2010s. In these 30 industries, ‘mega firms’ (which employ 10,000 or more) have seen a massive surge in employment. This column describes some of our main findings. Industries drive increasing inequality. Figure 1 reports our measure of inequality, which is the variance of log (annual) labour earnings. The contributions of different industries to between-industry inequality are shown in Figure 2. Recall from Figure 1 that between-industry inequality (the green region of Figure 1) accounts for 61.9% of the total increase in inequality. The 30 industries that contribute most to increasing inequality explain nearly all (98.1%) of the between-industry contribution to inequality - despite employing less than 40% of the workforce. Of these 30 industries, 19 are high-paying and account for 54.0% of the increase in between-industry inequality; 11 are low-paying and account for 44.1% of this increase. The other 271 industries account for only 1.9% of rising inequality. Is increasing inequality driven by changes in employment or earnings? Figure 3 shows that there are sizeable differences in the answer to this question depending on whether the industry is high- versus low-paying. Among high-paying industries, 83.9% of the contribution to inequality is due to strong increases in earnings, while any employment increases in these industries only account for only 16.1% of this rise. In contrast, the employment among the 11 most important low-paying industries surged, accounting for 68.3% of their contribution to between-industry inequality. A much more modest decline in earnings in these low-paying industries explains the remaining 31.7%. Industries drive recent changes in inequality in the US. Thirty out of the 301 industries in our classification can explain nearly all of the rise in between-industry inequality. Continued study of these industries can help shed light on the mechanisms by which inequality continues to increase.

We now address the question of how much of increasing inequality is due to employment versus earnings. Note that changes in either employment or earnings can contribute to earnings inequality. For example, restaurants tend to be among the lowest-paying employers. When (relatively) more workers are employed by restaurants, inequality will increase. A decline in the pay of this industry would also increase inequality.

Is increasing inequality driven by changes in employment or earnings? Figure 3 shows that there are sizeable differences in the answer to this question depending on whether the industry is high- versus low-paying. Among high-paying industries, 83.9% of the contribution to inequality is due to strong increases in earnings, while any employment increases in these industries only account for only 16.1% of this rise. In contrast, the employment among the 11 most important low-paying industries surged, accounting for 68.3% of their contribution to between-industry inequality. A much more modest decline in earnings in these low-paying industries explains the remaining 31.7%.

Following the pathbreaking work of Abowd et al. (1999), Card et al. (2013), and Song et al. (2019), we consider the question of the extent to which changes in inequality are driven by workers or firms, which is especially important for understanding the contribution of high-paying firms and industries to inequality. Do high-paying firms offer larger premia? Or are they hiring highly paid (i.e. more costly) workers? As Song et al. (2019) demonstrate, there are three channels through which differences among employers contribute to inequality. These are:

1. Pay premia: some firms offer greater earnings to any worker

2. Sorting: high-paying firms employ more highly paid workers

3. Segregation: more highly paid workers concentrate among each other

We explore the relative contributions of these three phenomena to between-industry inequality in Figure 4. Sorting has the greatest role in increasing inequality, and its contribution to the variance of log annual labour earnings increased from 0.078 to 0.112 from 1996-2002 to 2012-2018. In other words, high-paying industries increasingly employ highly paid workers. Highly paid workers tend to be employed in the same industries (to the exclusion of workers with low earnings), and this rises from 0.059 to 0.089. Industry-level pay premia have also widened somewhat over time. This latter channel has had a smaller contribution to the variance of log earnings and rose from 0.033 to 0.044.

‘Mega firm’ employment has surged in the 30 industries that drive increasing inequality

We now explore the role of ‘mega firms’ in increasing inequality. By mega firms we mean those that employ at least 10,000 workers. Mega firms have been increasing as a share of employment, as documented by Bloom et al. (2018) and Autor et al. (2020). In Figure 5, we break this increase down according to our four industry groups.

The rise in mega firm employment is quite dramatic in the 30 industries that drive increasing inequality. It is especially apparent in the low-paying firms that drive inequality through increasing employment. The employment share of mega firms in the 11 low-paying industries that drive between-industry inequality increased by 2.5 percentage points, from 3.1% to 5.6%. Note that this implies that millions of additional workers were employed in mega firms in low-paying industries: the total number of people employed in the US in 2018 was more than 150 million (Bureau of Labor Statistics 2019). The employment share of mega firms in the 19 high-paying industries increased by 1.4 percentage points, from 3.2% to 4.5%. The employment share of mega firms in the other 271 industries fell.

Industries drive recent changes in inequality in the US. Thirty out of the 301 industries in our classification can explain nearly all of the rise in between-industry inequality. Continued study of these industries can help shed light on the mechanisms by which inequality continues to increase.

References

Abowd, J, F Kramarz and D Margolis (1999), “High Wage Workers and High Wage Firms”, Econometrica 67(2): 251-333.

Autor, D, D Dorn, L F Katz, C C Patterson, and J Van Reenen (2020), “The Fall of the Labor Share and the Rise of Superstar Firms”, Quarterly Journal of Economics 135(2): 645-709.

Berlingieri, G, P Blanchenay, C Criscuolo (2017), “Great Divergences: The growing dispersion of wages and productivity in OECD countries”, VoxEU.org, 15 May.

Bloom, N, F Guvenen, B S Smith, J Song and T von Wachter (2018), “The Disappearing Large-Firm Premium”, AEA Papers and Proceedings 108: 317-322.

Bloom, N, T Hassan, A Kalyani, J Lerner and A Tahoun (2021), “How disruptive technologies diffuse”, VoxEU.org, 10 August.

Card, D, A Cardoso and P Kline (2016), “Bargaining, Sorting, and the Gender Wage Gap: Quantifying the Effect of Firms on the Relative Pay of Women”, Quarterly Journal of Economics 131(2): 633-686.

Card, D, J Heining and P Kline (2013), “Workplace Heterogeneity and the Rise of West German Wage Inequality”, Quarterly Journal of Economics 128(3): 967-1015.

Cooper, Z, S Craig, M Gaynor and J Van Reenen (2019), “The Price Ain’t Right? Hospital Prices and Health Spending on the Privately Insured”, Quarterly Journal of Economics 134(1): 51-107.

Decker, R A, A Flaaen and M D Tito (2016), “Unraveling the Oil Conundrum: Productivity Improvements and Cost Declines in the U.S. Shale Oil Industry”, FEDS Notes.

Dey, M, S Houseman and A E Polivka (2006), “Manufacturers’ Outsourcing to Employment Services”, Working paper.

Dey, M, S Houseman and A Polivka (2010), “What Do We Know About Contracting Out in the United States? Evidence from Household and Establishment Surveys” in K G Abraham, J R Spletzer and M Harper (eds), Labor in the New Economy, Chicago: University of Chicago Press.

Dorn, D, J Schmieder and J Spletzer (2018), “Domestic Outsourcing in the United States”, Unpublished.

Fernald, J (2014), “Productivity and Potential Output before, during, and after the Great Recession”, NBER Macroeconomics Annual 29(1): 1-51.

Foster, L, J Haltiwanger and C J Krizan (2006), “Market Selection, Reallocation, and Restructuring in the U.S. Retail Trade Sector in the 1990s”, Review of Economics and Statistics 88(4): 748-758.

Foster, L, J Haltiwanger, S Klimek and C J Krizan (2016), “The Evolution of National Retail Chains: How We Got Here”, in E Basker (ed), Handbook on the Economics of Retailing and Distribution, Northampton, MA: Edward Elgar Publishing.

Fulton, B (2017), “Health care market concentration trends in the United States: Evidence and policy responses”, Health Affairs 36(9): 1530-1538.

Goldschlag, N and J Miranda (2016), “Business Dynamics Statistics of High Tech Industries”, US Census Bureau Center for Economic Studies.

Hecker, D (2005), “High-technology employment: a NAICS-based update”, Monthly Labor Review 128(7).

Kroszner, R and P Strahan (2014), “Regulation and Deregulation of the US Banking Industry: Causes, Consequences, and Implications for the Future”, in N L Rose (ed), Economic Regulation and Its Reform: What Have We Learned?, Chicago: University of Chicago Press.

Luo, T, A Mann and R Holden (2010), “The expanding role of temporary help services from 1990 to 2008”, Monthly Labor Review 133(8): 3-16.

Marin, D (2016), “Inequality in Germany: How it differs from the US”, VoxEU.org, 23 June.

Mion, G, L D Opromolla and G Ottaviano (2020), “Dream jobs: A comparison of career prospects in Portuguese firms”, VoxEU.org, 28 August.

Song, J, D Price, F Guvenen, N Bloom, and T von Wachter (2019), “Firming Up Inequality”, Quarterly Journal of Economics 134(1): 1-50.

John Haltiwanger, Henry Hyatt, Jim Spletzer, "Rising inequality: Industries and mega firms,"Center For Economic And Policy Research, May 30, 2022, https://voxeu.org/article/rising-inequality-industries-and-mega-firms

Rising inequality: Industries and mega firms

Why has earnings inequality been increasing? Evidence from an increasing number of countries indicates that where people work is of key importance (Card et al. 2013, 2016, Marin 2016, Berlingieri et al. 2017, Song et al. 2019, Mion et al. 2020). Some employers pay more than others, and these pay differentials have widened in recent decades. An important question is why employer-driven inequality has been rising over time.

In a new paper, we document an important determinant of how much an employer pays: the industry (Haltiwanger et al. 2022). An employer’s industry is determined by its production processes or its products. Pay differentials across industries are large and have widened over time, leading to increases in labour earnings inequality. For example, restaurants tend to be low-paying employers. Our data indicate that in recent decades in the US, about one-tenth of the rise in labour earnings inequality can be attributed to restaurants employing more people and paying them less.

We use matched employer-employee data for 18 US states to study labour earnings inequality (hereafter, simply ‘inequality’; note that this excludes self-employment and capital income). Thirty industries (out of a total of 301) can account for nearly all of the rise in inequality from the late 1990s to the late 2010s. In these 30 industries, ‘mega firms’ (which employ 10,000 or more) have seen a massive surge in employment. This column describes some of our main findings.

Industry-level differences drive increasing inequality

Industries drive increasing inequality. Figure 1 reports our measure of inequality, which is the variance of log (annual) labour earnings. We estimate inequality in three seven-year intervals: 1996-2002, 2004-2010, and 2012-2018. This is broken down into the inequality within firms and between firms. The between-firm component of inequality is further broken down into that which occurs between firms in the same industry versus that which occurs at the industry level.

Inequality in the US has risen over time. In Figure 1, we show that the variance of log earnings has increased from 0.794 in 1996-2002 to 0.916 in 2012-2018. The variance of log earnings is the sum of three components. Most inequality occurs within firms, and this dispersion increased from 0.512 to 0.531, accounting for 14.9% of the increase in inequality. Among firms in the same industry, dispersion increased from 0.112 to 0.140, accounting for 23.1% of the rise in inequality. Dispersion between industries increased from 0.170 to 0.245, accounting for 61.9% of the increase in inequality. Therefore, most of the rise in inequality has occurred across industries.

About 10% of industries account for all of the between-industry rise in inequality. We consider 301 industries defined using the North American Industrial Classification System (NAICS) at the 4-digit level. We classify industries based on whether they tend to pay more or less than the average, as well as their contribution to inequality. Thirty industries account for at least 1% (positive) of the rise in between-industry inequality. Another 271 industries each contribute less than 1% to between-industry inequality.

The contributions of different industries to between-industry inequality are shown in Figure 2. Recall from Figure 1 that between-industry inequality (the green region of Figure 1) accounts for 61.9% of the total increase in inequality. The 30 industries that contribute most to increasing inequality explain nearly all (98.1%) of the between-industry contribution to inequality - despite employing less than 40% of the workforce. Of these 30 industries, 19 are high-paying and account for 54.0% of the increase in between-industry inequality; 11 are low-paying and account for 44.1% of this increase. The other 271 industries account for only 1.9% of rising inequality.

We start with the 19 high-paying industries. At least ten of these industries have been defined as high-tech in terms STEM intensity according to the criteria of Hecker (2005) and Goldschlag and Miranda (2016). Two high-paying industries are in mining, including support activities such as drilling oil wells. The contribution of these industries is likely related to the shale oil boom (e.g. Decker et al. 2016) Five high-paying industries involve finance, insurance, or corporate headquarters. The outsized role of these industries may reflect restructuring and consolidation that have followed financial deregulation (e.g. Krozner and Strahan 2014).

Health care and social assistance includes both low-paying and high-paying industries. Two high-paying industries include physician offices and hospitals, in which considerable consolidation has occurred (e.g. Fulton 2017, Cooper et al. 2019). Of the three low-paying industries, two deal primarily with care for the elderly, including both in-home care (e.g. hospice) and retirement homes. The remaining low-paying industry, Individual and Family Services, includes adoption and foster care, services for persons with disabilities, and crisis hotlines.

The nine remaining low-paying industries can be divided into roughly two groups. First, there are two industries that provide support to businesses and facilities. This includes temporary help, where employment has increased (e.g. Luo et al. 2010), as well as Professional Employee Organisations (Dey et al. 2006). It also includes cleaning and other support services in which there has been a substantial amount of outsourcing (e.g. Dorn et al. 2018). The remaining six industries include restaurants, retail trade, and gyms, which have been transformed in recent decades with the rise of national chains (e.g. Foster et al. 2016).

This note explores the substitutability between policy rate hikes and reductions in the size of the Federal Reserve's balance sheet for the removal of policy accommodation. We do so using a version of the FRB/US model augmented to incorporate the effects of changes in the Federal Reserve's asset holdings on term premiums. We illustrate how the equivalence between policy rate hikes and balance sheet reductions in the model depends on assumptions about the evolution of the balance sheet over an extended period. Overall, the model predicts that reducing the size of the balance sheet by about $2.5 trillion over the next few years, as opposed to maintaining the size at its peak level, would be roughly equivalent to raising the policy rate a little more than 50 basis points on a sustained basis. However, this estimate is associated with considerable uncertainty. Table 1 reports some key simulation statistics for 2024:Q3, an illustrative date that corresponds to the resumption of asset purchases under the full-runoff scenario. For the full-runoff scenario, total asset holdings in 2024:Q3 are roughly $0.8 trillion (2.8 percent of GDP) lower than for the baseline. The macroeconomic effects of this smaller balance sheet are offset through a small reduction in the level of the federal funds rate of only 9 basis points, reflecting the modest difference in term premiums between the two scenarios. In other words, the differences in interest rates and economic effects of allowing full runoff of the balance sheet as opposed to the capped runoff in the FOMC's plans are, by this calculation, very small. For the no-runoff scenario, the level of assets is $2.1 trillion (7.7 percent of GDP) higher in 2024:Q3 than in the baseline balance sheet scenario, a difference that widens to about 2.5 trillion by the time the balance sheet size under the baseline scenario reaches its trough in mid-2025; the policy rate is 56 basis points higher in our reference quarter. Put differently, the balance sheet reduction in the baseline scenario is equivalent to a bit more than two sustained 1/4 percentage-point hikes in the policy rate when measured against the no-runoff scenario

Edmund Crawley, Etienne Gagnon, James Hebden, and James Trevino, "Substitutability between Balance Sheet Reductions and Policy Rate Hikes: Some Illustrations and a Discussion,"Federal Reserve Board, June 3, 2022, https://www.federalreserve.gov/econres/notes/feds-notes/substitutability-between-balance-sheet-reductions-and-policy-rate-hikes-some-illustrations-20220603.htm

Substitutability between Balance Sheet Reductions and Policy Rate Hikes: Some Illustrations and a Discussion

This note explores the substitutability between policy rate hikes and reductions in the size of the Federal Reserve's balance sheet for the removal of policy accommodation. We do so using a version of the FRB/US model augmented to incorporate the effects of changes in the Federal Reserve's asset holdings on term premiums. We illustrate how the equivalence between policy rate hikes and balance sheet reductions in the model depends on assumptions about the evolution of the balance sheet over an extended period. Overall, the model predicts that reducing the size of the balance sheet by about $2.5 trillion over the next few years, as opposed to maintaining the size at its peak level, would be roughly equivalent to raising the policy rate a little more than 50 basis points on a sustained basis. However, this estimate is associated with considerable uncertainty.

Economic context

In response to the pandemic-driven recession, the FOMC lowered the target range for the federal funds rate rapidly to its effective lower bound (ELB) of between 0 and 0.25 percent. The Committee also purchased Treasury securities and agency mortgage-backed securities (MBS) to support smooth market functioning and progress toward its maximum employment and price stability goals. At the conclusion of the asset purchase program, the Federal Reserve's balance sheet size had increased by roughly $4.8 trillion (20 percent of annual GDP).

Policymakers have since turned to removing policy accommodation. In March 2022, they started raising the target range for the federal funds rate and conveyed their anticipation that ongoing increases in the policy rate would be appropriate at future meetings. In May 2022, the FOMC released its Plans for Reducing the Size of the Federal Reserve's Balance Sheet, with reductions beginning on June 1 of this year.2 Both the policy rate and balance sheet tools are expected to contribute to a firming of the stance of monetary policy.

The relationship between balance sheet reductions and policy rate hikes

In the model we use, the policy rate and balance sheet tools influence the economy primarily through their effects on the medium- and longer-term interest rates that drive economic activity.3 Policy rate actions and communications do so by affecting the cost of short-term borrowing and expectations about the path of short-term interest rates. Balance sheet policies primarily influence the term premiums embedded in medium- to longer-term yields by changing the supply—current and expected—of longer-term securities held by the public. In the model, the two tools are substitutes in terms of their ability to affect medium- and longer-term interest rates, employment, and inflation when the ELB constraint is not binding. Because of this substitutability, the model can be used to translate balance sheet reductions in terms of equivalent increases in the path of the federal funds rate that would lead to similar macroeconomic outcomes.

The model implies that a one-time permanent reduction in the Federal Reserve's holdings of 10-year equivalent Treasury securities equal to 1 percent of nominal GDP raises the term premium on a 10-year Treasury security by about 10 basis points, all else equal.4,5 In the model, this amount of policy tightening can also be achieved by raising the average expected path of the federal funds rate over the coming decade by about 10 basis points. Together, these relationships provide a simple rule of thumb for the substitutability between balance sheet reductions and policy rate hikes in the model when the economy is away from the ELB. However, as we discuss next, the translation of dollar amounts of balance sheet reductions into equivalent policy rate hikes depends on the evolution of the size and maturity composition of the balance sheet.

Illustrative balance sheet scenarios

We consider three illustrative balance sheet scenarios:

Baseline: Shrink the balance sheet starting in June 2022 in a manner consistent with the FOMC's Plans for Reducing the Size of the Federal Reserve's Balance Sheet, as captured by the illustrative baseline balance sheet scenario featured in the 2021 SOMA annual report.6 After a three-month phase-in period, let up to $60 billion in Treasury securities and up to $35 billion in agency MBS run off the balance sheet per month.
No runoff: Keep the balance sheet size constant in dollar terms by rolling over Treasury securities and agency MBS until growth in the demand for Federal Reserve liabilities provides a basis for resuming organic growth of the balance sheet.
Full runoff: Shrink the balance sheet through the full runoff of all Treasury securities as well as maturing and prepaid agency MBS starting in June 2022, without any phase-in period (about $105 billion per month, on average over the runoff period).

The top-left and top-right panels of figure 1 report the total value of assets held by the Federal Reserve in current dollar terms and as a share of nominal GDP, respectively, in the scenarios. In all three scenarios, it is assumed that the ratio of reserves to nominal GDP declines to the average level of reserves held in December 2019 as a share of nominal GDP. Net asset purchases then resume to maintain the ratio at this illustrative value and grow other Federal Reserve liabilities in line with nominal GDP, consistent with a ratio of total assets to nominal GDP of about 21 percent. The no-runoff scenario provides a counterfactual benchmark to assess the overall effect of balance sheet reductions in the other two scenarios.

Table 1 also shows that the differences in policy rate paths across scenarios are roughly equal (but with an opposite sign) to the corresponding differences in 10-year term premiums. These offsetting effects on term premiums leave the 10-year real and nominal Treasury yields little changed. However, as the addendum to table 1 shows, there is no fixed relationship in the model between dollar reductions in the balance sheet at a point in time (or, alternatively, in terms of the peak-to-trough reduction) and offsetting changes in the federal funds rate. The reason is that the effects of balance sheet policies in the model depend on the public's exposure to changes in longer-term interest rates over extended periods: The larger the public's holdings of long-duration assets, the greater the compensation that the public will require for holding the underlying duration risk. Therefore, the term premium effects of balance sheet policies in the model depend not only on the size of the Federal Reserve's asset holdings but also on how these asset holdings influence the maturity composition of assets held by the public. In the no-runoff scenario, the public holds less duration risk than in the baseline scenario through the end of the next decade, whereas the duration risk held by the public in the full-runoff scenario converges to that of the baseline scenario in 2027. Accordingly, a given dollar difference from the Federal Reserve's baseline asset holdings in the medium term is associated with larger macroeconomic effects—and, thus, calls for a larger policy rate offset—under the no-runoff scenario than under the full-runoff scenario because it has a larger effect on the public's exposure to duration risk.

Discussion of the results and related considerations

We note that balance sheet reductions in the model may not have the same effects as balance sheet expansions. At the time of runoff, the macroeconomic effects of asset holdings may be reduced because these holdings form a smaller share of nominal GDP than when they were purchased. In addition, as noted above, the effects of balance sheet actions in the model depend not only on the change in total assets, but more generally on the change in the duration of assets held by the public. For example, the duration of Treasury securities purchased in response to the Global Financial Crisis was notably longer than the duration of Treasury securities subsequently taken on by the public because of runoff.10 By contrast, in the current episode, the average duration of Treasury securities purchased by the Federal Reserve is expected to be similar to the duration of Treasury securities taken on by the public during runoff.

In a similar vein, we emphasize that our analysis only features portfolio balance effects (as captured by the duration risk channel) and, thus, does not take account of other possible transmission channels, such as those associated with flow effects or with signaling of the intended course of the policy rate.11 Notably, in the early months of the pandemic, asset purchases played an essential role in supporting smooth market functioning, an aspect not captured by the model. As policymakers turn to reducing asset holdings, market functioning and financial conditions are much improved, which, all else equal, could make the effects of asset reductions smaller than those of asset purchases. On the other hand, if financial markets were to find it difficult to absorb large amounts of runoff, then the effects on yields could be larger than we assume. Our modeling exercise also abstracts from possible complementarities between the policy rate and balance sheet tools. Researchers have notably stressed the importance of asset purchases for giving the public confidence that the federal funds rate will remain at the ELB for an extended period in response to economic downturns.12

Finally, we stress that there is significant uncertainty regarding the transmission of balance sheet and policy rate actions to medium- to longer-term interest rates, as well as the transmission of the resulting yield curve movements to the broader economy.13 For example, the term premium estimates used in the model are based on data prior to the Global Financial Crisis when interest rates were farther above the ELB than they are today.14 However, a number of authors, including King (2019) and Gagnon and Jeanne (2020), have argued that the effects of balance sheet purchases on term premiums are likely to be attenuated when medium- to longer-term interest rates are close to the ELB, in contrast to our modeling assumption that these effects are invariant to this factor. Furthermore, there is some evidence that increases in longer-term interest rates may have smaller effects on macroeconomic outcomes when they originate from increased term premiums than when they originate from increased expectations of the policy rate.15

References

Andrés, Javier, David López-Salido, and Edward Nelson (2004). "Tobin's Imperfect Asset Substitution in Optimizing General Equilibrium," Journal of Money, Credit, and Banking, vol. 36 (4),pp. 665‒90.

Bonis, Brian, Ihrig, Jane and Wei, Min (2017). "The Effect of the Federal Reserve's Securities Holdings on Longer-term Interest Rates," FEDS Notes. Washington: Board of Governors of the Federal Reserve System, April 20.

Chung, Hess, Cynthia Doniger, Cristina Fuentes-Albero, Bernd Schlusche, and Wei Zheng (2018). "Simulating the Macroeconomic Effects of Unconventional Monetary Policy," FEDS Notes. Washington: Board of Governors of the Federal Reserve System, July 20.

Doniger, Cynthia, James Hebden, Luke Pettit, and Arsenios Skaperdas (2019). "Substitutability of Monetary Policy Instruments," FEDS Notes. Washington: Board of Governors of the Federal Reserve System, July 19.

Fuhrer, Jeffrey C., and Glenn D. Rudebusch (2004). "Estimating the Euler Equation for Output," Journal of Monetary Economics, vol. 51 (6),pp. 1133‒53.

Gagnon, Joseph E. (2016). "Quantitative Easing: An Underappreciated Success," Policy Brief Number PB16-4. Washington: Peterson Institute for International Economics, April.

Gagnon, Joseph E. and Olivier Jeanne (2020). "Central Bank Policy Sets the Lower Bound on Bond Yields," Working Paper 20-2. Washington: Peterson Institute for International Economics, January.

Kiley, Michael T. (2014). "The Aggregate Demand Effects of Short- and Long-Term Interest Rates," International Journal of Central Banking, vol. 10 (4),pp. 69‒104.

King, Thomas (2019). "Expectation and Duration at the Effective Lower Bound." Journal of Financial Economics, vol. 134 (3),pp. 736‒760.

Krishnamurthy, Arvind and Annette Vissing-Jorgensen (2013). "The Ins and Outs of LSAPs (PDF)," in Global Dimensions of Unconventional Monetary Policy, proceedings of the Jackson Hole Economic Policy Symposium. Kansas City: Federal Reserve Bank of Kansas City,pp. 57-111.

Li, Canlin and Min Wei (2013). "Term Structure Modeling with Supply Factors and the Federal Reserve's Large Scale Asset Purchase Programs," International Journal of Central Banking, vol. 9 (1),pp. 3‒39.

Smith, A. Lee, and Victor J. Valcarcel (2022). "The Financial Market Effects of Unwinding the Federal Reserve's Balance Sheet," Research Working Paper no. 20‒23. Kansas City: Federal Reserve Bank of Kansas City, January.

Woodford, Michael (2012). "Methods of Policy Accommodation at the Interest-Rate Lower Bound (PDF)," in The Changing Policy Landscape, proceedings of the Jackson Hole Economic Policy Symposium. Kansas City: Federal Reserve Bank of Kansas City,pp. 185‒288.

________________________________

1. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. The authors thank Erik Bostrom, Ally Lamb, Philippa Marks, and Jake Scott for their research assistance. The note benefited from comments and suggestions by Alyssa Anderson, Jim Clouse, Rochelle Edge, Chris Gust, Ben Johannsen, David López-Salido, Trevor Reeve, Zeynep Senyuz, and Bob Tetlow. Return to text

2. The FOMC's Plans for Reducing the Size of the Federal Reserve's Balance Sheet, released on May 4, 2022, are available athttps://www.federalreserve.gov/newsevents/pressreleases/monetary20220504b.htm. Return to text

3. The model we use is a combination of the FRB/US model of the U.S. economy and a model of the Federal Reserve's balance sheet holdings and their effects on term premiums based on Li and Wei (2013). See Chung and others (2018) for a description of this augmented model and Bonis and others (2017) on the modeling of the balance sheet and associated financial effects. Return to text

4. The 10-year equivalent of a fixed-income portfolio is calculated as the par amount of on-the-run ten-year Treasury notes that would have the same par value times average duration as the portfolio under consideration; see Li and Wei (2013) for a mathematical characterization. Return to text

5. The strength of this effect in the model is within the range of estimates derived from time-series models and event studies of asset purchase program announcements. For a review of such estimates, see Gagnon (2016). Return to text

6. The assumptions underlying the report's baseline balance sheet scenario reflect the FOMC's Plans for Reducing the Size of the Federal Reserve's Balance Sheet; the median economic projections of respondents to the Survey of Primary Dealers and Survey of Market Participants conducted in March 2022 by the Open Market Trading Desk of Federal Reserve Bank of New York; and a number of illustrative auxiliary assumptions about elements such as usage of the Federal Reserve's liquidity facilities, the evolution of the Treasury General Account, and the longer-run size of the Federal Reserve's balance sheet. For details on the construction of the baseline balance sheet scenario, see Federal Reserve Bank of New York (2022), Open Market Operations during 2021, a report prepared for the Federal Open Market Committee by the Markets Group of the Federal Reserve Bank of New York, May 24, https://www.newyorkfed.org/markets/annual_reports. Because we use a different economic model and interpolation scheme, the values in the balance sheet and macroeconomic projections in our baseline scenario can differ slightly from the values reported in the annual report. Return to text

7. The degree of substitutability between balance sheet reductions and policy rate hikes in the model is somewhat insensitive to the economic projection, provided that the policy rate is not constrained by the ELB. Return to text

8. Specifically, we assume that the economic projection is underpinned by the assumptions in the baseline balance sheet scenario. Then, for each of the other two scenarios, we compute the term premium paths implied by the balance sheet side of the model when economic outcomes are identical to those in the economic projection. Treating the difference in term premium paths relative to the baseline balance sheet scenario as economic shocks, we then use the FRB/US model to find the path of the federal funds rate such that, in equilibrium, the outcomes for employment and inflation are close to those in the economic projection. In doing so, we constrain the path of the federal funds rate that supports those outcomes to be nearly parallel to the path in the baseline scenario from the simulation start in 2022:Q2 until 2030:Q4. The assumption of a uniform shift in the path of the federal funds rate offers a simple way to assess the average differences in policy rate space across scenarios over the next many years. In principle, other policy rate paths could underpin similar outcomes for employment and inflation in the model as long as these paths give rise to similar movements in medium- and longer-term real interest rates. We conduct the simulations assuming that the expectations determining financial variables, prices, and wages are model-consistent. Return to text

9. See Doniger and others (2019) for a previous illustration of the substitutability between policy rate hikes and balance sheet reductions using the FRB/US model augmented with term premium effects from the Federal Reserve's asset holdings. Our approach to computing a counterfactual policy rate path in the absence of balance sheet reductions differs modestly from the approach followed by these authors in that we consider a parallel shift in the baseline policy rate path whereas they penalize movements in the policy rate path over time, except for a one-time initial shift. The two approaches lead to similar estimates of the average difference in policy rates over the next many years across balance sheet scenarios. Return to text

10. For a discussion of the possibly smaller financial effects of balance sheet reductions than balance sheet expansions in the previous tightening episode, see Smith and Valcarcel (2022). Return to text

11. For a discussion of these other transmission channels, see, for example, Krishnamurthy and Vissing-Jorgensen (2013). Return to text

12. For an example, see Woodford (2012). Return to text

13. See Bonis and co-authors (2017) for an illustration of the sampling uncertainty attending the effect of asset purchases on term premiums in the Li and Wei (2013) model. While there is less uncertainty about the effects of policy rate decisions than balance sheet actions, the connection of policy rate actions to medium- to longer-term interest rates is loose. During the 1994‒95 tightening episode, the 10-year Treasury yield rose roughly one for one with the policy rate for a time, before retracing much of that rise. By contrast, during the 2004‒06 tightening episode, longer-term interest rates rose little as the policy rate was hiked, a situation known as the "Greenspan conundrum." In the current policy tightening episode, medium- and longer-term interest have so far risen more than the policy rate, reflecting in a large part the expectation that policymakers will raise the federal funds rate promptly toward neutral levels. Return to text

14. The subsequent period, which covers the Committee's large-scale asset purchase programs and maturity extension program, includes large public debt issuance, safe-haven flows associated with the Global Financial Crisis, the euro-area debt crisis, Brexit, and other factors that make estimation of the model challenging. Estimating the financial effects of purchases of agency mortgage-backed securities is further complicated by the difficulties in measuring the expected duration of the underlying securities because of mortgage prepayments. Return to text

15. Kiley (2014) considers both a micro-founded model with financial market segmentation (based on Andrés, López-Salido, and Nelson (2004)) and a reduced-form model (based on Fuhrer and Rudebusch (2004)). Both approaches suggest that movements in term premiums have approximately one-quarter to one-third of the effect on macroeconomic outcomes as movements in the expected path of the policy rate leading to equal changes in longer-term interest rates. Return to text

We built genome-wide polygenic scores for heritable traits, including height, body mass index, lipoprotein concentrations, cardiovascular disease, and intelligence, using summary statistics of genome-wide association studies in Europeans. Subsequently, we applied these scores to the genomes of ancient European populations. Our results revealed that after the Neolithic, European populations experienced an increase in height and intelligence scores, decreased their skin pigmentation, while the risk for coronary artery disease increased through a genetic trajectory favoring low HDLconcentrations. These results are a reflection of the continuous evolutionary processes in humans and highlight the impact that the Neolithic revolution had on our lifestyle and health.We used ancient DNA (aDNA) data of 827 ancient individuals from Western Eurasia, which were publicly available and aggregated into a database released 22 February 2019, by the David Reich lab (version 37.2) The aDNA data was used in conjunction with DNA data of 250 randomly selected modern individuals from Western Europe, which were downloaded from 1000G database Using GWAS summary statistics from GWAS catalogue and United Kingdom Biobank, which can be seen in Supplementary Table S1, we calculated PRS for each individual for various traits respectively, using a QTL p-value threshold of 10-6, in 250 kb windows. Multiple thresholds were further tested for assuring the robustness of the findings. We supplemented the effect of missing variants in a sample with the average of the entire population. We then scaled the PRS distribution to have a range of 1 to −1 before correlating them with the carbon-dated sample age. One general observation is the existence of a clear difference in the trajectories of the various traits before and after the Neolithic revolution: few changes are seen between the Early Upper Paleolithic until the Neolithic period, with a general acceleration of the evolutionary processes thereafter. Interestingly, while the period between the Early Upper Paleolithic and the Neolithic is characterized by stagnation or slight decrease in PRS related to intelligence, the genetic data show a clear increase in the scores for educational attainment, intelligence, and fluid intelligence from the Neolithic onwards, while the traits related with unipolar depression tend to decrease from that era on (Figures 3A,B). The most significant differences can be observed comparing the pre-Neolithic and Neolithic groups, as well as the post-Neolithic and modern groups, whereas the period between the Neolithic and post-Neolithic shows a very constant distribution of PRS scores (Figure 3C).In conclusion, using genetic models applied to ancient genomes, we describe important changes that have taken place in the physiological traits of modern Homo sapiens populations living in Europe since the Early Upper Paleolithic. Although the exact extent of the evolutionary changes Europeans underwent remain unclear, based on these genetic trajectories we can see that while the first 40,000 years in Europe represented a period of relative stagnation in the evolution of these traits, the spread of Neolithic populations and of the cultural complex associated with agriculture and animal domestication has led to important changes in height, weight, skin color, an increase in intellectual attainment, as well as in cholesterol metabolism and the risk for cardiovascular diseases. These new insights put into light the processes that have shaped human physiology since the migration of modern humans throughout Europe.

Yunus Kuijpers, Jorge Domínguez-Andrés, Olivier Bakker, Manoj Kumar Gupta, Martin Grasshoff, Cheng-Jian Xu, Leo A.B. Joosten, Jaume Bertranpetit, Mihai G. Netea and Yang Li, “Evolutionary Trajectories Of Complex Traits In European Populations Of Modern Humans,” Frontiers in Genetics,March 28, 2022, https://www.frontiersin.org/articles/10.3389/fgene.2022.833190/full

This paper synthesizes what we have learned about human capital into four stylized facts. Fact 1: Human capital explains a substantial share of the variation in labor earnings within and across countriesThe bottom line is that naïve cross-sectional comparisons and studies with strong quasi-experimental research designs yield very similar estimates of the economic return to education. Overall, studies that identify returns to education using instrumental variables, regression discontinuity, and other quasi- experimental approaches yield estimates of an additional year of education ranging between 6 and 18 percent, with a median in the 10-12 percent range. This is slightly higher than the 10 percent return from a “naïve” Mincer model, most likely because of some combination of measurement error and higher returns for marginal students (Card 1999). Across all OECD countries, the median earnings premium for a four-year college / tertiary education is 52 percent, or roughly 13 percent per year of education Table 1 uses data from the 1979 National Longitudinal Survey of Youth (NLSY79), which tracks a cohort of youth ages 14 to 22 in 1979 as they progress through the labor market. To estimate returns to education over the life-course, I compute the average inflation-adjusted hourly wage for individuals between the ages of 25 and 54 over multiple observations, and then regress log average hourly wages on years of education, race and gender indicators, and cognitive ability as measured by adolescent scores on the Armed Forces Qualifying Test (AFQT). Column 1 shows that the average return to a year of education over an individual’s prime working years is 10.9 percent. The R-squared of this regression is 30 percent. Controlling for AFQT to account for “ability bias” reduces the coefficient on years of education to 7.2 percent and increases the R-squared of the regression to 35 percent. Column 3 shows the average return for different levels of educational attainment, with less than high school as the left-out category. High school graduates and four-year college graduates earn an average of 13 percent and 48 percent higher wages than those with less than a high school education, respectively.These results are similar in magnitude to the quasi-experimental studies discussed above and to naïve cross-sectional estimates from other data sources. Basic measures of human capital such as education and cognitive ability can explain at least one-third of the variation in wages in a recent cohort of US workers. However, one-third is probably a lower bound for the impact of human capital on earnings, for three reasons. First, the calculation here does not include variation in education quality between workers with the same level of attainment. Quality adjustment is particularly important, because nearly all expansion of US postsecondary education over the last few decades has occurred within less-selective institutions. Carneiro and Lee (2011) estimate that the college premium would have grown an additional 30 percent between 1960 and 2000 if pre-college education quality were held constant for the marginal college graduate. Second, several studies find a larger role for human capital when it is measured in a way that includes education but also other attributes. For example, Smith et al. (2019) study the impact of owner death or retirement on private pass-through businesses and find that 75 percent of profits are attributable to the owner’s human capital, rather than physical or financial assets. Card et al. (2018) and Song et al. (2019) decompose the variance of earnings in matched employer-employee data and find that “worker effects” account for 40 percent of the variance in earnings in West Germany and 50 percent in the US respectively. Because worker effects are invariant to firm pay premia and occupational shifts by construction, we can reasonably consider them an estimate of workers’ human capital. I think the contribution of signaling is probably small, for two reasons. First, many studies find positive returns to education even when no degree or credential is earned. This is important because signaling theory requires employers to observe the signal, and most people don’t report years of education on a resume. For example, studies of compulsory schooling compare groups of students who all seek to drop out as soon as they can, but some are required to stay in school longer based on when they were born during a calendar year. Many of the youth staying in school for an extra year do not end up obtaining a high school degree at all - they drop out in 11th grade rather than 10th grade. Nonetheless, such studies show that additional education leads to gains in earnings.A school construction program in Indonesia studied by Duflo (2001) mostly worked by increasing primary school enrollment, not receipt of degrees—but still led to later gains in wages. Aryal, Bhuller and Lange (2022) cleverly exploit the differential observability of compulsory schooling laws across regions in Norway to separate returns to human capital from signaling, and find that human capital accounts for 70 percent of the private return to secondary school education. empirical support for signaling theory is scant. Clark and Martorell (2014) find no difference in earnings between high school students who barely pass or fail an exit exam, implying that there is no signaling value of a high school diploma.Some studies do find that the return to education decreases over time as employers learn workers’ true ability, which is a testable implication of the signaling model Yet a similar test of the employer learning model in a more recent cohort finds that the return to education does not diminish with experienceFact 2: Human capital investments have high economic returns throughout childhood and young adulthood. There is strong evidence supporting value of skill investments in early childhood.Perhaps best-known are two randomized evaluations of preschool interventions from the 1960s: the High/Scope Perry Preschool Project and the Carolina Abecedarian Project. These studies are from decades ago, involving small-scale and intensive interventions for highly disadvantaged families, and thus their results may not generalize to larger and more recent programs. However, several studies of more recent preschool interventions also find substantial impacts —although the evidence also provides a puzzle. Pre-K programs often provide only a short-term boost to test scores that fades out in a few years. Yet they have longer-run impacts on important life outcomes such as high school graduation and college attendance, as well as non-educational metrics like reductions in crime and teen pregnancy and improved health later in life (for example, Ludwig and Miller 2007, Deming 2009, Gray-Lobe, Pathak and Walters 2021). In addition, there are many other methods of early childhood investment: pre-natal care, early child health care, food and nutrition support, home visits to encourage practices like breast-feeding and smoking cessation, and others. In recent essays in this journal, Aizer, Hoynes, and Lleras-Muney (2022) describe the evidence of long-term benefits from policy interventions affecting low-income children like Medicaid and food stamps, while Wust (2022) presents the evidence from the Nordic countries about the benefits of universal provision of early childhood investments in pre-natal care, health care at time of birth, and early childhood health care. Again, any benefit-cost analysis of these programs needs to take a long-term view, because many of these benefits only become apparent later in life. In sum, the evidence suggests that human capital investments are, at least in rough terms, equally productive between the ages of 0 and 25. Fact 3: The technology for producing foundational skills such as numeracy and literacy is well- understood, and resources are the main constraint. The overall picture is that some specific input investments work very well, but many do not, and it is often hard to predict ahead of time. I interpret the evidence as follows. First, at least in the United States, increased school spending is productive on the margin. Increasing school spending from current levels would produce substantially more human capital and may even pay for itself in the long run. The technology for producing basic math and literacy skills in school-aged children is fairly well-understood. Smaller class sizes, better school facilities, and more instructional time all have reliable impacts on the development of foundational academic skills. The inputs with the best track record of effectiveness - high-dosage tutoring, extra instructional time, personalization and teaching to the right level - mostly deliver to students “more of the same”, rather than reinventing the learning process. Second, while sharpening incentives works in some contexts, achievement gains are often short-lived and there is not much evidence of long-run benefits. An important caveat is that resources plus incentives appear to be more effective than resources alone. Third, simply giving schools money - and allowing them to spend it flexibly - may be a more reliable way to increase human capital than pinning our collective hopes on any particular “silver bullet” approach that all schools would be required to follow.This makes education experts queasy, and rightfully so. In a perfect world, increases in resources are combined with transparency and accountability for results. Yet the evidence suggests that “helicopter drops” of money are spent well enough to be worth the investment, at least in developed countries and in schools with strong internal accountability. However, just because school spending is economically productive on the margin does not mean that the money is spent optimally. It can be simultaneously true that school spending is productive and that much of it is wasted. We can probably always do better, and so innovation and experimentation are critical for increasing the productivity of human capital investments. Fact 4: Higher-order skills such as problem-solving and teamwork are increasingly economically valuable, and the technology for producing them is not well understood

David Demings, “Four Facts About Human Capital,” National Bureau Of Economic Research, June 2022, https://www.nber.org/papers/w30149

Summery, “…Intelligence test scores and educational duration are positively correlated. This correlation could be interpreted in two ways: Students with greater propensity for intelligence go on to complete more education, or a longer education increases intelligence. We meta-analyzed three categories of quasi-experimental studies of educational effects on intelligence: those estimating education-intelligence associations after controlling for earlier intelligence, those using compulsory schooling policy changes as instrumental variables, and those using regression-discontinuity designs on school-entry age cutoffs. Across 142 effect sizes from 42 data sets involving over 600,000 participants,we found consistent evidence for beneficial effects of education on cognitive abilities of approximately 1 to 5 IQ points for an additional year of education.Moderator analyses indicated that the effects persisted across the life span and were present on all broad categories of cognitive ability studied. Education appears to be the most consistent, robust, and durable method yet to be identified for raising intelligence….”

Stuart Ritchie and Elliot Tucker-Drob, "How Much Does Education Improve Intelligence? A Meta-Analysis,"Psychological Science, 2018, https://labs.la.utexas.edu/tucker-drob/files/2019/08/Ritchie-Tucker-Drob-2018-Psych-Science-How-Much-Does-Education-Improve-Intelligence.pdf

“….Table 2 shows the results from each study design after the inclusion of multiple moderators. We included as many moderators as possible for each design, though we chose to include the “achievement” categorization of the outcome test for the school-age-cutoff design (instead of the alternative categorization involving composite, fluid, and crystallized tests, which we used for the other designs) because it had such a substantial effect in the single-moderator model. Including multiple moderators reduced the τ statistic—to a larger degree for the control-prior-intelligence design than for the others— though significant heterogeneity remained in all cases. The moderators that were individually significant (e.g., age in the control-prior-intelligence design or achievement tests in the school-age-cutoff design) were also significant in the multiple-moderator model, indicating that their effects were incremental of the other moderators that we included….”

“…We tested whether studies that included only male participants showed a differential educational effect. This was not the case for control prior intelligence (effect for the 2 male-only estimates: 2.261 IQ points, SE = 0.897, p =.012; effect for 24 mixed-sex estimates: 1.027 IQ points, SE = 0.110, p = 6.79×10−21; difference SE = 0.905, p =.173), or for policy change (effect for the 8 male-only estimates: 1.683 IQ points, SE = 0.507, p =.001; effect for 22 mixed-sex estimates: 2.215 IQ points, SE = 0.788, p =.005; difference SE = 0.941, p =.572). There were no male-only school-age-cutoff studies….”

Multiple moderators.

“….Splitting the outcome cognitive tests into three broad categories—composite, fluid, and crystallized tests—we tested whether the category moderated effect sizes. For the control-prior-intelligence design, there were stronger educational impacts on composite tests (1.876 IQ points, SE = 0.467, p = 5.94×10−05) than on fluid tests (0.836 points, SE = 0.097, p =.152), and the difference between composite and fluid was significant (1.039 points, SE = 0.496, p =.036).There was only one crystallized outcome test for the control-prior- intelligence design (1.893 points, SE = 0.348, p = 5.34×10−08), so we did not include it in the moderator comparison here. For the policy-change design, there were significant effects for both composite (2.314 IQ points, SE = 0.869, p =.008) and fluid (2.272 points, SE = 0.765, p =.003) but not crystallized (1.012 points, SE = 1.125, p =.368) tests; however, the effects on the three different categories were not significantly different from one another (all difference p values >.35). Finally, for the school-age-cutoff design, there were significant effects of a year of education on composite (6.534 points, SE = 2.433, p =.007), fluid (5.104 points, SE = 0.621, p = 2.05×10−16), and crystallized (5.428 points, SE = 0.170, p = 1.04×10−223) tests; there were, however, no significant differences between effect sizes across outcome types (difference p values >.5). We then split the outcome tests into “achievement” tests versus “other” tests. There was only one achievement test in the control-prior-intelligence design, so we did not run this analysis. For policy change, there was no significant difference in the educational effect on the 7 achievement tests (2.760 IQ points, SE = 0.968, p =.004) versus the 23 other tests (1.784 points, SE = 0.553, p =.001; difference SE = 1.011, p =.334). However, for school-age cutoff, which had the largest proportion of achievement tests (38 of the 86 tests were classed as achievement tests), achievement tests showed a substantially and significantly larger educational effect (6.231 points, SE = 0.339, p = 2.85×10−75) than other tests (3.839 points, SE = 0.412, p = 1.11×10−20; difference SE = 0.371, p = 1.19×10−10)….”

Male-only studies.

Outcome test category.

“…Figure 1 shows the effect sizes in the first two study designs as a function of the participants’ mean age at the outcome test. For the control- prior-intelligence studies, outcome age was a significant moderator: The effect size of education declined by −0.026 IQ points per year of age (SE = 0.012, p =.029). At the youngest age (18 years) the effect size of having had an additional year of education was 2.154 IQ points, whereas at the oldest age (83 years) the effect size was 0.485 IQ points. This effect was smaller but still significant if the largest effect size (> 3 IQ points for an additional year) was excluded (−0.011 IQ points per year of age, SE = 0.005, p =.018). There was no significant moderating effect of age at outcome for the policy-change studies (0.014 IQ points per year, SE = 0.022, p =.543): The effect at the youngest age (18 years; 1.690 IQ points) was not significantly different from the effect at the oldest age (71 years; 2.413 IQ points). There was comparatively little variation in the age at the outcome test for the school-age- cutoff design (SD = 1.6 years; see Table 1); there was no significant age moderation effect (−0.027 points per year, SE = 0.399, p =.947)….”

“…. For the policy-change design, we tested whether the age at which the educational policy change produced an increment in compulsory schooling moderated the intervention effect. This was not the case: The effect size increased by a nonsignificant 0.038 IQ points per year of age at the intervention (SE = 0.228, p =.867). The meta-regression model implied that at the youngest intervention age (7.5 years), the effect size was 1.765 IQ points, and at the oldest (19 years) it was 2.204 IQ points….”

Age at outcome test.

“…For the control-prior-intelligence design, we tested whether the age at which the participants had taken the initial (control) cognitive test, or the gap between this early test and the outcome test, moderated the effect size. The age at the early test, which did not vary substantially (see Table 1) was not significantly related to the effect size (−0.024 IQ points per year, SE = 0.064, p =.706). For the youngest early-test age (10 years), the meta-regression model indicated that the effect size of 1 additional year of education was 1.243 IQ points; for the oldest (16 years), the effect size was 1.099 IQ points. Conversely, the time lag between the tests was a significant moderator of the effect size (−0.031 IQ points per year, SE = 0.015, p =.033). This meta-regression indicated that at the smallest age gap (5 years), the effect size was 2.398 IQ points, whereas for the largest age gap (72.44 years), the effect size was a substantially smaller 0.317 IQ points. Note that this age gap is almost fully confounded with the age at the out- come test (r =.988), assessed as a moderator below….”

Age at intervention.

Age at early test and time lag between tests.

“…We analyzed results from the three most prominent quasi-experimental methods for testing the effects of education on intelligence. We defined intelligence as the score on a cognitive test; see below for consideration of how the test scores might relate to the underlying psychological processes. Each method implements a different approach to minimize effects stemming from selection processes. Full meta-analytic inclusion criteria are reported below, but first we describe each of the three designs, providing a canonical example of each..…. The first research design, which we label control prior intelligence, is a longitudinal study in which cognitive testing data are collected before and after variation in the duration of education. This allows the relation between education and the second test to be adjusted for by each participant’s earlier ability level. An example of this design is the study by Clouston et al. (2012), who analyzed data from three large U.S. and UK cohort studies, all of which had both an adolescent and a midlife cognitive test. Results indicated that completing a university education was linked to higher midlife cognitive ability, above and beyond adolescent intelligence….…The second design, policy change, relies on changes in educational duration that are, by all accounts, exogenous to the characteristics of the individuals. An example of this design is the study by Brinch and Galloway (2012), who used large-scale data from a 1960s educational reform in Norway. This reform increased compulsory education by 2 years; critically, it was staggered across municipalities in the country. This allowed the researchers to estimate the effect of an additional year of school on a later intelligence test, taken by males at entry to military service as part of Norway’s universal military draft. Under the assumption that the policy change affected intelligence only via increasing years of schooling, the authors used an instrumental-variables analysis to estimate the effect of 1 year of schooling on intelligence at approximately 3.7 points on a standard IQ scale (M = 100, SD = 15 in the population)….…The third design takes advantage of a school-age cutoff. These studies use regression-discontinuity analysis to leverage the fact that school districts implement a date-of-birth cutoff for school entry. The first study to use this method was by Baltes and Reinert (1969), but the most highly cited example is by Cahan and Cohen (1989), who, in a sample of over 12,000 children across three grades of the Israeli school system (between the ages of approximately 10-12 years), found that schooling exerted positive effects on all of 12 tests covering a variety of cognitive domains. These educational effects were around twice the effect of a year of age. The strict assumptions of this method are some- times not fully met (Cliffordson, 2010); methodological issues are discussed in more detail below….”

Their Evidence

“…In three separate unconditional random-effects meta- analytic models (one for each study design), we estimated the effect of 1 additional year of education on cognitive outcomes. For all three study designs, there was a significant effect of 1 additional year of education. For control prior intelligence, the effect was 1.197 IQ points (SE = 0.203, p = 3.84×10−09); for policy change, it was 2.056 IQ points (SE = 0.583, p = 4.23×10−04); and for school-age cutoff, it was 5.229 IQ points (SE = 0.530, p = 6.33×10−23). An overall model including all estimates from all three designs found an average effect size of 3.394 IQ points for 1 year of education (SE = 0.503, p = 1.55×10−11)….”

Types of Studies They Looked At