Inequality, Saving and Global Imbalances: A New Theory with Evidence from OECD and Asian Countries

3.1 A Positive Inequality-saving Link under Liquidity Constraints

A positive link of saving with inequality can be derived from the traditional post-Keynesian model, though with necessary changes to incorporate useful elements from other hypotheses for saving behaviour. Consider two income groups: {B for bottom, T for top}. Group B is composed of workers with stable, low labour income, and group T consists of agents with variable, high non-labour income or super-high executive pay. Non-labour income includes property income captured by capitalists, tax revenue by governments and illegal or immoral income by public and corporate officials as widely observed in developing countries or transition economies.

Consumption is constrained by income without borrowing opportunity as in China and other parts of Asia with limited or no consumer credit. Most of income Y_B in group B is permanent income used for consumption C_B, with a high propensity to consume: α_B = C_B/Y_B. Much of income Y_T in group T is transitory income ready for saving after covering consumption C_T,3 with a low propensity to consume: α_T = C_T/Y_T. Then, α_B > α_T is postulated as usual to reflect the long-term differences in behavioural patterns between the two groups. Consumption habits are persistent under behavioural inertia.

Given the aggregate income Y = Y_B + Y_T, the income share Y_T/Y is used as a proxy for income inequality. Analysing the ratio of aggregate consumption C = C_B + C_T to aggregate income Y yields the traditional post-Keynesian model:

(1)

Clearly, Y_T/Y affects C/Y negatively under α_B > α_T. Thus, we have:

The equation 1 result is applicable to emerging Asian economies, as depicted in Figure 1. Savings in China and other Asian countries are used for investment, not consumption, owing to the underdevelopment of their financial systems in general and the lack of consumer credit in particular. Individuals must save for large purchases due to no opportunity to borrow against future income growth. Additionally, it is a state financing policy to direct national saving to manufacturing investment for trade expansion, and this financial orientation has been an important part of industrial policies and growth strategies in many Asian economies (Stiglitz, 2001). However, even massive investment in Asia such as China cannot productively absorb all its enormous saving, so surplus saving has constituted for years a substantial surplus of current account and a direct source of trade frictions with the USA and other OECD countries. The Asian exportation of surplus saving is blamed for prolonged economic problems in other countries (Greenspan, 2009).

3.2 A Negative Inequality-saving Link Due to Consumer Credit

A negative link between inequality and saving is caused by the fact that the saving of the wealthy is used not for output growth by investing firms but for increased consumption by poor households through deficit spending. This saving by the rich, after becoming consumer credit for the poor, should be viewed as consumption spending rather than domestic saving in the national income accounts. In this case, spending is constrained not by income but by liquidity, as observed in OECD countries and modelled as follows.

Consider a closed economy first, where the top income group lends to the bottom income group, and both sides of the credit market are domestic residents. Consumption propensity is defined as α′_T = C′_T/Y_T for group T as usual, but α′_B = C′_B/(Y_B + D′_d) for group B under income illusion, where renewable cheap debt D′_d borrowed is illusively taken as the ‘permanent income’.5 Aggregate consumption is C′ = C′_B + C′_T. Assume α′_T < α′_B. Denote by D′_d/Y_T = β′_T the lending ratio of group T. Piecing together the above descriptions yields an extended post-Keynesian model below:

(2)

Clearly, C′/Y increases with a higher ratio of Y_T/Y if β′_T > 1 − α′_T/α′_B (≡ κ_d > 0 due to α′_T < α′_B).6 We thus arrive at:

If setting β′_T = 0 in equation (2), our extended model reduces to equation (1) that was derived under no consumer credit. Working with α_B = α′_B yields C_B < C′_B, showing that consumption is increased with borrowing opportunities than without them even though borrowers’ consumption propensity remains unchanged. In fact, this propensity is likely to be boosted by consumer credit, so that the consumption-to-GDP ratio may expand further. In this case, κ_d increases with a higher propensity of α′_B, that is, the top income group must raise its lending ratio to maintain the condition of β′_T > κ_d.

The equation 2 result applies largely to the situation of OECD economies, as depicted in Figure 2. The negative role of inequality for saving is jointly driven by the two sides of the credit market. On the lending side, more of liquid assets backed by consumer loans to workers are preferred by investors as stores of value. These non-producible assets result in no job creation for indebted workers. On the borrowing side, many people borrowing against the rising value of their homes live with income illusion arising from asset bubbles, with such an illusion created by uninformative but persuasive advertisements. Even poor people optimistic about the future have developed habitual profligacy through the easy use of cheap credit under strong consumerism. Under these circumstances, spending is constrained by credit limits regardless of actual incomes, as suggested by β_T > κ_d in our equation (2) model.7

The relationship between surging inequality and declining saving has a clear bearing on policy response in the USA and other OECD countries. Income inequality generates social pressure, but no intervention has been attempted to reverse it. Instead, political efforts were often made to prop up the living standards of the bottom income group through encouraging easy credit and keeping demand robust despite its falling income share (Rajan, 2010). With more saving at the top and more borrowing at the bottom, consumption inequality has increased much less than income inequality, but the domestic indebtedness of the poor and the middle class has increased substantially. The saving rate is then on the freefall as a result of domestic lending, which is directed by financial markets for consumption rather than for investment (Kumhof and Ranciere, 2010).

3.3 The Negative Inequality-saving Link Strengthened by Foreign Financing

The contribution of consumer credit to saving collapse in OECD countries was reinforced by the cross-border mobility of savings under financial globalisation. In this case, domestic spending is no longer constrained by national income as foreign saving becomes available in the long run for easy use at low costs. For example, the spread of global capital markets has enabled poorer US households to borrow more relative to their incomes for about three decades, as reported in the literature (Kumhof and Ranciere, 2010). We proceed with our modelling by considering foreign financing for consumer credit to make the analysis more realistic.

Now consider an open economy. Consumption propensity is specified as α″_T = C″_T/Y_T for group T and α″_B = C″_B/(Y_B + D) for group B, where C″_B + C″_T = C″ is aggregate consumption and debt D borrowed is treated again as the ‘illusive income’. Assume α″_T < α″_B as usual. The debt D is now borrowed from foreign sources by D_f as well as from domestic savers by D″_d, so that D″_d + D_f = D. The lending ratio is D″_d/Y_T = β″_T for group T. Denote by λ (=D/D″_d ≥ 1) the ratio of total debt to domestic debt (as an index for international capital mobility), so that D = λβ″_TY_T, with a higher ratio λ indicating more foreign financing given D″_d.

Substituting the above specifications into the ratio of aggregate consumption to national income yields a further extended post-Keynesian model as follows:

(3)

If setting λ = 1 in equation (3) (i.e. zero foreign financing for domestic credit), this extended model is reduced to equation (2) that was obtained under consumer credit but with no foreign financing.

Equation 3 shows that the inequality index Y_T/Y affects the consumption ratio C″/Y positively and hence the saving rate s″ = S″/Y negatively if β″_T > (1 − α″_T/α″_B)/λ (≡ κ_f). Since κ_f = κ_d/λ < κ_d under λ > 1, we know that the condition of β″_T > κ_f for the open economy is easier to satisfy than that of β′_T > κ_d for the closed economy.8 The economic intuition of such closed-open comparison is that the rich need not raise their lending ratio to maintain the condition of β″_T > κ_f as the poor can borrow foreign saving to increase their consumption. Moreover, this condition becomes even weaker as κ_f falls with greater capital mobility λ (i.e. more foreign financing is available to facilitate domestic credit). We then establish:

The negative inequality-saving (or positive inequality-credit) nexus is attributable to swelling financial services facilitated by international capital flows, as implied by Figure 3 for the USA. This issue received scant attention before the recent global crisis (Brown, 2008); only recently has the literature begun to establish the direct relation of rising income inequality to surging household leverage under consumer credit and foreign financing (Iacoviello, 2008). Four points are worth further discussing below.

First, as savers’ desire to lend and households’ need to borrow increase simultaneously irrespective of national boundaries, so do both cross-border flows of funds between different income groups and the size of the financial sector. The aggressive expansion of this sector is indicated by many measures (Philippon, 2008). Second, while intermediating foreign savings makes the top income group increasingly richer, the bottom group becomes much more indebted, heightening economic inequality. Third, widespread consumer credit with foreign financing is responsible for the falling and low rate of saving, as is implied by equation (3) with ∂s″/∂λ = −β″_Tα″_BY_T/Y < 0. When this fall starts to be perceived as unsustainable, it becomes the trigger for a financial crisis (Kumhof and Ranciere, 2010). Fourth, surging inequality renders the economy more dependent on consumption for growth, which drives down saving below investment, so that trade deficits inevitably follow and must be financed through net borrowing from foreign sources. If this situation persists, current account imbalances will keep on getting worse. While dollar depreciations are needed to reduce the US trade deficit, the fundamental change required to curb this problem is an increase in the US national saving rate (Feldstein, 2008).

Since consumption propensities are lower for the top than for the bottom income group, it follows that consumption inequality is inherently less than income inequality. With our extended models presented above, one can prove that this phenomenon is reinforced by consumer credit and that consumption inequality falls below income inequality to a larger extent if there is more foreign financing for domestic credit and if debt-to-income ratios are higher for poorer households (as has happened in the USA; see Brown, 2008). Such significant phenomenon has been documented for the USA in the literature (Kumhof and Ranciere, 2010; Tridico, 2012). On the contrary, consumption and income inequality may stay together if consumer credit is absent or limited, with strong evidence found from China (Cai et al., 2010).

4.1 Econometric Procedure

Three empirical issues need to be addressed in our estimation as in other similar studies (Loayza et al., 2000). First, inertia in saving rates is likely to be present in the original annual data we use; therefore, we need a dynamic specification for estimation if data are good enough in quality and availability. Second, some regressors may be jointly determined with the saving rate as the dependent variable, and we thus should control for their joint endogeneity. Third, we may need to handle the possible presence of unobserved country-specific effects correlated with explanatory variables.

To deal with these issues, we apply the GMM system estimator in dynamic panels to OECD countries with large sample size and high data quality. This estimator allows for saving inertia by including lagged values of the dependent variable as a regressor. The estimator can also be used to control for unobserved country-specific effects and potential endogeneity problems using ‘internal instruments’ based on lagged values of explanatory variables. The Sargan test is adopted to test the overall validity of these instruments. The GMM estimator retains all the information in both cross-country and time-series data.9

Unfortunately, the GMM estimator is not suitable for Asian economies due to their small sample size and limited data availability. We therefore have to compute fixed effect panel estimates after ruling out random effects models through the Hausman test. A cross-section weighted FGLS estimator is used to correct for heteroscedasticity because of large differences in size among Asian economies. We employ the panel corrected standard error methodology to compute robust estimators in both the saving and consumption regressions.

Our empirical analysis is based on a reduced-form approach encompassing a variety of saving determinants, while our attention is focused on the saving effects of inequality and finance. We attribute private consumption to a broad set of underlying factors: (i) lagged values of dependent variables; (ii) alternative indexes of income inequality; (iii) various measures of financial depth; (iv) demographic structure features; (v) income-related variables and (vi) special characteristics of different economies under consideration. All variables are expressed in logarithms with only a few inevitable exceptions.10 The dependent variable in some regressions is the ratio of private consumption to GDP, which mirrors aggregate saving because household saving takes up an overwhelming share of aggregate saving. In other regressions, the rate of aggregate saving is used as the dependent variable to check for the robustness of our estimation results.

The regressors are selected according to economic relevance as well as data availability. Their selection is reported below: (i) income inequality, characterised by the Gini index with large sample size and by the income shares of the top (0.5, 1, 5) per cent of the population; (ii) financial depth, measured by various price and quantity variables, such as the real interest rate, domestic credit to private sectors relative to GDP (denoted credit.ps), private credit supplied by depository banks and other financial institutions relative to GDP (denoted credit.do), the ratio of credit to deposits in banks, the value added of the financial sector as a share in GDP and the ratio of M2 money to GDP; (iii) demographic structure, including the old age, young age and total dependency ratios; (iv) income-related variables, represented by the level and growth of GDP per capita; and (v) the share of industrial output in GDP, reflecting preferences of Asian economies for trade expansion by building large manufacturing capacity.

The financial variables deserve more attention. The real interest rate is the only price variable used in our regressions to detect the effect of finance on saving. Higher values of quantity variables represent higher degrees of financial depth, and the first two such variables (credit.ps and credit.do) capture the access to borrowing opportunities. The ratio of credit to deposits in banks indicates the degree of leverage or viability in the whole banking system. The financial value added as a share of GDP implies the size of the financial sector in an economy. The M2-to-GDP ratio is an indicator widely used to measure the degree of financial depth in developing countries. Financial variables may enter into interaction terms to detect certain cross-effects on saving or consumption.

It is worth noting what kinds of impacts the above factors would have on saving or consumption.11 As predicted by our theory (developed earlier), the estimates for inequality coefficients in consumption regressions could be positive in OECD countries but negative in Asian nations. Moreover, financial depth would be expected to contribute positively to consumption in OECD countries, while the industrial-GDP ratio should have a negative influence on consumption in emerging Asia.

The expected signs for other coefficient estimates are supposed to be very similar to those in previous studies relevant to saving decisions. The estimated coefficient on the lagged dependent variable would be anticipated to have a positive value to reflect habit formation. The overall impact of real interest rates on consumption could be positive if their income effect outweighs their substitution effect. Dependency ratios must affect consumption positively in all economies in consistency with the life-cycle hypothesis. The positive saving effects of income levels or growth rates of per capita GDP would be expected to arise in all regressions as established in intertemporal consumption theory.

4.2 Data Description

Most information used in our empirical analysis is the cross-country, time-series macroeconomic annual data from the CEIC Database and the World Development Indicators documented by the World Bank. Our financial data draw from the Financial Development and Structure 2009 supplied by the World Bank. Income distribution data we use are collected from the World Top Income Database and the UNU-WIDER World Income Inequality Database. These panel data sets are heavily unbalanced, especially for Asian economies, with the number of time-series observations varying considerably across countries.

Twenty-three OECD countries are involved in our regressions after excluding potential outliers.12 Instead of using long time-series data that span several decades (Cook, 1995; Schmidt-Hebbel and Servén, 2000; Smith, 2001; Li and Zou, 2004), our empirical analysis focuses only on the period from 2000 to 2007. This is a special period of time, during which the developed world witnessed a dramatic run-up in income inequality, an accelerating pace of leverage hike or debt accumulation, a rapid expansion of the financial sector and a sharp decline in the saving rate (Glick and Lansing, 2010; Kumhof and Ranciere, 2010; McKinsey Global Institute, 2010). These changes led to worsening global imbalances and eventual financial meltdown in OECD economies.

Twelve Asian countries are included in our empirical study.13 Many of them, widely recognised as ‘miracle’ economies in emerging Asia, have achieved strong output growth facilitated by high savings, but experienced income inequality deterioration for a few decades. The determinants of their consumption ratios and saving rates are investigated in our regressions for the period between 1990 and 2007, which saw their active participation in economic globalisation and financial integration with the Western world.

4.3 Estimation Results

Our discussion about estimation results is organised around core specifications displayed in Table 1, which is followed by Tables 2 and 3 reporting some variants of regressions. We resort to many regression specifications to check for the robustness of testing for the empirical validity of our theory proposed earlier. The set of regressors designed for OECD in our estimation is not exactly the same as, but slightly different from, that for Asian economies to account for differences in socioeconomic policy and structure between the two groups of countries.

In Table 1, we report estimation results from basic regressions for the private consumption ratio in OECD and Asian economies based on unbalanced panel data. The Sargan statistics find no evidence against the validity of instruments specified for GMM system estimation. Note that the key measure of financial depth in our regressions is domestic credit to private sectors relative to GDP (credit.ps) for both OECD and Asian countries, with the growth of M2 money relative to GDP used only for emerging Asia. Let us interpret the results recorded in columns 1 and 4 first and then compare them with those listed in other columns and obtained under alternative measures of financial depth.

Different inequality-consumption links can be made clearer through comparison between OECD and Asian economies, so our discussion below revolves around such a comparison. The lagged private consumption ratio in OECD countries has a positive and significant coefficient estimate, whose size reveals the degree of persistence in consumption habit over time and is observed to be large in these mature economies.

The coefficient on the inequality variable carries a significant estimate, which is found to be positive for OECD yet negative for Asian economies. A rise in the Gini index is observed to boost consumption in OECD countries but depress it in Asia. This empirical observation is consistent with the assertion of our theory proposed earlier. Previous findings by other authors are similar to ours, in that consumption is not dampened by widening income gaps in advanced countries (Heathcote et al., 2010).

Demographic factors we estimate have the same effect as predicted by the life-cycle hypothesis. Our coefficient estimates for dependency ratios are significant in all economies. The impact of old dependency on consumption is strong in OECD countries with ageing societies for long,14 where the private consumption ratio increases with a rising old dependency ratio. For Asian economies with younger population, young dependency has an impact on consumption similar to that of old dependency, where a rise in young (/old) dependency leads to an increase in consumption.

We find that the consumption ratio is negatively affected by income levels in OECD countries but by income growth in emerging Asia, with most of these effects being statistically significant. The different effects of income levels and growth rates between the two groups of countries coincide with the fact that income growth is low in OECD economies but high in Asia while OECD's income level is higher than Asia's. It seems that the Keynesian theory is still valid in the West while the life-cycle hypothesis works better in Asia. Consumption growth usually lags behind income growth in Asia due to serious inequality. For example, China's household consumption as a percentage of GDP has actually fallen since 2001 despite its average 10 per cent GDP growth per year for decades.

Financial depth (measured by credit.ps) is found to significantly affect consumption in a positive (/negative) way for OECD (/Asian) countries. When domestic credit to private sectors rises relative to GDP, private consumption relative to GDP grows in OECD countries, but falls in Asian economies (columns 4 and 5). This finding reflects the fact that the financial sector is oriented to encourage consumption via cheap, easy credit in developed countries but to promote investment and industrialisation in developing countries. This fact is corroborated in Asian economies by their significantly negative relationship between the ratio of private consumption to GDP and the share of industrial output in GDP. An increase in such a share is observed to be associated with a decline in the private consumption ratio.

Similar results emerge from alternative measures of financial depth used for OECD economies. Private credit by depository banks and other financial institutions relative to GDP (credit.do) and the ratio of credit to deposits in banks are used for estimation (columns 2 and 3). These two measures are also associated positively and significantly with private consumption in OECD countries, while the significance and signs of estimated coefficients on other regressors remain unchanged. The two measures do not exist in most of Asian economies, but data are available for the growth rate of the M2-to-GDP ratio and the share of financial value added in GDP. Both these variables, thus used as alternative measures of financial depth for Asia, are found to impact negatively on private consumption (columns 6 and 7). This result arises from the emphasis of Asian finance on industrial activity rather than on consumer credit, so that Asian people have to spend out of their own income or previous savings under liquidity constraints.

Private consumption in Asia is positively related to the real interest rate (column 7), seemingly suggesting the dominance of income over substitution effects. A more realistic interpretation may exist for this positive relation. In some Asian economies, like China, the interest rate on bank deposits is often capped by governments to a very low level, even below the inflation rate, to encourage manufacturing investment for trade expansion. However, the lack of social safety net tends to force people to diversify away from uncertainty about future income and policy by consuming less and saving a larger fraction of their income. This precautionary motive for saving arises in China despite low or even negative real returns on deposited money, thereby resulting in a positive association of consumption with interest rates (Nabar, 2011; Lardy, 2012).

Tables 2 and 3 are presented to show the robustness of our empirical results using alternative indicators for income inequality and by incorporating into the regressions the effects of interactions between inequality and finance.15

We take the results of Tables 2 and 3 as supportive of our core regression findings in Table 1. The new results for OECD and Asian economies are summarised separately for Tables 2 and 3.

In Table 2, when income shares of the top (0.5, 1, 5) per cent of population are used as inequality indicators, the positive inequality-consumption link still holds for OECD countries. The estimated coefficients on these indicators appear to be larger in size than those on the Gini index. Column 3 reveals that a rise in the income share of the top 5 per cent group would increase the private consumption ratio because the wealthy in this group can lend more to poor people for consumption through credit markets. The impacts on private consumption from other regressors, such as demographic, financial and income-related variables, are similar to those in the Table 1 basic regressions. Moreover, when income inequality is allowed to interact with financial depth, we find that their positive effects on private consumption are intensified significantly in OECD countries because all interaction terms carry positive coefficient estimates. The negative sign of the estimated coefficient on (credit.ps × real interest rate) suggests that a lower real interest rate strengthens the positive effect of credit on consumption because people would be encouraged to use more credit for consumption as their cost of borrowing falls.

In Table 3, for most Asian economies, data on top groups’ income shares are not available. We thus continue to rely on the Gini index for estimation and let it interact with financial variables to examine the robustness of our empirical results. The new results for five variables (the Gini index, dependency ratios, income growth, the share of industrial output in GDP and the real interest rate) are consistent with those of the Table 1 basic regressions in terms of estimates’ signs and significance. Moreover, the significantly negative coefficient estimates for credit.ps × Gini, (M2/GDP.growth) × Gini and (financial value added/GDP) × Gini imply that financial systems in Asia reinforce the negative linkage between inequality and consumption. An additional interaction term shows that the negative effect of inequality on consumption becomes weaker with a lower real interest rate in emerging Asia.

Table 4 summarises the basic regressions for the aggregate saving rate and is used to check for the robustness of our empirical results. Clearly, the estimation results from saving regressions in Table 4 are largely consistent with those from consumption regressions in Table 1. The Table 4 results are briefly interpreted as follows.

By contrast, financial depth indicators (credit.ps, (M2/GDP) growth × Gini, and financial value added as a share in GDP) have positive and significant estimates in Asia, suggesting the role of finance in facilitating aggregate saving rather than promoting consumption. A significantly positive coefficient estimate is obtained for the industrial share in GDP, and this result reflects the fact that much of increased saving in Asia has been channelled into investment for industrialisation and trade expansion.

For OECD countries with high demographic maturity aggravated by low birth rates, the estimated coefficient on their old dependency ratio is substantially larger than that on their young dependency ratio, indicating that the positive saving effect of their high-income levels might be dampened by their ageing population structure.