China’s saving rate is extraordinarily high (National saving rate by region, %)

1. Why Is China’s Saving Rate So High? A Comparative Study of Cross-Country Panel DataBy Juann H. Hung and Rong QianJanuary 3, 2012 The analysis and conclusions expressed in this presentation are those of the authors and should not be interpreted as those of the Congressional Budget Office.

2. China’s saving rate is extraordinarily high(National saving rate by region, %)

3. Motivation for the Paper • China’s national saving rate is extraordinarily high, while the US rate is extraordinarily low. Why are national saving rates so different? • China’s high saving is a main source of the large global imbalance. Concerns about the global imbalance have mounted pressure on China to do something – increase consumption, appreciate its currency, etc. • What if China’s high saving rate is largely structural? If its real exchange rate is not a main cause of China’s high saving rate, then a large currency revaluation may do more harm than good. (Spence 2010; Rodrik 2009).

4. Goals of the paper • Use a large multinational panel data to estimate a model of national saving rates that can be used to ascertain three questions: • What are the main drivers of national saving rates? • To what extent does China’s saving rate exceed the projection of benchmark models of saving rates? • What are the factors primarily responsible for China’s extraordinarily high saving rates?

5. The remainder of presentation • Existing theories of China’s high saving rate. • Specification and estimation of national saving rate models. • Estimation results. • China’s saving rate compared to model projections. • Major factors responsible for China’s high saving rate. • China’s high saving and the East-Asian growth model. • Conclusions

6. Existing theories of China’s high saving rate • A frayed social safety net and an under-developed financial sector (Chamon and Prasad 2008; Blanchard and Giavazzi 2006; Kuijs 2006). • Policies that favor industry at the cost of jobs and consumer spending (Kuijs 2006; Kuijs and Wang 2006; Bosworth and Collins (2007). • Strong economic growth (Modigliani and Cao 2004). • An undervalued currency (Wolf 2010). • Others

7. The story suggested by existing theories • Those popular theories – frayed social safety net, under-developed financial sector, heavy-handed industrial policies, undervalued currency, and strong growth rate – suggest that a root cause of China’s high saving is its poverty and underdevelopment and its haste to grow and catch up. • China was a destitute country when its pro-market economic reform started. Despite its rapid economic growth over the past three decades, its real per capital GDP in 2007 was still only about less than half of an average upper-middle income country (such as Brazil, Mexico, and Turkey). <Table below> • However, those papers tend to focus on just one or two factors alone. Would most of those variables still be significant when they are included in the same model together with other traditional factors such as demography and urbanization?

9. Estimating models of national saving rate • Dependent variable is national saving rate because we are interested in whether the “global imbalance” can be largely accounted for by the vast difference in pre-existing economic and structural conditions among countries. • Explanatory variables are popular factors in the literature plus some traditional and less talked-about drivers of saving. • Estimation method is a dynamic generalized-method-of-moments (GMM) estimation system.

10. The Estimation method • The method of estimation - a dynamic generalized GMM estimation of panel data - is the same as that used in Loayza, Schmidt-Hebbel, and Serven (2000). • That method has been used by many researchers to estimate equations that include lagged dependent variables as well as explanatory variables that are potentially endogenous.

11. Explanatory variables included in the model • Growth of real income per capita. • The life-cycle hypothesis: When the economy is growing, saving of the working population will increase relative to non-working population’s dissaving, causing aggregate saving to rise. • The habit-formation hypothesis: people’s habits tend to keep their consumption close to their existing level. Thus, in a fast growing economy, the saving rate will rise as people take time to increase their consumption compatible with their new income.

12. Explanatory variables included in the model • Income-adjusted growth rate. Given a poor country’s desire to grow and move to a higher income and consumption steady state – i.e., the Golden-Rule state, the effect of economic growth on its saving rate may be stronger than that on a richer country’s. This “catch-up” motivation for saving is greater, the poorer is a country. Thus, an income-adjusted growth rate is included in the model to capture this effect of growth on saving rate. • The relative income-adjusted growth rate of real per capita GDP of country i = [moving average of (Yust-1,t-3/Yi t-1, t-3)] x [(Yi t/Yi t-1) -1], where the first term is the income-adjustment factor, and Yus is the real per capita GDP of the United States.

13. Explanatory variables included in the model • Real income per capita. In a standard Keynesian model, saving is a positive function of income after their income exceeds subsistence level of consumption. • Dependency Ratio. According to the life-cycle hypothesis, a country’s saving declines when its dependency ratio increases. We include both the old dependency ratio and the young dependency ratio in our regressions. • Social Safety Net (Government’s Social Spending).If individuals expect to get high social security benefits when they retire or enter bad times, they will tend to reduce saving during their working days for retirement and for bad times.

14. Explanatory variables included in the model • Urbanization (urban population/total population). • Rural households, who depend heavily on agricultural income, tend to save a larger share of their income than do city dwellers because precautionary saving tends to be higher for those subject to higher income volatility. • In many developing countries the rural population are trapped in poverty. Thus, a rise in urbanization amounts to a decline in income inequality. To the extent that saving rate is positively correlated with income inequality, it will be negatively correlated with urbanization.

15. Explanatory variables included in the model • Real Interest Rate. • The substitution effect on saving is positive; the income effect is negative. • Governments could force the national saving rate to rise by keeping the real interest rate low (thereby suppressing consumption), thereby providing cheap credit to industries. • Domestic Credit.Increase in availability of credit, by making borrowing constraints less stringent, could lead to a decline in saving. • Higher borrowing constraints raises private saving by constraining consumption smoothing. • Stringent borrowing constraints also increases precautionary saving.

16. Explanatory variables included in the model • Inflation.Inflation may increase saving by redistributing wealth from workers (who have a lower saving rate) toward capital-owners (who have a higher saving rate). • Government may use inflation to lower the amount of consumer goods that wage-earners can afford, thereby extracting resources for capital accumulation and growth. • Inflation is also used as a proxy for macroeconomic uncertainty, which has a positive effect on precautionary saving.   • Government saving.If taxpayers’ offset to government saving is smaller than predicted by the RE hypothesis, a decrease in government saving will decrease the national saving rate to some extent.

17. Explanatory variables included in the model • The Real Exchange Rate. • According to the mercantilist view, countries can boost its net exports – thus national income and saving – by undervaluing its currency. Though not a consensus, this is a popular explanation for China’s high saving rate. • An undervalued currency may spur growth and saving through non-mercantilist channels: • Levy-Yeyati and Sturzzenegger (2007) show that an undervalued currency boosts output growth by increasing savings and capital accumulation by shifting resources from workers to firms. • Rodrik (2008) argues that currency undervaluation spurs economic growth by attracting the reallocation of resources from non-tradables toward manufacturing.

18. Explanatory variables included in the model • East-Asia Dummy. • Countries in East Asia – namely, Japan, South Korea, Taiwan, China, Hong Kong, and Singapore – on average have a higher national saving rate than do other regions. • Some analysts have attributed this to East Asian’s cultural factors, while others attributed it to the “East-Asian growth model” which includes various policies designed to promote growth through capital accumulation, such as making credit cheaper and more accessible to industries than to consumers. • Since it is difficult to quantify countries’ culture and “growth model”, we include an East Asia Dummy to capture any marginal effect on saving of “being an East-Asian country”.

19. Data • World Bank’s World Development Indicator 2009. IMF’s International Financial Statistics, Asian Development Bank, and Data from United Nations. • After the removal of data considered as outliers, we ended up with a sample of 70 countries, from 1980 to 2007.

20. Estimation Strategy and Results • Regressions that excludeadjusted growthas a regressor: • Table 2A reports results that include China in dataset. • Table 2A’ reports results that exclude China from dataset. • Regressions that include adjusted growth as a regressor: • Table 2B reports results that include China in dataset. • Table 2B’ reports results that exclude China from dataset.

25. General Observations of results • All explanatory variables, except inflation, are statistically significant at the 95% confidence level in at least two tables. They all have the theoretically justifiable signs. • Estimated coefficients regressions that include adjusted income growth are somewhat more stable across models than those that did not. • The coefficient estimates on the lagged dependent variable are around 0.5 in all four tables, indicating a high degree of persistence in national saving.

26. Observations from comparing Tables 2A and 2B (Table 2B includes adjusted growth as a regressor) • The coefficient onadjusted growthrate is statistically significant, with an expected sign and a reasonable magnitude, in all 6 regressions in Table 2B. • This offers some evidence of the “catch-up effect hypothesis”: the marginal effect of economic growth on saving rate is higher for poorer countries than for richer countries. • The coefficient estimates are much more stable in regressions in Table 2B than those in 2A. • This suggests that the specifications that include adjusted growth rate are better than those without.

27. Observations from comparing Tables 2A and 2A’ (Regressions without adjusted growth rate) • The East-Asia dummy has a large, positive, and significant coefficient regardless whether China is in dataset. However, its coefficient is much smaller when Chia is excluded from dataset (4.3 versus 6.8). ›› Factors proxied by that dummy was even more forcefully at work to boost saving in China than in its East-Asian peers. • The coefficient on exchange-rate variables are significant and have similar magnitudes regardless whether China is in dataset. ›› Marginal impact of the real exchange rate on saving rate in China is not markedly different from that in other countries. • The magnitudes of those exchange-rate coefficients are stable regardless of whether East Asia dummy is included in the regression ›› Net effect of factors proxied by that dummy is largely independent of the exchange rate.

28. Observations from comparing Tables 2B and 2B’ (Regressions including adjusted growth rate) • The coefficients on adjusted growth rate in all regressions are somewhat larger in when China is in dataset. ››The catch-up effect is more powerfully at work in China than in other countries. • In the regression that includes both adjusted growth rate and East-Asia dummy: When China is excluded from dataset, the East-Asia dummy is quantitatively and statistically insignificant. In contrast, adjusted growth rate remains significant and unchanged in magnitude, regardless of whether China is in the dataset. ›› Factors proxied by East-Asia dummy are highly correlated with drivers of “catch-up” growth in East-Asian economies outside of China. ›› There are other factors proxied by the dummy that are more powerfully at work in China than in other East-Asian economies.

29. Choosing the benchmark model • By standard of least average unexplained saving, the best model includes the East-Asia dummy,the exchange rate variables, andadjusted-growth rate. • However, the preceding analysis suggests that the effect of East-Asia dummy beyond what’s already included in the adjusted growth rate is largely be due to China-specific factors – such as its government’s ability to command industrial policies through the state-owned enterprises and other means – not shared by all countries. Thus, the East-Asia dummy should not be included in benchmark model. • We end up choosing Model B5, the model that includes income-adjusted growth rate and the two exchange rate variables, along with all non-policy variables.

31. Observations in support of choosing Model B5 as the benchmark model • The average unexplained saving rate (=national saving rate - the long-term model forecast) by standard of B5 is smaller than that of all other models except A6 and B6. • Model B5’s ability to fit national saving rates of other countries remain relatively unchanged regardless of whether it is estimated with or without China in the dataset. A feature not shared by Model B6. • These results reinforce the interpretation that the East-Asia dummy mostly proxies factors that are unique to China, and thus should not be included as an explanatory variable in the benchmark model.

32. China’s saving rate significantly exceeded projections of benchmark model • Using Model B5 as the benchmark, China saving rate exceeded model projections by about 10 – 12 percentage points from 1990 to 2007, depending on whether China is included in the dataset. • Using Model A5 (which excludes adjusted growth rate) as the benchmark, China’s average unexplained saving rate would be about 14-16 percentage points, depending on whether China is in the dataset. • Either by the standard of B5 or A5, China had a higher level of unexplained saving rate than almost all other countries included in the sample; only Bhutan has higher unexplained saving rate than China by both standards.

33. Assessing relative contribution of explanatory variables to China’s higher saving rate • The average saving rate of China during 1990-2007 about 45%, considerably higher than that in OECD countries (22%). • We use the estimated models to assess the relative contributions of each explanatory variable to the saving-rate gap between China and OECD.

34. Main drivers of China’s high saving rate • Across all models, China’s much lower level of old dependency ratiostands out as most responsible for China’s much higher saving rate. • The rankings of other factors are somewhat model-dependent. • By standard of Model B5, the other important drivers, in a descending order of importance, are • China’s higher growth rate (including adjusted growth rate), • China’s weaker social safety net, • China’s lowerurbanization ratio. • China’s undervalued currency

40. Dependency Ratios: China versus OECD average

41. Our benchmark model has become less able to explain China’s saving rates since 2001

42. Possible factors outside our models that may have become more important drivers of China’s saving rate • The 1997-98 Asian crisis boosted precautionary saving in both private and public sector: Zhou Xiaochuan (2009) claimed that one factor behind East Asian countries’ high saving rates is “defensive reactions against predatory speculation” that had led to the Asian financial crisis. • The widening gender-imbalance: Wei and Zhang (2009) hypothesized that the saving rate started to shoot up around 2002 largely the gender ratio for the marriage-age cohort began to be seriously out of balance, forcing households with sons to raise their savings to increase the chance of winning a bride. • The increase in the transfer of income away from the household sector (to banks and businesses) as a result of policymakers’ schemes to resolve the NPLs crisis that began in the late 1990s (Pettis 2010).

43. China’s high saving and East-Asian growth model • When China is excluded from the dataset, the East-Asia dummy becomes insignificant in a model that already includes adjusted growth rate. • Comparison of Model B6’ (which includes adjusted growth rate) and Model A6’ (which excludes adjusted growth rate) suggests: • Those factors proxied by the East-Asia dummy are mostly captured by adjusted growth rate, which in turn proxies for factors behind the rapid “catch-up” growth. • What may those common factors be?

44. Features of East-Asian growth model • The East-Asian growth model can be loosely described as a “high saving- high investment-high growth-high saving” strategy modeled after Japan’s model of growth and development. • World Bank (1993) emphasizes that a “virtuous circle” – going from higher growth, to higher savings, to even higher growth – has played a central role in successful development experiences in East Asia. • Although specific policy mix varied across countries, that growth model basically relies on heavy government interventions that favor capital formation (i.e., industrialization) at the expense of consumer spending, besides the well-known strategies of relying on exports for growth.

45. Observations suggesting that China has adopted an Variant of East-Asian growth model • China’s policies are known to favor capital-intensive investment, which arguably is less effective in absorbing its large pool of excess labor than labor-intensive investment. • Growth has been capital-intensive, contributing to depress household consumption relative to national income. • Corporate saving is high. Capital-intensive production has been encouraged by low interest rates and by the fact that most state-owned firms do not pay dividends. • The government has favored manufacturing over services by policies such as holding down its exchange rate and suppressing the prices of inputs such as land and energy. • Households’ disposable income had grown more slowly relative to GDP (figure below).

46. Disposable income & consumption as % share of GNI

47. China’s household saving rate has been rising

48. Implications of East-Asian experiences • The experience of Japan suggests that once the Chinese economy grows richer and better developed, we are likely to see a slow “normalization” of China’s saving rate. • China’s real per capita GDP in 2007 was still slightly below that of Japan in the early 1960s; thus, it is likely that China’s policy of financial repression will ease gradually, not rapidly. • There are signs that that process of “normalization” has begun in China. • The government allowed its currency to fall gradually against the US dollar from June 2005 to July 2008. After a pause during the 2008-2009 financial turmoil, it resumed that policy again. • The government has also begun to strengthen the social safety net in recent years (next figure).

49. Real per capita GDP of selected countries/regions(constant 2005 int’l dollars based on PPP rates)

50. Conclusions • Our benchmark model explains about 72-76 percent of China’s saving rate during 1990-2007, depending on whether China is in dataset. On average, China’s saving rate exceeded model predictions by about 10 to 12 percentage points. • The predominant drivers of China’s higher saving rates are its low old dependency ratio and, to a less extent, its strong growth rate,weak social safety net, low urbanization, and undervalued currency. The contribution of China’s currency undervaluation to its saving rate is relatively modest. • The East-Asian model of growth have also contributed to China’s saving rate, beyond and above its currency policy. • Our results imply that as the average Chinese population becomes older and its national income converges to its potential, its saving rate will also begin to decline.