Decomposing Effects of Population Aging on Economic Growth in OECD Countries^*

In recent years, most developed countries have been experiencing high rates of population aging. A number of researchers have found that population aging has negative effects on economic growth. For example, using a sample of 142 countries for the period 1960–2017, Lee and Shin (2019) find that population aging leads to lower growth of per capita GDP in both the short run and the long run, particularly in more aged countries, which are mostly developed countries. Similarly, Cooley and Henriksen (2018) find that population aging is a major driver of slower growth in most advanced countries. Utilizing state-level U.S. data for the period 1980–2010, Maestas, Powell, and Mullen (2016) also find that state-level population aging decreases the growth rate of the state output per capita.

Theoretically, the negative effects of population aging on economic growth operate through a number of channels. First, aging may lead to diminishing working population, which lowers labor input growth. Bloom and Williamson (1998) argue that the demographic dividend—the demographic transition resulting in much-faster growth of working-age population—was essential to East Asian countries’ so-called economic miracle: its extraordinary rapid growth. Hence, population aging that reverses this demographic transition is expected to lower economic growth by imposing a significant demographic deficit.¹

Second, aging, by lowering the saving rate (Park and Shin 2012; Horioka and Niimi 2017), can lead to less capital accumulation and hence a slowdown of growth. The theoretical background for the relationship between aging and saving goes back to the life-cycle hypothesis pioneered by Modigliani and coauthors.² According to them, because the aged rely on dissaving of wealth accumulated at earlier ages, an increase in the share of aged population lowers the saving rate.

Third, aging can also lower labor productivity growth. In a sample of 87 non-oil-exporting countries, Feyrer (2007) finds that productivity is highest for the workforce between the ages of 40 and 49 years. Aging, by decreasing this proportion of workers, lowers the aggregate productivity growth. Maestas, Powell, and Mullen (2016) and Aiyar, Ebeke, and Shao (2016) confirm that aging lowers the aggregate productivity growth based on state-level U.S. data and country-level European data, respectively.

Finally, aging can also have negative effects on the growth rate of total factor productivity (TFP). According to Jones (2010), aged people are less innovative and hence aged economies suffer from lower technological progress. Derrien, Kecskés, and Nguyen (2018) and Aksoy et al. (2019) provide supporting evidence based on the local labor force in the United States and OECD country samples, respectively. Engbom (2019) also develops a model that shows that a more aged labor force leads to a decline in firm and worker dynamics, which reduces economic growth due to less creative destruction.

However, there are also studies providing evidence that the impact of aging on economic growth is not so adverse. For example, using a panel data set for the period 1960–2005, Bloom, Canning, and Finlay (2008) find that the negative effect of old age on growth is minimal; they explain that the negative effects of diminishing working population can be mitigated by behavioral responses such as higher savings for longer lifetime, greater labor force participation, and increased immigration from foreign, relatively younger, countries. Acemoglu and Restrepo (2017) even further argue that population aging can promote higher growth by encouraging more active adoption of automation technology and provide some supporting empirical evidence.

Lee and Shin (2019) explain these contradictory findings by nonlinear effects of aging on growth. They argue that the negative effects of population aging on economic growth are not realized until the old dependency ratio reaches a certain high level; this is because when the rate of population aging is relatively low, an increase in the rate of population aging (in terms of old dependency ratio or old-age population share) does not coincide with a decrease in the share of working-age population. Only when the rate of population aging reaches a certain high level does increasing population aging coincide with a decrease in the share of working population, which then leads to the slowdown of economic growth rate. Eggertson, Lancastre, and Summers (2019) also show that in a more recent period (2008–15) and for different reasons, compared with the period 1990–2008, the impact of aging on output growth per capita became negative.³

In this paper, we identify the channels through which aging affects output growth per capita. We focus on 35 OECD countries in which population aging is more advanced and hence the negative effects of aging on output growth per capita are more pronounced.⁴ Another reason to focus on OECD countries is data restrictions: The labor force participation rate that is essential in our empirical study is not widely available for non-OECD countries.

There are researchers in the literature who decompose output growth into several channels and investigate how the determinants operate. For example, Wong (2007) decomposes output growth per capita into capital accumulation, human capital accumulation, and TFP growth and finds that TFP growth is the main channel through which determinants of output growth per capita operate. In particular, Maestas, Mullen, and Powell (2016) decompose state output growth per capita into growth in labor productivity and labor force growth and focus on the role of aging as a determinant of growth. They find that two-thirds of the negative effect of aging works through slower growth in the labor productivity and the remaining one-third arises through slower labor force growth. And, as noted earlier, whereas previous researchers had examined output growth per capita, Feyrer (2007) and Aiyar, Ebeke, and Shao (2016) focus on labor productivity growth and both find that aging of the workforce by decreasing the proportion of the workforce between the ages of 40 and 49 years lowers labor productivity growth mainly through TFP channel.

In this paper, we extend the previous studies in several dimensions. First, while most previous researchers consider three channels, we decompose GDP growth per capita into six: changes in (i) capital accumulation; (ii) human capital accumulation; (iii) average working hours; (iv) labor participation rate; (v) age composition; and (vi) TFP growth. By adding (iii), (iv), and (v), we can further investigate whether the negative effects of population aging on economic growth can be mitigated by increases in average working hours or labor force participation rate. We can also quantify the importance of demographic deficits by examining the impact on changes in age composition.

Second, while a number of previous researchers who focused on labor productivity were mostly interested in demographic changes in the workforce, we are more interested in the age distribution of the whole population, specifically, the old and youth dependency ratios (or the old and youth population shares). For example, Feyrer (2007) emphasizes that the share of cohorts in its 40s contributes most to the aggregate labor productivity, and Aiyar, Ebeke, and Shao (2016) find that the workforce share aged 55–64 years negatively affects labor productivity growth.

Finally, whereas most previous researchers on population aging's contemporaneous effects, we focus on its future effects. That is, previous researchers often measured the impact of aging by change in the old population share (or the old dependency ratio) in the same interval, that is, the change in the old population share from the beginning to the end of the period. Thus, they investigated the impact of rapid increase in the old population share on the growth rate of labor productivity during the same period; in their doing so, however, there was a high possibility of reverse causality.

In contrast, in the current study, we measure the extent of aging by the old population share (or the old dependency ratio) at the beginning of the period and examine its impact on future intermediate-run growth (i.e., 10-year average growth rate). In this way, we minimize the possibility of reverse causality. However, we also show the robustness of our results using instrumental-variable estimations.

We first confirm that aging in OECD countries has negative effects on GDP growth per capita. We then find that the most important channel through which the negative effects operate is lowered TFP growth; in many of our empirical specifications, the negative impact of aging on GDP per capita growth is explained solely by the channel of lowered TFP growth. We also find evidence of demographic deficit that works through lowered growth of age composition of the working population (ages 15 to 64 years), but this negative effect is more than neutralized by increases in the growth rates of average working hours and the labor force participation rate. In contrast, we find little evidence of a negative impact of aging on physical and human capital accumulations. While we examine intermediate-run growth, because TFP growth rate can be permanently lowered, population aging's negative effects on GDP growth per capita are expected to be permanent.

The remainder of this paper is organized as follows. Section 2 explains our empirical specification for the six channels through which population aging affects output growth per capita. Section 3 provides main empirical results. Section 4 concludes.

We collect GDP, capital stock, human capital stock, average working hours, and TFP from the Penn World Table (PWT) 9.1.⁸ Population and its age distribution are collected from the United Nations population database. The old-age dependency ratio is retrieved from the World Bank's World Development Indicators. The labor force participation rates are obtained from ILOSTAT provided by International Labor Organization. The definitions and sources of these variables are summarized in Appendix Table A-1.

To minimize the influence of business cycle fluctuations, we calculate average annual growth rate of GDP per capita at 10-year intervals from 1960 to 2017: (Period 1: 1960–69); (Period 2: 1970–79); (Period 3: 1980–89); (Period 4: 1990–99); (Period 5: 2000–09); and (Period 6: 2010–17). We calculate other growth rates corresponding to the six channels in the same way. For calculating GDP growth, we use PWT's national-accounts real GDP (RGDP^NA).⁹ We use the initial level of per capita GDP as a regressor, for which we use another PWT measure of real GDP, RGDP^O, that is, output side real GDP at chained purchasing power parities, calculated using 2011 prices that are constant across countries and time.

Table 1 reports summary statistics of the variables considered in the empirical analysis. The average growth rate of per capita GDP is 2.3 percent. The average per capita GDP in 2011 constant prices is $21,410. The average old and youth dependency ratios are 0.18 and 0.37, respectively. The average share of working population is 0.65. The average labor force participation rate is 59.9. The average annual growth rate of TFP is 1.2 percent.

Figure 1 also presents how each component of GDP per capita growth relates to the old dependency ratio. Figure 1-2 shows no clear relationship between the rate of capital accumulation and the old dependency ratio; note that we measure capital accumulation based on capital-output ratio. Figure 1-3 illustrates that the relationship between the rate of human capital accumulation and population aging is negative. Figure 1-4 presents no clear relationship between the growth rate of average hours and the old dependency ratio. Figure 1-5 is for the growth rate of labor force participation. Again, there is no clear relationship of the growth rate of labor force participation with the old dependency ratio. Figure 1-6 presents a clear negative relationship between the age 15+ composition and the old dependency ratio. Lastly, Figure 1-7 shows a negative relationship between TFP growth and the old dependency ratio.

Table 2-1 presents the estimation results when we regress change in each of the six channels on the old and youth dependency ratios; to verify the effects of demographic deficit based on the share of working-age population, we estimate equation (4) rather than equation (3). In the first column, we report also the estimation results when the dependent variable is the growth rate of GDP per capita. As explained in Section 2, we use the old and youth dependency ratios at the beginning of each period—that is, in 1960 for period 1, and so on. In addition to the dependency ratios, we include country fixed effects; we also include period dummies so as to capture OECD-wide shocks. As stressed by the previous studies on channel regressions, the estimated coefficient of the old dependency ratio in column (1) is supposed to be equal to the sum the estimated coefficients in columns (2)–(8) that represent the six channels.¹⁰

In column (1) of the table, the estimated coefficient of the old dependency ratio is negative and statistically significant, suggesting that the impact of population aging on growth is negative. The impact is sizable in the sense that increasing the old dependency ratio by 0.01 lowers the growth rate of GDP per capita by 0.18 percentage points. Among the six channels, the estimated coefficients of the old dependency ratio are statistically significant in columns (5), (6), and (7), and the estimated coefficient of the old dependency ratio in column (1) is almost the same as the sum of those in columns (5), (6), and (7). According to the estimates, the negative effect of aging on GDP per capita growth rate operates mainly through the channel of lowered TFP growth. The estimated coefficients indicate that increasing the old dependency ratio by 0.01 decreases the GDP per capita growth rate by 0.18 percentage points, which is more than fully explained by the 0.20 percentage point decrease in the TFP growth rate. It also decreases the growth rate of the share of working-age population by 0.02 percentage points, but this is more than nullified by an increase in the labor force participation rate of 0.05 percentage points.

In Table 2-2, we replace old and youth dependency ratios with old and youth population shares, and the coefficients of the old population share are statistically significant in columns (1), (4), (6), (7), and (8). The estimated coefficient in column (1) shows that increasing the old population share by 0.01 lowers the growth rate of GDP per capita by 0.32 percentage points; again, the sum of the estimated coefficients of the old dependency share in columns (4), (6), (7), and (8) is close to that in column (1). The table also shows that the negative impact of population aging is mostly explained by lowered TFP growth rate by 0.38 percentage points. While the effect of the demographic deficit in column (6) is negative, it is more than fully mitigated by increased average working hours in column (4). Overall, our estimates in Tables 2-1 and 2-2 suggest that population aging negatively affects GDP per capita growth rate mainly through the channel of lowered TFP growth rate. Aging also affects economic growth negatively through demographic deficit, but this purely demographic change is more than nullified by increased labor participation rate or average working hours.

In Table 3, we add the initial per capita GDP as a regressor and report panel regression results with country as well as period fixed effects.^11,12 In the regressions that use old and youth dependency ratios as regressors (Table 3-1), the coefficients of the initial GDP per capita are also statistically significant in columns (1), (2), (4), and (8). Overall, the results are similar to those reported in Table 2-1, where the initial per capita GDP is not included. In column (1) of Table 3-1, all the coefficients of the old and youth dependency ratios are statistically significant with negative signs; the estimated coefficients of the old dependency ratio are also statistically significant in columns (5), (6), and (8). The estimated coefficient is slightly smaller in absolute value in column (1): An increase in the old dependency ratio by 0.01 decreases the GDP per capita growth rate by 0.16 percentage points. Again, the coefficient of the old dependency ratio in column (1) is almost the same as the sum of the coefficients in columns (5), (6), and (8) where they are statistically significant. In fact, the negative effect of aging on GDP per capita growth is more than fully explained by its negative impact on TFP growth: An increase in the old dependency ratio by 0.01 decreases the TFP growth rate by 0.18 percentage points. It also decreases the share of working age population by 0.02 percentage points, but this is more than nullified by the 0.048 percentage point increase in the labor participation rate.

In Table 3-2, when we replace old and youth dependency rates with old and youth population shares, the coefficients of the initial GDP per capita are statistically significant in columns (1), (4), and (8); the coefficients of the old dependency ratio are statistically significant in columns (1), (4), and (8), and the estimated coefficient in column (1) is approximately the same as the sum of those in columns (4) and (8). Again, the negative impact of aging on GDP per capita growth is mainly explained by its negative impact on TFP growth. There is no evidence of demographic deficit, but population aging increases the average working hours by 0.04 percentage points.

In sum, our empirical evidence strongly suggests that population aging in terms of both higher old dependency ratio and old population share is negatively associated with lower GDP growth per capita in the next ten years. When we examine six channels that potentially drive this negative correlation, our evidence further suggests that this is mainly because population aging results in lower TFP growth. We also find that the negative association of high old dependency ratio and old dependency share with working age population is more than compensated by increases in the labor force participation rate and average working hours.

Although we make the old dependency ratio and the old population share a predetermined variable by choosing its value at the beginning of each period, there is still a possibility that the causality runs reversely. For example, if workers emigrate from countries that have more pessimistic prospects, both the old dependency ratio and the old population share can increase due to emigration of young workers expecting lower GDP growth in the future. In Table 4, to overcome this problem, we present instrumental-variable (IV) estimates. We use 10-year lagged values of the old and youth dependency ratios (population shares) and the birth rate as instruments. In both Tables 4-1 (old and youth dependency ratios) and 4-2 (old and youth population shares), the first-stage F-statistics and the overidentification-test p values indicate that the IVs are quite satisfactory in most columns.

The IV estimation results of column (1) in Tables 4-1 and 4-2 are quite consistent with those in other tables; all three coefficients are statistically significant with a negative sign at the 1 percent or 5 percent level. The coefficient of the old dependency ratio is larger in absolute value than that of the youth dependency ratio. The estimated coefficients of the old dependency ratio and the old population shares are slightly larger in absolute value than they are in Table 3: The estimated coefficients indicate that a one percentage point increase in the old dependency ratio (the old population share) lowers annual GDP growth rate per capita by 0.27 (0.50) percent. Looking at the channels, in Table 4-1, the coefficient of the old dependency ratio is also statistically significant in columns (7) and (8). The sum of the estimated coefficients of the old dependency ratio in columns (7) and (8) explains only about a little more than half of the negative impact of aging in column (1); still, the negative impact of aging on the GDP per capita growth rate is mainly explained through the channel of lowered TFP growth. Interestingly the negative impact of aging on demography is through the ratio of population ages 15+ and the working age population in column (7). In contrast, in Table 4-2, the coefficient of the old dependency ratio is statistically significant only in columns (1) and (8), and more than 60 percent of the negative impact of aging on growth can be explained solely by lowered TFP growth. Overall, the IV estimation results confirm that aging affects economic growth negatively and that this negative effect operates mainly through lowered TFP growth.

Previous studies such as Feyrer (2007) and Aiyar, Ebeke, and Shao (2016) highlighted lowered TFP growth as the most important channel in explaining the negative effect of aging on labor productivity growth. Although with our study we focus on GDP per capita and not GDP per worker, we also find that TFP growth is the most important and statistically significant channel. These previous studies also emphasize the importance of the share of workers aged 55–64 years; the idea is that because the productivity of workers in this age group is lower, increase in its share contributes negatively to labor productivity growth. In Table 5, following Aiyar, Ebeke, and Shao (2016), we add the population share aged 55–64 years as an additional variable and report panel regression results with country/period fixed effects. As before, we use old and youth dependency ratios in Table 5-1 and old and youth population shares in Table 5-2. As expected, the coefficient of the population share aged 55–64 years in column (1) is negative and statistically significant at the 5 percent level in both tables.¹³ Regarding the coefficient of the old dependency ratio and the old population share, the results in Tables 5-1 and 5-2 are quite consistent with those in Tables 3-1 and 3-2. In Table 5-1, while the coefficient of the old dependency ratio is negative and statistically significant in columns (7) and (8), it is positive and statistically significant in column (5). The estimated coefficient of the old dependency ratio in column (1) is almost the same as the sum of its coefficients in columns (5), (7), and (8). In Table 5-2, it is negative and statistically significant at the 1 percent level in columns (7) and (8) and positive and statistically significant at the 10 percent level in column (4). Again, its estimated coefficient is almost the same as the sum of its coefficients in columns (4), (7), and (8); the only difference is that the coefficient in column (7) rather than column (6) is statistically significant in Tables 5-1 and 5-2.

As explained, previous studies that examine the impact of aging on labor productivity growth ignore aging of the whole population and focus on aging of the workforce. However, if aging of the whole population affects TFP growth negatively, it will also negatively affect labor productivity growth. In Table 6,⁰ in order to evaluate the importance of the old dependency ratio on labor productivity, we adopt the growth rate of labor productivity (i.e., per worker GDP) as a dependent variable and show the panel regressions results with country fixed effects for various specifications. In columns (1) and (2) in Table 6-1, we use only the old and youth dependency ratios as a regression. In columns (3) and (4), we add the population share aged 55–64 years as an additional regressor. In columns (5) and (6), we instead add the initial per capita GDP as an additional regressor. In columns (7) and (8), we add both the population share aged 55–64 years and the initial per capita GDP as regressors. Finally, in columns (9) and (10), we report IV panel regression results. For each pair, the first one does not include period dummies and the second one does. In all the specifications from (1) to (10), the coefficient of the old dependency ratio is negative and statistically significant. In Table 6-2, we replace old and youth dependency ratios with old and youth population shares. Again, in all the specifications from (1) to (10), the coefficient of the old population share is negative and statistically significant. Our results in Table 6 suggest that aging of the whole population negatively affects not only on GDP per capita growth but also on labor productivity growth.

In this paper, we investigate six channels through which aging affects per capita GDP growth based on 35 OECD countries where population aging has reached remarkably high levels. The six channels we considered are increases in: (i) capital accumulation; (ii) human capital accumulation; (iii) average working hours; (iv) the labor force participation rate; (v) the share of population aged 15 years and over; and (vi) TFP. We confirm the findings from previous studies that population aging in OECD countries has negative effects on GDP growth per capita. We find that the most important channel through which the negative effects operate is lowered TFP growth. In most of our empirical specifications, lowered TFP growth associated with aging exceeds the lowered growth rate GDP per capita. We also find evidence of demographic deficit, but this negative effect of aging is more than nullified by compensating increases in the labor force participation rate or average working hours. We conclude that because TFP growth rate can be permanently lowered, aging's negative effects on GDP growth per capita may last permanently.