Household Deleveraging and Saving Rates: A Cross-Country Analysis
Romain Bouis* International Monetary Fund This draft: February 2015 Abstract A number of economists argue that household deleveraging can exert a significant drag on the economic recovery by weighing on consumption. Using a sample of OECD countries, this paper finds evidence of a negative relationship between household saving rates and changes in debt-to-income ratios. This relationship is however asymmetric, being statistically and economically more significant during debt build-ups than in deleveraging periods. Moreover, the effect of declining debt ratios on saving is only significant in economies where consumer credit is widespread. Results therefore suggest that the upward adjustment in saving rates associated with household deleveraging may widely differ across countries and may be overestimated for several economies. JEL codes: G01; E21; E44. Keywords: Household debt; saving rates; deleveraging; consumer credit; housing equity withdrawal.
─────────────────────────
*E-mail address:
[email protected]. I would like to thank participants to internal seminars at IMF, OECD, and Banque de France, to the CEMLA in Mexico City, and to the 3rd UECE conference in Lisbon, as well as Daniel Cooper and Christian Upper for useful comments and suggestions. The usual disclaimer applies.
1. Introduction Following a sharp increase in the run-up to the crisis, household debt-to-disposable income ratios have started to decline in some OECD countries and are expected to reverse in other countries as credit conditions tighten. According to the current consensus, such developments should bring significant costs for economic growth as deleveraging tends historically to be accompanied by higher household saving rates (Figure 1). In the United States, where debt ratios have declined by more than 20 percentage points since end 2007, household deleveraging has, for instance, been considered as responsible for the spectacular drop in consumption in the wake of the crisis (Glick and Lansing, 2011 or Dynan, 2012).1 Recent analyses have therefore warned of potential macroeconomic costs associated with the household deleveraging expected to take place in other countries in the near future (e.g. Cuerpo et al., 2013 for European economies or Krugman, 2013 for Canada). – Insert Figure 1 about here – The relationship between saving and deleveraging is however not clear-cut on theoretical grounds. The usual argument put forward by economists to justify a dampening effect of deleveraging on consumption is that by increasing debt repayments to reduce their indebtedness, households prioritise saving over consumption (see e.g. McCarthy and McQuinn, 2014 or Dombret, 2013). In practice, however, the bulk of the reduction in household debt is not driven by higher debt repayments, but by lower new borrowings and rising debt defaults (Bhutta, 2014), implying that there is no automatic link between household deleveraging and saving. Expanding defaults tend actually to support consumption, acting as a “financial decelerator” (Elul, 2008), while the impact of lower new borrowings on saving may vary depending on the importance of consumer credit in the economy. In this context, the rise in saving rates usually observed during deleveraging episodes may essentially reflect higher precautionary saving in an environment of heightened economic uncertainty, higher unemployment, and depressed valuations of real and financial assets, without any direct connection to changes in household debt. Against this backdrop, this paper examines the relationship between household saving rates and the change of debt ratios for a sample of 28 OECD countries over the period 1980-2012 to shed light on the macroeconomic costs of household deleveraging. Several studies have already examined the role of 1
A growing theoretical literature (e.g. Eggertson and Krugman, 2012; Guerrieri and Lorenzoni, 2011; and Hall, 2011) also points to significant negative effects of household deleveraging on private consumption and output. Guerrieri and Lorenzoni (2011) estimate for instance that a decline in the household debt-to-GDP ratio by 10 percentage points generates a 1% drop in output. 2
credit in explaining household saving rates. This paper differs from previous literature in two respects. First, it uses household debt series compiled for a large sample of countries from national balance sheet statistics (see Bouis et al., 2013 for details). Other panel studies use instead a broader measure of total credit to the non-financial sector, i.e. including households and private non-financial companies (e.g. Lyoza et al., 2000 or Mody et al., 2012). Second, while existing literature focuses on the negative effect of credit on saving stemming from credit market liberalisation, the change in credit conditions considered in this paper is decomposed into rising and declining debt ratios. By looking specifically at the effects of debt ratio declines on saving for a sub-sample of observations covering a full boom-bust credit cycle, this paper provides the first cross-country estimates of the relationship between credit and saving during deleveraging episodes. The empirical analysis yields two main results. First, the negative relationship between saving rates and changes in debt ratios is statistically and economically much more significant for rising than for declining indebtedness. Even when limiting the estimation sample to observations covering a full boombust credit cycle, the effect from declining debt ratios, although non-negligible, remains smaller than the effect from rising indebtedness or from other determinants of saving like disposable income growth, changes in asset prices, or economic uncertainty. The asymmetric impact of credit on saving may reflect a financial-deepening effect of credit market liberalisation, while financial reform reversals are scarce and business-cycle changes of saving rates are weakly related to changes of debt ratios in the average OECD economy. Second, panel estimates of the saving-credit relationship hide substantial cross-country heterogeneity. Declining debt ratios translate into higher saving rates in economies where consumer credit is prevalent (e.g. in Australia, Canada, Korea, the United Kingdom, and the United States) but do not impact saving in economies with poorly-developed consumer credit markets, including during historical deleveraging episodes. Estimates for a sub-sample of countries also show that the change in consumer debt is significantly related to saving while the change in housing debt (which however represents the bulk of household debt) is not, confirming the importance of consumer credit in driving the negative creditsaving relationship. This result could reflect a reverse causality issue between consumer credit and saving. Empirical evidence for the United States however suggests that the contraction in consumer credit observed during the deleveraging was mainly due to some cutbacks in the provision of credit by banks (see e.g. Gropp et al., 2014), rather than to a reduced credit demand from households stemming from standard wealth effects.2 With consumption of liquidity-constrained households being affected by this
2
As noted by Gropp et al. (2014), new consumer borrowings in the United States may have declined with the housing bust as households optimally reduced their demand for debt and implicitly, for consumption, in response to a 3
credit tightening, the macroeconomic impact of deleveraging crucially depends on the size of the consumer credit market in the economy. Empirical analysis therefore suggests that the costs associated with household deleveraging may be overestimated. The effect of changes in debt ratios on saving is asymmetric for the average OECD economy, and even in countries where consumer credit is important, the impact of deleveraging on saving is small in comparison to effects from uncertainty or asset prices. These findings are in line with the conclusion reached by Carroll et al. (2012) for the United States that increased credit availability accounts for most of the long-term decline in saving, while fluctuations in wealth and uncertainty capture the bulk of the business-cycle variation.3 They also lend support to findings by Takáts and Upper (2012, 2013) that the strength of economic recovery is poorly correlated with private debt deleveraging. The cushioning effect of defaults on the decline of consumption may partly explain the weak impact of deleveraging on saving. While rising debt is associated with higher spending, defaults accompanying deleveraging could mitigate the fall in consumption resulting from lower credit growth as households’ cessation of debt repayments raises the amount of income available for consumption. Policies facilitating household debt write-downs could therefore limit the macroeconomic costs associated with deleveraging. Another explanation may lie in the relatively modest response of consumption to short-term credit developments in some countries. In particular, in economies with poorly-developed consumer credit markets and where rising debt ratios had therefore little effects on consumption over the past fifteen years, there may be no reason to expect any large effect of deleveraging on saving rates. Deleveraging in these countries may be merely associated with lower housing prices, without any significant impact on saving given the controversial effect of housing wealth on consumption (see e.g. Cooper and Dynan, 2015 for a recent survey on wealth effects on household consumption). This paper complements recent empirical research analysing the response of consumption to the level of household debt (the debt overhang effect) instead of to the change in the debt ratio (the leveraging and deleveraging effects). Mian and Sufi (2011a), for instance, find that high household debt built up in some U.S. counties during the boom led to weaker economic conditions in those counties in the early part of the recovery. Mian et al. (2013) estimate a larger response of consumption to negative shocks to household wealth for households with higher leverage and for those who are more likely to be in negative negative shock on wealth (demand-side explanation). Alternatively, banks may have tightened their credit standards in areas where real estate prices declined more sharply (supply-side explanation). Evidence tends to support the supply-side explanation. 3
The authors find that the largest contributor to the decline in consumption in the U.S. during the Great Recession was the collapse in household wealth, followed by the increase in precautionary saving while credit availability played a substantially smaller role. 4
equity. Dynan (2012) and Dynan and Edelberg (2013) find that excessive leverage has contributed to the weakness in consumption in the wake of the crisis, even after controlling for other factors expected to influence spending, such as changes in income and wealth.4 Finally, Cooper (2012) finds little evidence that deleveraging affected household consumption in the United States. While the consumption growth of households who reduced their debt was lower than the consumption growth of other households, this pattern is essentially the same prior to and during the Great Recession. None of these papers has however directly examined the impact of declining debt ratios on the level of consumption or saving. Without challenging the view that too much debt may weigh on consumption, results presented in this paper (which are robust to controlling for the level of debt or a proxy of the debt-service burden) suggest that deleveraging does not necessarily translate into higher saving rates for the average economy. Importantly, the response of saving to credit can vary greatly across countries, depending on the size of the consumer credit market. The rest of the paper proceeds as follows. Section 2 presents a brief definition of household deleveraging. Section 3 discusses the relationship between saving and credit, explaining why saving rates and changes in debt ratios may be empirically correlated and why the strength of this relationship may vary across economies. Section 4 introduces the empirical approach. Econometric results are then presented and discussed in Section 5. Section 6 concludes the paper.
2. Definition of household deleveraging Episodes of household deleveraging are traditionally defined as periods of declining debt-toincome ratios over several years.5 Historically, household deleveraging in OECD economies lasted more than 4 years for a decline in the debt ratio of 16 percentage points on average. These numbers however hide a significant variation across episodes (Table 1). For instance, in the early nineties, Japan experienced a decline in its debt ratio of only 4 percentage points over two years. In contrast, in Denmark, Finland, Norway, and Sweden, household deleveraging lasted between 6 and 8 years, with drops of debt ratios of between 25 and 50 percentage points of disposable income. Despite significant downward adjustments, in the vast majority of cases, debt ratios did not return to their pre-boom levels, possibly reflecting structural effects from financial market liberalisation. 4
Dynan (2012) however recognizes that the economic impact of leverage is relatively modest while the econometric estimates have large standard errors. An increase in the household’s mortgage leverage ratio of 0.1 is for instance associated with a decline in annual consumption growth of 0.3 percentage point. 5
Strictly speaking, household deleveraging should be analysed on the basis of debt-to-asset ratios. The value of assets is however much more volatile than the value of debt, and debt-to-asset ratios can offer a misleading picture of household debt sustainability. Data on household non-financial assets are besides only available for a limited number of countries so that deleveraging episodes are generally identified using debt-to-income ratios. 5
– Insert Table 1 about here – Deleveraging can occur through lower debt, higher economic growth, or higher inflation, as noted by Tang and Upper (2010). It is however quite rare that household debt-to-disposable income ratios decrease because of a fall in the absolute level of debt. Deleveraging occurs instead through credit growth lagging behind nominal income growth as shown in Table 1 where the change of the debt-to-income ratio is decomposed into a net credit effect and a nominal growth effect.6 With lower inflation rates today than in the past, the contribution of nominal income growth to deleveraging may be less important. Stagnant or declining debt levels are indeed more common in the on-going deleveraging processes taking place in some OECD economies since 2007 (Table 2),7 although this may reflect the fact that many of these countries are still in their early phase of deleveraging with a sharp contraction of credit and low nominal income growth. – Insert Table 2 about here – The change in the level of household debt can itself be decomposed into inflows, i.e. new borrowings, and outflows, which include debt repayments and defaults: ΔDebt = New Loans – Debt Repayments – Debt Defaults.
[1]
In general, available data do not allow distinguishing between these three components. One exception is the United States where the Federal Reserve Bank of New York’s Consumer Credit Panel offers detailed information on consumer debt and loan performance. According to these individual credit records data, new loans appear as the main driver of the evolution of debt. The recent decline in household debt in the United States is for instance found to be mainly explained by lower credit growth rather than by expanding debt repayments and mortgage defaults. Even if debt repayments and defaults had not grown since 2007, mortgage debt in the United States would have declined over the 2009-11 period because of a material drop in new borrowings, mainly reflecting a dramatic decline in first-time homebuying (Bhutta, 2014). The key role of new borrowings in driving debt levels is confirmed by
6
The change of the debt-to-income ratio between the dates t and t+T can indeed be decomposed into a change in the level of debt (a net credit effect) and a change in nominal income growth (a nominal income growth effect) as follows:
Debt Debt d d Y t T Y t
Debt t T Debt t Y d t T Y d t Debt t d . Y d t T Y dt Y t T
7
Note that the debt ratio in Switzerland experienced a downward adjustment over 2006-2008 as reported in Table 1, but has since then increased again and is now exceeding its 2005 peak. 6
Knotek and Braxton (2012). Differentiating between the number of consumers taking on more debt and the dollar amount by which individual consumers are changing their debt levels, the authors report that the bulk of deleveraging has occurred through a sharp decline in the number of consumers taking on additional debt. Finally, although defaults are playing a smaller role than new borrowings in the evolution of debt, they still explain nearly 30% of the debt decline in the United States over 2008-2011 (Bouis et al., 2013).8 The overall contribution to household deleveraging may be even larger as rising charge-offs increase losses of the banking sector, leading to a tightening of credit conditions and a reduction of new loans (Koo, 2011 and Li and Patwari, 2012).
3. The relationship between credit and saving Historically, episodes of credit booms and busts are often associated with changes in saving rates, as shown in Figure 1. The nature of the relationship between the change in household debt and saving remains however largely unexplored and it is useful to discuss the influence on saving of each of the three drivers of debt appearing in relation [1]. First, it must be noted that rising debt defaults cannot account for the positive relationship between deleveraging and saving rates as defaults may actually increase rather than dampen consumption by freeing up resources available for households’ expenditures (Cooper, 2012). Second, although higher debt repayments mechanically translate into higher saving, they are unlikely to explain why saving increases during deleveraging, as in practice the decline in debt ratios is not explained by higher debt repayments, as seen above. Moreover, arguing that households significantly raise their debt repayments in bad times, as claimed by some observers,9 seems counterintuitive. During the deleveraging phase, households (who do not default) are instead more likely to keep paying down debt at a similar pace as they do in normal times and to significantly reduce their new borrowings. This implies that understanding the effect of deleveraging on saving rates requires examining the relationship between new borrowings and saving.
8
This estimate of the magnitude of the contribution of defaults to debt reduction is lower than the two-thirds number reported in MGI (2012) or Cooper (2012). This is because it only considers the contribution from the crisis-related increase in charge-offs (see Bouis et al., 2013, footnote 8, for details). 9
For instance, Svensson (2012) notes that Riksbank advocates of “leaning against the wind” have stated that to restore debt-to-asset ratios in face of declining housing prices, households may “(…) want to pay off some of their debt to come down to a more suitable level. This means they will prioritise saving over consumption.” Likewise, Dombret (2013) claims that when “the value of the assets held by the private sector fall, while the value of the liabilities remains the same (…) households and enterprises are compelled to increase their saving in order to pay off their debts.” 7
As pointed out by Dynan (2012), in traditional models saving is determined by income, wealth, preferences, uncertainty, and the return on saving, but credit does not exert an independent effect. On the accounting side, the following identity still suggests a possible relationship between saving and the change in debt: S – I = ∆Financial Assets – ∆Financial Liabilities,
[2]
where S denotes household saving, I household gross fixed capital formation (namely housing investment and the acquisition by households of existing houses and lands from other institutional sectors), ∆Financial Assets the net acquisition by households of financial assets (excluding valuation effects), and ∆Financial Liabilities net borrowings (i.e. new borrowings minus debt repayments). Identity [2] indicates that at the aggregate level, an increase in net borrowings can finance consumption (i.e. lower saving), investment or the acquisition of financial assets. Nothing ensures however that rising debt is used to finance higher consumption as it can instead simply be used for the acquisition of financial assets or the financing of investment. For instance, in the case of the United Kingdom, Nickel (2004) reports that both the rate of accumulation of financial liabilities and of financial assets rose together over 1998-2004. Consequently, the proportion of post-tax household income which is consumed remained stable over the same period, suggesting that there is no strong relationship between aggregate consumption growth and aggregate debt accumulation. 3.1 Why deleveraging and saving rates are empirically correlated Even in the absence of a mechanical link between saving rates and changes in debt ratios, several factors, directly or indirectly related to the evolution of indebtedness, can explain why saving rates rise in deleveraging periods:
First, credit has a direct influence on saving under the hypothesis of liquidity-constrained households borrowing to smooth transitory income shocks. When credit conditions tighten, as is typically the case in deleveraging periods (see Bhutta, 2014 and Gropp et al., 2014 for empirical evidence for the United States), households cannot borrow as easily as before to offset negative income shocks and have to increase their buffer-stock saving (Caroll, 2001). In line with this hypothesis, empirical analyses find that credit market liberalisation played a significant role in the
8
steady decline of saving rates observed since the early eighties (e.g. Loyaza et al., 2000).10 Studies also find that short-term changes in credit availability partly explain business-cycle fluctuations of saving (e.g. Caroll et al., 2012);
Second, household deleveraging tends to be accompanied by house price declines and other asset depreciations (Figure 2) that in turn impact consumption through wealth effects. Such effects, which do not appear in identity [2] as national accounting aggregates are measured net of valuation effects, are estimated to significantly affect saving (although consensus on the economic importance of these wealth effects has not yet been reached, see Cooper and Dynan, 2015, for a survey). Carroll et al. (2012) for instance identify the collapse in household wealth as the largest contributor to the decline in consumption in the United States during the Great Recession;
Third, deleveraging is associated with declining house prices and therefore, with a lower availability of home equity-based borrowing, reducing consumption (Mian and Sufi, 2011b). The economic impact of housing equity withdrawal (HEW) on consumption is however controversial. Whether the proceeds from HEW drive consumption or whether the correlation between HEW and consumption results from the wealth effect of rising property prices remains uncertain. In addition, the share of home-equity based borrowing used for consumption purpose may be relatively marginal in some countries (Menegatti and Roubini, 2007);11
Finally, deleveraging and credit busts are often associated with depressed labour markets and higher economic uncertainty (Figure 2) leading to higher precautionary saving. Mody et al. (2012) estimate for instance that at least two-fifths of the increase in saving observed in 2007-2009 in OECD countries is due to higher unemployment risk and GDP volatility. – Insert Figure 2 about here –
10
Mortgage credit liberalisation may also have contributed to the long-term decline of saving rates by reducing the minimum deposit required for first-time homebuyers (Muellbauer, 2008).
11
Klyuev and Mills (2007) argue for instance that the recent fall in the U.S. saving rate cannot be attributed to HEW but mainly to housing wealth effects. Financial liberalisation has increased the liquidity of home equity by making its withdrawal easier, but HEW itself does not explain changes in saving rates. Likewise, Benito et al. (2006) document that the relationship between HEW and consumption in the United Kingdom has weakened in the beginning of the years 2000 as agents benefited from greater access to unsecured credit during the 1990s. In contrast, Bailliu et al. (2012) report a significant effect of HEW on consumption in Canada based on micro data, estimating an average share of home-equity extraction used to finance consumption and home renovation of about 40 per cent. 9
While the first factor (the liquidity-constrained households’ hypothesis) implies a causal relationship running from credit to saving, credit does not play a direct role in the case of the three other factors. Identifying the direct influence of credit on saving therefore requires to properly control for effects indirectly related to changes in debt ratios. 3.2 Implications in terms of cross-country heterogeneity Several credit market features and other country-specific settings potentially affect the transmission of credit to saving implying that the strength of the credit-saving relationship may vary across economies. These settings include: -
The share of consumer credit in disposable income. Expansion in consumer credit is strongly associated with consumption growth (see e.g. Bacchetta and Gerlach, 1997) and its importance in the economy is expected to strengthen the link between saving and credit. Although the evolution of household debt is mainly driven by mortgage debt, consumer and mortgage credits are highly correlated (Chmelar, 2013) and the change of total debt is likely to be accompanied by a change in consumer debt in the same direction, with immediate effects on the consumption of liquidityconstrained households.
-
The availability of housing equity withdrawal to borrow against accumulated house equity. Greater “liquidity” of housing equity is indeed associated with a higher impact of housing sector activity on the rest of the economy by increasing the financial accelerator effect of rising house prices on consumption (see IMF, 2008, for some cross-country evidence).
-
Homeownership rates may raise the macroeconomic effects of housing wealth effects and HEW. Campbell and Cocco (2007) find microeconomic evidence of a positive effect of house prices on consumption for the cohort of old households who are homeowners, and an effect that is close to zero for the cohort of young households who are renters. The level of homeownership is, however, not a sufficient condition, as noted by Catte et al. (2004) who report a weak crosscountry correspondence between owner-occupation and the sensitivity of consumption to real house prices.
-
The level of the saving rate in conjunction with the degree of liquidity of household assets (as reflected by the presence of “illiquid” tax-deferred saving/pension accounts) may impact the relationship between borrowing and saving. As outlined by Menegatti and Roubini (2007), when households have low or negative saving and limited liquid assets, consumption cannot be increased in response to a positive wealth effect (or any exogenous shock) by reducing saving or 10
financial assets and the only way for households to raise consumption is to increase financial liabilities. -
Arrangements to limit the costs for households to default on their debt may cushion the impact of deleveraging on saving rates. As stressed by Cooper (2012), households’ cessation of mortgage payments actually raises the amount of income that is available for non-housing related expenditures and households may choose to default rather than reduce consumption. This “financial decelerator” mechanism (Elul, 2008) may partly explain why the damping effect of deleveraging on consumption estimated for the United States during the Great Recession, although non-negligible, is relatively modest (Dynan, 2012), in particular in comparison with the consumer spending collapse observed in 1930, before institutional changes lowered the cost of default by 1938 (Olney, 1999). As shown in Table 3, OECD countries can differ significantly with respect to the above-
mentioned characteristics, possibly implying a different response of saving to changing debt ratios across economies.12 – Insert Table 3 about here –
4. Empirical approach This section introduces the empirical approach, presenting the variables and the econometric specification. Variables The dependent variable is the household gross saving rate. The explanatory variable of interest is the change of the household ratio of total financial liabilities to gross disposable income.13 Regressions include as controls the usual determinants of saving identified in previous literature: -
Income-related variables, namely the level and the growth rate of real disposable income and the terms of trade, are expected to be positively related to saving rates as households tend to increase their saving when they become richer or their income grows faster (Lyoza et al., 2000);
12
Unfortunately, the lack of cross-country information on household debt write-down arrangements precludes investigating the importance of the interaction effect of credit with the cost of default. Coletta et al. (2013) consider the impact of the quality of bankruptcy law on household indebtedness but the indicator used by the authors, which comes from the World Bank Doing Business database, is related to the corporate sector.
13
A detailed description of the construction of the debt variable is presented in the appendix of Bouis et al. (2013). The household sector refers in this paper to the aggregate account of households and non-profit institutions serving households (NPISHs) as data for the household sector alone are rarely available. 11
-
The old and young-age dependency ratios are included to account for the existence of demographic structure effects;
-
The cyclically-adjusted government net lending as a percentage of potential GDP, which is found to significantly explain OECD saving rates in previous panel regressions (see e.g. Röhn, 2010), is incorporated to control for Ricardian equivalence effects;
-
The unemployment rate, the inflation rate, and GDP growth volatility (based on a GARCH model measure, following Mody et al., 2012), used as proxies of income and macroeconomic uncertainty, are expected to increase saving rates for precautionary motives;
-
Financial variables include the real short-term interest rate and household wealth, proxied by household financial net worth as a share of disposable income (lagged by one period to avoid reverse causality from household saving to wealth). Due to limited cross-country data coverage, this measure of household wealth does not include housing wealth, which yet represents a significant share of household assets. To address this issue, regressions incorporate the one-period lagged change in house prices. Alternative specifications also use the lagged change in stock market valuations given limited coverage of the household financial net worth variable. Data cover 28 OECD countries over the period 1980-2012. Annual data of saving rates are of
better quality than quarterly data while annual information on household debt offers a larger coverage across time, in particular for the deleveraging episodes of the 1980s. In alternative specifications, I also consider quarterly data to increase the number of observations per country or to analyse cross-country heterogeneity and the role of credit market institutions over a more recent period. Description and sources of the data are provided in the appendix. Econometric specification I estimate the following dynamic fixed-effects panel data equation:
si ,t 0 si ,t 1 1 ln RY d i ,t 2 GRY d i ,t 3Tradei ,t 4Old i ,t 5Young i ,t 6UNRi ,t Debt 7 Inflationi ,t 8Volatilityi ,t 9 IRS i ,t 10 d i t i ,t , Y i ,t
[3]
where s denotes the household gross saving rate, RYd the real household gross disposable income (in PPPs), GRYd the growth rate of the real household gross disposable income, Trade the terms of trade ratio, Old and Young are the old and young-dependency ratios, UNR the unemployment rate, Inflation the annual growth rate of the consumer price index, Volatility is a GARCH (1,1) measure of GDP growth 12
volatility, IRS the real short-term interest rate, Debt is the stock of household debt (or total financial liabilities), Yd the nominal household gross disposable income, ηi and γt are country and year fixed effects and ε the error term. Household financial net worth and government net lending are considered in a second step of the analysis as including these variables in the regressions reduces the sample size due to limited time coverage in the 1980s. Given persistence in the saving rate, the lagged value of this latter is included as an explanatory variable, implying that the estimates from using the least squares dummy variable (LSDV) estimator may be biased (Nickel, 1981). Alternative approaches like the instrumental variable (IV) or the Arellano and Bover (1995) General Method of Moments (GMM) estimators can be employed to address this problem. As underlined by Attanasio et al. (2000) or Cecchetti et al. (2011), these two estimators work well when the number of panel units is large and the time dimension is small. But for the typical size of a macroeconomic panel (T greater than 30 and N equal to 20), Monte Carlo simulations indicate that the bias created by using the LSDV estimator is more than offset by its greater precision compared to IV and GMM estimators. Equation [3] is therefore estimated by using the ordinary least squares (OLS) countryfixed effects estimator. As a robustness check, I also employ the bias-corrected LSDV estimator proposed by Bruno (2005) extending the approach of Bun and Kiviet (2003) to unbalanced panels. Results (not reported) are qualitatively similar. Simultaneity or reverse causality may also be a concern. Saving may for example impact households’ decision to take a credit, insofar as an exogenous decrease in consumer spending depresses the need to borrow to finance that spending. In this case, OLS estimates would be biased toward finding a more negative relationship between saving rates and changes in debt ratios. This would however also imply that an insignificant estimated effect of debt reduction on saving rates could be considered as a conservative result with respect to the hypothesis of no relationship between deleveraging and saving.
5. Estimation results This section presents the econometric results. It first discusses estimates of the baseline specification. The effect of the change in the debt ratio is then decomposed into a rising and a declining indebtedness effect in sub-section 5.2. The respective impacts of debt build-ups and of deleveraging for a sub-sample of credit boom-and-bust observations are analysed in sub-section 5.3. Finally, cross-country heterogeneity and the role of institutions are examined in sub-section 5.4.
13
5.1 Baseline equation Table 4, column (1) presents results of the baseline equation. The lagged household saving rate has a positive and significant coefficient with a large degree of persistence (estimated coefficient of 0.79), implying that the long-run effects of other explanatory variables are almost fifth as large as their respective short-run effects, in case these variables change permanently. Estimates of coefficients of explanatory variables are broadly in line with previous findings of the literature. Revenue-related variables – the log of real disposable income (in PPPs), its growth rate and the terms of trade – have the expected positive sign. The growth rate of the real disposable income has a material impact on saving with a one-standard deviation increase in this variable being associated to a rise of the household saving rate in the short run of 0.9 percentage points, representing the highest effect among the explanatory variables considered here. Unreported regressions indicate that both positive and negative growth rates of real disposable income are positively related to saving rates (with coefficients statistically highly significant), suggesting that households increase their saving when disposable income grows faster but dissave as disposable income declines. This latter result is consistent with the hypothesis that households run down saving in face of a negative income shock to smooth consumption. It is also in line with the observation in several European economies (e.g. Austria, Belgium, Greece, Italy or Spain) of a sharp decline in household saving rates in the wake of the Great Recession, despite higher economic uncertainty, in a context of falling disposable income. The negative coefficient on the old dependency ratio is consistent with standard life-cycle models of consumption. A one-standard deviation increase in the old-age dependency ratio leads to a saving rate rise of 0.5 percentage points. As expected, higher economic uncertainty, as proxied by unemployment, inflation, and GDP growth volatility, translates into larger saving rates, probably reflecting some precautionary saving motives. The saving rate increases by 0.4 percentage points following a one-standard deviation increase in unemployment or in inflation, and by 0.5 percentage points in response to a one-standard deviation increase in GDP volatility. The real interest rate is positively related to the saving rate, suggesting that its substitution effects outweigh its income effects. This result contradicts Loyaza et al.’s (2000) finding of a negative effect of the real interest rate on household saving but is consistent with Mody et al. (2012) whose sample of countries is closer to mine. In alternative specifications (results reported below), the sign of the coefficient however turns negative, confirming the unambiguous effect of the interest rate on saving. The real interest rate variable has the smallest economic effect, with a one-standard deviation increase in this variable being associated with a rise of the saving rate of 0.2 percentage points. 14
Finally, the estimated coefficient of the change of the household debt-to-disposable income ratio is highly significant (under the 0.1% statistical level), with the expected negative sign. The economic impact of the variable is also important. A one-standard deviation increase in the change of the debt ratio – corresponding to more than 5 percentage points of gross disposable income – is associated with a 0.7 decline of the saving rate in the short run, representing the second largest economic impact on saving, after the effect from real disposable income growth. In unreported regressions, I also decompose the effect of the annual change of the debt-todisposable income ratio into a net credit effect and a nominal income growth effect as follows:
Debt d Y t
Debtt Y d t Debtt 1 d . Y dt Y t 1 Y dt
[4]
The first term on the right-hand side – the net credit-to-disposable income ratio Debt d Y
t
– appears
t
in the accounting identity [2] linking household saving to the change in financial liabilities (after dividing all terms of the identity by household disposable income) and is therefore expected to be an important driver of saving rates.14 The second term on the right-hand side – the nominal income growth effect
Y d t Debt t 1 – also matters for the analysis of the effect of deleveraging on saving, given that several Y d t 1 Y dt episodes of household deleveraging occur through credit growth lagging behind income growth as seen in Section 2. Results show that both coefficients are statistically highly significant, suggesting that household saving decisions are influenced by changes in debt ratios whether these latter are driven by changes in the absolute or in the relative level of credit. The economic impacts on saving of the net borrowing and of the nominal income growth effects are besides roughly similar. Additional control variables In column (2), I report the impact from lagged net financial wealth (scaled by disposable income). The coefficient of this variable is not statistically significant. The change of the lagged net financial wealth is not significant neither (results not reported).
14
Previous analyses of the determinants of household saving rates use a similar ratio of credit flow to GDP or to disposable income as a proxy of financial depth (e.g. Lyoza et al., 2000 or Mody et al., 2012). One notable difference of these studies compared with the present paper is that they consider a broad measure of credit to the non-financial private sector, instead of credit to households. 15
Column (3) shows the effect from changes in house prices and stock market valuations, lagged by one year. The coefficients of both variables have the expected negative sign, consistent with a wealth effect explanation but the housing price variable is not statistically significant. On theoretical grounds, the impact of house prices on saving from wealth effects is not clear-cut as an increase in house prices redistributes wealth within the household sector rather than boosting net aggregate wealth (IMF, 2008). The absence of a significant effect of house price growth on saving may besides result from the high correlation between the change in house prices and the change in debt. Dropping this latter variable from the regression indeed leads to a highly significant negative effect of the growth rate of house prices, confirming the hypothesis of a common causality between house prices and consumption (Attanasio et al., 2005).15 Economic effects, although non-negligible, are quite modest, with a one-standard deviation increase in annual stock market performance (+28.2 p.p.) translating into a decline of the saving ratio of 0.3 percentage points. The statistically insignificant effect from real house price growth is even smaller. A one-standard deviation increase (+7.5 p.p.) of the variable leads to a 0.2 percentage point decline of the saving rate. Results concerning the coefficient of the change in the debt ratio are also qualitatively similar when attempting to take into account Ricardian equivalence effects through the ratio of cyclically-adjusted government net lending to potential GDP (column (4)). The sample is then reduced to 19 countries due to limited time coverage of this variable. In line with the accounting identity [2], the housing investment ratio is positively related to saving, as shown in column (5). Considering all the variables together for the whole sample of countries (column (6)) or for a sub-sample excluding countries with limited time coverage (column (7)) does not modify the economic and statistical significance of the change in the debt ratio.16, 17 Results (not reported) are also robust to including the IMF index of financial liberalisation to capture the trend effect of financial liberalisation on saving, the change in money M2, the level of debt (whose coefficient is statistically insignificant, in line with results obtained by Carroll et al., 2012 for the United States), a rough proxy of
15
Results presented here therefore contrast with Case et al. (2005) who find for a panel of 14 OECD countries a larger effect on consumption of housing wealth than stock market wealth. As outlined by Muellbauer (2008), Case et al.’s (2005) study however omits to control for the effects of important drivers of consumption like income growth expectations, the unemployment rate, interest rates, and importantly, shifts in credit conditions which are highly correlated with changes in house prices. More generally, the empirical literature on the relative sizes of the financial wealth and housing wealth effects reaches mixed results (see Cooper and Dynan, 2015, for a survey).
16
Countries with limited time coverage (i.e. with less than 20 years of observations) include Czech Republic, Estonia, Greece, Hungary, Ireland, Poland, Portugal, Slovak Republic, and Slovenia.
17
I do not consider in the reported regressions the lagged net financial wealth as including this variable, which is never significant (in level or in first difference), reduces the sample size. 16
interest expenses (i.e. the level of debt times the interest rate) or the growth rate of real disposable income one year ahead (as in Mody et al., 2012) to capture expected income growth effects. Estimates are thus broadly in line with previous literature. Importantly, the economic impacts of most explanatory variables are very similar to the ones reported in Loayza et al. (2000) who are yet employing different estimation techniques (GMM and IV), suggesting that biases from using the LSDV estimator may be relatively limited. – Insert Table 4 about here – 5.2 Decomposing the change of the debt ratio into rising and declining indebtedness Table 5 shows estimates of the effects of positive versus negative changes in the debt-todisposable income ratio. The two coefficients have the expected negative sign but differ widely in terms of statistical and economic significances (column (1)). Only rising debt ratios are statistically significant at the conventional level (below 0.1%). They have besides a much larger economic impact on the saving rate than declining debt ratios. A one-standard deviation increase in the change of the debt ratio (+4.3 percentage points) is associated with a decline in the household saving ratio of 0.6 percentage points, while a decline in the debt ratio of a one-standard deviation (-1.7 percentage points) leads to a statistically insignificant increase in the saving rate of only 0.15 percentage points. Results are qualitatively similar when including other control variables or excluding countries with limited time coverage (columns (2) to (4)). Rising indebtedness significantly reduces saving rates but declining debt ratios are weakly associated with saving rates. This asymmetric effect of credit may reflect a number of factors. First, saving rates may decline with rising indebtedness because of a financial-deepening effect and apart from this structural effect, the response of saving to credit could be relatively weak. Second, the response of saving rates to declining debt could be highly heterogeneous across economies, implying large standard errors of panel estimates of the coefficient of declining debt ratios. Finally, if the bulk of countries have experienced a boom in credit but not yet a deleveraging phase, only the positive change of the debt ratio can be significantly related to the saving rate (observations with declining debt ratios represent only one-fourth of the estimation sample). The next sub-section addresses this latter issue by focusing on observations with a full boom and bust cycle in the household debt ratio. Sub-section 5.4 will deal with the cross-country heterogeneity in the saving–credit relationship. – Insert Table 5 about here –
17
5.3 Historical episodes of booms and busts in debt ratios To further investigate the respective impacts of rising indebtedness and of deleveraging on saving rates, equation [3] is re-estimated by restricting the sample to observations with a full boom and bust cycle of the debt-to-income ratio as reported in Table 1 (for each episode, the estimation period starts ten years before the peak and ends the year the debt ratio reaches its trough).18 Observations with declining debt ratios now represent almost 40% of the sample. Due to the small size of the sample and limited data coverage, explanatory variables like government net lending or net financial wealth cannot be considered in the analysis as this would imply excluding some important deleveraging episodes of the 1980s. As shown in Table 6, the change in the debt ratio is now significantly related to saving rates both in debt build-up and in deleveraging years (column (1)). The coefficient of the positive change of the debt ratio remains however statistically and economically more significant than the coefficient of the negative change. The effect from rising debt ratios on saving rates is actually three times as large as the effect from declining debt ratios. Historically, during these boom-bust credit cycles, a one-standard deviation increase in the annual change of the debt ratio (+4.5 percentage points) was associated with a decline in the saving rate of 1 percentage point while a one-standard deviation of the negative debt ratio change (-2.7 percentage points) led to a rise in the saving rate of only 0.3 percentage points. In alternative regressions (columns (2) to (6)), I control for the changes in house prices and stock market valuations. The one-year lagged growth rate of house prices is again non-significant (column (2)) but this result only holds true for declining house prices (column (3)). Using instead the contemporaneous growth rate of house prices, coefficients of both rising and declining house prices are statistically significant (results not reported). One-year lagged changes in stock market valuations are highly significant (column (4)). Controlling for these housing and equity valuation effects (columns (6) and (7)), the economic and statistical significance of declining debt ratios is smaller than before and barely or no more significant at conventional levels. Based on estimates of equation (6), a one-standard deviation increase in the growth rate of real house prices (+8.4 p.p.) and a one-standard deviation increase in the growth rate of real equity prices (+27.7 p.p.) are respectively associated with a decline in the saving rate of 0.35 and 0.5 percentage points. Economic impacts from real disposable income growth and from GDP volatility are respectively of 0.6 and 0.5 percentage points. In contrast, the impact of deleveraging represents a statistically insignificant 0.3
18
The deleveraging episodes of the Netherlands and of Switzerland in the early nineties are not included in the estimations because of a lack of data. 18
percentage point increase in the saving rate. These estimates therefore suggest that debt changes play only a limited role in the adjustment of saving rates during deleveraging phases in comparison to wealth effects (particularly those related to stock market valuations), disposable income growth and economic uncertainty. – Insert Table 6 about here – With a limited number of years per country (12 on average), LSDV estimates reported in Table 6 are more likely to be biased than for the whole sample. As a robustness test, alternative regressions are estimated with quarterly data from the BIS credit to non-financial private sector database.19 Results presented in Table 7 are broadly in line with those obtained with annual data. In particular, both the rise and the decline of the debt ratios are negatively related to saving rates but the positive change in the debt ratio appears economically and statistically much more significant than the negative one. Results are qualitatively similar when considering observations from five years before the peak of the debt ratio (columns (1) to (2)), or ten years before (columns (3) to (4)), to the trough. – Insert Table 7 about here – Overall, estimates suggest only some tentative evidence of an effect from declining household debt ratios on saving rates for the average OECD economy, even when limiting the estimation to observations covering a full boom-bust credit cycle. While the effect from rising debt ratios is robust to various specifications and control variables, the impact from declining debt ratios appears more fragile, lending support to the hypothesis of an asymmetric effect of credit on saving. Financial market liberalisation may partly account for this asymmetry, as household indebtedness rarely returns to its preboom level and reversals of financial reforms are scares (Abiad et al., 2008). Household debt defaults could also explain why for a given change of the debt ratio, the economic impact on saving rates of rising indebtedness is larger than the effect from deleveraging. While households may take advantage of looser credit conditions to finance consumption during the boom, defaulting on debt during the bust could mitigate the negative effect from lower credit and makes consumption smoother. Finally, cross-country 19
This database (described in Dembiermont et al., 2013) offers a more restrictive measure of household indebtedness than the total financial liabilities variable considered until now as it does not incorporate a number of items (namely “Securities other than shares” i.e. Financial derivatives; “Shares and other equity”; “Insurance technical reserves” which includes “Net equity of households in life insurance and pension funds reserve”, and “Other account payables”). Also, for some countries, only loans from domestic financial institutions are covered. Still, the BIS debt variable represents on average more than 90% of total financial liabilities and the evolutions of the two debt measures are strongly correlated. Three deleveraging episodes – Denmark, the Netherlands and Switzerland – are lost for data availability reasons. 19
heterogeneity in the credit-saving relationship could also account for the large standard errors of the estimated coefficient of declining debt ratios, as examined in the following sub-section. 5.4 Cross-country heterogeneity and the role of credit market institutions Although panel data estimates indicate a significant effect of credit on saving rates, this effect is unlikely to be identical across economies given the heterogeneity in credit market institutions and other country–specific institutional settings possibly shaping the response of household saving to credit. The rise in household debt observed over the past 15 years has been accompanied by a significant decline of the saving rate in some countries while in some other economies, household saving rates have remained quite stable despite a large increase in debt ratios. Scatter plots between quarterly net credit-to-disposable income ratios and household saving rates over 1995-2012 confirm the presence of a high cross-country heterogeneity in the strength of the credit-saving relationship (Figure 3). This latter is particularly important in Australia, Canada, Korea, the United Kingdom or the United States while some other countries show virtually no relationship between saving and credit.20 – Insert Figure 3 about here – Against this background, this section investigates the role played by credit market features and other institutional settings in explaining cross-country differences in the credit-saving relationship. Ideally, this issue should be addressed by using a sample of observations with a full boom-bust credit cycle so as to estimate the interaction effect of institutions during the debt build-up and in the deleveraging phase. Data on institutional settings for the deleveraging episodes reported in Table 1 are however available for only a few observations while the small number of countries composing this sub-sample implies a limited variation in institutional settings. As regards the recent period, only a few OECD economies have significantly engaged into a deleveraging process since 2007 and debt ratios in these countries are still declining (Table 2). The effect of institutions on the credit-saving relationship is therefore examined for
20
The correlations between saving rates and the changes in debt ratios are even stronger. As previously outlined, this is because the change in the debt ratio is by construction negatively related to the nominal growth rate of disposable income (see equation [4]) which is itself positively related to saving rates. The relationship between saving rates and the changes in debt ratios is therefore negative for most countries when the nominal income growth rate is not controlled for. For this reason, scatter plots between net credit-to-disposable income ratios and saving rates, as shown in Figure 3, are more informative. 20
the whole sample of observations over the period 1995-2012.21 Given the short time span of the sample, estimations of the interaction effects are carried out with quarterly data. Interaction effects Following the discussion in Section 3, I consider several interaction variables: the share of consumer credit in disposable income (dummy variable equal to one if the average ratio of consumer credit to disposable income is larger than 15%; zero otherwise); the availability of housing equity withdrawal; homeownership rates; and the level of the saving rate in conjunction with the share of pension funds in the economy (dummy variable equal to one if the country has an average saving rate below 10% and a share of pension funds in GDP above 50%; zero otherwise). Table 8 reports results of the regressions. The prevalence of consumer credit seems to have a significant impact on the transmission of credit to saving as shown by the estimated coefficient on the interaction between the “high consumer credit” dummy and the change in the debt ratio (column (1)). Actually, only countries where consumer credit is relatively important show a statistically significant relationship between saving and changes in debt ratios. In contrast, in countries where housing equity withdrawal is available, the impact of debt changes on saving rates does not seem to be stronger than in other economies. The interaction of the dummy variable for HEW with the change of the debt ratio has even a surprisingly positive sign (column (2)). This result holds true even when weighting the interaction term by the homeownership rate (results not reported). Likewise, estimating separate regressions for the two groups of countries does not provide results supporting the hypothesis of a stronger credit-saving relationship in countries with HEW (not shown). This finding is however not surprising given mixed evidence in the literature on the effect of HEW on consumption, with several studies showing a relatively modest short-term impact of changes in housing equity withdrawal on consumption growth (see e.g. Smith, 2006 for New Zealand). Results are also consistent with studies reporting that only a small share of HEW is used for consumption. Survey evidence for Australia, the United Kingdom, or the Netherlands (see Ebner, 2013) suggests for instance that housing equity withdrawal is not used to finance consumption but mostly to pay off old debt or to
21
The choice of the mid-nineties as a starting point of the sample is motivated by several factors: time-varying information on institutional settings is in some cases only available over the past 15 years; the middle of the nineties roughly corresponds to the starting years of the housing boom in most OECD countries (see André, 2010); and effects from credit market liberalisation should be less a concern with mortgage market liberalisation in European countries having ended in the mid-1990s (see Carstensen et al., 2009). Results are anyway qualitatively similar when considering the whole period although non-time-varying information on institutional settings may not capture the evolution of institutions since 1980. 21
make home improvements while in the U.S., only 16% of HEW was devoted to consumption in 20012002, the rest being allocated among home improvements (i.e. residential investment), repayment of debt and acquisition of real assets (Menegatti and Roubini, 2007).22 Finally, there is no evidence of a significant effect of pension funds and/or of the average level of saving rates on the saving-credit relationship. The interaction term of the debt change with the dummy variable for countries with low levels of saving rates has the expected negative sign but is not significant. The coefficient of the interaction between the change in the debt ratio and the dummy for countries with pension funds is not significant neither (results not reported). The interaction of the two dummy variables with the change in the debt ratio has the expected negative sign but is again non-significant (column (4)). – Insert Table 8 about here – Splitting the estimation sample As a robustness check, Table 9 shows estimates of the effects of debt on saving rates by splitting the sample of countries based on the average level of the consumer credit-to-disposable income ratio. Consistent with results reported in Table 8, the effects of the changes in the debt ratio or of the credit-todisposable income ratio on household saving rates are economically and statistically significant only for the group of countries with a high ratio of consumer credit in disposable income. Results are qualitatively similar when excluding from the estimation Korea – which belongs to the high consumer credit group of countries and experienced two deleveraging episodes over 1995-2012 (results not reported).23 They are also robust to excluding from the estimation the Great Recession period (columns (5) to (8)). Estimates therefore lend support to the hypothesis that the presence of consumer credit in the economy plays a significant role in the transmission of credit to saving rates in contrast to other institutional settings and in particular HEW. – Insert Table 9 about here –
22
In contrast, using household-level data over 2002-2006, Mian and Sufi (2011b) find little evidence that home equity-based borrowing in the United States is used to pay down credit card balances, to purchase new homes or investment properties. While these authors do not have data on real outlays, they conclude that consumption and home improvement are possible uses of the increased borrowing in response to rising house prices.
23
In particular, Korea saw distressed credit card debt rising from 7.5% of total credit card receivables in 2000 to 34% in 2003 while household delinquencies reached at the end of 2003 about 17% of the economically active population, triggering a contraction of GDP in the first half of the year (see e.g. He et al., 2005). 22
In light of this latter finding, the equations of Table 5 are re-estimated by splitting the sample according to the importance of consumer credit.24 Results shown in Table 10 indicate that declining indebtedness has a significant effect on saving rates in the group of countries with a high share of consumer credit. Based on estimates of column (2) excluding Korea, a one-standard deviation increase in rising indebtedness (+3 percentage points) translates into a decline in the saving rate of 0.25 percentage points while a one-standard deviation shock on the declining debt ratio variable (-0.6 percentage point) leads to an increase in the saving rate of less than 0.1 percentage point. Using instead estimates of column (1) including Korea still indicates an effect from rising indebtedness on saving rates more than twice as large as the effect from declining debt ratios. In contrast, in the group of countries where consumer credit is relatively marginal, declining debt ratios do not have any significant effect on saving rates. Only positive changes in debt ratios impact saving rates: a one-standard deviation increase in rising indebtedness (+3.7 percentage points) is associated with a decline in the saving rate of 0.5 percentage points. Importantly, the impact of indebtedness on saving is significantly asymmetric. The null hypothesis of equality of the coefficients of rising and declining indebtedness is rejected at a 1% confidence level (columns (4) and (5)). This asymmetry holds true even when excluding from the estimation changes in housing and stock market prices which may capture indirect effects from changing indebtedness (results not reported). – Insert Table 10 about here – Finally, Table 11 shows estimates of the effects on saving rates of rising and declining debt ratios during the historical boom and bust credit episodes reported in Table 1, after excluding the few countries with a high share of consumer credit in disposable income (i.e. Australia, Japan, Korea, and the United Kingdom). In contrast to results obtained for the whole sample of boom-bust credit cycles (Tables 6 and 7), deleveraging does not have any significant effect on saving rates in countries with poorly developed consumer credit markets. Tests of equality of the coefficients of rising and declining debt ratios again suggest that the impact of debt ratios on saving rates is significantly asymmetric: credit booms translate into lower saving rates but deleveraging is not associated with higher saving rates (columns (3) and (4)). This result also holds true when using quarterly data (columns (5) and (6)) or when excluding proxies for housing and stock market wealth effects (results not reported). – Insert Table 11 about here – 24
Countries with limited time coverage (Austria, Czech Republic, Estonia, Greece, Hungary, Ireland, Poland, Portugal, Slovak Republic, and Slovenia) are excluded from the estimation but results are anyway qualitatively similar. 23
Consumer versus housing debt ratios A more robust approach to test for the importance of consumer credit in driving the credit-saving relationship is to differentiate the effects of consumer credit and of housing credit on saving. Long series of debt for consumption and for housing are however available for only eight countries (Australia, Belgium, Canada, France, Greece, Japan, United Kingdom, United States), reducing dramatically the size of the sample. Results of estimations, reported in Table 12, still confirm the presence of a negative creditsaving relationship for this restricted sample of countries (column (1)). Interestingly, the negative creditsaving relationship vanishes once controlling for consumer credit (i.e. the change in the consumer debt ratio in column (2)) but not when controlling for housing credit alone (column (3)). Actually, only consumer credit is significantly related to saving rates (columns (4) to (6)) while the null hypothesis of equality of the coefficients of consumer credit and of housing credit is rejected at the 3% confidence level (not reported). This result further supports the view that the negative relationship between saving and changes in debt ratios is driven by consumer credit. This consumer credit-saving relationship could of course reflect reverse causality, as deleveraging episodes tend to take place in a subdued demand environment. Still, bi-directional causation between consumer credit and consumption during deleveraging cannot be excluded, insofar as deleveraging and the associated reduction in consumer borrowings result from credit tightening by banks. For instance, Gropp et al. (2014) provide evidence for the recent US deleveraging that the reduction in non-mortgage and credit card borrowing was more driven by cutbacks in the provision of credit by banks than by a demand-based response to lower housing wealth. During deleveraging, banks tighten credit conditions, including for consumer credit with some direct implications for consumption of liquidity-constrained households. In this context, it is not surprising that the macroeconomic impact of deleveraging on consumption depends on the importance of consumer credit in the economy. – Insert Table 12 about here – Overall, econometric estimates confirm the presence of an asymmetric effect of credit on saving rates together with a high cross-country heterogeneity in the response of saving rates to changes in debt ratios. Rising indebtedness is associated with lower saving rates – possibly reflecting a structural financial deepening effect – but declining indebtedness translates into higher saving rates only in countries where debt is likely to directly finance consumption through consumer credit. Estimates for a subsample of countries further confirm the importance of consumer credit in driving the relationship between saving and changes in debt ratios. In contrast, changes in debt for housing, which however represents the bulk of household debt, are not related to saving rates. 24
6. Conclusion In this paper, I find evidence that the impact of household deleveraging on saving rates is relatively modest after controlling for standard determinants of saving. Although the impact of declining indebtedness on saving is statistically significant in some countries, it is much smaller than the effects from other saving determinants like income growth, household wealth or economic uncertainty. This suggests that the bulk of the increase in saving rates usually observed around household deleveraging episodes is not driven by a debt decline effect but instead by developments of other macroeconomic variables usually accompanying the downward adjustment of debt ratios. These findings are in line with evidence reported by Carroll et al. (2012) who show that, in the United States, the long-term expansion of credit supply since the early-1980s (probably reflecting financial innovation and liberalisation) encouraged households to save less while business cycle fluctuations in saving are mainly explained by changes in household wealth and labour income uncertainty. Results also indicate that the relationship between household saving rates and changes in debt ratios is highly heterogeneous across countries, as declining debt ratios impact saving rates only in countries where consumer credit is widespread. Actually, the negative relationship between saving and credit seems to be explained by consumer credit rather than by housing credit. This suggests that the upward adjustment in saving rates associated with deleveraging could be marginal in economies where rising indebtedness has not financed consumption directly over the past fifteen years. In contrast, the adjustment could be material in countries like Canada or Korea where consumer credit is relatively important.25 Other channels associated with household deleveraging and not present in historical data could also weigh on consumption in the coming years. In particular, in countries where interest-only mortgages have sharply expanded in the run-up to the crisis, public authorities may be tempted to modify the redemption profile of loans to reduce the risk exposure of banks and accelerate the deleveraging process.26 Bringing forward the repayment of principal of interest-only loans in the context of a renegotiation of the contracts or imposing regular amortising requirements for all new mortgages would mechanically translate 25
Deleveraging in Canada has not yet started but the country experienced a dip in non-mortgage credit to households in 2008-09, when several banks faced funding difficulties on the U.S. money market. As evidenced by Damar et al. (2014), this tightening however did not translate into lower levels of consumption as the credit shock was temporary and most households compensated by drawing down liquid assets to smooth consumption. In the context of a more prolonged adverse lending shock – as is typically the case during a deleveraging phase – consumption would probably react more negatively. 26
In the Netherlands, for instance, interest-only loans represent more than 50% of total mortgages, making the deleveraging process in the country particularly slow. 25
into higher saving rates. Likewise, in countries with flexible mortgage rates, a rise in interest rates would reduce new borrowings and contribute to the deleveraging process but also increase the debt payment ratio to the detriment of consumption.27 This channel could be particularly harmful in economies where bankruptcy laws are not debtor-friendly. In these countries, policies reducing the costs for households to default on their debt, as implemented by the United States in the wake of the Great Depression, would speed up the deleveraging and promote a stronger recovery.
27
A rough proxy of interest expenses is found to explain the saving rates positively but only after excluding the interest rate variable from the estimations. Results concerning the impact of the change in the debt ratio are qualitatively unchanged. 26
References Abiad, A., E. Detragiache, and T., Tressel (2008), “A New Database of Financial Reforms,” IMF Working Paper, No. 266. Andrews, D. (2010), “Real House Prices in OECD Countries – The Role of Demand Shocks and Structural and Policy Factors,” OECD Economics Department Working Papers, No. 831. André, C. (2010), “A Bird’s Eye View of OECD Housing Markets,” OECD Economics Department Working Papers, No. 746. Attanasio, O., L. Blow, R. Hamilton, and A. Leicester (2005), “Booms and Busts: Consumption, House Prices and Expectations,” Institute for Fiscal Studies Working Paper No. 24. Attanasio, O., L. Picci, and A.E. Scorcu (2000), “Saving, Growth, and Investment: A Macroeconomic Analysis Using a Panel of Countries,” The Review of Economics and Statistics Vol. 82, No. 2, pp. 182-211. Bacchetta, P. and S. Gerlach (1997), “Consumption and Credit Constraints: International Evidence,” Journal of Monetary Economics, Vol. 40, No. 2, pp. 207–238. Bailliu, J., K. Kartashova, and C. Meh (2012), “Household Borrowing and Spending in Canada,” Bank of Canada Review, Winter 2011–2012, 16-29. Benito, A., J. Thompson, M. Waldron, and R. Wood (2006), “House Prices and Consumer Spending,” Bank of England Quarterly Bulletin, Summer 2006. Bhutta, N. (2014), “ The Ins and Outs of Mortgage Debt During the Housing Boom and Bust,” Federal Reserve Board FEDS Working Paper No. 2014-91. Bouis, R., A. K. Christensen, and B. Cournède (2013), “Deleveraging: Challenges, Progress and Policies,” OECD Economics Department Working Papers, No. 1077. Bruno, G. (2005), “Estimation and Inference in Dynamic Unbalanced Panel-Data Models with a Small Number of Individuals,” The Stata Journal, Vol. 5, No. 4, 473-500. Bun, M. and J. Kiviet (2003), “On the Diminishing Returns of Higher Order Terms in Asymptotic Expansions of Bias,” Economics Letters, 79, 145-152. Campbell, J. and J. Cocco, (2007), “How Do House Prices Affect Consumption? Evidence from Micro Data,” Journal of Monetary Economics, vol. 54, No. 3, 591-621. Carroll, C. (2001), “A Theory of the Consumption Function, With and Without Liquidity Constraints,” Journal of Economic Perspectives Vol. 15, No. 3, 23-45. Carroll, C., J. Slacalek, and M. Sommer (2012), “Dissecting Saving Dynamics: Measuring Wealth, Precautionary, and Credit Effects,” IMF Working Paper No. 219.
27
Carstensen, K., O. Hülsewig, and T. Wollmershäuser (2009), “Monetary Policy Transmission and House Prices: European Cross-Country Evidence,” CESifo Working Paper Series No. 2750. Case, K, J. Quigley, and R. Shiller (2005), “Comparing Wealth Effects: The Stock Market versus the Housing Market,” Advances in Macroeconomics, Vol. 5, No. 1, 1-32. Catte, P., N. Girouard, R. Price, and C. André (2004), “Housing Markets, Wealth and the Business Cycle,” OECD Economics Department Working Paper, No. 394. Cecchetti, S., M. Mohanty, and F. Zampolli (2011), “The Real Effects of Debt,” BIS Working Papers No. 352. Chmelar, A. (2013), Household Debt and the European Crisis, ECRI Research Reports. Coletta, M., R. De Bonis, and S. Piermattei (2013), “Why Does the Level of Household Debt Vary among Industrial Countries? Evidence from the Flow-of-Funds,” mimeo, Bank of Italy. Cooper, D. (2012), “U.S. Household Deleveraging: What Do the Aggregate and Household-Level Data Tell Us?,” Federal Reserve Bank of Boston Public Policy Briefs No. 2. Cooper, D. and K. Dynan (2015), “Wealth Effects and Macroeconomic Dynamics,” Journal of Economic Surveys, forthcoming. Cuerpo, C., I. Drumond, J. Lendvai, P. Pontuch, and R. Raciborski (2013), “Indebtedness, Deleveraging Dynamics and Macroeconomic Adjustment,” Economic Papers No. 477, April, Directorate-General for Economic and Financial Affairs, European Commission, Brussels. Damar, E., Gropp, R. and Mordel, A. (2014), “Banks’ Financial Distress, Lending Supply and Consumption Expenditure,” ECB Working Paper No. 1687. Dembiermont, C., M. Drehmann, and S. Muksakunratana (2013), “How Much Does the Private Sector Really Borrow? A New Database for Total Credit to the Private Non-Financial Sector,” BIS Quarterly Review, March 2013, 65-81. Dombret, A. (2013), “The Debt Crisis and its Consequences for the Real Economy,” Speech at the Handelsbatt CFO Congress, Königstien. Dynan, K. (2012), “Is a Household Debt Overhang Holding Back Consumption?,” Brookings Papers on Economic Activity, Spring 2012, 299-344. Dynan, K. and W. Edelberg (2013), “The Relationship Between Leverage and Household Spending Behavior: Evidence from the 2007-2009 Survey of Consumer Finances,” Federal Reserve Bank of St. Louis Review, September/October, 425-448. Ebner, A. (2013), “A Micro View on Home Equity Withdrawal and its Determinants: Evidence from Dutch Households,” Journal of Housing Economics, Vol. 22, No. 4, 321-337. Eggertsson, B. and P. Krugman (2012), “Debt, Deleveraging, and the Liquidity. Trap: A Fisher-MinskyKoo Approach,” Quarterly Journal of Economics, Vol. 127, No. 3, 1469-1513. 28
Elul, R. (2008), “Collateral, Credit History, and the Financial Decelerator,” Journal of Financial Intermediation, Vol. 17, No. 1, 63-88. Glick, R. and K. Lansing (2011), “Consumers and the Economy, Part I: Household Credit and Personal Saving,” FRBSF Economic Letter No. 1, January. Gropp, R., J. Krainer, and E. Laderman (2014), “Did Consumers Want Less Debt? Consumer Credit Demand versus Supply in the Wake of the 2008-2009 Financial Crisis,” Federal Reserve Bank of San Francisco Working Paper No. 8. Guerrieri, V. and G., Lorenzoni (2011), “Credit Crises, Precautionary Savings, and the Liquidity Trap,” NBER Working Papers, No. 17583. Hall, R. (2011), “The Long Slump,” American Economic Review Vol. 101, No. 2, 431–469. He, D., E. Yao, and K. Li (2005), “The Growth of Consumer Credit in Asia,” Hong Kong Monetary Authority Quarterly Bulletin, March. Igan, D. and P. Loungani (2012), “Global Housing Cycles,” IMF Working Paper No. 217. IMF (2008), “The Changing Housing Cycle and the Implications for Monetary Policy,” World Economic Outlook, Chapter 3. Knotek, E. and J.C. Braxton (2012), “What Drives Consumer Debt Dynamics?,” Federal Reserve Bank of Kansas City Economic Review, No. 4, 31-54. Koo, R. (2011), “The World in Balance Sheet Recession: Causes, Cure, and Politics,” Real World Economics Review No. 58, 19-37. Krugman, P. (2013), “Worthwhile Canadian Comparison,” New York Times Blog, June 15. Li, W. and S. Patwari (2012), “The Economics of Household Leveraging and Deleveraging,” Federal Reserve Bank of Philadelphia Business Review (Third Quarter). Loayza, N., K. Schmidt-Hebbel, and L. Servén (2000), “What Drives Saving across the World?,” The Review of Economics and Statistics, Vol. 82, No. 2, pp. 165-181. McCarthy, Y. and K. McQuinn (2014), “Deleveraging in a Highly Indebted Property Market: Who Does it and are There Implications for Household Consumption?,” Research Technical Paper Central Bank of Ireland, No. 5. Menegatti, C. and N. Roubini (2007), “The Direct Link between Housing and Consumption: Wealth Effect and Home Equity Withdrawal,” mimeo. MGI (2012), Debt and Deleveraging: Uneven Progress on the Path to Growth, McKinsey Global Institute, January. Mian, A. and A. Sufi (2011a), “Consumers and the Economy, Part II: Household Debt and the Weak U.S. Recovery,” Federal Reserve Bank of San Francisco Economic Letter No. 02. 29
Mian, A. and A. Sufi (2011b), “House Prices, Home Equity Based Borrowing, and the U.S. Household Leverage Crisis,” American Economic Review 101, 2132-2156. Mody, A., F. Ohnsorge, and D. Sandri (2012), “Precautionary Savings in the Great Recession,” IMF Working Paper No. 42. Muellbauer, J. (2008), “Housing, Credit and Consumer Expenditure,” CEPR Discussion Paper No. 6782. Nickell, S. (1981), “Biases in Dynamic Models with Fixed Effects,” Econometrica, Vol. 49, No. 6, 14171426. Nickell, S. (2004), “Household Debt, House Prices and Consumption Growth,” Speech given at Bloomberg, London, Tuesday 14 September 2004. Olney, M. (1999), “Avoiding Default: The Role of Credit in the Consumption Collapse of 1930,” Quarterly Journal of Economics, Vol. 114, No. 1, 319–335. Röhn, O. (2010), “New Evidence on the Private Saving Offset and Ricardian Equivalence,” OECD Economics Department Working Papers, No. 762. Smith, M. (2006) “What Do We Know about Equity Withdrawal by Households in New Zealand?,” Reserve Bank of New Zealand workshop Housing, Savings, and the Household Balance Sheet, Wellington, 14 November. Svensson, L. (2012), “Monetary Policy, Debt and Unemployment,” Speech at a meeting at SNS – Centre for Business and Policy Studies, Stockholm, 14 November 2012. Takáts E. and C. Upper (2012), “Deleveraging and Recovery,” in The ESRB at 1, S. Gerlach, E. Gnan and J. Ulbrich (editors), SUERF, Vienna. Takáts, E. and C. Upper (2013), “Credit and Growth after Financial Crises,” BIS Working Papers, No. 416. Tang, G. and C. Upper (2010), “Debt Reduction after Crises,” BIS Quarterly Review, September, 25-38.
30
Figure 1 – Comparison of average saving rates in historical credit booms and busts (in percent of gross disposable income) 28
23
18
13
8
3
Average saving rate in credit boom
Average saving rate in deleveraging
Note: Years refer to the peak of the household debt-to-disposable income ratio. Average saving rates are computed over 5 years before the peak of the debt ratio (peak year included) for the boom phase, and from the peak to the trough of the debt ratio (peak year excluded) for the deleveraging phase. In the case of Korea 2002 and Netherlands 1982 episodes, the boom phase is defined over the two years preceding the peak (respectively, to avoid an overlap with the previous deleveraging episode, and for data availability reasons).
31
Figure 2 – Changes of macroeconomic variables around historical debt turning points Change of real house prices during the deleveraging (per cent)
Change of average annual growth rate of real stock prices between boom and bust (percentage points)
Change of the unemployment rate during the deleveraging (percentage points)
Change of average annual GDP growth volatility (declining growth) between boom and bust (percentage points)
Note: Changes in real house prices and in unemployment rates are computed from the peak to the trough of the debt ratio. Averages of annual growth rates of real stock prices and of GDP growth volatility (declining growth) are computed over 5 years before the peak of the debt ratio (boom phase) and from the peak to the trough (bust phase). GDP growth volatility is multiplied by a dummy equal to one if real GDP growth is declining, and zero otherwise. The sample is composed of the historical credit boom and bust episodes reported in Table 1 (changes for the two Korean episodes are not reported due to an overlap in the data), with years referring to peaks of the debt ratios.
32
Figure 3 – Scatter plots of net credit-to-disposable income ratios Debt (x-axis, percentage points) Yd
and saving rates (y-axis, per cent), 1995-2012.
14 12 10
20
2
4
6
8
0
1
3
4
-2
0
4
0
1
2
DEU
3
4
5
DNK
2
4
6
-5
0
6 -1
0
1
FIN
3
-5
0
10
0
2
4
6
-1
0
1
3
2 0 -1
0
2
8
10 5
6
-10
-2
0
2
4
-5
0
5
10
-5
0
5
10
20
1
2
-5
0
5
10
0
1
2
4
6
10 5 0
5
10
-2
0
2
11
10
12
12
13
USA
15
14
SWE
0
2
4
6
-2
0
2
4
6
8
0
6
0
9
8
5
5
10
10
15
-1
0 -5
PRT
20
POL
10
-2
15
18 14 12 0
6
NOR
10
0 -1
4
16
20 10
15 10 5 -2
2
NLD
30
KOR
20
JPN
0
20
-4
-2
ITA
10 12 14 16
15
10 0
1
IRL
20
HUN
20
GRC
2
12 14 16 18 20 22
-2
13
8
6
10
14
8
4
12
15
10
6
16
12
5
GBR
17
18 16 14
2
FRA
14
ESP
5
10
0
8
-2
2
13
8
15
4
16
14 15
10
8
17
12
16 17
10
18
2
18
CZE 14
CHE
2
6
14 -1
12
0
8
16
14
5
12
10
18
16
15
CAN
22
BEL
18
AUT
20
AUS
0
33
1
2
3
4
-2
0
2
4
Table 1 – Household debt ratios and saving rates during full boom-bust credit cycles (Percentage points of disposable income) 5-year period before debt peak Country
Peak
Trough
Δ Debt-toincome ratio
AUS BEL CHE CHE DNK ESP FIN GBR JPN KOR KOR* NLD* NOR SWE
1988 1999 1987 2005 1987 1989 1989 1989 1990 1997 2002 1982 1988 1989
1990 2001 1993 2008 1994 1995 1997 1996 1992 1998 2004 1985 1995 1995
20.7 6.9 39.3 18.6 66.4 17.7 28.9 33.6 27.2 18.5 31.7 3.1 54.6 26.9
Net credit effect
Nominal growth effect
40.9 15.1 65.8 30.7 95.5 35.1 48.0 62.3 47.9 51.5 41.0 10.3 92.1 60.3
-20.2 -8.2 -26.4 -12.1 -29.1 -17.3 -19.1 -28.7 -20.7 -33.0 -9.3 -7.2 -37.5 -33.4
Average gross saving-toincome ratio 18.6 18.5 n.a. 19.6 5.9 12.3 8.2 7.2 18.3 23.9 11.6 17.5 6.8 3.6
From peak to trough Δ Debt-toincome ratio
Net credit effect
Nominal growth effect
-6.0 -6.0 -17.7 -9.9 -24.4 -3.4 -27.0 -12.4 -4.3 -12.3 -8.0 -4.2 -48.3 -44.9
8.3 0.3 36.8 11.3 20.6 19.9 -7.6 33.6 6.6 -8.9 9.0 3.2 3.2 -3.3
-14.2 -6.3 -54.5 -21.2 -45.0 -23.3 -19.3 -46.0 -11.0 -3.4 -17.0 -7.4 -51.5 -41.6
Average gross saving-toincome ratio 17.3 17.3 n.a. 21.3 7.1 14.6 11.4 9.1 19.4 27.7 13.3 18.1 9.9 8.8
Notes: *: Changes and averages during the boom phase are computed two years before the peak. The change in the debt-to-income ratio is decomposed into a net credit effect Debtt T Debtt and a nominal growth effect Y d t T d d Y t T Y t Debtt d . Y dt Y t T Sources: OECD national accounts, national central banks, and author’s calculations.
34
Table 2 – Household debt ratios and saving rates in the run-up to the Great Recession and during on-going deleveraging processes (or from 2007 for still debt booming economies) (Percentage points of disposable income) From 1997 to debt peak (or to 2007) Country
Peak
Δ Debt-toincome ratio
Net credit effect
Nominal growth effect
AUS AUT BEL CAN CHE CZE DEU DNK ESP EST FIN FRA GBR GRC HUN IRL ITA JPN KOR NLD NOR NZL POL PRT SVK SVN SWE USA
2010 2000 2009 2007 2010 2007 2010 2009 2006 2008 2009 2010 2007
74.9 21.3 15.6 37.3 8.4 32.2 10.4 122.5 77.7 97.0 47.2 23.5 66.7 55.2 62.2 158.8 35.1 1.6 41.6 117.5 72.2 70.1 18.4 71.6 20.0 24.3 60.0 39.8
121.1 45.3 34.3 76.5 48.6 40.0 15.7 184.0 106.7 106.4 69.3 44.7 103.4 63.4 69.9 196.5 43.9 -5.7 79.5 157.9 122.5 111.1 22.0 101.6 30.3 36.3 94.2 77.3
-46.3 -24.1 -18.7 -39.2 -40.2 -7.8 -5.4 -61.5 -29.0 -9.4 -22.1 -21.3 -36.6 -8.2 -7.8 -37.7 -8.7 7.3 -37.9 -40.4 -50.3 -40.9 -3.6 -30.0 -10.2 -12.0 -34.2 -37.5
Average gross savingto-income ratio 10.9 14.3 16.7 7.5 19.7 10.1 15.6 6.2 11.9 1.3 8.4 15.2 4.6 8.2 11.4 11.0 16.3 11.3 14.8 13.9 9.9 -0.1 11.0 9.8 8.8 14.7 7.3 9.5
From debt peak (or from 2007) to 2012 Δ Debt-toincome ratio
Net credit effect
Nominal growth effect
2.3 -2.9 12.2 20.2 15.2 12.0 -22.1 -19.3 -7.3 -21.4 8.5 13.9 -25.4 32.3 -15.2 -13.5 9.0 -10.1 17.2 46.1 8.9 -13.0 19.9 -7.6 14.6 -1.4 11.7 -23.4
47.8 2.4 22.5 44.4 30.2 17.7 2.9 12.6 -5.0 -12.0 27.4 21.8 3.3 13.0 -9.5 -30.0 9.2 -12.8 47.2 52.7 56.3 11.7 29.1 -7.7 21.0 -1.7 41.8 -5.9
-45.6 -5.3 -10.2 -24.1 -15.0 -5.7 -25.1 -31.9 -2.3 -9.4 -18.9 -7.9 -28.7 19.3 -5.6 16.4 -0.2 2.8 -30.1 -6.6 -47.4 -24.7 -9.2 0.1 -6.4 0.3 -30.1 -17.5
Average gross savingto-income ratio 17.5 12.3 15.9 9.1 22.2 10.4 16.4 7.3 13.7 7.8 9.5 15.9 6.1 -1.3 9.1 10.8 13.1 8.2 9.7 11.5 11.6 3.4 5.3 10.6 8.0 12.0 13.0 10.9
Note: The change in the debt-to-income ratio is decomposed into a net credit effect Debtt T Debtt and a nominal growth effect Y d t T d d Y t T Y t Debtt d . Y dt Y t T Sources: OECD national accounts, national central banks, and author’s calculations.
35
Table 3 – Differences in credit market features, homeownership rates, and pension funds (Numbers in percent)
Country
Australia Austria Belgium Canada Czech Republic Denmark Estonia Finland France Germany Greece Hungary Ireland Italy Japan Korea Netherlands New Zealand Norway Poland Portugal Slovak Republic Slovenia Spain Sweden Switzerland United Kingdom United States
Consumer credit-todisposable income ratio (2000-2007 average) 19.4 15.7 6.4 36.3 4.9 14.7 9.2 12.2 11.1 15.4 9.8 21.3 3.7 15.5 30.4 7.7 14.2 9.0 9.5 3.7 11.5 12.4 7.8 4.4 23.6 22.9
Housing equity withdrawal
Homeownership rate (1999-2005 average)
Share of pension funds in GDP, 2011
Yes No No Yes No Yes Yes Yes No No No Yes Yes No No Yes Yes Yes Yes No No Yes Yes No Yes Yes
70.4 52.3 68.7 65.7 51.6 82.1 65.2 54.8 39.9 80.3 91.4 79.9 66.9 50.3 55.4 63.6 92.7 83 48.2 37.5 68.1 67.2
92.8 4.9 4.2 63.7 6.5 49.7 5.3 75 0.2 5.5 0.04 3.8 46.2 4.9 25.1 4.5 138.2 7.4 15.8 15 7.7 8.4 2.9 7.8 8.0 110.8 88.2 70.5
Sources: OECD, national central banks, Luxembourg Income Study, Andrews (2010), Igan and Loungani (2012) and author’s calculations.
36
Table 4 – Saving rates and changes in debt ratios, baseline equations
Lagged saving rate Log (Real disposable income Yd) Real growth rate of Yd Terms of trade(a) Old dependency ratio Young dependency ratio Unemployment rate Inflation rate(a) GDP growth volatility Real interest rate(a) Δ(Debt/Yd)
Dependent variable: Household gross saving rate (3) (4) (5) (6)
(1)
(2)
0.789*** (26.71) 2.142 (1.66) 0.355*** (12.27) 0.600 (0.38) -0.120** (-2.41) -0.001 (-0.03) 0.081** (2.22) 13.286*** (3.38) 0.301*** (2.92) 11.353** (2.72) -0.127*** (-5.87)
0.780*** (27.78) 3.362** (2.47) 0.356*** (13.11) -0.152 (-0.08) -0.113** (-2.14) -0.021 (-0.46) 0.107*** (2.78) 14.383*** (3.68) 0.303*** (3.03) 12.961*** (3.14) -0.122*** (-4.76) 0.000 (0.08)
Net financial wealth/Yd, lagged
0.795*** (24.78) 1.026 (0.89) 0.392*** (10.97) 0.837 (0.65) -0.138** (-2.59) -0.012 (-0.24) 0.074* (1.84) 14.949*** (2.80) 0.256** (2.61) 12.101** (2.27) -0.111*** (-4.01)
House prices, real growth lagged
0.798*** (24.66) 0.868 (0.55) 0.406*** (9.25) 1.514 (0.68) -0.114 (-1.61) -0.041 (-0.47) 0.031 (0.66) 17.589* (1.91) 0.383*** (3.77) 6.767 (0.96) -0.119*** (-4.39)
Cyclically adj. government net lending
0.775*** (23.04) 2.489 (1.44) 0.387*** (8.45) 0.231 (0.11) -0.091 (-1.24) 0.035 (0.38) 0.154** (2.53) 16.588** (2.49) 0.264** (2.61) 14.828** (2.34) -0.116*** (-3.51)
0.794*** (26.20) 1.038 (0.61) 0.419*** (8.02) 0.573 (0.29) -0.119 (-1.57) -0.038 (-0.46) 0.106* (1.83) 18.408** (2.64) 0.361*** (4.07) 6.936 (1.04) -0.119*** (-3.40)
0.129*** (2.89)
-0.023 (-1.26) -0.013*** (-2.83) -0.044 (-1.37) 0.124** (2.32)
-0.011 (-0.61) -0.012** (-2.73) -0.093** (-2.50) 0.120 (1.57)
Yes Yes 0.565 698 28
Yes Yes 0.493 542 26
Yes Yes 0.772 476 19
-0.099** (-2.72)
Housing investment/Yd
Country fixed effects Time fixed effects R-squared Number of observations Number of countries
0.790*** (27.95) 2.207* (1.88) 0.348*** (12.06) -0.465 (-0.32) -0.112** (-2.23) 0.016 (0.37) 0.131*** (3.58) 11.433*** (3.09) 0.308*** (3.15) 12.611*** (2.98) -0.143*** (-5.99)
-0.022 (-1.21) -0.011** (-2.72)
Stock prices, real growth lagged
Yes Yes 0.565 698 28
Yes Yes 0.424 670 28
Yes Yes 0.780 606 26
Yes Yes 0.792 493 19
Notes: (a) Expressed in logs (log of (1 + x) for the real interest rate and the inflation rate). Constant term not reported. ***, **, * denote significativity at the 1%, 5%, and 10% level, respectively. Robust t-statistics appear in parentheses.
37
(7)
Table 5 – Positive versus negative changes of debt ratios Dependent variable: Household gross saving rate (1) (2) (3) (4) Lagged saving rate Log (Real disposable income Yd) Real growth rate of Yd Terms of trade(a) Old dependency ratio Young dependency ratio Unemployment rate Inflation rate(a) GDP growth volatility Real interest rate(a) Δ(Debt/Yd)>0 Δ(Debt/Yd)<0
0.786*** (26.04) 2.208* (1.75) 0.352*** (12.29) 0.537 (0.33) -0.124** (-2.55) -0.007 (-0.16) 0.083** (2.37) 14.012*** (3.74) 0.298*** (2.87) 11.047** (2.58) -0.149*** (-7.02) -0.059 (-0.91)
0.819*** (27.42) 0.717 (0.63) 0.387*** (11.89) 0.700 (0.48) -0.122** (-2.43) -0.020 (-0.49) 0.053 (1.47) 16.593** (2.73) 0.357*** (3.71) 6.464 (1.22) -0.146*** (-7.09) -0.092 (-1.32)
0.772*** (22.78) 2.625 (1.59) 0.378*** (8.30) 0.031 (0.02) -0.094 (-1.37) 0.033 (0.36) 0.164** (2.76) 17.996** (2.71) 0.261** (2.54) 14.254** (2.16) -0.143*** (-5.58) -0.049 (-0.71) -0.024 (-1.27) -0.013*** (-2.86) -0.042 (-1.34) 0.138** (2.74)
0.791*** (25.30) 1.157 (0.70) 0.412*** (7.84) 0.453 (0.23) -0.121 (-1.69) -0.039 (-0.47) 0.115** (2.11) 19.847*** (2.96) 0.368*** (4.13) 6.536 (0.93) -0.143*** (-5.06) -0.062 (-0.85) -0.011 (-0.63) -0.012** (-2.83) -0.092** (-2.53) 0.133* (1.75)
Yes Yes 0.557 698 28
Yes Yes 0.841 565 19
Yes Yes 0.477 542 26
Yes Yes 0.748 476 19
House prices, real growth lagged Stock prices, real growth lagged Cyclically adj. government net lending Housing investment/Yd Country fixed effects Time fixed effects R-squared Number of observations Number of countries
Notes: (a) Expressed in logs (log of (1 + x) for the real interest rate and the inflation rate). Constant term not reported. ***, **, * denote significativity at the 1%, 5%, and 10% level, respectively. Robust t-statistics appear in parentheses.
38
Table 6 – Full boom-bust cycles of debt ratios
Lagged saving rate Log (Real disposable income Yd) Real growth rate of Yd Terms of trade(a) Old dependency ratio Young dependency ratio Unemployment rate Inflation rate(a) GDP growth volatility Real interest rate(a) Δ(Debt/Yd)>0 Δ(Debt/Yd)<0
Dependent variable: Household gross saving rate (3) (4) (5) (6)
(1)
(2)
0.573*** (9.15) 4.302 (1.17) 0.142 (1.71) 7.381 (1.19) 0.060 (0.30) 0.094 (0.85) 0.112 (1.34) 19.239 (1.23) 0.319** (2.51) 24.824*** (3.59) -0.241*** (-10.63) -0.141** (-2.64)
0.558*** (8.76) 2.983 (0.87) 0.173* (2.23) 7.554 (1.23) 0.088 (0.43) 0.093 (0.78) 0.077 (1.05) 22.376 (1.56) 0.340** (3.13) 27.139*** (4.31) -0.224*** (-11.18) -0.128* (-2.08) -0.032 (-1.53)
House prices, real growth lagged Rising house prices, lagged
0.580*** (7.85) 5.043 (1.33) 0.161* (2.09) 5.922 (0.97) 0.202 (0.79) 0.133 (0.95) 0.111 (1.45) 20.856 (1.59) 0.355** (3.06) 28.600*** (4.46) -0.225*** (-14.06) -0.135** (-2.43)
0.600*** (10.96) 4.928 (1.38) 0.186* (2.22) 5.419 (0.91) -0.000 (-0.00) 0.043 (0.36) 0.149 (1.64) 13.099 (1.17) 0.290** (2.43) 19.089** (3.03) -0.252*** (-9.52) -0.121* (-2.16)
0.578*** (10.42) 3.160 (0.93) 0.234** (2.76) 5.477 (0.93) 0.006 (0.02) 0.030 (0.24) 0.107 (1.29) 16.746 (1.74) 0.320** (3.07) 21.446*** (3.53) -0.229*** (-8.64) -0.102 (-1.53) -0.041* (-2.08)
-0.068*** (-3.66) 0.050 (0.85)
Declining house prices, lagged Stock prices, real growth lagged
-0.016** (-2.83)
-0.017** (-3.06) -0.009 (-1.65) -0.033** (-2.83)
Declining stock prices, lagged
Yes Yes 0.378 158 11
Yes Yes 0.447 158 11
Yes Yes 0.344 158 11
Yes Yes 0.361 155 11
Yes Yes 0.399 155 11
-0.015** (-2.44) -0.022* (-1.86) Yes Yes 0.449 155 11
Notes: (a) Expressed in logs (log of (1 + x) for the real interest rate and the inflation rate). Constant term not reported. ***, **, * denote significativity at the 1%, 5%, and 10% level, respectively. Robust t-statistics appear in parentheses.
39
0.599*** (9.18) 4.736 (1.19) 0.225** (2.75) 3.817 (0.65) 0.130 (0.36) 0.071 (0.47) 0.129 (1.38) 14.783 (1.58) 0.335** (3.08) 22.597*** (3.42) -0.232*** (-8.12) -0.111* (-1.89)
-0.074*** (-3.73) 0.033 (0.60)
Rising stock prices, lagged
Country fixed effects Time fixed effects R-squared Number of observations Number of countries
0.606*** (11.45) 4.042 (1.03) 0.196** (2.35) 4.753 (0.79) 0.061 (0.20) 0.056 (0.45) 0.122 (1.25) 11.128 (1.01) 0.295** (2.49) 18.515** (2.49) -0.257*** (-8.71) -0.128** (-2.32)
(7)
Yes Yes 0.364 155 11
Table 7 – Full boom-bust cycles of debt ratios, quarterly data Dependent variable: Household gross saving rate
Lagged saving rate Log (Real disposable income Yd) Real growth rate of Yd Terms of trade(a) Old dependency ratio Young dependency ratio Unemployment rate Inflation rate(a) GDP growth volatility Real interest rate(a) Δ(Debt/Yd)>0 Δ(Debt/Yd)<0 House prices, real growth lagged Stock prices, real growth lagged Country fixed effects Time fixed effects R-squared Number of observations Number of countries
5 years before peak to trough (1) (2) 0.889*** 0.875*** (36.09) (38.70) 4.682 4.216* (1.78) (1.86) 0.332*** 0.370*** (4.66) (5.44) -2.268 -1.605 (-0.89) (-0.75) -0.032 -0.038 (-0.16) (-0.26) 0.095 0.084 (1.39) (1.43) 0.064 0.061 (1.26) (1.42) 65.506* 61.851* (2.05) (2.06) 0.013 0.011 (0.23) (0.21) 7.095 6.071 (1.63) (1.50) -0.300*** -0.283*** (-4.68) (-4.01) -0.192* -0.159 (-1.88) (-1.84) -0.056** (-3.28) -0.010 (-1.75) Yes Yes 0.425 355 9
Yes Yes 0.448 355 9
10 years before peak to trough (3) (4) 0.910*** 0.897*** (51.30) (47.66) 2.475 1.593 (1.09) (0.70) 0.374*** 0.405*** (6.05) (7.43) -0.229 0.305 (-0.12) (0.18) -0.036 -0.091 (-0.24) (-0.72) 0.022 -0.004 (0.46) (-0.09) 0.062 0.055 (1.59) (1.40) 61.682* 58.489* (2.28) (2.23) 0.025 0.018 (0.48) (0.38) 4.334 3.438 (1.17) (1.05) -0.284*** -0.264*** (-6.13) (-5.48) -0.166 -0.131 (-1.84) (-1.81) -0.058** (-3.19) -0.010** (-2.41) Yes Yes 0.649 486 9
Yes Yes 0.749 479 9
Notes: (a) Expressed in logs (log of (1 + x) for the real interest rate and the inflation rate). Constant term not reported. ***, **, * denote significativity at the 1%, 5%, and 10% level, respectively. Robust t-statistics appear in parentheses.
40
Table 8 – Saving rates, changes in debt ratios, and interaction effects, quarterly data Dependent variable: Household gross saving rate (1) (2) (3) (4) Lagged saving rate
0.892*** (59.04) 2.874*** (3.06) 0.505*** (8.72) -0.277 (-0.31) -0.015 (-0.67) 0.108** (2.26) 0.015 (1.21) 16.735 (0.79) 0.047 (1.63) -6.381** (-2.56) -0.010 (-0.64) -0.016** (-2.25) 0.011 (0.63) -0.049 (-1.27) -0.138* (-1.81)
Log (Real disposable income Yd) Real growth rate of Yd Terms of trade(a) Old dependency ratio Young dependency ratio Unemployment rate Inflation rate(a) GDP growth volatility Real interest rate(a) House prices, real growth lagged Stock prices, real growth lagged Cyclically adj. government net lending Δ(Debt/Yd) Δ(Debt/Yd)*High Consumer Credit dummy Δ(Debt/Yd)*HEW dummy
0.891*** (81.74) 2.961*** (4.08) 0.500*** (8.42) -0.295 (-0.36) -0.023 (-1.01) 0.124*** (3.29) 0.015 (1.56) 18.721 (0.85) 0.048* (1.81) -6.293** (-2.79) -0.014 (-0.92) -0.016** (-2.10) 0.000 (0.03) -0.251** (-2.42)
0.896*** (61.83) 2.763** (2.64) 0.511*** (9.94) 0.025 (0.03) -0.024 (-0.52) 0.108** (2.34) 0.015 (1.34) 25.790 (1.00) 0.050** (2.17) -6.641** (-2.47) -0.008 (-0.47) -0.017* (-2.09) 0.006 (0.34) -0.025 (-0.12)
0.894*** (67.67) 3.086*** (3.77) 0.528*** (10.37) -0.370 (-0.40) -0.013 (-0.52) 0.128** (2.72) 0.018* (1.81) 19.164 (0.87) 0.053** (2.23) -6.805** (-2.53) -0.015 (-0.96) -0.016** (-2.14) 0.005 (0.28) -0.099* (-1.99)
0.165* (1.73)
Δ(Debt/Yd)*Homeownership
-0.113 (-0.37)
Δ(Debt/Yd)*High Pension dummy*Low Saving dummy
-0.019 (-0.38)
Country fixed effects Time fixed effects R-squared Number of observations Number of countries
Yes Yes 0.280 1390 22
Yes Yes 0.262 1390 22
Yes Yes 0.359 1223 19
Yes Yes 0.285 1390 22
Notes: (a) Expressed in logs (log of (1 + x) for the real interest rate and the inflation rate). Constant term not reported. ***, **, * denote significativity at the 1%, 5%, and 10% level, respectively. Robust t-statistics appear in parentheses.
41
Table 9 – Saving rates, changes in debt ratios, and consumer credit – Robustness checks, quarterly data Dependent variable: Household gross saving rate 1995Q1-2012Q4 High Consumer Credit Low Consumer Countries Credit Countries (1) (2) (3) (4) Lagged saving rate Log (Real disposable income Yd) Real growth rate of Yd Terms of trade(a) Old dependency ratio Young dependency ratio Unemployment rate Inflation rate(a) GDP growth volatility Real interest rate(a) House prices, real growth lagged Stock prices, real growth lagged Cyclically adj. gov. net lending Δ(Debt/Yd)
0.875*** (31.47) 6.689*** (3.84) 0.460*** (7.59) -0.561 (-0.65) 0.042 (0.75) 0.130** (2.48) 0.069 (1.72) 28.058 (0.86) 0.070 (1.81) -7.543*** (-4.23) 0.012 (0.46) -0.030*** (-5.53) 0.036 (1.33) -0.204** (-2.87)
Credit effect, ΔDebt/Yd Country fixed effects Time fixed effects R-squared Number of observations Number of countries
0.873*** (33.37) 6.430*** (3.78) 0.632*** (18.76) -0.186 (-0.20) 0.036 (0.61) 0.143* (2.23) 0.045 (1.31) 41.390 (1.20) 0.069 (1.81) -10.565*** (-6.73) 0.004 (0.15) -0.028*** (-5.28) 0.042 (1.47)
0.859*** (43.83) 2.940** (2.80) 0.564*** (6.16) -1.542* (-1.90) 0.008 (0.18) -0.015 (-0.23) 0.053*** (3.37) 16.686* (1.92) 0.073** (2.51) 0.477 (0.14) -0.031 (-1.40) 0.004 (1.60) -0.016 (-1.39) -0.024 (-1.22)
-0.188*** (-3.44) Yes Yes 0.158 576 9
Yes Yes 0.163 576 9
0.859*** (43.27) 3.015** (2.79) 0.592*** (5.95) -1.535* (-1.89) 0.009 (0.20) -0.009 (-0.14) 0.053** (2.62) 16.855* (1.87) 0.075** (2.55) -0.200 (-0.06) -0.032 (-1.41) 0.004 (1.60) -0.014 (-1.20)
1995Q1-2006Q4 High Consumer Credit Low Consumer Credit Countries Countries (5) (6) (7) (8) 0.846*** (12.70) 7.613* (2.10) 0.239* (2.16) 0.974 (0.39) 0.034 (0.41) 0.199* (1.95) 0.167 (1.25) 42.762 (1.18) 0.022 (0.39) -4.479* (-2.04) 0.011 (0.59) -0.021*** (-4.97) 0.016 (0.72) -0.309*** (-4.68)
-0.024 (-0.78) Yes Yes 0.490 814 13
Yes Yes 0.479 814 13
0.858*** (11.18) 7.439 (1.67) 0.471*** (4.91) 0.979 (0.37) 0.020 (0.19) 0.254* (2.16) 0.160 (1.10) 58.145 (1.51) 0.019 (0.27) -8.527*** (-4.52) 0.005 (0.25) -0.020*** (-3.87) 0.028 (1.19) -0.273*** (-3.42)
Yes Yes 0.268 380 9
Yes Yes 0.271 380 9
Notes: (a) Expressed in logs (log of (1 + x) for the real interest rate and the inflation rate). Constant term not reported. ***, **, * denote significativity at the 1%, 5%, and 10% level, respectively. Robust t-statistics appear in parentheses.
42
0.816*** (33.74) 4.820 (1.66) 0.551*** (4.99) -4.358** (-2.46) 0.010 (0.18) 0.099 (1.06) 0.119* (1.81) 21.472* (1.94) -0.027 (-0.49) 0.467 (0.09) -0.060* (-1.93) 0.004 (1.35) -0.071** (-2.83) -0.014 (-0.39)
0.815*** (34.84) 5.001 (1.74) 0.569*** (4.98) -4.305** (-2.34) 0.006 (0.10) 0.107 (1.18) 0.115 (1.65) 21.124* (1.91) -0.029 (-0.54) -0.003 (-0.00) -0.059* (-1.79) 0.004 (1.35) -0.067** (-2.71) -0.039 (-1.00)
Yes Yes 0.275 537 13
Yes Yes 0.260 537 13
Table 10 – Positive versus negative changes of debt ratios and consumer credit Dependent variable: Household gross saving rate High Consumer Credit Countries 1980-2012 1980-2012, 1980-2006, KOR KOR excluded excluded (1) (2) (3) Lagged saving rate
Low Consumer Credit Countries 1980-2012 1980-2006 (4)
(5)
0.601*** (42.64) 8.385*** (5.50) 0.182** (2.85) 7.940*** (6.96) 0.096* (2.39) 0.231*** (6.84) 0.190*** (4.65) 22.428*** (6.36) 0.157* (2.51) 4.178 (0.57) -0.012 (-1.99) 0.014 (0.51) -0.116* (-2.03) -0.261** (-3.85)
0.561*** (25.83) 13.931*** (5.40) 0.195*** (5.46) 2.050 (1.24) 0.137** (3.66) 0.111 (2.10) 0.180** (3.17) 16.418*** (4.89) 0.114** (2.93) -5.570 (-1.21) 0.001 (0.24) -0.017 (-0.88) -0.079** (-2.92) -0.109* (-2.53)
0.499*** (12.02) 16.737*** (15.82) 0.151** (3.66) 2.616 (0.74) 0.129** (2.88) 0.127* (2.47) 0.306*** (7.19) 19.934** (3.63) 0.040 (0.44) -0.626 (-0.16) 0.003 (0.72) -0.018 (-0.72) -0.081* (-2.44) -0.066 (-1.31)
0.772*** (21.88) 2.676* (1.95) 0.441*** (12.09) -1.541 (-1.36) -0.086 (-0.89) -0.041 (-0.76) 0.071 (1.40) 17.389*** (3.17) 0.263* (1.84) 3.916 (0.72) -0.006 (-1.61) -0.037*** (-4.06) -0.137*** (-8.06) 0.045 (0.91)
0.750*** (13.67) 2.520* (2.02) 0.439*** (9.92) -0.799 (-0.43) -0.075 (-0.79) -0.020 (-0.34) 0.061 (0.89) 19.695** (2.77) 0.310* (1.95) 2.324 (0.41) -0.006 (-1.52) -0.047*** (-3.87) -0.149*** (-6.53) 0.083 (1.48)
Test Coeff. Δ(Debt/Yd)>0 = Coeff. Δ(Debt/Yd)<0
0.02
0.68
0.83
0.01
0.01
Country fixed effects Time fixed effects R-squared Number of observations Number of countries
Yes Yes 0.210 182 6
Yes Yes 0.136 161 5
Yes Yes 0.178 132 5
Yes Yes 0.827 348 12
Yes Yes 0.840 284 12
Log (Real disposable income Yd) Real growth rate of Yd Terms of trade(a) Old dependency ratio Young dependency ratio Unemployment rate Inflation rate(a) GDP growth volatility Real interest rate(a) House prices, real growth lagged Stock prices, real growth lagged Δ(Debt/Yd)>0 Δ(Debt/Yd)<0
Notes: (a) Expressed in logs (log of (1 + x) for the real interest rate and the inflation rate). Constant term not reported. ***, **, * denote significativity at the 1%, 5%, and 10% level, respectively. Robust t-statistics appear in parentheses.
43
Table 11 – Full boom-bust cycles of debt ratios and consumer credit Dependent variable: Household gross saving rate Annual data
(2)
KOR, GBR, JPN excl. (3)
KOR, GBR, JPN, AUS excl. (4)
0.418*** (6.60) 7.516** (2.52) 0.184** (2.27) 0.845 (0.26) -0.114 (-0.45) -0.111 (-1.58) 0.150* (2.08) 2.790 (0.54) 0.351*** (3.45) 16.888* (2.23) -0.009** (-2.32) -0.057*** (-5.34) -0.192*** (-5.76) -0.064 (-0.81)
0.372*** (5.55) 17.269*** (5.36) 0.152* (1.94) 2.250 (0.99) -0.339 (-1.71) 0.007 (0.08) 0.189** (2.41) 1.169 (0.15) 0.260* (1.96) 17.822* (2.18) -0.008 (-1.55) -0.059*** (-5.69) -0.204*** (-8.40) -0.061 (-0.72)
0.394*** (5.31) 18.568*** (4.15) 0.207** (2.53) 1.475 (0.48) -0.282 (-1.09) -0.011 (-0.16) 0.213* (2.34) 10.540 (1.43) 0.305** (2.57) 17.519* (2.08) -0.007 (-1.17) -0.067*** (-7.23) -0.200*** (-9.50) -0.019 (-0.25)
0.355*** (4.69) 17.707*** (4.26) 0.224** (3.05) 2.575 (1.16) -0.432 (-1.76) -0.044 (-0.57) 0.242* (2.26) 14.020 (1.27) 0.376** (2.80) 12.909 (1.23) -0.010** (-2.59) -0.069*** (-11.00) -0.193*** (-8.43) -0.009 (-0.13)
0.801*** (27.41) 8.395** (3.10) 0.478*** (6.05) -7.862** (-2.74) 0.422 (1.28) 0.091 (0.97) 0.044 (0.62) 0.980 (0.05) 0.115** (3.13) -2.488 (-1.36) -0.047 (-1.93) 0.002 (0.38) -0.250** (-3.07) -0.023 (-0.47)
0.811*** (24.14) 7.446*** (4.16) 0.479*** (45.78) -1.103 (-1.79) 0.144 (1.20) 0.041 (0.85) 0.073 (1.89) 20.030 (0.98) 0.117** (2.69) -3.423* (-2.09) -0.064* (-2.50) -0.001 (-0.35) -0.276** (-3.38) -0.011 (-0.21)
Test Coeff. Δ(Debt/Yd)>0 = Coeff. Δ(Debt/Yd)<0
0.25
0.21
0.10
0.07
0.02
0.06
Country fixed effects Time fixed effects R-squared Number of observations Number of countries
Yes Yes 0.22 141 10
Yes Yes 0.16 124 9
Yes Yes 0.01 112 8
Yes Yes 0.03 99 7
Yes Yes 0.01 234 6
Yes Yes 0.03 311 6
Lagged saving rate Log (Real disposable income Yd) Real growth rate of Yd Terms of trade(a) Old dependency ratio Young dependency ratio Unemployment rate Inflation rate(a) GDP growth volatility Real interest rate(a) House prices, real growth lagged Stock prices, real growth lagged Δ(Debt/Yd)>0 Δ(Debt/Yd)<0
KOR excluded
KOR, GBR excl.
(1)
Quarterly data, KOR, GBR, JPN, AUS excluded 5 years 10 years before peak before peak to trough to trough (5) (6)
Notes: (a) Expressed in logs (log of (1 + x) for the real interest rate and the inflation rate). Constant term not reported. ***, **, * denote significativity at the 1%, 5%, and 10% level, respectively. Robust t-statistics appear in parentheses.
44
Table 12 – Saving rates, consumer credit, and housing credit Dependent variable: Household gross saving rate
Lagged saving rate Log (Real disposable income Yd) Real growth rate of Yd Terms of trade(a) Old dependency ratio Young dependency ratio Unemployment rate Inflation rate(a) GDP growth volatility Real interest rate(a) House prices, real growth lagged Stock prices, real growth lagged Δ(Debt/Yd)
(1)
(2)
(3)
(4)
(5)
(6)
0.803*** (32.42) 5.031** (2.69) 0.376*** (7.65) 0.928 (0.45) -0.082 (-1.85) 0.064 (0.88) 0.109 (1.22) 18.335** (2.65) 0.222*** (4.61) -5.411* (-1.93) 0.024 (0.85) 0.005 (1.34) -0.096** (-3.27)
0.807*** (30.23) 3.684* (2.23) 0.410*** (8.05) 0.783 (0.49) -0.108** (-2.90) 0.083 (1.01) 0.083 (1.36) 11.276* (1.91) 0.224*** (5.50) -7.928* (-2.36) 0.023 (0.84) 0.006 (1.32) -0.056 (-1.87) -0.373** (-2.64)
0.803*** (32.80) 5.035** (2.71) 0.375*** (7.86) 1.006 (0.47) -0.081 (-1.66) 0.067 (0.96) 0.109 (1.23) 18.558** (2.98) 0.219*** (4.80) -5.437* (-1.90) 0.023 (0.75) 0.005 (1.34) -0.101*** (-4.15)
0.814*** (30.79) 3.608* (2.06) 0.431*** (8.82) 0.254 (0.15) -0.121** (-2.84) 0.071 (0.85) 0.097 (1.54) 10.024 (1.65) 0.243*** (5.26) -8.130** (-2.47) 0.025 (0.80) 0.006 (1.41)
0.808*** (28.17) 3.640* (2.07) 0.412*** (8.29) 0.563 (0.34) -0.113** (-2.50) 0.075 (0.89) 0.082 (1.32) 10.494 (1.81) 0.231*** (5.26) -7.912* (-2.33) 0.025 (0.80) 0.006 (1.32) -0.041 (-1.87) -0.381** (-2.68) -0.022 (-0.49)
0.810*** (31.32) 4.403** (2.43) 0.436*** (8.54) -0.754 (-0.41) -0.138** (-3.47) 0.071 (0.86) 0.124 (1.66) 8.996 (1.23) 0.238*** (5.63) -8.266* (-1.99) 0.026 (0.74) 0.005 (1.28)
Δ(Consumer Debt/Yd) Δ(Housing Debt/Yd)
0.008 (0.21)
-0.422** (-3.22) -0.056 (-1.26)
Δ(Consumer Debt/Yd)>0
-0.365 (-1.66) -0.529*** (-4.87) -0.080 (-1.47) 0.027 (0.34)
Δ(Consumer Debt/Yd)<0 Δ(Housing Debt/Yd)>0 Δ(Housing Debt/Yd)<0 Country fixed effects Time fixed effects R-squared Number of observations Number of countries
Yes Yes 0.95 195 8
Yes Yes 0.95 195 8
Yes Yes 0.95 195 8
Yes Yes 0.95 195 8
Yes Yes 0.95 195 8
Yes Yes 0.95 195 8
Notes: (a) Expressed in logs (log of (1 + x) for the real interest rate and the inflation rate). Constant term not reported. ***, **, * denote significativity at the 1%, 5%, and 10% level, respectively. Robust t-statistics appear in parentheses.
45
Appendix – Variable definitions and data sources Variable Saving rate
Definition Ratio of household gross saving to gross disposable income.
Source OECD National Accounts.
Household debt
Annual data: Total financial liabilities of households. Quarterly data: debt (loans) of households.
Real disposable income
Household disposable income deflated by the Consumer Price Index (CPI) and converted in US dollars using PPPs. Ratio of export and import prices.
Annual data: OECD National Accounts and national central banks (see Bouis et al., 2013 for details). Quarterly data: BIS (see Dembiermont et al., 2013). OECD National Accounts.
Terms of trade Old and young-age dependency ratios Unemployment rate Inflation rate GDP growth volatility Real short-term interest rate Household net financial wealth-todisposable income ratio Real house price growth rate
Real stock price growth rate
Cyclically adjusted government net lending Housing investment ratio Money-to-GDP ratio IMF Financial Liberalisation index
Homeownership rates Housing equity withdrawal Consumer debt Housing debt Pension funds
Proportions of the population aged 65 and above, and of the population aged below 15 in the working age population (population aged 15-64), respectively. Unemployed persons as a percentage of the labour force. Growth rate of the CPI. GARCH (1,1) measure of GDP growth volatility. Real three-month interest rate based on private consumption deflator. Household financial assets minus liabilities, divided by gross disposable income. Growth rate of real house price indexes, deflated by CPI.
Growth rate of the MSCI equity index, except for Estonia (OMX Tallinn), Korea (Index Korea Exchange Composite), Portugal (Index PSI-20), and Slovakia (Slovakia SAX 16), deflated by CPI. Cyclically adjusted government net lending as a percentage of potential GDP (in general available from 1986Q4). Growth fixed capital formation in housing to disposable income ratio. Money supply M2, national definition, divided by nominal GDP. Index of financial regulation across seven different dimensions: credit controls and reserve requirements, interest rate controls, entry barriers, state ownership, policies on securities markets, banking regulations, and restrictions on the capital account. Proportion of the population aged between 25 and 84 owing a house in total population aged between 25 and 84. Dummy variable equal to one of housing equity withdrawal is available in the country, zero otherwise. Debt used for consumption of durable and non-durable goods divided by household gross disposable income. Debt used for the purchase of a house divided by household gross disposable income. GDP share of pension funds in 2011.
OECD National Accounts. OECD Labour Force Statistics. OECD Main Economic Indicators. OECD National Accounts. OECD National Accounts and author’s calculations. OECD National Accounts. OECD National Accounts and author’s calculations. OECD and BIS (for Austria, Czech Republic, Estonia, Greece, Hungary, New Zealand, Slovak Republic, and Slovenia). Datastream and OECD National Accounts. OECD Economic Outlook database. OECD National Accounts. Datastream and OECD National Accounts. IMF (see Abiad et al., 2008).
Luxembourg Income Study. Andrews (2010), Igan and Loungani (2012). National central banks, OECD National Accounts. National central banks, OECD National Accounts. OECD Global Pension Statistics.
Note: The household sector refers to the aggregate account of households and non-profit institutions serving households (NPISHs).
46
