Golden Ages: A Tale of the Labor Markets in China and the United States
- Date Posted:
- Is Database:
- Database
The age of peak earnings, or “golden age,” has remained stable at 45-50 years old in the U.S. over the past 30 years, but decreased from 55 to 33 years old in China.
Over the past 30 years the age of peak earnings has been almost constant in the US (40-50) but had a major decline in the PRC from ~ 55 to 35. China has seen lower returns to human capital then the US. Three core findings, “….In this paper, we first document stark differences in the cross-sectional age-earnings profiles between the U.S. and China, the two largest economies in the world, during the past thirty years. We find that, first, the peak age in cross-sectional age-earnings profiles, which we refer to as the “golden age,” stayed almost constant at around45-50years old in the U.S., but decreased sharply from55to around35years old in China; second,the age-specific real earnings grew drastically in China, but stayed almost stagnant in the U.S.; andthird, the cross-sectional and life-cycle age-earnings profiles looked remarkably similar in the U.S., but differed substantially in China….”
Hanming Fang and Xincheng Qiu, "“Golden Ages”: A Tale of the Labor Markets in China and the United States," National Bureau Of Economic Research, November 2021, https://www.nber.org/papers/w29523
Korea’s path“…Is this a realistic prediction? Only history will tell for sure, but interestingly, Figure11shows that such a pattern of increasing “golden ages” actually happened in Korea during the past ten years, using data from the Korean Labor and Income Panel Study (KLIPS). Korea experienced its fastest growth during the1960s to1990s. After that, it began to slowdown. Appendix Figure A.6depicts the decomposition for Korea, together with the decomposition for U.S. and China. It is worth noting that the cohort effects are particularly large from cohort1945to cohort1960, but starts to decelerate afterwards. This is consistent with our explanation of the race between inter-cohort productivity growth and returns to experience. As inter-cohort productivity growth starts to give its way to experience in Korea, the “golden age” comes back to older ages, as in our hypothetical scenario in Figure10…”
Potential implications, “…We also use the inferred components, particularly the series of the quantities of human capital, to revisit several important and classical applications in macroeconomics and labor economics, including the growth accounting and the estimation of the TFP growth, and the college wage premium and the skill-biased technical change. We find that once we adjust for the changes in the quantities of human capital, the estimated contribution of the TFP to GDP per capita growth is smaller than the previous estimates in the literature. We also find that the skill-biased technical change played an important role in the rising college premium to ensure that the relative price of college human capital does not drop as much as it would otherwise do when there is a large increase in the quantity of college human capital. A simple simulation exercise using our framework also suggests that, as the Chinese economy slows down to a “new normal” growth rate similar to that in the U.S., the golden ages of the cross-sectional age-earnings profile in China will start to increase to older ages, similar to what has happened in Korea in the last ten years….”
Theory, “…To explain these striking differences, we propose and empirically implement a unified decomposition framework to infer from the repeated cross-sectional earnings data the life-cycle human capital accumulation (the experience effect), the inter-cohort productivity growth (the cohort effect), and the human capital price changes over time (the time effect), under an identifying assumption that the growth of the experience effect stops at the end of one’s working career. The decomposition suggests that China has experienced a much larger inter-cohort productivity growth and higher increase in the rental price to human capital compared to the U.S.; but the return to experience is higher in the U.S….”
Datapoint suggesting underlying talent might be constrained, “…The contributions of relative labor supply, relative human capital per worker, and skill-biased technical change to the evolution of relative human capital prices are depicted in Figure9. Itshows clearly that in both U.S. and China, the relative quantity of college human capital grows rapidly, which would have led to sharp declines in the price of college human capital relative to non-college human capital. Due to skill-biased technical changes, the relative price of collegehuman capital did not decline even more in the U.S. and actually increased in China in the last thirty years….”
“…For example, college and high school graduates may possess different types of skills that are not perfect substitutes. To do so, we perform the decomposition as discussed in Section4separately for college workers and high school workers. College and high school workers are allowed to have different paths of life-cycle human capital growth, different inter-cohort human capital growth, and different time series of human capital price changes. The only restriction is that for both college workers and high school workers, there is no additional skill accumulation from experience in the last two experience bins towards the end of working life. Since our imputation of potential experience assumes that college graduates start to gain experience from22years old and high school graduates start to gain experience from18years old, effectively it is assumed that college graduates do not have additional returns to experience in52-61years old and high school graduates in48-57years old. This is largely overlapped with the “flat spot” proposed by Bowlus and Robinson (2012). After detailed investigation of the U.S. data, they conclude that a reasonable choice for the flat spot of the experience effect is around50-59for college graduates and46-55for high school graduates The results are presented in Figure7. First, within an education group, the returns to experience are still higher in the U.S. than in China. Within a country, the experience effects are larger for college workers than high school workers. This is consistent with findings documented by the previous literature that life-cycle wage growth tends to be faster for workers with more education (see Bagger et al., 2014, for example). The difference between the two education groups in their experience effect profiles, however, is much smaller compared to the difference in the cohort effects that we are turning to….”
Controlling For Education The Returns To Experience Are Still Higher In US Relative TO PRC
“…We present the contribution of each source — physical capital per worker, human capital per worker, and the residual — to the growth of GDP per worker in Figure5. We find that all three sources contribute almost equally to the U.S. growth, with the contribution of human capital slightly exceeding the other two sources. The picture is quite different in China. Although the absolute level of the growth in human capital is larger in China than in the U.S., the relative contribution of human capital turns out be the least important to China’s growth. But this is merely a result of an even faster speed at which the physical capital and TFP grow in China. In fact, physical capital is responsible for almost60% of the growth in GDP per worker, and TFP for almost another30% in China….”
Chinese GDP Per Worker Driven By Physical Investment, Not Human Capital Or TFP
“…Conceptually, across-sectionalage-earnings profile, which summarizes earnings of workers of different ages at a given point of time, is a different notion to thelife-cycleearnings profile, which tracks the earnings of a typical person over his life course. Thus one should not expect the crosssectional age-earnings profiles to coincide with the life-cycle ones. In Figure3, we reproduce the cross-sectional profiles from Figure1on the left, and plot the life-cycle earnings path of various birth cohorts on the right, with each curve representing a10-year cohort bin. The top panel is for the U.S., and the bottom panel for China. In the U.S. (Figure3a), cohorts with year of birth expanding half a century share remarkably similar life-cycle earnings paths. Furthermore, life-cycle profiles on the right of Figure3a closely resemble the cross-sectional profiles on the left (which is reproduced from the right panel of Figure1a), in both its shape and level. In a stationary environment where the life-cycle profile does not vary across cohorts, the cross-sectional profiles and the life-cycle profiles essentially coincide with each other. In such an economy, a30-year-old worker who wants to predict his (real) earnings10years later can simply take a look at the contemporary earnings of a40-year-old worker. This provides a justification for the voluminous prior literature that use cross-sectional profiles as approximations to life-cycle patterns. Although conceptually it is not correct to interpret crosssectional age-earnings profiles as life-cycle patterns, in practice they are close to each other for the U.S. case. In other words, stationarity is an reasonable assumption when studying the U.S. earnings profiles. However, as shown in Figure3b, the life-cycle patterns of different cohorts differ drasticallyfor China. More recent cohorts enjoy both much higher earnings and steeper life-cycle earnings growth. These life-cycle profiles also demonstrate no resemblance at all to the cross-sectional profiles, although they are actually linked to each other. Note that both the left and the right panels are just different ways to visualize the same underlying data.13It is perhaps not surprising that in a fast-growing economy such as China, stationarity is not a valid approximation…”
Cross Section versus Life Earnings Age-Earnings Profiles
“…To sum up, Figure2plots the evolution of the cross-sectional “golden ages” in the U.S. and China during1986-2012. For each country and each year, we run a kernel regression of log earnings on age to predict age-specific earnings, and obtain an estimated golden age in that year as the age achieving the maximal predicted earnings. Furthermore, we fit a linear time trend of the estimated golden age for each country. Figure2shows clearly that in the U.S., the golden age has stayed constant at around48years old in the past thirty years, while in China there exhibits a strong downward trend in the golden ages from1986to2012, decreasing from more than55years old to around35years old….”
PRC, “…In Figure1b, we plot the cross-sectional age-earnings profile for Chinese male workers, using the same procedure as discussed before. There are several striking contrasts between Figure1a and Figure1b. First, Chinese workers have experienced a dramatic increase in real earnings in the past30years for all age groups. It is reflected in the large vertical upward shifts of the ageearnings profiles for later cross sections. The earnings of Chinese urban male workers increased by nearly six folds. This is in marked contrast to the earnings stagnation in the U.S. Second, while the shape of the cross-sectional age-earnings profiles and hence the corresponding “golden ages” have stayed more or less constant in the U.S., the “golden age” in China is continuously evolving to younger ages. Prior to2000, the age-earnings profiles of China had a familiar hump-shape with the “golden age” at around55, although there already were some signs of a declining “golden age” in1996-2000. Between2001and2004, the age-earnings profile is almost flat and humps at around age40-45. After2005, the “golden age” is35years old…”
US, “…Figure1a depicts the cross-sectional age-earnings profiles for male workers in the U.S. Each curve represents a cross section that pools five or four adjacent years. In the construction of each curve, we first perform a nonparametric kernel regression of annual labor earnings on age separately for each cross section, where the Epanechnikov kernel function and rule-of-thumb bandwidth estimator are applied, and then display the smoothed values with the95% confidence intervals. To avoid potential impacts of extreme values, we drop outliers defined as earnings in the top2.5% and bottom2.5% in each year. We normalize all earnings to the2015year using CPI. Individuals are weighted by the person-level ASEC weight. Figure1a reveals that, first, the “golden age” in the U.S. is relatively stable at around50years old during the past three decades;second, the U.S. has witnessed little growth in age-specific mean real earnings. That is, both the shape and the level of the age-earnings profiles are largely unchanged…”
The Evidence
Ed Comment:“We saw in recent paper that the quality of researchers declines as the share of researchers increases. For the same reason (.i.e., a shortage of talent) every increase in a research is a decrease in the quality of people employed to other critical endeavors such as commercialization.”
New paper replicates Bloom'sAre Ideas Getting Harder To Findfor China and Germany and finds evidence of a decline in research productivity in both countries providing support to Bloom's work
What they did, “….Following Bloom et al., we calculate the research productivity parameter,𝛼𝛼, in equation (1),by taking the average of output growth per firm and decade (1990s, 2000s, and 2010s), and dividing by average input levels. As measures for output we use sales revenue, employment, revenue labor productivity, and market capitalization (monetary units deflated by the GDP implicit price deflator). Market capitalization is not available for Germany’s predominantly privately owned companies and we substitute it with sales revenue from innovative products and services. Regarding inputs, Bloom et al. (2020) show theoretically that research inputs in (1) can be measured by𝑆𝑆̃𝑡𝑡, the“effective number of researchers”, by deflating a firm’s R&D expenditures, 𝑆𝑆with the nominal wage rate for high-skilled workers in the economy…”
Bottom line, “….Table 1 depicts our results. In Germany, the effective number of researchers grows at an annual rate of 1.5% to 4.9%. Like Bloom et al.’s findings for the U.S., however, such input growth is not met with a proportional growth in output.As a result, we find declines in research productivity ranging from3.7% to 7.8% per year. The average of the four estimates, equal to -5.225%, implies that research productivity halves every fourteen years, which is very close to the estimated halflife of thirteen years for the U.S. (Bloom et al., 2020). In China, we observe an extremely rapid expansion of research activities during the first and second decades of the 21st century, with growth rates for effective researchers ranging between 21% and 24%.5 The resulting output growth, again, is not proportional to such inputs, which is reflected in a decrease in research productivity estimated between 15.4% and 29.3%. Averaged across estimates, this amounts to a decline of -23.775% per year, or a half-life of around 3 years….”
Note they theorize that China might see a quicker decline in research productivity due to internal constraints, “….Overall, ideas are not only getting harder to find in the U.S., but that the same holds true for the largest R&D-spending countries in Europe and Asia respectively. Although estimates are difficult to compare, due to differences between data sources, negative growth rates are, in fact, remarkably similar across Germany and the U.S. China has undergone an even larger decline in research productivity in the last two decades, which reflects its rapid transformation from principally capital-driven growth toward more innovation-led growth. It remains to be seen whether China will start to follow productivity trends of advanced economies. The increasingly inward looking and mission-driven nature of Chinese innovation policy (Chinese State Council, 2020), however, suggests that research productivity might continue to decline faster in China than elsewhere. Knowledge production at the technology frontier crucially relies on creative freedom, serendipitous discovery, and exchange…”
