The macroeconomic role of consumer credit : a comparison between the role of consumer credit in the British debt-led and the Dutch export-led economy

(1)

The macroeconomic role of consumer credit

A comparison between the role of consumer credit in the British debt-led and the Dutch

export-led economy

Tom van de Haar

10379029

23-06-17

(2)

2

1: Introduction

The credit card is, with 68% of the US households having one or more, firmly embedded in the economic culture of the US (Ackerman, Fries & Windle, 2012: 67). In the last few decades the credit card has not only become a legitimate tool of finance for US households, but also became the symbol of an American lifestyle associated with consumerism and private indebtedness (Trumbull, 2012: 10; Ritzer, 1995: 11). Credit cards are, as an instrument that combines revolving credit and electronic payment, the embodiment of ongoing technological innovations in the credit industry looking for ways to reduce the costs associated with offering credit (Durkin, Elliehausen & Zywicki, 2015: 283) and to improve the convenience for its users.

The credit card is one of the many types of consumer credit available to individuals and households. From the last part of the 20th century onwards the access to consumer credit as well as the demand for consumer credit grew rapidly (Crouch, 2009: 390; Jentzsch & Riestra, 2006:28). Different types of consumer credit became more easily available, which in turn led to the diffusion and distribution of consumer credit amongst all segments of the population, but especially the low and middle-income households (Vandone, 2009: 2; Crouch, 2009: 390). The rise in the prevalence of credit cards and other forms of consumer credit is closely connected to the increase in the amount of consumer debt. Households increasingly rely on credit to fuel their consumption. The US consumer debt as a

percentage of disposable of personal income increased from 17.8% in 1980 to 25.1% in 2006 (Barba & Pivetti, 2009: 115). Consumer debt to GDP increased as well in the US; from 12.9% in 1992 to 16.9% in 2001 (Dutt, 2006: 341-342).

The rise in consumer credit is not solely a phenomenon that occurred in the US. The demand for consumer credit is expanding throughout Europe as well (Vandone, 2009: 1). The outstanding amount of consumer credit has been rising in Europe since the early 1980s. However, the growth levels of consumer credit, and the absolute- and relative amount of consumer indebtedness vary widely across European countries (Guardia, 2000: 6-12). The growth rate of consumer credit is especially high in the countries that recently joined the European Union. Between 2000 and 2006 the average annual growth rate of consumer credit in Lithuania was 424%, in Estonia 285%, and in Latvia 165% (Vandone, 2009: 29).

The literature offers different explanations as to why the use of consumer credit has increased in the last few decades. Although a clear consensus does not seem to exist, the explanations put forward are not all necessarily mutually exclusive. These arguments range from financial liberalization and a

(5)

5

decrease in interest rates to government policies and wage stagnation and explain respectively the supply-side and the demand-side dynamics of consumer credit growth.

The literature seems also ambivalent about the consequences of consumer credit on the performance of an economy. Consumer credit is often identified as an important, or even main, driver of economic growth in the US and other advanced countries in the last few decades (See for example Cohen, 2007; Crouch, 2009; or Krippner, 2011). However, a group of, mostly

post-Keynesian, scholars theorize that a growth in the use of consumer credit could, in the long term, be negatively related to future household demand, and thus economic growth.

Bertola, Disney & Grant (2006: 2) note that when the debt levels of households are discussed it is often in a tone of moral disapproval. However, as they correctly argue, borrowing can, depending on the situation, be the sensible thing to do from an economic point of view, and is not necessarily less rational than saving. Nevertheless, high levels of household debt can have devastating effects on an economy, as the financial crisis of 2007-2008 has shown. The dependence of an economy on high levels of consumption financed with expensive types of consumer credit could turn out to be a recipe for trouble in the long run.

What role does consumer credit play exactly in Western developed economies? Since high levels of indebtedness can have devastating effects, it is important to gain more insight into the role of consumer credit in the economy. Furthermore, clear and broad research into the role of consumer credit outside of the US is lacking. The literature offers different reasons as to why consumer credit has grown in recent years, and what the effect of the run-up in consumer credit is for an economy. However, these explanations have the tendency to make very broad claims based mainly on the experience of the US. This thesis focusses therefore on the macroeconomic role of consumer credit, and the consequence thereof, in different types of Western developed economies. The thesis consists of two complementary parts. The first part focusses on explanations as to why consumer credit has risen in the last few decades; while the second part determines what the impact of consumer credit is on economic growth.

The effect consumer credit has on an economy depends, according to post-Keynesian scholars, on the growth regime a country has adopted. It is argued that two extreme types of growth regimes have emerged in the era of neoliberalism: the debt-led growth regime and the export-led growth regime. It is important to make a distinction between the different growth regimes, because the kind of growth regime a country has adopted influences not only the effect of consumer credit, but also the related policy decisions. These policy decisions could in turn have an effect on the factors that influence the growth of consumer credit. This thesis focusses therefore on the role of consumer

(6)

6

credit in a debt-led and an export-led economy, respectively the United Kingdom and the Netherlands.

For both countries a multiple linear regression model was estimated to examine different variables that the literature has identified as contributing factors to the growth of consumer credit. The estimations of the linear regression model for the United Kingdom indicate that both real wages and the level of unemployment are significant predictors of consumer credit growth; while the results of the Netherlands indicate that the credit-supply side factors may be less important.

To determine the short-run and long-run impact of consumer credit on economic growth, this thesis used multiple equation regression models. The estimations of the vector autoregression model for the United Kingdom indicate that a tridirectional positive feedback relationship exists between GDP, household consumption, and consumer credit. The estimations of the vector error correction model for the Netherlands do not indicate that consumer credit has a positive effect on the Dutch economy in the short run, but do provide evidence for a negative long-term relationship between consumer credit and GDP. The results of the second part of the analysis confirm the idea of a two-sided character of consumer credit. Consumer credit may provide economic stimulus in short run, but could, in the long-term have a negative impact on economic growth. However, the precise impact of consumer credit differs per economy. The results suggest that consumer credit affects the British debt-led and the Dutch export-led economy in the short run differently.

The thesis is structured as follows. The second chapter of this thesis examines the debates in the literature and develops a framework of analysis that shapes the empirical arguments of the thesis. The third chapter presents the research objective, the research questions, and the hypotheses. The fourth chapter presents the methodology, the case selection, and an overview of the data used in this thesis. Chapter five presents the data preparation, the analysis, and the results. Lastly, chapter six offers a conclusion.

(7)

7

2: Review of the literature

2.1: Introduction

In the last few decades the demand for consumer credit grew significantly in OECD-countries. The total outstanding amount of consumer credit not only increased in absolute terms, but also relative to GDP. Furthermore, between 2000 and 2006 the consumer credit to disposable income ratio and the consumer credit to household consumer expenditure ratio grew in most European countries. However, broad differences exist between EU-countries. Consumer credit as a percentage of disposable income was in 2006 over 25% in the UK, while it was around 10% in the Netherlands (Vandone, 2009: 27-28).

The increase in the use of consumer credit and rising levels of household debt in general are social phenomena that have grasped the attention of many scholars in the last few decades. Many scholars have begun to offer important and interesting theoretical and empirical evidence and insights with regard to the role of consumer credit in the economy, and the relationship between economic growth and consumer credit. This has led to the emergence of a range of theories as to why people increasingly resort to credit. Although the use of consumer credit clearly offers economic advantages (see Durkin, Elliehausen, Staten & Zywicki, 2014), the corresponding levels of indebtedness vary widely among the economically diverse population.

Economic analysis related to consumption, saving, or credit, often use the life-cycle hypothesis and the similar permanent-income hypothesis as a framework to interpret these differences (Vandone, 2009: 21). The life-cycle hypothesis and the permanent-income hypothesis argue that households plan their consumption and savings behaviour over their lifetimes. In order to obtain maximum utility, households smooth their consumption pattern by rearranging their income flows (Barba & Pivetti, 2009: 119). The life-cycle hypothesis suggests the consumption of households is based on the average lifetime income, while the permanent-income hypothesis supposes that the level of

consumption of a household is, in part, determined by their expected future income. The need for credit depends therefore on the income and the expected future income of households. Because earnings tend to be hump-shaped, younger people have a greater need for consumer credit, while middle age people will, on average, save more (Hall, 1978: 971; Bertola, Disney & Grant, 2006: 6). However, the life-cycle hypothesis and the permanent-income hypothesis are not uncontested (Fornell, Rust & Dekimpe, 2010: 29). Both are classic examples of economic theorizing, but cannot fully explain the rise of consumer credit or the consumption behaviour of modern day societies

(8)

8

(Cynamon & Fazarri, 2008: 1). The life-cycle hypothesis and the permanent-income hypothesis only explain microeconomic behaviour, and not necessarily differences among countries.

In order to understand why the use of consumer credit has risen, to account for variation in the patterns we see, and to understand the impact of this increase on the economic performance, a more diverse range of macroeconomic literature is useful. This chapter will therefore examine the debates in the literature and develop a framework of analysis that shapes the empirical arguments of the thesis. This literature review is composed of two parts. The first part focuses on the growth of consumer credit; while the second part explores the different arguments related to the impact of consumer credit on economic growth.

2.2: How to explain the growth in the use of consumer credit?

The general consensus seems to be that the growth in the use of consumer credit by households can be traced back to the 1970s and the 1980s. The literature offers competing explanations for the growth in the use of consumer credit. These explanations can broadly be divided into two categories: supply-side and demand-side explanations. The credit supply-side connects the rise of consumer credit to an increase in the supply of credit as a result of, for example, improvements in the financial sector and processes of deregulation (Barba & Pivetti, 2009: 119). The underlying assumption of the credit supply-side arguments seems to be that the availability of consumer credit is and/or was the limiting factor holding back growth in the total amount of credit used. The credit demand-side argues in contrast that the constraining element is household demand.

2.2.1: Supply-side explanations

Supply-side explanations point to financial liberalization to clarify the growth in consumer credit. In the US, for instance, financial innovations and reforms during the 1980s abolished constraints that limited credit offers and the profitability for financial institutions. As a result the deregulation heightened the competition and reduced specialization among financial institutions, which in turn led to rapid growth in the use of consumer credit (Gelpi & Julien-Labruyère, 2000: 107; Krippner, 2011: 143). According to Coco (2012: 21) the financial liberalization, including, among other things, the market deregulation, Wall Street’s securitization inventions, and low interest rates, created the conditions for the promotion of consumer debt. A similar story can be found in the European countries. Jentzsch & Riestra (2006: 28-29) argue that the growth in the use of consumer credit throughout Europe was driven by, among other things, deregulation, competition, and declining interest rates. Changes such as the abolition of the interest rate ceilings in the European countries

(9)

9

resulted in a favourable macroeconomic climate in which consumer credit, and especially the supply-side of credit, was able to flourish.

Looking at the abovementioned developments through the lens of the life-cycle hypothesis and the permanent-income hypothesis, one could argue that households were constrained to insufficient levels of consumer credit. Before the financial liberalization, households that wanted to smooth their consumption pattern were not always able to borrow enough to achieve the desired level of

consumption. Financial liberalization led to an easing of liquidity constraints on households, which allowed them to borrow more, and thus maximize the utility of their average lifetime income / expected future income (Barba & Pivetti, 2009: 119).

According to this view the rising levels of indebtedness is a rational outcome of the process of financial liberalization. However, Barba & Pivetti (2009: 119) note that evidence to back up the interpretation of the life-cycle hypothesis and the permanent-income hypothesis for the run-up in consumer credit is rather weak.

2.2.2: Demand-side explanations

The credit demand-side offers a broad range of explanations to clarify the increasing demand for consumer credit. Consumers increasingly need, or feel the need, to use consumer credit. However, consumers resort to consumer credit not only to smooth their consumption pattern, but also to supplement their income, and to manage difficult financial situations (Vandone, 2009: 4). Some scholars connect the increasing consumer indebtedness and household indebtedness in Western developed economies to deliberate public policy. Crouch (2011: 109, 114) argues, for instance, that in several countries, and in particular the US and the UK, governments encouraged a rise in household debt in order to stimulate the economy. Furthermore, Soederberg (2014: 1-12) claims that the unemployed and the working poor in the US are being made dependent on expensive types of consumer credit in order to overcome barriers to capital accumulation. Around the 1980s, a period characterized by stagnant real wages and high levels of unemployment, a new form of

accumulation emerged: credit-led accumulation; a central form of accumulation that is fuelled by the extraction of interest and fee-based revenue (Soederberg, 2013: 495).The dependence of the unemployed and the working poor on credit is facilitated and promoted by a form of neoliberal governance, which she refers to as ‘debtfarism’. The rise of consumer credit is, in the case of these two groups, a consequence of a diverse set of neoliberal policies aimed at the normalization and naturalization of high levels of private debt. The result is that consumer credit is being used to supplement or replace the living wage or the government benefit cheque, which benefits both the

(10)

10

employers and the financial institutions. The debtfare state maintains the current situation through the ongoing process of financial and legal deregulation (Soederberg, 2014: 1-12).

Other scholars connect the growth in consumer debt and household debt to wage stagnation and rising inequality (see also van der Zwan, 2014). Weller (2007: 583-584) argues that the run-up in consumer debt in the US was a consequence of slow income growth and rising prices for consumer items. The rise in debt was an economic necessity for households, and not the result of profligate spending. Wisman (2013: 922-923, 931) connects the credit growth in the US prior to the financial crisis to consumption externalities resulting from wage stagnation and rising inequality. Between the mid-1970s and 2006 wages were relatively stagnant, while during the same period income inequality and wealth inequality substantially increased. Wealthy households were as a result able to consume more. In order to maintain their relative social standing, other households also needed to increase their level of consumption, which prompted households to react in three different ways: they worked longer hours, reduced their savings, and took on debt.

Wisman is not the only one who argues that social norms and social status are important drivers of the rising indebtedness of consumers. Cynamon & Fazzari (2008: 4-7) argue that behavioural patterns based on social norms drove the level of indebtedness. The preferences that drive

consumer behaviour are shaped by social forces. Dynamics of contemporary social forces led to new social norms and consumption behaviours, which resulted in a rise in household spending relative to income, and caused consumer credit to explode. Barba & Pivetti (2009: 122, 126) suggest that the rising levels of household debt should be seen as a conspicuous response to wage stagnation and the rising levels of inequality. In excess of current income, consumers rely on credit to improve their standard of living, and to maintain their relative social standing; the idea of ‘keeping up with the Joneses’.

It is important to note that credit supply-side and the credit demand-side explanations are not mutually exclusive. In order for the total outstanding amount of consumer credit to grow, both the supply and the demand for credit need to be sufficient. However, there is, according to Dutt (2006: 343), reason to assume that the credit-supply side factors were more important in the US. Results from a survey show that the willingness to borrow to finance luxury or necessity-type goods or services did not increase (Pollin, 1988, as cited in Dutt, 2006: 343). The abovementioned factors that explain the growth of consumer credit do not automatically generate determined outcomes, because institutional settings and government policies vary in nontrivial ways. Nevertheless, in general the story in the literature seems to be that consumer credit became, as a result of a process of deregulation around the 1980s, cheaper and more easily available in both the US and Europe. In

(11)

11

order to maintain their relative standards of consumption and their social status in an era characterized by wage stagnation and rising levels of inequality, households adopted debt as a substitute for wages.

2.3: What is the macroeconomic effect of consumer credit?

Debt and its risks to the economy are of course a major concern in the post-crisis world. Consumer credit enhances current income resources based on the calculation that future income (growth) can lead both to improved prospects and successful reimbursement of the loan. One of the big concerns about consumer credit growth is that where this calculation goes wrong, it could be negatively related to future household demand and GDP growth. The economic importance of consumer spending can hardly be overestimated according to Fornell, Rust & Dekimpe (2010: 28), since it accounts for the majority of the GDP of a country. Any risks associated with the continuation of consumer spending are therefore important (Durkin et al., 2014: 42). Whether consumer credit is, or could become, such a risk is unclear, since general consensus on the effect of consumer debt and household debt on the growth of an economy does not seem to exist.

In recent decades Western developed countries are characterized by a process of wage stagnation, rising levels of inequality, and a fall in the labour’s income share, which fell from 68% in 1980 to 61.5% in 2005 (IMF, 2007, as cited in Guttmann, 2010: 3). According to post-Keynesian literature the pro-capital redistribution ought to have dampened economic growth in most countries

(Stockhammer & Onaran, 2012: 13). However, despite the redistribution, a drag on consumer demand did not take place, and economic growth did not slow down in the advanced capitalist countries.

To explain the growth performance of economies in the last few decades, a diverse group of scholars point to consumer debt and household debt as important drivers of economic growth (i.e. Cohen, 2007; Barba & Pivetti, 2008; Cynamon & Fazari, 2008; Crouch, 2009; Guttmann, 2010; Krippner, 2011; and Hein & Mundt, 2012). Debt-financed consumption played, in the last decades, a key-role in the growth of a number of economies. Households, instead of governments, took on debt to

stimulate the economy; a growth model that Crouch (2009: 390) describes as ‘privatized

Keynesianism’. However, consumer debt and household debt did not play the same role in all of the advanced capitalist countries in terms of boosting economic growth.

Empirical observations of variation across economies have also affected the debates in the literature. Post-Keynesian scholars argue that two extreme types of growth regimes have emerged in the era of

(12)

12

neoliberalism: the debt-led growth regime and the export-led growth regime. Economic growth in both regimes relies on external drivers to stimulate the demand. The debt-led growth regime is dependent on debt-financed (household) consumption as a stimulus for economic growth; whereas economic growth in the export-led growth regime is based on export surpluses (Lavoie &

Stockhammer, 2012: 32; Stockhammer & Onaran, 2012: 13). Furthermore, the debt-led growth regime is characterized by current account deficits, while the export-led growth regime has a positive current account. Because the growth regimes are necessary counterparts of each other on the global level, both growth regimes rely directly or indirectly on the willingness of households in debt-led countries to indebted themselves (Hein & Mundt, 2012: 34).

Hein & Mundt (2012: 34, 41-42) develop a more nuanced taxonomy to classify empirical growth regimes. In their empirical study they distinguish between four different growth regimes in period 2000-2008: the debt-led consumption boom type, the strongly export-led mercantilist as the two extreme types, and the domestic demand-led type, and weakly export-led type as the intermediate types of capitalism. The debt-led growth regime can be found in several economies in the 2000’s, among which the economy of the US and the UK; while the export-led growth regime can be found in, among others, Germany, Japan and the Netherlands (Hein, 2014: 426).

The contrasting growth regimes each rely on external stimulation of demand. Although stimulating economic growth via growing levels of indebtedness or export surpluses works in the short run, it is argued that is unsustainable in the long run because the growth regimes are built on internal contradictions and require ever-growing current account imbalances (Hein & Mundt, 2012: 43; Lavoie & Stockhammer, 2012: 35; Stockhammer & Onaran, 2012: 14; Hein, 2014: 437). According to Lavoie & Stockhammer (2012: 35) the crisis of ’08 and the subsequent deleveraging process illustrate the limits of these types of growth models. However, after the crisis there is again a tendency

towards rising imbalances (Hein, 2014: 424).

More scholars have their doubts on the long-term sustainability of debt-financed consumption as an economic driver. Cynamon & Fazarri (2008: 21) apply Minsky’s financial instability hypothesis to analyse the dynamics of household debt and economic growth in the US. Although Minsky primarily focused on corporate debt accumulation (Palley, 1994: 371), he did suggest that household debt could contribute to the business-cycle dynamics (Minsky, 1992, as cited in Isaac & Kim, 2013: 246). From this Minskyan perspective, Cynamon & Fazzari (2008: 21-23) argue that household debt initially boosted economic growth in the US. However, the level of indebtedness eventually became

(13)

13

A group of scholars use economic models to research the macrodynamic effects of consumer debt. A number of these models make a distinction between the short-run dynamics and the long-term impact of household debt on the GDP of a country. Dutt (2006), for example, uses an extended version of a Steindlian model of growth and income distribution to show that consumer credit can fuel economic growth in the short run, but could depress aggregate demand and economic growth in the long run. Palley (1994) develops a linear multiplier-accelerator model to analyse the cyclical aspects of consumer debt. His conclusion is similar to Dutt’s: consumer debt could initially increase aggregate demand, but the debt service payments reduce the consumption and output level in the long run.

The empirical literature provides mixed evidence on the link between consumer debt and/or household debt and the subsequent economic performance. Durkin, Elliehausen & Zywicki (2015: 289) argue that economic studies have failed to provide convincing evidence for the idea that growth in consumer credit will lead to a decrease in future spending. Available evidence suggests quite the opposite: there is a positive relation between the growth in consumer credit and future spending. A possible explanation for this relation is that consumer credit rises when households are optimistic about the current conditions, instead of pessimistic. However, Johnson (2007, as cited in Johnson & Li, 2007: 4, 6) found a negative relationship between the growth in credit card debt and total household expenditures, and a negative link between the rise in the growth of revolving consumer debt and the consumption growth rate. Maki (2002: 43, 50) notes that not a lot of evidence supports the idea that household debt service burdens negatively influence future consumption. Several papers found a positive link between consumer credit growth and the growth of future consumption. Based on a panel of 30 countries from 1960 to 2012, Mian, Sufi & Verner (2015: 35) conclude that there is a negative relationship between an increase in the household-debt-to-GDP ratio and subsequent GDP growth. Using econometric methods, Kim (2016: 127, 150) finds a positive relationship between household debt and GDP growth in the short-term. However, in the long run the debt has a negative impact on GDP. The results, which are similar to the outcome of several economic models, suggest that in short-term debt-financed household spending could stimulate an economy, but in the long term it could, if the level of debt becomes excessive, generate a negative effect.

Although the empirical evidence is mixed, the emphasis in the literature seems to be on the two-sided character of consumer credit. Consumer credit could initially provide a temporary solution to the tension between the pro-capital redistribution and the necessity of rising consumption levels for economic growth. However, in the long run consumer credit could have a negative impact on the economic performance. Not every country became dependent on debt-financed consumer spending

(14)

14

for economic growth. Guttmann (2010: 4) argues that a constellation of institutional

complementarities is the reason why some countries are in favour of higher levels of indebtedness, while other countries are not. The type of growth regime a country has also plays a role in the subsequent policy decisions. The governments of the export-oriented countries were, compared to the debt-led countries, concerned with other types of policies, such as keeping export prices low (Crouch, 2011: 115). The growth model not only influences the level of indebtedness, and the short-term and long-short-term impact of the level of indebtedness, but also the associated policy choices. The macroeconomic effect of consumer credit will therefore likely differ between countries.

In an era characterized by wage stagnation and rising levels of inequality, households increasingly decided to resort to consumer credit. However, because institutional settings and government policies vary in nontrivial ways, the level of household indebtedness, the factors that gave rise to the consumer credit growth, and the possible macroeconomic consequences differ widely among countries. The precise role of consumer credit is thus unclear. More research into the role of consumer credit that highlights the applicability of the suggested explanatory factors, and the macroeconomic effect of consumer credit for the different types of economies is therefore necessary.

(15)

15

3: Research Design

3.1: Research objective

In the last few decades the relative and absolute amount of outstanding consumer debt grew significantly in a lot of countries. The literature offers different reasons as to why consumer credit has grown in recent years. However, these explanations have the tendency to focus mainly on the experience of the US. Empirical research into the question whether these trends can also be found in other countries is limited.

Furthermore, most literature on the relationship between consumer credit and subsequent economic performance is solely focused on the US (Mian et al., 2015: 1). In terms of boosting economic growth, however, consumer credit did not play the same role in all countries according to the literature. The effect of consumer credit in countries that adopted the debt-led growth regime will most likely differ from the effect it has on countries that resemble the export-led growth regime. In addition, the empirical evidence is ambivalent about the effect of consumer credit on economic growth. It is therefore important to gain a more broad insight into the role of consumer credit. According to the literature, the effect of consumer credit is different in countries with a debt-led and export-led growth regime. It is important to make a distinction between the different growth

regimes, because the kind of growth regime a country has adopted influences not only the effect of consumer credit, but also the related policy decisions. These policy decisions could in turn have an effect on the factors that influence the growth of consumer credit. Hein & Mundt (2012) offer in their study the most detailed empirical classification of growth regimes. They argue that in the business cycle of 2000-2008 the debt-led growth regime can be found in the United Kingdom, while the Netherlands is identified as one of the export-led growth regimes (Hein & Mundt, 2012: 41; Hein, 2014: 426). This thesis therefore focusses on the role of consumer credit in the United Kingdom and the Netherlands. The United Kingdom will be used as an example of a country with a debt-led growth regime, while the Netherlands will serve as an example of a country with an export-led growth regime.

(16)

16 3.2 Researchquestions

To fully understand why the use of consumer credit grew in a particular economy, one should look at country-specific developments, since institutional settings vary in nontrivial ways. However, the objective of this thesis is not to dissect the growth pattern of a single country; the goal is to examine whether the explanations in the literature are applicable to Western developed countries other than the US, and to figure out what the impact of consumer credit is on economic performance.

This thesis focusses on the macroeconomic role of consumer credit, and the consequence thereof, in Western developed economies. The role consumer credit plays in an economy depends, according to the literature, on the growth regime a country has adopted. Both the impact of consumer credit and the associated policy decisions differ between the two growth regimes. It is therefore essential to make a distinction between the two growth regimes.

To accomplish the objective of this thesis the following research question has been formulated: What is the role of consumer credit in a debt-led and an export-led economy?

The answer to the research question will be composed of two parts. Firstly, the contributing factors that a play a role in the run-up of consumer debt will be identified. The literature identifies different factors as contributors to the run-up in consumer debt. The first part will focus on these factors: wage development, the level of unemployment, the level of inequality, the charged interest rate levels associated with consumer credit and changes in the availability of consumer credit. The goal is to figure out whether these explanations are applicable to both the debt-led economy of the United Kingdom and export-led economy of the Netherlands.

To answer the first part properly five sub-questions have been formulated: 1.1 Can wage development significantly predict growth in consumer credit?

1.2 Can the level of unemployment significantly predict growth in consumer credit? 1.3 Can the level of inequality significantly predict growth in consumer credit? 1.4 Can the interest rate level significantly predict growth in consumer credit?

1.5 Can a change in the availability of consumer credit significantly predict growth in consumer credit?

Secondly, the impact of consumer credit on the debt-led economy and export-led economy will be measured. The second part will examine the short-term interaction between credit and economic growth as well as the long-term impact on the economic performance.

(17)

17

To answer the second part properly two further sub-questions have been formulated: 2.1 What is the impact of consumer credit on GDP growth in the short term?

2.2 What is the impact of consumer credit on GDP growth in the long term?

3.3 Hypotheses

Based on the literature seven hypotheses have been formulated. These hypotheses will be tested using quantitative methods .The first part will focus on five factors that the literature identifies as contributors to the growth in consumer debt. These factors can further be divided into two categories. The first three factors are related to the credit demand-side explanations, whereas the last two factors are connected to the supply-side explanations. The second part will focus on the short-term and long-term impact of consumer credit.

A number of scholars argue that the run-up in consumer debt in the US was a consequence of wage stagnation (i.e. Weller, 2007; Barba & Pivetti, 2009; Wisman, 2013; van der Zwan, 2014). Some explanations see the growth as a conspicuous response to wage stagnation, while others focus more on the economic necessity of debt-financed spending.

H1.1: Wage development can significantly predict growth in consumer credit With H0: Wage development cannot significantly predict growth in consumer credit

According to Soederberg (2013; 2014) credit-led accumulation emerged as an attempt to overcome the barriers to capital accumulation in a period characterized by stagnant real wages and high levels of unemployment. The unemployed and the working poor in the US were as a result actively being made dependent on consumer credit. Vandone (2009) argues that households use consumer credit to augment their income in order to overcome difficult financial situations.

H1.2: The level of unemployment can significantly predict growth in consumer credit. With H0: The level of unemployment cannot significantly predict growth in consumer credit. Rising inequality is also identified as a cause of the rising levels of consumer credit. A number of these explanations emphasize the role of social norms and status (Cynamon & Fazzari, 2008; Barba & Pivetti, 2009; Wisman, 2013; Van der Zwan, 2014).

H1.3: The level of inequality can significantly predict the growth in consumer credit. With H0: The level of inequality cannot significantly predict growth in consumer credit.

(18)

18

A number of scholars argue that the decline in the interest rates of consumer credit is one of the factors that positively influence the total outstanding amount of consumer credit. Coco (2012) and Jentzsch & Riestra (2006), for instance, make the abovementioned connection. One would expect that lower costs associated with the use of consumer credit results in a higher demand, and vice versa.

H1.4: The interest rate level can significantly predict growth in consumer credit. With H0: The interest rate level cannot significantly predict growth in consumer credit. Consumer credit became as a result of the process of financial deregulation more easily available in both the US and Europe (Gelpi & Julien-Labruyère, 2000; Krippner, 2011; Jentzsch & Riestra, 2006). Furthermore, Dutt (2006) argues that there is reason to assume that the credit-supply side factors were more important than the demand for credit to clarify the run-up in consumer credit in the US. The credit demand-side explanations emphasize the increase in the supply of credit in order to explain the run-up in consumer credit.

H1.5: A change in the availability of consumer credit can significantly predict growth in consumer credit.

With H0: A change in the availability of consumer credit cannot significantly predict growth in consumer credit.

Consumer credit is often identified as an important driver of economic growth (i.e. Cohen, 2007; Crouch, 2009; Guttmann, 2010; Krippner, 2011). Furthermore, empirical evidence (Maki, 2002; Durkin, Elliehausen & Zywicki, 2015) suggests that there is a positive relationship between consumer credit and future spending.

H2.1: Consumer credit has a positive effect on GDP growth in the short-term.

With H0: Consumer credit does not have a positive effect on GDP growth in the short-term Post-Keynesian scholars argue that economic growth dependent on debt-financed spending is unsustainable (i.e. Hein & Mundt, 2012; Lavoie & Stockhammer, 2012; Stockhammer & Onaran, 2012). Mian, Sufi & Verner (2015) show how household debt and subsequent GDP growth are negatively related. Kim (2016) finds a positive relationship between household debt and GDP growth in the short run, but a negative one in the long run.

H2.2: Consumer credit has a negative effect on GDP growth in the long term.

(19)

19

4 Methodology

4.1: Research methods

This thesis will use quantitative methods to research the role of consumer credit in a debt-led and export-led country. The answer to the research question will be composed of two parts. Firstly, the contributing factors that, according to the literature, play a role in the run-up of consumer debt will be examined. A multiple linear regression model will be estimated to examine whether the different factors are applicable to both a debt-led and export-led country.

Secondly, multiple equation regression models will be used to measure the impact of consumer credit on economic growth. Multiple equation regression models offer the possibility to jointly model a set of mutually dependent time series variables (Heij, de Boer, Franses, Kloek & van Dijk, 2004: 681). Either a vector autoregression (VAR) model or a vector error correction (VEC) model will be estimated. A VAR model captures the short-run dynamic interrelationships between time series variables (Lütkepohl, 2006: 73; Kim, 2016: 131). A VEC model is particular type of VAR that takes the cointegration structure of the variables into account (Lütkepohl, 2006: 73), and incorporates both the short-run and long-run relations between the variables (Heij et al., 2004: 652).

The choice for either a VAR model or a VEC model depends on the outcome of preliminary tests. If the variables are (trend) stationary, one could estimate a VAR model in levels. However, if the variables contain stochastic trends a VAR model in levels is no longer valid, because it may lead to spurious regressions. In that case one should remove the stochastic trends by taking the first differences of the data, and estimate a VAR model in first differences. Time series variables that contain a stochastic trend can also be cointegrated, which means the variables are integrated of order 1 (and thus contain stochastic trends), but the linear combination of the variables is stationary. If the variables with stochastic trends are also cointegrated, one should choose to estimate a VEC model (Heij et al., 2004: 652-665, 681).

VAR models are often used in macroeconomics (Greene, 2003: 586), and have also been used in relation to household debt. A VAR model and a VEC model have been applied to study the short-term and long-term relationship between household debt and aggregate output in the US by Kim (2016). Furthermore, Mian, Sufi & Verner (2015) use a VAR to determine the relation between the

(20)

20 4.2: Case selection

Hein & Mundt (2012) analyse the cyclical average data for the trade cycle of the early 2000s with the aim of distinguishing between the debt-led and export-led growth regime. In their study they look primarily at the contribution of private consumption to GDP growth, the contribution of the balance of goods and services to GDP growth, and the net financial balances, to decide whether a country had a debt-led or export-led growth regime. Hein & Mundt (2012: 41-42) and Hein (2014: 424-426) conclude that in the business cycle of 2000 until 2008 the debt-led growth regime can be found in the United Kingdom, while the Netherlands is identified as one of the export-led growth regimes. The debt-led economies grew because of strong domestic demand, and in particular private

consumption. However, the economic growth was accompanied by negative growth contributions of the balance of goods and services. The export-led economies were characterized by weak domestic demand and private consumption, but gained in this period from positive net export contributions. However, there is no reason to assume that the British and Dutch economies only followed

respectively a debt-led or an export-led growth regime in the business cycle of the early 2000s. The cyclical averages on the abovementioned macroeconomic variables are a higher in the period 2000-2008, but nonetheless quite similar compared to the averages in the period 1995-2016 for the UK, and 1993-2013 for the Netherlands. Furthermore, the tendency towards rising current account imbalances is clearly visible (see appendix 1, figure A).

The averages of the United Kingdom in the period 1995-2016 are similar to the averages in 2000-2008. Table 1 shows the yearly average contribution of private consumption to the increase of GDP and the yearly average contribution of the balance of goods and services to the increase of GDP for both periods of time. The yearly contribution of private consumption is in percentage points somewhat lower in the period 1995-2016. However, the contribution of goods and services is even higher compared to 2000-2008. Furthermore, the United Kingdom had a negative current account balance in all years since 1995 (see appendix 1, figure A).

Table 1. Yearly contribution to GDP (in pp) in the United Kingdom

Yearly contribution to GDP 2000-2008 1995-2016

Private consumption 1,91 1,57

Balance of goods and services -0,13 -0,21

Source: Authors' own calculation based on AMECO (2017 A). Note: The period 1995-2016 is based on the availability of consumer credit data in the United Kingdom

(21)

21

The yearly averages of the Netherlands in the period 1993-2013 are also similar to the averages in 2000-2008. Table 2 shows the yearly average contribution of both private consumption and the balance of goods and services. In the years preceding the business cycle of the early 2000s the Netherlands did enjoy substantial contributions of private consumption to GDP growth. However, on average the 2000-2008 and the 1993-2013 periods are quite similar. Furthermore, the Netherlands had in all the years between 1993 and 2013 a positive current account balance (see appendix 1, figure A).

Table 2. Yearly contribution to GDP (in pp) in the Netherlands

Yearly contribution to GDP 2000-2008 1993-2013

Private consumption 0,58 0,77

Balance of goods and services 0,46 0,40

Source: Authors' own calculation based on AMECO (2017 A). Note: The period 1993-2013 is based on the availability of consumer credit data in the Netherlands

Because of the similarities between the averages in both periods, this thesis will assume that both the British and the Dutch economy followed a pattern similar to the ideal type of the debt-led and export-led growth regime during respectively 1995-2016 and 1993-2013.

(22)

22 4.3 Data

A multiple linear regression model will be estimated to examine the five variables that the literature has identified as contributing factors. The goal is to research whether the different factors are applicable to both the debt-led economy of the United Kingdom and the export-led economy of the Netherlands. For this part of the analysis six time series variables are relevant: consumer credit, wage development, level of unemployment, level of inequality (Gini coefficient), the consumer credit interest rate levels, and changes in the availability of consumer credit.

It is not possible to take the changes in the availability of consumer credit in the United Kingdom into account because of the lack of an extensive data set. Changes in the availability of consumer credit is available for the Netherlands. Because the data set consists of net percentages reflecting the

proportion of banks indicating tighter credit standards, the data will be transformed into a dummy variable. Furthermore, it is not possible to measure the effect of wage development on the growth of consumer credit in the Netherlands because there is a lack of data that is quarterly available.

Multiple equation regression models will be used to measure the impact of consumer credit on economic growth. A maximum of three time series variables will be used to estimate a model, because adding more variables means that more parameters need to be estimated, which will lead to a gigantic reduction in the degrees of freedom. The second part of the analysis will focus on the relationship between consumer credit and economic growth. The channel through which consumer credit influences GDP is the level of household consumption. For the second part of the analysis three time series variables are therefore relevant: GDP, total household consumption, and consumer credit.

Appendix 2 presents detailed information about all the data used in the analysis. All the data related to the UK is either from the Office of National Statistics or the Bank of England. All the data related to the Netherlands is either from Statistics Netherlands (CBS) or De Nederlandsche Bank (DNB, The Dutch Bank). All the data is either monthly or quarterly available, except the level of inequality. All the monthly data will be transformed into quarterly data by taking the average of every three-month period. The Gini coefficient of both countries is only available on a yearly basis. The data will

(23)

23

5 Results

5.1 Introduction

The results are dived into four sections. The first two parts will show the data preparation and the estimation of the multiple linear regression models for respectively the United Kingdom and the Netherlands. The last two sections of this chapter will show the data preparation and the results of the two multiple equation regression models.

5.2 Multiple linear regression analysis UK

5.2.1 Data preparation

The data for the United Kingdom covers the period 2000Q1 – 2016Q4. Four independent time series variables and one dependent variable will be included in the multiple linear regression model: wages, the level of unemployment, the level of inequality, the average interest rate level, and the

outstanding amount of consumer credit. All the variables are log-transformed, because the relative changes between the variables are more relevant than the absolute changes.

3.45 3.5 3.55 3.6 93 99 05 11 17 Inequality_log 1.6 1.8 2 2.2 2.4 93 99 05 11 17 Unemployment_log 2.7 2.8 2.9 3 93 99 05 11 17 Interest_log 6.05 6.1 6.15 6.2 6.25 6.3 93 99 05 11 17 Wages_log 10.5 11 11.5 12 12.5 93 99 05 11 17 Credit_log

(24)

24

Figure 3 depicts the log-transformed variables. It seems like all variables except Credit_log follow a random walk. To avoid spurious regressions, it is important that the time series are (trend)

stationary. Before it is possible to estimate a regression model, the stationary properties of the data need to be investigated. The Augmented Dickey-Fuller (ADF) test and the Phillips-Perron (PP) test will be used to investigate whether the time series variables are (trend) stationary.

Table 4. Unit root tests for the log-transformed variables

ADF PP Credit_log 3.11 (0.00) -1.41 (0.86) Wages_log -2.10 (0.24) -2.99 (0.04) Unemployment_log -2.01 (0.28) -1.71 (0.43) Inequality_log -1.13 (0.70) -0.99 (0.76) Interest_log -1.89 (0.34) -1.43 (0.57)

Notes: Test statistics and the corresponding p-values in parentheses. The null hypothesis for both tests is that the variable contains a unit root. Lag lengths for the ADF are selected by the majority of the information criteria, except for Credit_log. See section 5.4 within this chapter for details. Output of the information criteria is available upon request.

Table 4 shows the results of the unit root tests. According to the ADF test all variables except Credit_log contain a unit root. According to the PP test all variables except Wages_log contain a unit root. Because there is ambiguity in the results of Credit_log and Wages_log more formal testing is required. The modified augmented Dickey-Fuller (DF-GLS) test is used to determine whether the variables are stationary. The results of the DF-GLS tests can be found in appendix 3 (figure A) for Wages_log, and in appendix 7 (figure A) for Credit_log. It is not possible to reject the null hypothesis of a unit root for any of the 11 lags included in both tests. The conclusion is that all the

log-transformed variables are not stationary. The stochastic trends in the variables will therefore be removed by taking the first differences of the data.

Again, the ADF test and the PP test have been used to formally investigate whether the variables in their first differences exhibit a unit root. The plotted graphs of the variables can be found in figure 5. The test statistics and the corresponding p-values are shown in table 6.

(25)

25

Note: The inequality variable is transformed into quarterly data using a simple linear interpolation method. The result is

that the graph of the first differences of the variable looks different compared to the other variables.

Table 6. Unit root tests for the variables in their first differences

ADF PP Creditdif -1.84 (0.03) -5.78 (0.00) Wagesdif -9.51 (0.00) -9.40 (0.00) Unemploymentdif -3.10 (0.03) -5.56 (0.00) Inequalitydif -2.85 (0.05) -4.64 (0.00) Interestdif -2.55 (0.10) -5.01 (0.00)

Notes: Test statistics and the corresponding p-values in parentheses. The null hypothesis for both tests is that the variable contains a unit root. Lag lengths for the ADF are selected

by the majority of the information criteria, except for Credit_log. See section 5.4 within this chapter for details. Output of the information criteria is available upon request.

Table 6 summarizes the results of both tests. Creditdif, Wagesdif and Unemploymentdif are, according to both the ADF and the PP test, stationary. The conclusion is therefore that these variables are stationary in their first differences. The results of Inequalitydif and Interestdif are ambiguous. More formal testing is therefore required. The DF-GLS test is used to determine whether the variables are stationary. The results are reported in Appendix 3 (Table B & C). The results indicate that both variables contain a unit root.

-.1 -.05 0 .05 .1 93 99 05 11 17 Unemploymentdif -.06 -.04 -.02 0 .02 93 99 05 11 17 Interestdif -.02 -.01 0 .01 93 99 05 11 17 Inequalitydif -.04 -.02 0 .02 93 99 05 11 17 Wagesdif -.1 -.05 0 .05 .1 93 99 05 11 17 Creditdif

(26)

26

To avoid difficult interpretations that are not relevant for this thesis, it is decided to solely include variables that are stationary in their first differences. This thesis therefore proceeds to estimate a multiple linear regression model with the variables that are I(1).

5.2.2 Multiple linear regression model

The literature assumes that the two independent variables are leading indicators of consumer credit growth. The regression model should therefore include the lagged values of the independent variables. A cross-correlation analysis will help to decide the optimal amount of lag to include in the model.

The cross-correlations between the independent variables and the dependent variable are shown in appendix 4 (figure A). The strongest correlation between Creditdif and Wagesdif can be found in the first 4 lags. The regression model will therefore include the first 4 lags of Wagesdif. The correlation between Creditdif and Unemploymentdif has a clearly defined peak around 2 lags. The estimation of the regression model will therefore include first 2 lags of Unemploymentdif.

In first estimation of the multiple linear regression model there were no signs of autocorrelation. However, the assumption of homoscedasticity was violated. Although heteroscedasticity does not bias the coefficients estimates, it does affect the standard errors, and the associated t-statistics. The linear regression model is therefore re-estimated, this time adjusting the standard errors to allow for heteroscedasticity.

Table 7. Summary of the estimated linear regression model

Coefficient Std. Error P Wagesdif -- .519 ( .099, .939) .210 .002 L1. .329 (-.091, .749) .210 .123 L2. .671 ( .312, 1.03) .179 .000 L3. .640 ( .167, 1.11) .236 .009 Unemploymentdif -- -.042 (-.171, .085) .064 .506 L1. -.084 (-.192, .025) .054 .127 L2. -.107 (-.202,-.011) .048 .029 Constant .002 (-.002, .005) .002 .430

Notes: R2 =.63. Number of observations is 64. Robust / White-Huber standard errors. The confidence intervals are in the parentheses.

(27)

27

The estimated linear regression model with robust standard errors is summarized in table 7. The model is significant, and fits the data reasonably well (R2 =.63). Furthermore, Wagesdif is without lag and with a lag length of 2 and 3 quarters significant. Unemploymentdif is with a lag of 2 significant as well. Although several variables were excluded from the multiple linear regression model, it is possible to give an answer to the following two sub-questions:

Sub-question 1.1: Can wage development significantly predict growth in consumer credit?

The results of the multiple linear regression model indicate that a change in the wage development is, with and without lagged values, able to significantly predict the growth in the total outstanding amount of consumer credit. Furthermore, the positive coefficient suggests that there is positive relationship between the variables. It is therefore possible to reject the null hypothesis, and to accept the alternative hypothesis H1.1: Wage development can significantly predict growth in consumer credit.

Sub-question 1.2: Can the level of unemployment significantly predict growth in consumer credit? The results of the multiple linear regression model indicate that a change in the level of

unemployment is, with a lag length of two, able to significantly predict the growth in the total outstanding amount of consumer credit. Furthermore, the coefficient suggests that there is a

negative relationship between the variables. It is therefore possible to reject the null hypothesis, and to accept the alternative hypothesis H1.2: The level of unemployment can significantly predict growth in consumer credit.

(28)

28 5.3 Multiple linear regression analysis NL

The data for the Netherlands covers the period 2003Q1 – 2013Q4. Three independent time series variables and one dependent variable will be included in the multiple linear regression model: the level of unemployment, the level of inequality, the average interest rate level, and the outstanding amount of consumer credit. Furthermore, changes in the availability of consumer credit will also be included in the model as a dummy variable.

ADF PP

Credit_log -2.81 (0.00) 0.80 (1.00)

Unemployment_log -1.26 (0.65) -0.77 (0.83)

Inequality_log -1.09 (0.72) -0.74 (0.84)

Interest_log -0.49 (0.89) -1.23 (0.66)

Notes: Test statistics and the corresponding p-values in parentheses. The null hypothesis for both tests is that the variable contains a unit root. Lag lengths for the ADF are selected by the majority of the information criteria, except for Credit_log. See section 5.5 within this chapter for details. Output of the information criteria is available upon request. 1 1.2 1.4 1.6 1.8 90 95 00 05 10 15 Interest_log 1.4 1.6 1.8 2 2.2 90 95 00 05 10 15 Unemployment_log -1.12 -1.1 -1.08 -1.06 -1.04 90 95 00 05 10 15 Inequality_log 9 9.2 9.4 9.6 9.8 90 95 00 05 10 15 Credit_log

(29)

29

Figure 8 depicts the log-transformed variables. It is decided that all variables except Credit_log follow a random walk. The Augmented Dickey-Fuller (ADF) test and the Phillips-Perron (PP) test have been used to investigate the stationary properties of the variables. The results are summarized in table 9. According to the ADF test all variables except Credit_log contain a unit root. The results of the PP test indicate that none of the log-transformed variables are stationary. Because of the ambiguity in the results of Credit_log more formal testing is required. The modified augmented Dickey-Fuller (DF-GLS) test is used to determine whether the variables are stationary. The results of the DF-GLS test are reported in appendix 9 (figure A). It is not possible to reject the null hypothesis of a unit root for any of the 11 lags included in both tests. The conclusion is therefore that all the log-transformed

variables are not stationary. The stochastic trends in the variables will therefore be removed by taking the first differences of the data.

Again, the ADF test and the PP test have been used to formally investigate whether the variables in their first differences exhibit a unit root. The plotted graphs of the differenced variables can be found in figure 10. The test statistics and the corresponding p-values are shown in table 11.

-.3 -.2 -.1 0 .1 90 95 00 05 10 15 Interestdif -.005 0 .005 .01 .015 90 95 00 05 10 15 Inequalitydif -.1 -.05 0 .05 .1 .15 90 95 00 05 10 15 Unemploymentdif -.02 -.01 0 .01 .02 .03 90 95 00 05 10 15 Creditdif

(30)

30

Table 11. Unit root tests for the variables in their first differences

ADF PP

Creditdif -2,81 (0,03) -4.70 (0,00)

Unemploymentdif -2,57 (0,10) -3,39 (0,01)

Inequalitydif -2,52 (0,11) -3,42 (0,01)

Interestdif -3,55 (0,01) -3,70 (0,00)

Notes: Test statistics and the corresponding p-values in parentheses. The null hypothesis for both tests is that the variable contains a unit root. Lag lengths for the ADF are selected

by the majority of the information criteria, except for Credit_log. See section 5.5 within this chapter for details. Output of the information criteria is available upon request.

Table 11 summarizes the results of both tests. Creditdif and Interestdif are, according to both the ADF and the PP test, stationary. The conclusion is therefore that these variables are stationary in their first differences. The p-values of Unemploymentdif and Inequalitydif are in the ADF test not significant. More formal testing is required to determine the stationary properties of the differenced variables. The DF-GLS test is used to determine whether the variables are stationary. The results are reported in Appendix 5 (Table A & B), and suggest both variables contain a unit root.

It is possible to difference the variables once more; however, adding I(2) variables into the multiple linear regression model will cause unnecessary difficulties in regard to the interpretation of the results. It is therefore decided to solely include variables in the regression model that are stationary in their first differences.

5.3.2 Multiple linear regression model

The regression model will include the independent variable Interestdif, the dummy variable that represents the changes in the availability of consumer credit, and the dependent variable Creditdif. The amount of lag to include in the regression model will, again, be decided with the help of a cross-correlation analysis.

The correlation between Creditdif and Interestdif has a clearly defined peak around 4 lags. However, it is not reasonable to expect that a change in interest rate level for credit has an effect on the growth of consumer credit 2 years later. The maximum lag length that seems plausible is 1. Only 1 lag will therefore be included in the regression model.

In first estimation of the multiple linear regression model there were no signs of heteroscedasticity. However, the assumption of no autocorrelation in the error terms was violated. The linear

(31)

31

regression model is therefore re-estimated; this time using the Cochrane-Orcutt estimation to correct for autocorrelation.

Table 12. Summary of the estimated linear regression model

Coefficient Std. Error P

Interestdif -- .003 ( .026, .033) .817 .002

L1. -.002 (-.030, .007) .880 .123

Credit availability -- .001 (-.006, .007) .783 .506

Constant -.003 (-.009, .003) .341 .430

Notes: R2 = -.08. Number of observations is 41. The confidence intervals are in the parentheses. Original Durbin-Watson statistic = .64. Transformed Durbin-Watson statistic = 1.96

Table 12 summarizes the estimated linear regression model. The model is not significant, and does not fit the data well (R2 =-.08). None of the independent variables has a significant p-value. Although several variables were excluded from the linear regression model, it is possible to give an answer to the following two sub-questions:

Sub-question 1.4: Can the interest rate level significantly predict growth in consumer credit? The results of the linear regression model indicate a change in the interest rate level is, with and without lagged values, not able to significantly predict the growth in the total outstanding amount of consumer credit. It is therefore not possible to reject the null hypothesis, and to accept the

alternative hypothesis H1.4: The interest rate level can significantly predict growth in consumer credit.

Sub-question 1.5: Can a change in the availability of consumer credit significantly predict growth in consumer credit?

The results of the linear regression model indicate a change in the availability of consumer credit is not able to significantly predict the growth in the total outstanding amount of consumer credit. The null hypothesis can therefore not be rejected, while the alternative hypothesis H1.5: A change in the availability of consumer credit can significantly predict growth in consumer credit cannot be

(32)

32 5.4 Multiple equation regression analysis UK

The data for the United Kingdom covers the period 1995Q2 – 2016Q4. Three time series variables will be included in the multiple equation regression analysis: GDP, household consumption, and the outstanding amount of consumer credit. All the variables are log-transformed, because the relative changes between the variables are more relevant than the absolute values.

Figure 13 presents the graphs of the log of the three variables. A clear trend is visible in GDP_log and Consumption_log1. However, Credit_log follows a random walk with drift. The Augmented Dickey-Fuller (ADF) test and the Phillips-Perron (PP) test will be used to investigate whether the time series variables are stationary. Similar to Heij et al. (2004: 602) four lags will be included in the ADF test to pick up possible seasonally effects. For variables with a clear direction it is important to include a trend term in both tests. Because the variables GDP_log and Consumption_log show a clear direction, it is important to include a trend term in both tests. The ADF test for Credit_log will include a drift term in the regression, while the PP test will include a trend term (because in the PP test a random walk with drift is a special case of case 4, see Hamilton, 1994: 497-502).

Table 14 shows the results of the unit root tests. According to the ADF test and the PP test, GDP_log and Consumption_log both contain a unit root. The p-value of Credit_log is in the ADF test

significant. However, it is only significant when a drift term is included. Furthermore, the p-value of Credit_log is in the PP not significant. Because of the ambiguity of the results more formal testing is required. The modified augmented Dickey-Fuller (DF-GLS) test is used to determine whether the variable contains a unit root. The results of the DF-GLS test are in the in appendix 7 (figure A). It is not possible to reject the null hypothesis of a unit root for all 11 lags included in the test. The conclusion is therefore that all the log-transformed variables are not stationary.

1

Even if it was decided GDP_log and Consumption_log follow a random walk with drift, the conclusion would ultimately be the same: all variables are I(1). However, this thesis did not account for structural breaks in the data.

11.6 11.8 12 12.2 12.4 12.6 93 99 05 11 17 Consumption_log 12.2 12.4 12.6 12.8 13 13.2 93 99 05 11 17 GDP_log 10.5 11 11.5 12 12.5 93 99 05 11 17 Credit_log

(33)

33

ADF PP

GDP_log -1.70 (0.75) -1.50 (0.82)

Consumption_log -2.19 (0.50) -1.80 (0.71)

Credit_log 3.11 (0.00) -1.41 (0.86)

Notes: Test statistics and the corresponding p-values in parentheses. The null hypothesis for both tests is that the variable contains a unit root. The p-values of GDP_log and Consumption_log are in the ADF test with and without the time trend not significant. The p-value of GDP_log is in the PP test with or without the time trend not significant. The p-value of Consumption_log is in the PP test with a trend not significant, but without a trend it is. The p-value of Credit_log is in the ADF test only significant when a drift term is included. The p-value of Credit_log is in the PP test with trend not significant, but without it is.

The Johansen test will be used to test for cointegration between the three variables. Different information criteria will help to determine how many lags should be included in the cointegration test. The information criteria include the Final Predictor Error (FPE), Akaike’s Information Criteria (AIC), the Hannan-Quinn Information Criterion (HQIC) and the Bayesian Information Criterion (SBIC). The FPE and the AIC both select 7 lags, while HQIC and SBIC both choose 2 lags2. Appendix 7 shows the results of two Johansen cointegration tests (Table B & C). The cointegration test results with 2 and 7 lags show no sign of cointegration. The results of both tests are the same when a linear trend term is included in the cointegrating equations. The conclusion is therefore that the variables are not cointegrated.

Due to the fact that there is no cointegration it is possible to estimate a VAR model. However, a VAR model in levels is in this case not appropriate, because it may lead to spurious regressions. The stochastic trends in the variables need be removed by taking the first differences of the data.

2_{The output of the information criteria is available upon request.} -.02 -.01 0 .01 .02 .03 93 99 05 11 17 GDPdif -.1 -.05 0 .05 .1 93 99 05 11 17 Creditdif -.02 -.01 0 .01 .02 .03 93 99 05 11 17 Consumptiondif

The macroeconomic role of consumer credit : a comparison between the role of consumer credit in the British debt-led and the Dutch export-led economy