Debt-Driven Growth? Wealth, Distribution and Demand in OECD Countries

Debt-driven Growth?
Wealth, Distribution and Demand in
OECD Countries
Abstract: The paper investigates the effects of changes in the distribution of income and in
wealth on aggregate demand and its components. We extend the Bhaduri and Marglin (1990)
model to include personal income inequality as well as asset prices and debt. This allows for an
evaluation of the wage or profit-led nature of demand regimes, of the expenditure cascade
argument (Frank et al. 2010) and several hypotheses regarding the effects of wealth and debt.
Our estimates are based on a panel of 18 OECD countries covering the period 1980-2013. For
the full panel the average demand regime is found to be wage led. We fail to find effects of
personal inequality, but do find strong effects of debt and property prices which have been the
major drivers of aggregate demand in the decade prior to the 2007 crisis.
Keywords: Post-Keynesian economics, wage-led growth, Bhaduri-Marglin model, demand

regimes, wealth effect, Veblen effect, expenditure cascade, aggregate demand.
JEL classifications: E11, E12, E21
This paper is part of the research project ‘Rising inequality as a structural cause of the present
financial and economic crisis’ with financial support by the Institute for New Economic
Thinking. An earlier version of this paper was presented at the workshop Inequality, debt and
demand, Kingston University, Sept 2014. We are grateful to the participants of the workshop for
helpful comments. The usual disclaimers apply.
Engelbert Stockhammer
Kingston University
Engelbert Stockhammer, Kingston University, Penrhyn Road, KT1 2EE Surrey, UK
Email: [email protected]
Phone: +44 208 417 7774
Rafael Wildauer
Kingston University London
[email protected]
PKSG Debt-driven growth? Wealth, distribution and demand in OECD countries
1 Introduction
The effects of the changes in income distribution on aggregate demand and the role of debt and
financial assets are two intensely debated issues in modern heterodox macroeconomics. This
corresponds to important changes in contemporary capitalism. Since 1980 dramatic shifts in the
distribution of income as well as in the valuation of assets have occurred. Figure 1 shows the evolution
of the wage and top income share, real house prices and household debt for five major OECD countries.
Wage shares have been falling in all countries. Top incomes as well as property prices and household
debt have increased in most but not all countries. Notably Germany and Japan do not experience a
property price boom in the decade prior to 2008. While there is agreement on the trends and their
importance, there is disagreement in heterodox macroeconomics on their effects.
Figure 1. Macroeconomic trends 1980-2010
As regards the role of income distribution there have been two important debates. First, on changes in
the functional distribution of income, Kaleckians have argued that an increase in the wage share will
increase aggregate demand (called a wage-led demand regime) because the propensities to consume
March 2015 Page | 1 Stockhammer and Wildauer

out of wages is higher than that out of capital incomes (Kalecki 1954). On the other hand, Marxists
(Goodwin 1967) argue that higher profits translate into higher investment (called a profit-led demand
regime). Bhaduri and Marglin (1990) presented a unifying framework, which allows for both wage-led
and profit-led regimes that has become a widely used tool within post-Keynesian economics and has
inspired a rich empirical literature (Bowles and Boyer 1995; Stockhammer and Onaran 2004; Naastepad
and Storm 2007; Hein and Vogel 2008; Stockhammer and Stehrer 2011; Onaran and Galanis 2012). Most
of this literature focuses on the effects of changes in the functional income distribution, but pays little
attention to other factors. The second debate about income distribution focused on the sharp rise in top
incomes (Atkinson et al. 2011). The standard Kaleckian hypothesis is that rising inequality will lead to
lower consumption expenditures as the rich will have a lower consumption propensity than the poor.
However, Frank et al. (2010) argue that people will emulate consumption patterns of richer peers in an
attempt to climb up the social ladder. Rising personal income inequality will thus lead to expenditure
cascades and increase consumption. 1
Post-Keynesian (PK) macroeconomics has long recognised the importance of finance in general and
asset prices and debt specifically. Minsky (1995) regarded debt cycles as the driver of economic
fluctuations. Recently there have been several attempts to formalise his model (Charles 2008; Fazzari et
al. 2008; Keen 1995; Ryoo 2013). There is a surge in interest in stock-flow consistent (SFC) models
(Godley and Lavoie 2007), which highlight the impact of stock variables such as debt and net wealth on
macroeconomic aggregates. We note two shortcomings of the debate. First, while there is an agreement
on the key role of debt and wealth, there is no agreed-upon model. In Minsky’s analysis business debt is
central. This clearly does not fit the recent experience of rising household debt and a consumption
boom. In contrast the stock-flow consistent modelling (SFC) literature typically highlights household
debt and it often allows for different stock and flow effects of debt or asset prices. Second, the debate
has so far motivated more theoretical than empirical work (Zezza 2009 is one of the few exceptions).
Overall, post-Keynesian macroeconomics here lags behind the mainstream literature, which has built a
substantial body of empirical research on wealth effects and consumption, which is ironic, given that
most mainstream macro models have given little role to debt. Onaran et al. (2011) on the USA and Nishi
(2012) on Japan are some of the few post-Keynesian studies analysing the effects of functional income
1
We use the terms expenditure cascades and Veblen effects synonymously.

distribution and household wealth and debt, but they do not investigate the role of personal income
distribution and, importantly, they only analyse a single country.
The aim of the paper is to analyse the effects of changes in functional and personal income distribution
as well as changes of household and business wealth and debt on aggregate demand and its
components. We are interested in the sign of the effects, which will allow us to evaluate whether
demand is wage-led or profit-led and whether expenditure cascades or ‘the rich save more’ effects
dominate. But we also want to analyse the growth contribution of these effects, in order to determine
to what extent the characterisation of growth as ‘debt-driven’ or ‘finance-led’ (Stockhammer 2012; Hein
2012b) is a useful description.
The paper extends a Bhaduri-Marglin model for measures of personal income inequality as well as
measures of property and financial wealth and private debt. We estimate it based on a panel of 18 OECD
countries covering the period 1980 to 2013. Our contribution is threefold: First, we provide a neo-
Kaleckian framework to analyse the effects of distribution and wealth on aggregate demand and its
components. This allows us to assess different hypotheses on the effects of wealth and debt variables.
Second, we provide an empirical assessment of the relative growth contributions of these effects for
different country groups. Third, we estimate our model by means of panel analysis whereas most of the
relevant literature relies on time series evidence for individual countries.
The panel approach does impose the restriction that parameters are identical across countries, which
clearly will only hold as an approximation. Panel analysis comes with costs as well as benefits. Its main
advantage is that it allows for including a rich set of country experiences, i.e. more variation in the
variables. In particular we include many European countries, whereas most of the research on wealth
effects is on Anglo-Saxon countries. Given that data for household and business debt is only available
since 1980 (for most countries) and that top income data is only available annually, our research
question can only be explored by panel analysis. However, we will provide some evidence that the
pooling assumption is a reasonable approximation and that heterogeneous outcomes across the
countries are driven primarily by differences in the time paths of the explanatory variables rather than
by differences in parameters across countries.
The remainder of the paper is organized as follows: Section 2 presents the theoretical framework used.
Section 3 reviews the existing empirical literature dealing with the Bhaduri-Marglin framework and the

empirical literature estimating wealth and debt effects. Section 4 discusses the data set and the
econometric issues of using panel data methods in a macroeconomic context. Section 5 presents the
results and a final section concludes.
2 The theoretical framework

The starting point for our investigation is the Bhaduri and Marglin (1990) model that has become a
standard reference point in modern post-Keynesian economics. We extend the model to include
expenditure cascades and wealth and debt effects in both investment and consumption. Real aggregate
demand (𝑌) consists of consumption spending (𝐶), investment expenditures (𝐼), net exports (𝑁𝑋) and
government spending (𝐺). We abstract from government and thus write aggregate demand as:
𝐴𝐷 = 𝐶 + 𝐼 + 𝑁𝑋 (1)
Consumption is
𝐶 = 𝐶(𝑌, 𝑊𝑆, 𝑄, 𝑊𝐻, 𝑊𝐹, 𝐷𝐻, ∆𝐷𝐻), (2)
With CY, CWS, CWH, CWF, C∆DH >0, CDH<0, CQ=?
where 𝑌, 𝑊𝑆, 𝑄, 𝑊𝐻, 𝑊𝐹, and 𝐷𝐻 are income, the wage share, personal income inequality, housing
wealth, financial wealth and household debt, respectively. We expect 𝑌 to have a positive effect and 𝑊𝑆
to have a negative effect because the marginal propensity to consume (MPC) out of wage income is
likely to be higher than that out of capital income. The personal distribution of income is relevant for
two reasons. First, the marginal as well as average propensities to consume vary across income groups.
The standard assumption here is that the poor have a higher MPC, which would imply a negative effect
of inequality on consumption < 0 . Second, if households care about consumption and income
relative to their peers, an increase in inequality has a positive effect on consumption >0 .
Following the work of Duesenberry (1949), Frank (1985) and Frank et al. (2010) developed the theory of
consumption cascades, which can occur when people have upward-looking consumption norms, i.e. if
they try to keep up with those above them in the income distribution. Several authors have
incorporated these assumptions in macroeconomic models (Kapeller and Schütz 2014; Belabed et al.
2013).

The role of household wealth and household debt has recently gained prominence, however there are
several competing hypotheses, which are summarised in Table 1. Keynesians have long highlighted the
importance of financial factors, but consumption has traditionally not featured prominently in this. Early
contributions highlighted that changes in the liquidity preference could cause financial crises (Keynes
1973; Davidson 1972). The main transmission mechanism for this was interest rates. In PK models
household debt has a dual influence on consumption since it provides a source of finance, thus having a
positive impact on consumption but also leads to servicing costs which depresses consumption if the
MPC out of interest income is low (Dutt 2006; Nishi 2012; Hein 2012a). In our context this implies the
hypothesis of < 0 because higher debt levels also lead to higher interest payments and thus
decrease disposable income and thus consumption. On the other hand ∆

> 0 since taking on
additional debt initially increases disposable income and the ability to finance consumption
expenditures. Therefore the overall effect is not a priory clear. However, a shortcoming of these models
is that they do not explicitly assign a role to asset prices (or net wealth). However, the most important
reason for households to go into debt is not consumption related, but asset transaction related, namely
the acquisition of homes. The Minskyian stream within post-Keynesian economics has long emphasised
the role of asset prices in borrowing (and lending). In these models (Ryoo 2013), optimistic investors will
drive up asset prices during boom phases, lowering corporate financing costs and thus encouraging
businesses to take on more debt. However, Minsky’s original writings analysed businesses and their
debt rather than households and mortgage debt. Extending his argument to households, we would
expect a strong effect of housing wealth, which underwrites household debt and we would expect
autonomous movements in housing wealth to drive both, debt and consumption, > 0.2
Table 1. Different consumption theories

Hypothesis Theoretical Argument Predicted signs
Inequality lowers consumption because
the rich save more richer households have a higher <0
propensity to save.
Households make consumption
expenditure cascades >0
decisions with respect to richer peers.
housing wealth is no wealth Rising house prices lead to wealth =0
2
This means that in our context the hypotheses derived from Minsky for household behaviour, namely a positive
partial derivative of C with respect to property prices, is equivalent to that of Muellbauer. There are theoretical
differences however. Muellbauer (2007) is based on rational life-cycle consumption while Minskyian households
are becoming more optimistic due to endogenous animal spirits based on asset price cycles.

effects for home owners and higher

savings of willing-to-be-homeowners.
Net wealth (NW) is the relevant wealth
net wealth effect ( )
=−
measure for consumption decisions.
Due to credit constraints, changes in
credit constraints housing wealth effect consumption even > 0 and >
if shocks are anticipated.
Rising asset prices lead to increasingly
Minskyian households >0
optimistic lending and spending.
The stock of debt implies interest rate
payments which affect consumption
stock and flow effects of debt < 0 and ∆
>0
negatively whereas the flow of debt
affects consumption positively.
In mainstream consumption theory households maximise utility over the life-cycle. Thus net wealth,
which is assets minus liabilities (𝑁𝑊 = 𝑊𝐻 + 𝑊𝐹 − 𝐷𝐻), plays a key role. If net wealth is the variable
to affect consumption, this implies ( )

=− . However, it is not straight forward that
measured net wealth is the relevant variable. Buiter (2010) argues that housing wealth does not
constitute wealth since rising prices only make consumers who are long in housing better off, whereas
those who rent are worse off. He shows that in a representative agent model the net effect is zero, i.e.
= 0. New Keynesian modifications of the neoclassical model highlight the possibility of credit
rationing (Muellbauer 2007). In these models housing wealth can relax credit constraints because it
serves as collateral > 0 and we would expect > because housing wealth is more likely
to be accepted as collateral than financial wealth.
Investment is
𝐼 = 𝐼(𝑌, 𝑊𝑆, 𝑖, 𝑄, 𝑊𝐻, 𝑊𝐹, 𝐷𝐻, 𝐷𝐵) (3)
With IY, IWH, IWF, I∆DB, I∆DH >0, Ii, IDB, IDH<0, IWS, IQ=?
where 𝑖 and 𝐷𝐵 are the long term real interest rate and business debt respectively. Aggregate demand
and long-term real interest rates are standard in investment functions. The wage share may indicate
future profitability and retained earnings are an important source of funding. Stock markets represent
funding conditions for firms and are considered a leading business cycle indicator. We expect a positive
effect. Total investment consists of business investment and residential investment. We regard

residential investment as determined by a similar set of variables as consumption expenditures, i.e. our
investment function will also depend on the wage share, income inequality, housing and financial
wealth, and household debt. Three remarks are in order. First, while business investment will depend
negatively on the wage share, residential investment may also react positively to changes in the wage
share if wage earners own homes. The overall effect of the wage share on investment is thus
ambiguous. Second, since housing is an especially visible expenditure, it is likely to be influenced by
status comparison behaviour. Thus if there are strong consumption cascades, we would also expect
them to show in investment expenditures. Third, property prices are a cost for residential investment
and thus rising housing wealth may have a negative effect. However, increasing property prices raise
household wealth may improve access to credit (because of the rising value of collateral). This will have
a positive effect on residential investment. Theoretically, the effect of housing wealth on investment is
thus ambiguous.
Net exports are
𝑁𝑋 = 𝑁𝑋 𝑌, 𝑌 , 𝐸𝑋, 𝑊𝑆, 𝑊𝐻 (4)
𝑌 represents real foreign income and 𝐸𝑋 is the nominal effective exchange rate. For net exports the
close relationship of real unit labour costs and the wage share, justifies including the latter3. Since wages
are driving the domestic price level and thus the country’s international competitiveness, net exports
are expected to depend negatively on the wage share. The influence of domestic and foreign income as
well as the effective exchange rate is straight forward. Beside of that, rising housing wealth, via rising
property prices, potentially influences domestic price competitiveness and thus exports.
Substituting equation 2, 3, and 4 into 1, we can solve for equilibrium income, 𝑌 ∗ . The effect of a change
in the wage share on 𝑌 ∗ then is:
∗
= (5)
where 𝑓 = ( + + ) and 𝑓 = ( + + ). The short run effect is determined by 𝑓
which is private excess demand and represents the effect of a change in the functional income
distribution given a certain level of income. is the multiplier that also includes the marginal effects
3
In fact the AMECO database defines and computes real unit labour costs and the wage share at market prices
identically.

of income on investment. If 𝑓 > 0 then the demand regime is called wage-led and profit-led if the
effect is negative.
The effect of a change in 𝑊𝐻 is expected to be positive,4 thus we do not distinguish between different
regimes. However, we may wish to assess the relative actual impact of changes in income distribution
and of changes in wealth variables.
In the empirical analysis we will identify these regimes, which are based on the partial effects. In
addition we will also identify estimated contributions. These are calculated as the estimated coefficient
times the actual change in the explanatory variable, e.g. 𝛽 ∆𝑊𝐻, where 𝛽 is the estimated
coefficient of 𝑊𝐻 on 𝐶, which is an estimate for the partial effect .
3 The related empirical literature

As our approach integrates considerations of functional income distribution, personal income
distribution and wealth and debt, there is a potentially large and diverse literature that it is relevant. The
first debate is that on wage-led or profit-led demand regimes. Bhaduri and Marglin (1990) has become a
standard point of reference for the empirical literature. There are differences in the behavioural
equations estimated as well as in econometric methodology. The first one relies on a system approach
based on VAR models and often focuses on the mutual interaction between distribution and demand,
but typically do not distinguish between effects on consumption and investment. Stockhammer and
Onaran (2004), estimate a five variable VAR and find weak evidence for wage-led demand. Barbosa-Filho
and Taylor (2006), Kiefer and Rada (2014), Carvalho and Rezai (2014) estimate two variable VARs and
find profit-led demand.5 A second group applies a single equation approach where consumption,
investment and the external sector functions are estimated separately. Papers covering several
countries include Bowles and Boyer (1995), Naastepad and Storm (2007), Hein and Vogel (2008),
Stockhammer and Stehrer (2011) and Onaran and Galanis (2012).6 All these studies find wage-led
domestic demand regimes for most countries. Third, Hartwig (2014) is the only study to use panel data
4
Theoretically the effect of housing wealth on investment could be negative. In this case a negative total effect
could arise.
5
Among these only Barbosa-Filho and Taylor present results for investment and consumption. They find large
negative wage share results in the consumption function, which is at odds with their theoretical model.
6
A series of later papers puts more focus on the estimation of the net export effects where real unit labour costs
are driving price levels and thus are affecting exports and imports (Stockhammer et al. 2008; Onaran et al. 2011;
Stockhammer et al. 2011).

to estimate a single equation approach and finds a slightly wage-led demand regime. Kiefer and Rada
(2014) estimate demand and distribution equations for a panel of OECD countries with a set of control
variables that shift income distribution and find that demand is profit-led. Neither Hartwig (2014) nor
Kiefer and Rada (2014) control for wealth variables or personal income distribution.Most of the
literature uses relatively simple specifications including disposable income, interest rates and the wage
share as determinants in capital and investment functions. Onaran et al. (2011) and Nishi (2012) are
among the few exceptions and will be discussed below.
Second, there is a growing theoretical literature that is employing the relative income hypothesis.
Nevertheless there is of yet little empirical research that analyses its impact on aggregate consumption.7
Behringer and van Treeck (2013) use inequality as a variable explaining current account positions as well
as household saving rates and find a negative effect for the G7 countries. Brown (2004) offers a time
series analysis of consumption expenditures for the USA, where he controls for current income and for
inequality and finds a negative effect of inequality on consumption, which is at odds with Frank’s
argument. Carvalho and Rezai (2014) offer a theoretical model along the Bhaduri-Marglin lines with
personal income distribution effects based on consumption cascades. Their empirical estimation uses a
Threshold VAR, i.e. they split the sample according to periods of high and low personal inequality, but
do not control for personal inequality directly, nor do they control for financial variables.
The large moves in financial as well as housing wealth, especially in the US, have led to renewed interest
in the size of wealth effects, much of it is inspired by a neoclassical framework. In the basic formulations
either financial wealth, housing wealth and debt or share and property prices are added to standard
control variables (Girouard et al. 2006; Ludwig and Sløk 2004; Slacalek 2009). They find that the MPC out
of housing wealth is higher compared to financial wealth in the US and UK, but that MPC out of housing
wealth is often small and/or statistically insignificant in European countries. Moreover, wealth effects
have been increasing with financial deregulation. In a variation emphasising the importance of credit
availability Muellbauer (2007) and Aron et al. (2012) stress the role of housing wealth in relaxing credit
7
Neumark and Postlewaite (1998) and Bowles and Park (2005) use income and inequality measures to explain
labour supply decisions. McBride (2001) reports the effects of relative income on self-reported happiness. Again on
the micro level, Alvarez-Cuadrado et al. (2012) report that consumption is a positive function of average
consumption of a geographical reference group. Thus an increase in inequality would decrease aggregate
consumption.

constraints. They show that for the USA and the UK measures of credit availability are a key driver of
consumption.
Wealth and debt considerations have not played a major role in post-Keynesian analysis of consumption
until recently. Indeed while there has been resurgence in the role of debt and financial instability, most
contributions do not explicitly address consumption dynamics. For example Zhang and Bezemer (2014)
investigate the effects of debt on growth and disaggregate debt by sector and according to whether it is
stock-transaction related. They find a negative effect of debt. Kim et al. (2015) is one of the few
exceptions that addresses consumption directly. They develop a post-Keynesian theory of consumption
based on social norms, relative income considerations and rule of thumb behaviour. They estimate an
aggregate consumption function for the USA 1952-2011 as a function of income, wealth, borrowing and
other controls and find that borrowing has positive effects. Based on an SFC framework Zezza (2009)
reports results for a private expenditure function, i.e. consumption plus investment, that is explained by
financial assets, stock prices, housing wealth and the change in household and business debt. The
specification does not include the level of household debt, nor does it include distributional variables.
The closest to our research question are Onaran et al. (2011) and Nishi (2012). Onaran et al. (2011)
introduce housing and financial wealth in a Bharduri-Marglin model and also distinguish between rentier
and non-rentier profit incomes in order to control for the effects of financialisation. They find the US
economy to be modestly wage-led and that growth has relied on wealth effects during periods of a
declining wage share. Nishi (2012) extends the standard Bhaduri Marglin framework for household
borrowing and estimates a structural VAR including the profit share, capital accumulation and the debt
ratio with quarterly data on Japan for the period 1992-2010, he finds Japan being profit-led and debt
burdened. Both do not control for personal income distribution and their studies are restricted to one
country each. Nishi only considers debt, but not financial or housing wealth.
The empirical analyses of the effects of changes in income distribution and of changes in wealth have
proceeded separately within post-Keynesian macroeconomics. While there is a high degree of
coherence in modelling of functional income distribution, there is disagreement over the effects of
changes in the wage share and of personal income distribution. There is as of yet hardly any empirical
results on expenditure cascades at the macroeconomic level. Although Post-Keynesian economics has
produced some highly original works on the role of debt and wealth, in its empirical analysis it does lag
behind the mainstream, which has produced a substantial literature on wealth effects and consumption.

4 Data and econometric method

Our data set covers 18 OECD member countries from 1980 to 2013 on an annual basis8. Definitions and
data sources are provided in Table A1 in the Appendix. We use real GDP (𝑌) as well as private final
consumption (𝐶), gross fixed capital formation (𝐼), exports of goods and services (𝑋) and imports of
goods and services (𝑀) at 2005 market prices in billions of national currency. These variables, the
corresponding deflators and the adjusted wage share at current factor costs (𝑊𝑆) are taken from the
Annual Macro-Economic database (AMECO). Real long term interest rates (𝑖) are taken from AMECO and
OECD’s Main Economic Indicator (MEI) database. Credit to households (𝐷𝐻), credit to the business
sector (𝐷𝐵), real property prices (𝑃𝑃) and trade weighted effective exchange rates (𝐸𝑋) are from the
Bank for International Settlements (BIS). The Theil index of wage dispersion (𝑇ℎ𝑒𝑖𝑙) is from the
University of Texas Inequality Project. The Gini coefficient (𝐺𝑖𝑛𝑖) and the share of richest 1% of
households (𝑇𝑜𝑝1) are taken from the Standardized World Income Inequality Database. The IMF’s
International Financial Statistics and the OECD’s MEI database are the sources for the stock price series
(𝑆𝑃). The latter is deflated using the CPI from the AMECO database. Real GDP of OECD countries (𝑌 ) is
also taken from OECD’s MEI database.
We use real property and stock price indices as proxies for housing wealth and financial wealth of the
household sector, because wealth data is not available (for sufficiently long time periods) for most
countries. This is common in the literature estimating wealth effects9, but it only captures price indices
but not quantity changes.
Our panel has a small N as well as a small T (N=18, T=33), which leads to econometric issues that are
distinct from much of the panel literature which assumes a very large N and small T. Our econometric
baseline specification is thus a first difference (FD) estimator and we perform a several robustness
checks. In our panel time series issues such as non-stationary regressors and unit roots are important.
Indeed, panel unit root tests (Choi 2001) indicate that the logarithmized data in levels exhibit unit roots.
After first differencing, the null hypothesis that all series contain a unit root can be rejected for all
variables (see Appendix Table A2). Based on these results we prefer the FD estimator to the standard
within-panel transformation since both allow for country fixed effects but the former is more reliable
with non-stationary data. The non-stationarity of our data set is also a reason not to use the widely used
8
The countries included are: Austria, Australia, Belgium, Canada, Switzerland, Germany, Denmark, Spain, Finland,
France, Ireland, Italy, Japan, Netherlands, Norway, Sweden, United Kingdom and the US.
9
See Paiella (2009), Attanasio and Weber (2010) and Cooper and Dynan (2014) for recent surveys.

dynamic system GMM procedure (Blundell and Bond 1998) since it requires mean-stationary series
(Baltagi 2013, p.167).
To address potential problems of autocorrelation in dynamic specifications we apply the Anderson and
Hsiao (1982) (A&H) estimator as well as restricted versions of one-step Arellano and Bond (1991) as
robustness checks. With the popular difference (Arellano and Bond 1991) and system GMM (Blundell
and Bond 1998) estimators the set of instruments required to handle the correlation of the lagged
dependent variable with the error term (Nickell 1981) exhibits quadratic increase in 𝑇 and thus these
methods become unfeasible when 𝑇 gets large in relation to 𝑁. We do not use the system GMM
estimator because it requires mean stationarity in levels (Baltagi 2013, p.167), which is not satisfied in
our dataset. We also experimented with cointegration specifications (available upon request). Results
were qualitatively similar to the FD estimator, but the results are not robust and cointegration relations
are usually very weak.
To investigate the sensitivity of our results to the pooling assumption we compare our results to those
of the mean group (MG) estimator of Pesaran and Smith (1995). The MG approach estimates an
individual model for each unit and averages across them. Thus one obtains an average estimate as with
a pooled procedure but without a priori restricting the coefficients to be identical for each country. If
the estimated parameters were strongly heterogeneous across units, the MG estimator and its standard
errors would differ from a pooled regression.
5 Results
5.1 Consumption function
The consumption function we are estimating is of the following form:
ln(𝐶 ) = 𝛽 ln(𝑌 ) + 𝛽 ln(𝑊𝑆 ) + 𝛽 ln(𝐷𝐻 ) + 𝛽 ln(𝑃𝑃 ) +

𝛽 ln(𝑆𝑃 ) + 𝛽 ln(𝑄 ) + 𝜇 + 𝑣 (7)
where 𝜇 are country fixed effects10. Heteroskedasticity and autocorrelation robust standard errors are
used in all specifications. We estimate equation 7 using different techniques and the results are
summarised in Table 2. The first four columns report variations of a FD estimator with varying control
10
Due to our limited sample size we were not able to include country and time fixed effects simultaneously,
especially with the dynamic specifications. Adding time dummies only proved to be relevant for the crisis years
anyway and including them did not change our results. See Table A5 in the Appendix.

variables. Specification 2 uses a Gini index and specification 3 a Theil index of wage dispersion instead of
the income share of the richest 1% of households in order to measure the personal income distribution.
Specification 4 focuses on the role of debt rather than assets and includes household debt in differences
as well as in levels in order to allow for negative stock and positive flow effects. Specification 5 employs
the MG estimator in order to assess the robustness of the FD estimator with respect to the assumption
of homogeneous coefficients across countries. Specification 6 reports a dynamic specification using the
A&H estimator and specification 7 the difference GMM estimator.
Table 2. Consumption function, dependent variable: log(Ct)

(1) (2) (3) (4) (5) (6) (7)
FD FD FD FD MG A&H GMM
Log(C)t-1 0.248*** 0.373***
(0.09) (0.12)
Log(Y)t 0.710*** 0.716*** 0.806*** 0.718*** 0.633*** 0.593*** 0.569***
(0.07) (0.06) (0.08) (0.06) (0.09) (0.10) (0.15)
Log(WS)t 0.136* 0.136* 0.141 0.154* 0.144** 0.072 0.078
(0.08) (0.08) (0.09) (0.08) (0.07) (0.07) (0.08)
Log(PP)t 0.017 0.021** 0.011 0.013 0.027*** 0.027**
(0.01) (0.01) (0.01) (0.03) (0.01) (0.01)
Log(SP)t -0.002 -0.001 -0.003 0.009 0.003 0.004
(0.00) (0.00) (0.00) (0.01) (0.00) (0.00)
Log(DH)t 0.110*** 0.105*** 0.085*** 0.114*** 0.128*** 0.049** 0.011
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.03)
ΔLog(DH)t 0.038*
(0.02)
Log(TOP1)t 0.008 0.008 -0.017 0.008 0.007
(0.01) (0.01) (0.03) (0.01) (0.01)
Log(GINI)t 0.027
(0.02)
Log(THEIL)t 0.018
(0.05)
cons 1.098**
(0.46)
N 474 485 361 456 492 448 387
cent. R2 0.859 0.860 0.879 0.863 0.866
F-stat 349 342 503 345 602
p-value Sargan 0.024

p-value Hansen 0.108 0.041

p-value AR(1) 0.200
p-value AR(2) 0.481
* p<0.1, ** p<0.05, *** p<0.01, heteroskedastic and autocorrelation robust standard errors reported in brackets.
FD refers to first difference estimators, MG stands for Mean Group estimator, A&H is the Anderson and Hsiao
(1982) estimator and GMM refers to the difference GMM estimator. p-value Sargan/Hansen refer to
overidentification tests and p-value AR represent Arellano-Bond autocorrelation tests.
Both have been restricted to two lags for instrumenting Ct-1 in order to keep the number of instruments
in an acceptable relation to our limited sample size. In general, results are robust across specifications,
however the difference GMM estimator does not pass the overidentification tests, pointing to
endogeneity problems with the instruments. Also the simple OLS version of the A&H estimator barely
passes these diagnostic tests. Therefor our preferred estimator is first differences (column 1). Overall,
the main results are similar across the static specifications, and we interpret the similar results of our
preferred specification (1) and the MG estimator (5) as support for our decision to pool the data.
The most robust result is the (long-run11) income elasticity of about 0.7, which is of an expected
magnitude. A 1% increase in the wage share has a direct (long-run) effect on consumption of about
0.14% across the static specifications. Household debt, as well as property prices have positive impacts
on consumption with elasticities of about 0.1 and 0.02, respectively. When testing for differences
between stock and flow effects in household debt (specification 4), we do find statistically significant
positive effects for debt levels, representing the stock effect, as well as for changes, representing the
flow effect. While the latter result is in line with the hypothesis of debt stock and flow effects, a positive
stock effect is not and we therefore focus on the specifications not distinguishing between the two.
Since mortgage debt dominates household debt measures, we interpret the pronounced effect of the
latter as evidence for the importance of housing wealth and equity withdrawals in financing consumer
spending. Surprisingly the variable which captures these effects directly, the property price index, turns
out statistically insignificant and with a much smaller estimated partial effect compared to household
debt levels. The reason might be that rising property prices are a prerequisite for equity withdrawals but
that the actual decision of households to withdraw equity for consumption purposes are influenced by
other factors independent of property prices. Stock prices have no statistically significant effect on
consumption. Estimated effects are small in all specifications and turn negative in some cases.
11
The coefficients in the dynamic specification are transformed to long run effects by multiplying them with 1/(1-
βC) where βC is the coefficient on Ct-1.

To assess consumption cascades, we include three different measures of the personal income
distribution. However neither the income share of the top 1% of households nor an income Gini-
coefficient or the Theil index are statistically significant in any of our specifications. Hence, we do not
find evidence for consumption cascades.
The main findings from the analysis of the consumption function can be summarized as follows: First,
the wage share has statistically significant positive effects on consumption expenditures. The size of that
effect is modest but robust across specifications. Second, household debt seems to be the most
important financial variable in explaining consumer behaviour. This result is also robust across
specifications. We fail to find evidence for different stock and flow effects of debt. Third, property prices
have small and often statistically insignificant effects. Housing wealth effects seem to be captured by the
debt measure since consuming housing wealth requires taking on additional mortgages. Fourth, share
prices have no statistically significant effect on consumption. Fifth, we fail to find evidence for an effect
of personal income inequality on aggregate consumption spending.
5.2 Investment function

The investment function takes the following form:
ln(𝐼 ) = 𝛽 ln(𝑌 ) + 𝛽 𝑖 + 𝛽 ln(𝑊𝑆 ) + 𝛽 ln(𝐷𝐻 ) + 𝛽 ln(𝐷𝐵 ) +

𝛽 ln(𝑃𝑃 ) + 𝛽 ln(𝑆𝑃 ) + 𝛽 ln(𝑄 ) + 𝜇 + 𝑣 (8)
again with country fixed effects 𝜇 and in addition to those variables already used in the consumption
function also a long-term real interest rate (𝑖) and non-financial corporate business debt (𝐷𝐵). Equation
(8) is estimated including only 16 instead of 18 countries since our data set does not contain information
on business debt for Switzerland and Ireland. The specification above is augmented by lags of the
dependent variable as well as lags of exogenous regressors (Table 3). In estimating equation 8 we start
with a baseline first difference approach (specification 1) and then add additional controls or use
different estimation techniques. Specification 2 uses a Gini coefficient instead of top income shares, in
specification 3 asset prices are dropped in favour of debt stock and flow effects. Specification 4 applies
the MG estimator and specifications 5 and 6 the A&H and the difference GMM estimator, again with
limited lag length for instruments. A&H does not pass the Hansen overidentification restriction test and
we thus do not consider it further. The GMM estimator passes the overidentification restriction test and
first order autocorrelated errors are reported as required in the case of a properly specified dynamic
first difference estimator. However the estimated coefficient on lagged investment is almost 1.1

implying not only a unit root problem but also an explosive dynamic. Overall, the results are reasonably
robust across specifications, however not as stable as in the case of the consumption function. Due to
the weak performance of the dynamic specifications, our preferred estimator is again the static FD
estimator (specification 1).
Results reported in Table 3 are as follows: Income has a very strong impact on investment spending with
an elasticity well above 1. This finding is robust across all specifications. The (long-run) elasticity with
respect to the wage share is about 0 in the first differences specifications and negative in the dynamic
ones. Long term real interest rates affect investment expenditures negatively in all specifications.
Property prices have a positive impact in all specifications ranging from 0.04 to 0.28, pointing to the
importance of property prices for residential investment spending. Household as well as business debt
have mostly negative effects. If the change in household and business debt is added (specification 3),
the estimated coefficients on debt changes are positive, and in the case of household debt statistically
significant, whereas the level effect for household debt turns positive. Thus we do not find evidence for
negative stock effects of either household or business debt in the investment function. In contrast to the
income share of the richest 1% of households, the Gini index turns out to be statistically significant with
a negative effect. We conclude that relative consumption, with respect to housing, does not feed
through to aggregate (residential) investment spending
Table 3. Investment function, dependent variable: log(It)

(1) (2) (3) (4) (5) (6)
FD FD FD MG A&H GMM
Log(I)t-1 0.803*** 1.083***

(0.05) (0.10)
Log(Y)t 1.910*** 1.896*** 2.169*** 1.826*** 2.155*** 2.154***
(0.12) (0.10) (0.11) (0.26) (0.16) (0.19)
Log(Y)t-1 -0.558*** -0.490*** -0.602*** -0.832*** -1.845*** -2.400***
(0.17) (0.14) (0.17) (0.26) (0.17) (0.29)
Log(WS)t 0.501*** 0.424*** 0.745*** 0.815*** 0.186 -0.126
(0.16) (0.14) (0.15) (0.21) (0.26) (0.21)
Log(WS)t-1 -0.482*** -0.422*** -0.430*** -1.093*** -0.533*** -0.647***
(0.11) (0.10) (0.12) (0.28) (0.12) (0.15)
it -0.513*** -0.432*** -0.674*** -0.914*** -0.315** -0.239
(0.13) (0.12) (0.13) (0.24) (0.13) (0.16)
Log(PP)t 0.276*** 0.264*** 0.221*** 0.106*** 0.037

(0.02) (0.03) (0.05) (0.03) (0.06)

Log(SP)t 0.007 0.006 0.045** -0.014 -0.020*
(0.01) (0.01) (0.02) (0.01) (0.01)
Log(DB)t -0.003 -0.001 -0.062 0.031 -0.065** -0.065*
(0.02) (0.02) (0.04) (0.06) (0.03) (0.03)
ΔLog(DB)t 0.045
(0.04)
Log(DH)t -0.105*** -0.106*** 0.005 -0.085 -0.063 0.011
(0.04) (0.04) (0.05) (0.11) (0.04) (0.09)
ΔLog(DH)t 0.214***
(0.07)
Log(TOP1)t -0.002 -0.006 -0.049 0.051*** 0.063***
(0.02) (0.02) (0.09) (0.01) (0.02)
Log(GINI)t -0.191***
(0.07)
cons -2.258*
(1.19)
N 407 403 404 423 395 387
2
cent. R 0.746 0.755 0.711 0.728
F-stat 337 252 804 1629 5927
p-value Sargan 0.705
p-value AR(1) 0.004
p-value AR(2) 0.440
* p<0.1, ** p<0.05, *** p<0.01, heteroskedastic and autocorrelation robust standard errors reported in
brackets. FD refers to first difference estimators, MG stands for Mean Group estimator, A&H is the
Anderson and Hsiao (1982) estimator and GMM refers to the difference GMM estimator. p-value
Sargan/Hansen refer to overidentification tests and p-value AR represent Arellano-Bond autocorrelation
tests.
decisions. Stock prices have small and statistically insignificant effects in the static specifications. A
considerable part of total capital formation relies on household spending decisions and therefor partially
explains zero or positive effects of the wage share on total investment as well as negative effects of
household debt.
The key findings regarding the investment function are the following: First, it is not straightforward to
find negative effects of the wage share on investment. In some specifications we obtain positive
elasticities which seem to be the result of positive wage share effects on capital formation in the

construction sector. Our preferred specification yields effectively zero long-run effects of the wage
share. Second, property prices influence investment spending positively, pointing to the positive effect
of property price booms on residential investment. Third, the negative effect of household debt
indicates that higher debt levels prevent households from residential investment spending. Fourth, the
personal distribution of income, measured by the income share of the richest 1% of households and the
income Gini coefficient, is either not statistically significant or exhibits a negative effects, not supporting
Veblen effects.
5.3 External sector

The external sector was modelled by estimating an export as well as an import equation. Both proved to
be robust with respect to different estimation procedures. For brevity Table 4 only reports the FD
estimator and the dynamic A&H estimator. The long-run elasticity of exports with respect to foreign
demand (𝑌 ) is about 2, independent of the specification. These export elasticities seem to be
influenced by a trend of globalization since income elasticities of exports well beyond 1 are not plausible
in the long run. As expected, the elasticity of exports with respect to the wage share is negative and lies
between -0.16 and -0.25. The elasticity with respect to the nominal effective exchange rate is negative
and thus is in line with expectations (i.e. appreciations of the domestic currency leading to lower
exports). Property prices have a negative impact on exports of about 0.1 which can be interpreted as
their contribution to domestic inflation and thus increasing export costs.
The income elasticity of imports is about 1.3. The effect of the wage share is not statistically significant
and about 0. Exchange rates have a statistically significant positive effect, as expected. Property prices
also statistically significantly affect imports in a positive way. Rising property prices might drive up the
domestic price level and thus encourage imports, ceteris paribus. The export elasticity of imports is
about 0.35. The import equation includes exports to reflect the dependence of exports on imported raw
materials and intermediary goods. Results are similar if exports are excluded.
Table 4. Foreign sector, dependent variables: log(Xt) and log(Mt)

(1) (2) (3) (4)
X: FD X: A&H M: FD M: A&H
lagged dep. var. 0.249 -0.170
(0.25) (0.21)
Log(Yf)t 2.450*** 2.585***
(0.18) (0.27)

Log(Yf)t-1 -0.493*** -1.098*

(0.18) (0.65)
Log(Y)t 1.306*** 1.419***
(0.19) (0.31)
Log(WS)t 0.09 0.176
(0.09) (0.13)
Log(WS)t-1 -0.247** -0.161 -0.087 -0.124
(0.11) (0.10) (0.06) (0.08)
Log(EX)t -0.185*** -0.196*** 0.159** 0.142*
(0.02) (0.03) (0.06) (0.08)
Log(EX)t-1 -0.115*** -0.053
(0.04) (0.07)
Log(PP)t -0.104*** -0.089*** 0.186*** 0.152**
(0.03) (0.03) (0.04) (0.06)
Log(PP)t-1 -0.098***
(0.03)
Log(X)t 0.443*** 0.421***
(0.05) (0.08)
Log(X)t-1 -0.091** 0.016
(0.04) (0.15)
N 481 468 496 483
cent. R2 0.728 0.547 0.806 0.791
F-stat 198 373 265 328
* p<0.1, ** p<0.05, *** p<0.01, heteroskedastic and autocorrelation
robust standard errors reported in brackets.
5.4 Demand regimes and the drivers of growth

Table 5 summarises the key results regarding demand regimes. It is based on the preferred FD estimates
(specification 1 in Table 3, 4 and 5). Elasticities have been transformed into marginal effects and
normalized by income to make them comparable.12 Row 4 adds the effect of a 1%-point change in the
12
The marginal effect of the wage share reported in Table 5 is computed in the following way:
= =𝛽 , ∅ ∅
+𝛽, ∅ ∅
+𝛽 , ∅ ∅
+𝛽 , ∅ ∅

wage share on private excess demand, 𝑌 , which is numerator of equation (5), 𝑓 , and determines the
sign of the effect of changes in distribution on equilibrium demand. It can be thought of as the first
round effect or the sum of the partial effect, given a certain level of income. The second round effects
include the indirect effect as the first round effects increase income and thus induce additional
expenditures.
Table 5. Marginal effect of 1 percentage point shift of WS in % of GDP on private excess demand
PANEL US FR DE AT NL
C 0.12 0.14 0.11 0.12 0.11 0.10
I 0.01 0.01 0.01 0.01 0.01 0.01
NX -0.05 -0.02 -0.06 -0.08 -0.10 -0.15
YPED 0.08 0.12 0.06 0.04 0.02 -0.05
openness 21% 11% 24% 33% 42% 61%
Effects are based on coefficients from specification (1) in Table 1 and Table 2 and specifications (1) and (3) in Table
3. Elasticities are converted into marginal effects using GDP weighted sample averages. Openness is computed as
the average of nominal import and export shares: (𝑃 𝑀 + 𝑃 𝑋)/2𝑃 𝑌.
There are several interesting patterns. First, the domestic effects of the wage share on consumption and
investment are similar across countries. Demand regimes in all countries are domestically wage led.
Second, there is a substantial difference of the net export effects that directly corresponds to the degree
of openness, i.e. exports plus import relative to GDP13. A large and relatively closed economy like the
USA has a small net export effect and is overall strongly wage led. Medium sized open economies like
France, or Germany have substantially smaller effects. In small open economies like Austria or the
Netherlands the negative external sector effects become so large that the total demand regime can
become profit led.
Finally we want turn to the question which variables have been the main drivers of growth in the decade
prior to the 2007 crisis. Table 6 reports to what extent the change of explanatory variables, explain the
change in consumption and investment spending in the period 1997-2007. In addition to the full panel
𝛽_, is the estimated elasticity of consumption, investment spending, exports or imports with respect to the wage
share. ∅ represents the average over the sample period of the GDP weighted average of the consumption-to-
income ratio of the 18 countries included in the panel and similarly ∅𝑊𝑆 is the sample period average of the GDP
weighted average of the wage share. So first GDP weighted averages of 𝐶/𝑌, 𝐼/𝑌, 𝑋/𝑌 and 𝑀/𝑌 over the panel
countries are computed for each year. In a second step GDP weighted averages (based on PPPs) of these yearly
averages is computed.
13
Openness is computed as the average of nominal exports and imports to nominal GDP. We evaluate export and
import shares at sample average and assume that current account is in balance.

we also take a look at four country groups: Anglo-Saxon (Australia, Canada, the United Kingdom and the
United States), Euro-North (Austria, Belgium, Germany, Finland, and the Netherlands), Euro-South
(Spain, Italy and Ireland) and non-Euro-North (Denmark, Switzerland, Norway and Sweden). These
country groupings are motivated by the hypothesis that distinct growth models have emerged in the
form of a debt-driven model in the Anglo-Saxon countries and the southern European countries and an
export-driven model in the Nordic countries (Stockhammer 2009; Hein 2012b). This distinction should
lead to distinct pattern of demand formation.
Table 6. Growth contributions

change 1997-2007 Panel Anglo €-North Non-€-North €-South
ΔC 31% 42% 13% 29% 29%
ΔI 34% 46% 18% 48% 52%
ΔY 28% 35% 22% 28% 29%
(1) ΔC-βYΔY 10.7% 17.4% -2.4% 8.8% 8.8%
(2) βWSΔWS -0.5% -0.2% -0.8% -0.4% -0.6%
C
(3) βAΔA 9.4% 12.8% 3.6% 10.2% 20.4%
(4) βQΔQ 0.2% 0.2% 0.1% 0.1% 0.1%
(5) ΔI-βYΔY-βiΔi -4.9% -2.5% -12.5% 8.5% 11.8%
(6) βWSΔWS -0.1% 0.0% -0.1% 0.0% -0.1%
I
(7) βAΔA 6.3% 11.9% 0.3% 11.4% -2.2%
(8) βQΔQ 0.0% 0.0% 0.0% 0.0% 0.0%
(9) βYfΔYf-βYΔY -0.4% -9.2% 8.1% -0.7% -1.2%
NX (10) βWSΔWS 0.5% 0.2% 1.0% 0.4% 0.7%
(11) βPPΔPP -8.1% -12.8% -2.0% -11.1% -9.1%
(12) βWSΔWS -0.17% -0.08% -0.07% -0.02% -0.16%
Y-SR (13) βAΔA 5.32% 8.72% 1.25% 2.78% 8.64%
(14) βQΔQ 0.10% 0.11% 0.06% 0.04% 0.06%
(15) βWSΔWS 0.14% 0.24% 0.02% 0.01% 0.10%
Y-LR (16) βAΔA 9.42% 18.84% 1.39% 3.03% 13.21%
(17) βQΔQ 0.17% 0.24% 0.06% 0.05% 0.10%
Rows (1) and (5) report the change of 𝐶 and 𝐼 between 1997 and 2007 not explained by changes in income (and
the interest rate), respectively. Rows (2), (6) and (10) report the predicted change in 𝐶, 𝐼 and 𝑁𝑋, based on the
change in 𝑊𝑆 and the estimates from specifications (1) in Tables 2, 3 and 4. Rows (3) and (7) report the predicted
change in 𝐶 and 𝐼 based on the changes and the corresponding coefficients of property and stock prices as well as
household debt: 𝛽 𝛥𝐴 = 𝛽 𝛥𝑃𝑃 + 𝛽 𝛥𝑆𝑃 + 𝛽 𝛥𝐷𝐻. Rows (4) and (8) report the predicted change in 𝐶
and 𝐼 based on the change in Top 1% income share changes. Row (9) contain the predicted change in 𝑁𝑋 due to
the changes in 𝑌 and 𝑌 . Row (11) reports similar results but due to changes in 𝑃𝑃. Rows (12) to (14) report short

run total GDP effects based on 𝑓 of equation (5) and rows (15) to (17) report the long run equivalents taking also
into account the multiplier effect of 𝑓 from equation (5).
Consumption and investment grew much slower in the Euro-North group than in the other groups. Rows
1 and 5 calculate the growth in consumption and investment that is not explained by income growth.
While consumption expanded more slowly than income in the Euro-North group, there is a substantial
unexplained gap in consumer spending by income for the other groups (17.4%, 8.8% and 8.8% of GDP
for the Anglo-Saxon, Euro-South and non-Euro-North groups respectively). The unexplained gap
spending for investment (row 5) amounts to 11.8% and 8.5% of GDP in case of Euro-South and non-
Euro-North.14 Changes in the wage share hardly explain any of these dynamics. The contributions are
less than 1% for consumption (row 2) and effectively 0 for investment (row 6). The same holds true for
the income share of the top 1% (rows 4 and 8). In contrast changes in property prices, stock prices and
debt (row 3) explains a rise in consumer spending of 12.8%, 10.2% and 20.4% for the Anglo-Saxon, non-
Euro-North and Euro-South groups. In the case of investment (row 7) these asset variables explain 11.9%
and 11.4% for Anglo-Saxon and non-Euro-North while they did not affect or even diminished investment
spending in the other two groups. Rows 9, 10 and 11 calculate the contributions of the differential
between domestic and foreign demand, the wage share and property prices on net export. Rows 12-14
combine the consumption, investment and net export effects to compute a short-run private excess
demand effect15 that can be attributed to asset dynamics (row 13), the wage share (row 12) and
personal income inequality (row 13). Rows 15-17, (based on equation 5)16, do the same for equilibrium
income, which takes into account the multiplier mechanism. Asset effects contributed almost 20% to
GDP growth in the Anglo-Saxon economies and 13% in Euro-South, but only 1.4% in Euro-North and 3%
in non-Euro-North. The differences in the underlying multipliers are driven by the varying degrees of
openness to trade. Anglo-Saxon and Euro-South which are more closed in comparison to the two North
groups have larger multipliers.
The results summarised in Table 6 illustrate that the panel results that pool parameters can explain very
different country group performances. These are due to varying degrees of openness and different asset
14
Row 5 calculates the part of investment growth that is not explained changes in income and the interest rate.
15
Y-SR is constructed by summing up the consumption, investment and net export effects weighted by
consumption, investment and net exports relative to GDP.
16
The multipliers are 1.77 for the overall panel and 2.16, 1.11, 1.09 and 1.53 for Anglo-Saxon, Euro-North, non-
Euro-North and Euro-South, respectively. The long run effects (Y-LR) then result from the short run (Y-SR) effects
times the corresponding multipliers.

price and debt dynamics. The direct effects of distributional shifts were negligible. Property prices and
household debt played the dominant role in explaining growth prior to the crisis.
6 Conclusions
The paper has investigated the role of functional and personal income distribution as well as the role of
wealth and debt in consumption and investment. The basis for this was an extended Bhaduri and
Marglin (1990) model. The econometric analysis was based on a sample of 18 OECD countries for the
period 1980-2013. We have four major findings. First, we do find statistically significant effects of the
functional income distribution on consumption and investment. These effects are modest in size, but
qualitatively, we find wage-led domestic demand.
Second, we fail to find effects for personal income distribution, measured by top incomes shares, Gini
coefficients or a Theil index of sectoral wage dispersion. It is possible that negative effects of inequality
due to lower consumption propensities of the rich are offset by positive imitation effects stressed by the
Veblen tradition. Alternatively it might also be the case that consumers do care about consumption of
their peers but rising property prices provided the necessary collateral to take on debt for financing
these expenditure cascades.
Third, we find statistically significant and robust positive effects of household debt on consumption. This
is at odds with the standard view of the role of wealth, which would expect a negative partial effect of
household debt. We fail to find different effects for debt as a stock and as flow variable. We do find
negative effects of household debt on investment (which includes residential investment). Real property
prices have strong positive and statistically significant effects in the investment function, whereas they
only play a limited role for consumption.
Forth, to analyse economic significance we have calculated the contributions of key variables to
consumption and investment growth in the decade prior to the crisis (1998-2007). This indicated that
functional and personal income distribution have negligible effects, whereas property prices and
household debt have had strong positive contributions. This is in line with the hypothesis of an asset
price-driven (or debt-driven) growth model in explaining growth prior to the 2007 crisis.

References
Alvarez-Cuadrado, F. et al., 2012. Envy and habits: Panel data estimates of interdependent preferences,
Bank of Spain Working Paper 1213.
Anderson, T. & Hsiao, C., 1982. Formulation and estimation of dynamic models using panel data. Journal
of Econometrics, 18(1), pp.47–82.
Arellano, M. & Bond, S., 1991. Some Tests of Specification for Panel Data: Monte Carlo Evidence and an
Application to Employment Equations. Review of Economic Studies, 58(2), pp.277–297.
Aron, J. et al., 2012. Credit, Housing Collateral, and Consumption: Evidence From Japan, the U.K., and
the U.S. Review of Income and Wealth, 58(3), pp.397–423.
Atkinson, A.B., Piketty, T. & Saez, E., 2011. Top Incomes in the Long Run of History. Journal of Economic
Literature, 49(1), pp.3–71.
Attanasio, O.P. & Weber, G., 2010. Consumption and Saving: Models of Intertemporal Allocation and
Their Implications for Public Policy. Journal of Economic Literature, 48(3), pp.693–751.
Baltagi, B., 2013. Econometric Analysis of Panel Data 5th ed., Wiley.
Barbosa-Filho, N.H. & Taylor, L., 2006. Distributive and demand cycles in the US economy—a
structuralist Goodwin model. Metroeconomica, 57(3), pp.389–411.
Behringer, J. & Treeck, T. van, 2013. Income distribution and current account: A sectoral perspective,
Macroeconomic Policy Institute Working Paper 125.
Belabed, C.A., Theobald, T. & van Treeck, T., 2013. Income distribution and current account imbalances,
Macroeconomic Policy Institute Working Paper 126.
Bhaduri, A. & Marglin, S., 1990. Unemployment and the real wage: the economic basis for contesting
political ideologies. Cambridge Journal of Economics, 14(4), pp.375–393.
Blundell, R. & Bond, S., 1998. Initial conditions and moment restrictions in dynamic panel data models.
Journal of Econometrics, 87(1), pp.115–143.
Bowles, S. & Boyer, R., 1995. Wages, aggregate demand, and employment in an open economy: an
empirical investigation. In G. Epstein & H. Gintis, eds. Macroeconomic policy after the conservative
era: Studies in investment, saving and finance.
Bowles, S. & Park, Y., 2005. Emulation, Inequality, and Work Hours: Was Thorsten Veblen Right?*. The
Economic Journal, 115(1996), pp.397–412.

Brown, C., 2004. Does income distribution matter for effective demand? Evidence from the United
States. Review of Political Economy, 16(3), pp.291–307.
Buiter, W.H., 2010. Housing Wealth Isn’t Wealth. Economics: The Open-Access, Open-Assessment E-
Journal, 4(2010-22).
Carvalho, L. & Rezai, A., 2014. Personal Income Inequality and Aggregate Demand, FEA-USP Working
Paper 2014-23.
Charles, S., 2008. Teaching Minsky’s financial instability hypothesis: a manageable suggestion. Journal of
Post Keynesian Economics, 31(1), pp.125–138.
Choi, I., 2001. Unit root tests for panel data. Journal of International Money and Finance, 20(2), pp.249–
272.
Cooper, D. & Dynan, K., 2014. Wealth Effects and Macroeconomic Dynamics. Journal of Economic
Surveys, Advance Access.
Davidson, P., 1972. Money and the real world. Economic Journal, 82(325), pp.101–115.
Duesenberry, J.S., 1949. Income, Saving and the Theory of Consumer Behavior, Harvard University Press.
Dutt, A.K., 2006. Maturity, Stagnation and Consumer Debt: a Steindlian Approach. Metroeconomica,
57(3), pp.339–364.
Fazzari, S., Ferri, P. & Greenberg, E., 2008. Cash flow, investment, and Keynes-Minsky cycles. Journal of
Economic Behavior and Organization, 65(3), pp.555–572.
Frank, R., 1985. The demand for unobservable and other nonpositional goods. The American Economic
Review, 75(1), pp.101–116.
Frank, R.H., Levine, A.S. & Dijk, O., 2010. Expenditure Cascades, available at SSRN:
http://ssrn.com/abstract=1690612.
Girouard, N., Kennedy, M. & André, C., 2006. Has the rise in debt made households more vulnerable?
OECD Working Paper 2006-63.
Godley, W. & Lavoie, M., 2007. Monetary Economics, New York: Palgrave Macmillan.
Goodwin, R.M., 1967. A Growth Cycle. In C. H. Feinstein, ed. Socialism, Capitalism and Economic Growth.
Cambridge: Cambridge University Press.
Hartwig, J., 2014. Testing the Bhaduri–Marglin model with OECD panel data. International Review of
Applied Economics, 28(4), pp.419–435.

Hein, E., 2012a. Finance-dominated Capitalism, Re-distribution, Household Debt And Financial Fragility
In A Kaleckian Distribution And Growth Model. PSL Quarterly Review, 65(260), pp.11–51.
Hein, E., 2012b. The Macroeconomics of Finance-Dominated Capitalism - And Its Crisis, Cheltenham:
Edward Elgar.
Hein, E. & Vogel, L., 2008. Distribution and growth reconsidered: empirical results for six OECD
countries. Cambridge Journal of Economics, 32(3), pp.479–511.
Kalecki, M., 1954. Theory of Economic Dynamics Collected . J. Osiatynski, ed., Oxford: Clarendon Press.
Kapeller, J. & Schütz, B., 2014. Debt, boom, bust: a theory of Minsky-Veblen cycles. Journal of Post
Keynesian Economics, 36(4), pp.781–814.
Keen, S., 1995. Finance and economic breakdown: modeling Minsky’s “financial instability hypothesis.”
Journal of Post Keynesian Economics, 17(4), pp.607–635.
Keynes, J.M., 1973. The General Theory of Employment, Interest and Money. The collected writings of
John Maynard Keynes volume VII, Cambridge: Macmillan.
Kiefer, D. & Rada, C., 2014. Profit maximising goes global: the race to the bottom. Cambridge Journal of
Economics, Advance Access.
Kim, Y.K., Setterfield, M. & Mei, Y., 2015. Aggregate consumption and debt accumulation: an empirical
examination of US household behaviour. Cambridge Journal of Economics, Advance Access.
Ludwig, A. & Sløk, T., 2004. The relationship between stock prices, house prices and consumption in
OECD countries, available at:
http://www.mea.mpisoc.mpg.de/uploads/user_mea_discussionpapers/umgw36cg2qs6mo20_dp4
4.pdf.
McBride, M., 2001. Relative-income effects on subjective well-being in the cross-section. Journal of
Economic Behavior & Organization, 45(3), pp.251–278.
Minsky, H.P., 1995. The Financial Instability Hypothesis: A Restatement. In P. Arestis & T. Skouras, eds.
Post Keynesian Economic Theory. Sussex: Wheatsheaf Books, pp. 24–55.
Muellbauer, J.N., 2007. Housing , Credit and Consumer Expenditure. In Housing, Housing Finance,
Monetary Policy. Jackson Hole, pp. 267–334.
Naastepad, C. & Storm, S., 2007. OECD demand regimes (1960-2000). Journal of Post Keynesian
Economics, 29(2), pp.211–246.
Neumark, D. & Postlewaite, A., 1998. Relative income concerns and the rise in married women’s
employment. Journal of Public Economics, 70(1), pp.157–183.

Nickell, S., 1981. Biases in Dynamic Models with Fixed Effects. Econometrica, 49(6), pp.1417–1426.
Nishi, H., 2012. Structural VAR analysis of debt, capital accumulation, and income distribution in the
Japanese economy: a Post Keynesian perspective. Journal of Post Keynesian Economics, 34(4),
pp.685–712.
Onaran, Ö. & Galanis, G., 2012. Is aggregate demand wage-led or profit-led? National and global effects,
ILO Conditions of Work and Employment Series No. 40.
Onaran, Ö., Stockhammer, E. & Grafl, L., 2011. Financialisation, income distribution and aggregate
demand in the USA. Cambridge Journal of Economics, 35(4), pp.637–661.
Paiella, M., 2009. The Stock Market, Housing and Consumer Spending: a Survey of the Evidence on
Wealth Effects. Journal of Economic Surveys, 23(5), pp.947–973.
Pesaran, H.M. & Smith, R., 1995. Estimating long-run relationships from dynamic heterogeneous panels.
Journal of Econometrics, 68(1), pp.79–113.
Ryoo, S., 2013. Minsky cycles in Keynesian models of growth and distribution. Review of Keynesian
Economics, 1(1), pp.37–60.
Slacalek, J., 2009. What drives personal consumption? The role of housing and financial wealth. The BE
Journal of Macroeconomics, 9(1), pp.1–35.
Stockhammer, E., 2012. Financialization, income distribution and the crisis. Investigación Económica,
71(279), pp.39–70.
Stockhammer, E., 2009. The finance-dominated accumulation regime, income distribution and the
present crisis. Papeles de Europa, 19, pp.58–81.
Stockhammer, E., Hein, E. & Grafl, L., 2011. Globalization and the effects of changes in functional income
distribution on aggregate demand in Germany. International Review of Applied Economics, 25(1),
pp.1–23.
Stockhammer, E. & Onaran, Ö., 2004. Accumulation, distribution and employment: a structural VAR
approach to a Kaleckian macro model. Structural Change and Economic Dynamics, 15(4), pp.421–
447.
Stockhammer, E., Onaran, O. & Ederer, S., 2008. Functional income distribution and aggregate demand
in the Euro area. Cambridge Journal of Economics, 33(1), pp.139–159.
Stockhammer, E. & Stehrer, R., 2011. Goodwin or Kalecki in Demand? Functional Income Distribution
and Aggregate Demand in the Short Run. Review of Radical Political Economics, 43(4), pp.506–522.
Zezza, G., 2009. Fiscal policy and the economics of financial balances. European Journal of Economics
and Economic Policies, 6(2), pp.289–310.

Zhang, L. & Bezemer, D., 2014. How the Credit Cycle Affects Growth: The Role of Bank Balance Sheets,
University of Groningen Faculty of Economics and Business Working Paper 14026-GEM.

Appendix
Table A1. Data definitions and sources
abbreviation full variable name unit source
Adjusted wage share: total economy: as
WS percentage of GDP at current factor cost
(ALCD2) % GDP AMECO
Y Gross domestic product at 2005 market Billion, national
prices (OVGD) currency AMECO
PY Price deflator gross domestic product at
market prices (PVGD) 2005=100 AMECO
C Private final consumption expenditure at Billion, national
2005 prices (OCPH) currency AMECO
PC Price deflator private final consumption
expenditure (PCPH) 2005=100 AMECO
I Gross fixed capital formation at 2005 prices: Billion, national
total economy (OIGT) currency AMECO
PI Price deflator gross fixed capital formation:
total economy (PIGT) 2005=100 AMECO
PM Price deflator imports of goods and services
(PMGS) 2005=100 AMECO
M Imports of goods and services at 2005 prices Billion, national
(OMGS) currency AMECO
X Exports of goods and services at 2005 prices Billion, national
(OXGS) currency AMECO
PX Price deflator exports of goods and services
(PXGS) 2005=100 AMECO
i AMECO and OECD
Real long-term interest rates, deflator GDP % (MEI)
DH Household and NPISH, all liabilities % GDP BIS
DB Non-financial corporate, all liabilities less
shares and other equity % GDP BIS
PP real property prices BIS (exact definitions
vary, deflated with CPI) 2005=100 BIS and OECD
SP IMF (International
Financial Statistics)
share price index; CPI deflated 2005=100 and OECD (MEI)
THEIL University of Texas
estimated household income inequality Theil Index Inequality Project
f
Y OECD real GDP 2005=100 OECD
IC Gross fixed capital formation at 2005 prices: Billion, national
construction (OIGCO) currency AMECO
GINI gini coefficient (pre tax and post transfer) of
the Standardized World Income Inequality SWIID

Database
TOP1 top 1% income share of the SWIID SWIID
EX Nominal effective exchange rate 2005=100 BIS

Table A2. Unit root tests following Choi (2001)

variable inv. chi-squared inverse logit inverse normal specification transformation
C 0.511 0.756 0.768 trend log-level
Y 0.989 1.000 1.000 trend log-level
PP 0.156 0.408 0.455 trend log-level
SP 0.043 0.108 0.094 trend log-level
DHH 0.217 0.411 0.392 trend log-level
TOP1 0.237 0.467 0.498 trend log-level
DBUS 0.362 0.303 0.272 trend log-level
inv 0.152 0.472 0.578 trend log-level
X 0.991 1.000 1.000 trend log-level
M 0.784 0.989 0.983 trend log-level
WSfc 0.235 0.456 0.494 no trend log-level
i 0.723 0.559 0.560 no trend level
exnom 0.063 0.027 0.022 no trend log-level
C 0.010 0.005 0.003 no trend log-dif
Y (2 lags) 0.019 0.034 0.055 no trend log-dif
PP 0.000 0.000 0.000 no trend log-dif
SP 0.000 0.000 0.000 no trend log-dif
DHH 0.062 0.026 0.021 no trend log-dif
Top1 0.000 0.000 0.000 no trend log-dif
DBUS 0.001 0.000 0.000 no trend log-dif
inv 0.001 0.001 0.000 no trend log-dif
X (2 lags) 0.071 0.054 0.047 no trend log-dif
M 0.000 0.000 0.000 no trend log-dif
WSfc 0.000 0.000 0.000 no trend log-dif
i 0.000 0.000 0.000 no trend dif
exnoml 0.000 0.000 0.000 no trend log-dif

Debt-Driven Growth? Wealth, Distribution and Demand in OECD Countries

Uploaded by

Copyright:

Available Formats

Debt-Driven Growth? Wealth, Distribution and Demand in OECD Countries

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Debt-Driven Growth? Wealth, Distribution and Demand in OECD Countries

Uploaded by

Copyright:

Available Formats

Debt-driven Growth?

Wealth, Distribution and Demand in

Keywords: Post-Keynesian economics, wage-led growth, Bhaduri-Marglin model, demand

JEL classifications: E11, E12, E21

Figure 1. Macroeconomic trends 1980-2010

March 2015 Page | 1 Stockhammer and Wildauer

March 2015 Page | 2 Stockhammer and Wildauer

March 2015 Page | 3 Stockhammer and Wildauer

2 The theoretical framework

𝐶 = 𝐶(𝑌, 𝑊𝑆, 𝑄, 𝑊𝐻, 𝑊𝐹, 𝐷𝐻, ∆𝐷𝐻), (2)

With CY, CWS, CWH, CWF, C∆DH >0, CDH<0, CQ=?

March 2015 Page | 4 Stockhammer and Wildauer

decrease disposable income and thus consumption. On the other hand ∆

Table 1. Different consumption theories

March 2015 Page | 5 Stockhammer and Wildauer

effects for home owners and higher

to affect consumption, this implies ( )

to be accepted as collateral than financial wealth.

𝐼 = 𝐼(𝑌, 𝑊𝑆, 𝑖, 𝑄, 𝑊𝐻, 𝑊𝐹, 𝐷𝐻, 𝐷𝐵) (3)

March 2015 Page | 6 Stockhammer and Wildauer

Net exports are

𝑁𝑋 = 𝑁𝑋 𝑌, 𝑌 , 𝐸𝑋, 𝑊𝑆, 𝑊𝐻 (4)

where 𝑓 = ( + + ) and 𝑓 = ( + + ). The short run effect is determined by 𝑓

March 2015 Page | 7 Stockhammer and Wildauer

coefficient of 𝑊𝐻 on 𝐶, which is an estimate for the partial effect .

3 The related empirical literature

March 2015 Page | 8 Stockhammer and Wildauer

March 2015 Page | 9 Stockhammer and Wildauer

March 2015 Page | 10 Stockhammer and Wildauer

4 Data and econometric method

March 2015 Page | 11 Stockhammer and Wildauer

ln(𝐶 ) = 𝛽 ln(𝑌 ) + 𝛽 ln(𝑊𝑆 ) + 𝛽 ln(𝐷𝐻 ) + 𝛽 ln(𝑃𝑃 ) +

March 2015 Page | 12 Stockhammer and Wildauer

Table 2. Consumption function, dependent variable: log(Ct)

March 2015 Page | 13 Stockhammer and Wildauer

p-value Hansen 0.108 0.041

March 2015 Page | 14 Stockhammer and Wildauer

5.2 Investment function

ln(𝐼 ) = 𝛽 ln(𝑌 ) + 𝛽 𝑖 + 𝛽 ln(𝑊𝑆 ) + 𝛽 ln(𝐷𝐻 ) + 𝛽 ln(𝐷𝐵 ) +

March 2015 Page | 15 Stockhammer and Wildauer

Table 3. Investment function, dependent variable: log(It)

Log(I)t-1 0.803*** 1.083***

March 2015 Page | 16 Stockhammer and Wildauer

(0.02) (0.03) (0.05) (0.03) (0.06)

March 2015 Page | 17 Stockhammer and Wildauer

5.3 External sector

Table 4. Foreign sector, dependent variables: log(Xt) and log(Mt)

March 2015 Page | 18 Stockhammer and Wildauer

Log(Yf)t-1 -0.493*** -1.098*

5.4 Demand regimes and the drivers of growth

March 2015 Page | 19 Stockhammer and Wildauer

March 2015 Page | 20 Stockhammer and Wildauer

Table 6. Growth contributions

March 2015 Page | 21 Stockhammer and Wildauer

March 2015 Page | 22 Stockhammer and Wildauer

March 2015 Page | 23 Stockhammer and Wildauer

March 2015 Page | 24 Stockhammer and Wildauer

March 2015 Page | 25 Stockhammer and Wildauer

Log(I)t-1 0.803* 1.083*