RDP 2025-08: Ageing and Economic Growth in China Appendix A: China Data and Regressions

Patrick Hendy

November 2025

A.1 Geographic variation

The following maps illustrate the log differences in old-age ratio by province. Figure A1 shows the data for the 1990s, Figure A2 for the 2000s and Figure A3 for the 2010s.

Figure A1: 1990s Ageing Patterns
For mainland China
Figure A1: 1990s Ageing Patterns - A map of mainland China in the 1990s where the colour of each province indicates the relative log difference change in the old-age ratio. The map shows that the largest increases in the old-age ratio were in central and northeast China, while the increase in old-age population was smaller in coastal regions.

Sources: Author's calculations; CEIC Data; National Bureau of Statistics of China.

Figure A2: 2000s Ageing Patterns
For mainland China
Figure A2: 2000s Ageing Patterns - A map of mainland China in the 2000s where the colour of each province indicates the relative log difference change in the old-age ratio. The map shows that the largest increases in the old-age ratio were in central and northeast China, while the increase in old-age population was smaller in coastal regions.

Sources: Author's calculations; CEIC Data; National Bureau of Statistics of China.

Figure A3: 2010s Ageing Patterns
For mainland China
Figure A3: 2010s Ageing Patterns - A map of mainland China in the 2010s where the colour of each province indicates the relative log difference change in the old-age ratio. The map shows that the old-age ratio increased significantly in northeast China, while coastal regions also saw an increase in the old-age ratio.

Sources: Author's calculations; CEIC Data; National Bureau of Statistics of China.

A.2 Intertemporal growth correlations

As discussed in Section 2, the correlation between provincial growth over the 1990s and over the 2000s is negative. However, for the 2000s and 2010s, this correlation is positive.

Figure A4: Economic Growth Correlation
Between 1990s and 2000s, by province
Figure A4: Economic Growth Correlation - A scatter plot showing log differences in GDP per capita from 1990 to 2000 against log differences in GDP per capita from 2000 to 2010. Each dot indicates a province. The graph shows that there was a slight negative correlation between per capita GDP growth in the 1990s and 2000s.

Note: Line of best fit shown.

Sources: Author's calculations; CEIC Data; National Bureau of Statistics of China.

Figure A5: Economic Growth Correlation
Between 2000s and 2010s, by province
Figure A5: Economic Growth Correlation - A scatter plot showing log differences in GDP per capita from 2000 to 2010 against log differences in GDP per capita from 2010 to 2020. Each dot indicates a province. The graph shows that there was a positive correlation between per capita GDP growth in the 2000s and 2010s.

Note: Line of best fit shown.

Sources: Author's calculations; CEIC Data; National Bureau of Statistics of China.

A.3 Full regression results

The first-stage regression summary statistics for the main specification are shown in Table A1.

Table A1: First-stage Regression Summary Statistics
  R-squared Adjusted R-squared Robust F (1,27) Prob > F
Δlog(old-age ratio) 0.7675 0.7493 70.6168 0.0000

Note: F-statistic adjusted for 28 clusters of provinces.

The full table for the regressions described in Tables 2 and 3 is shown in Table A2.

Table A2: Regression Results
Effect on GDP per capita growth
  OLS Instrumental variables
20-year lag 20-year lag 20-year lag 20-year lag 30-year lag 40-year lag
Δlog(old-age ratio) −0.270
(0.181)		 −0.716***
(0.210)		 −0.899***
(0.202)		 −0.569**
(0.287)		 −0.868***
(0.199)		 −0.615***
(0.182)		 −0.806***
(0.180)
Controls
Year
2000 0.017
(0.059)		 0.651
(0.074)		 −0.114**
(0.057)		 −0.120**
(0.056)		 −0.101
(0.240)		    
2010 −0.427***
(0.073)		 −0.337***
(0.074)		 −0.427***
(0.068)		 −0.506***
(0.092)		 −0.889***
(0.177)		 −0.382***
(0.053)		  
Industry
Agriculture         0.159
(0.121)		    
Mining         −0.027
(0.021)		    
Manufacturing         0.198***
(0.065)		    
Industry#Year
Agriculture#2000         0.206*
(0.111)		    
Agriculture#2010         −0.167
(0.122)		    
Mining#2000         0.043
(0.043)		    
Mining#2010         0.044
(0.033)		    
Manufacturing#2000         −0.143*
(0.075)		    
Manufacturing#2010         −0.254***
(0.070)		    
Coastal
Coastal 0.229***
(0.049)		 0.207***
(0.047)		       −0.158**
(0.074)		 −0.130***
(0.041)
Coastal#2000 −0.363***
(0.079)		 −0.372***
(0.084)		          
Coastal#2010 −0.382***
(0.068)		 −0.340***
(0.063)		       0.022
(0.075)		  
Province fixed effects N N N Y N N N
Constant 1.416***
(0.027)		 1.468***
(0.032)		 1.567***
(0.036)		 1.634***
(0.034)		 1.969***
(0.144)		 1.482***
(0.058)		 1.160***
(0.077)
Observations: clusters 84: 28 84: 28 84: 28 84: 28 84: 28 58: 30 28: 28
Notes: ***, ** and * denote statistical significance at the 1, 5 and 10 per cent levels, respectively. Standard errors are in parentheses and clustered at a provincial level.

Running the main specification of the instrumental variable regression individually for each decade shows that the coefficient on ageing is large and negative for the 2000s and 2010s, and smaller for the 1990s (Table A3). The sample size when running the regression using individual decades is relatively small.

Table A3: Regression Results for Main Specification
By decade
Decade Effect of GDP per capita growth
1990s −0.053
(0.785)
2000s −0.768
(0.693)
2010s −0.895***
(0.171)

Notes: ***, ** and * denote statistical significance at the 1, 5 and 10 per cent levels, respectively. Standard errors are in parentheses and clustered at a provincial level.

The full table for the regression described in Table 4 is shown in Table A4.

Table A4: Alternative Functional Form
Effect on GDP per capita growth by instrumental variables
  20-year lag
Δlog(old-age ratio) −3.710***
(0.985)
Controls
Year
2000 0.036
(0.062)
2010 −0.306***
(0.079)
Coastal
Coastal 0.226***
(0.046)
Coastal#2000 −0.376***
(0.083)
Coastal#2010 −0.336***
(0.067)
Constant 1.439***
(0.029)
Observations: clusters 84: 28

Notes: ***, ** and * denote statistical significance at the 1, 5 and 10 per cent levels, respectively. Standard errors are in parentheses and clustered at a provincial level.