STOCK MARKETS, BANKS AND ECONOMIC GROWTH

(1)

UNIVERSITY OF GRONINGEN

Faculty of Economics and Business

MSc Business Administration, Specialization Finance

STOCK MARKETS, BANKS AND ECONOMIC

GROWTH

Supervising professor:

Prof. Dr. Kasper Roszbach

Author:

Oana Peia (s1942743)

(2)

2 Abstract:

This paper investigates the empirical relationship between financial development and

economic growth for eight developed European countries. Using time series data within a

VECM framework, I find evidence of a robust long-term link between both stock markets and

banks and real GDP growth in all the countries analysed. The causality of this relationship is

examined through weak exogeneity tests. The results suggest that causality patterns differ

with the type of financial institution analysed: in most countries, stock market development

causes, in the long run, economic growth, while the relationship between banking sector

development and real GDP growth appears to be bi-directional.

(3)

3 1. Introduction

2. Literature review

2.1 Theoretical considerations

2.2 Empirical evidence

3. Variables, Data and Hypotheses

3.1 Variable definitions

3.2 Data sources

3.3 Hypotheses

4. Methodology

5. Results

5.1 Germany

5.2 The Netherlands

5.3 Belgium

5.4 France

5.5 Italy

(4)

4 1. Introduction

The importance of the relationship between financial development and economic growth is well recognized in the growth and financial literature alike. Ever since Schumpeter (1912), Patrick (1966) or Goldsmith (1969), a burgeoning literature, both theoretical and empirical, has been concerned with the finance-growth nexus, and, while much of the existing work confirms the positive correlation between financial development and economic growth, the debate over the causality of this relationship is still an ongoing issue.

Advances in computational capacity and the availability of longer time series, in particular with regards to stock market data, have recently enabled researchers to explore the issue of causality between finance and growth in a more rigorous way (Beck, 2008: 4). Two different views pervade with regards to this debate. On the one hand, well-functioning financial intermediaries and markets reduce information asymmetries and transaction costs, which leads to a more efficient resource allocation and thus may foster long-term growth (Beck & Levine, 2004). An extensive cross-country literature has, indeed, found a positive correlation between the level of financial development and economic growth. On the other hand, however, it has been argued that this positive correlation does not necessarily imply that financial development causes economic growth, but rather it may simply follow real output growth which is generated by other factors. According to this view, it is economic development that spurs the demand for different financial contracts, markets and institutions and that the financial system responds automatically to these demands (Levine, 1997: 688). In settling this debate, many scholars (Demetriades & Hussein, 1998; Arestis & Demetriades, 1997; Arestis et al., 2001, to name a few) have drawn caution to the broad-brush picture of causality obtained from cross-country analysis and have argued that a time series approach is better able to reveal important insights which cannot be inferred from the averaged-out results of cross-sectional studies.

This paper examines the relationship between financial development and economic growth within a VAR framework using time series data from eight European countries. The main research question can be formulated as follows: What is the long-term relationship between financial development and economic growth and does the causality of this relationship go from financial development to economic growth or in the opposite direction? In answering this question, I look both at stock markets and the banking sector and, as such, I bring evidence on the relative merits of the bank-based and market-based financial systems. The cointegration results show a stable long-term relationship between both types of financial intermediaries and output growth, with the impact of banks being, as expected, larger in economies traditionally categorized as bank-based.

(5)

5

Johansen cointegration procedure. The results show an interesting pattern. In seven of the eight countries analysed there is strong evidence of causality going from stock market development to economic growth, while the relationship between output growth and banking system development is, in most cases, bi-directional. Moreover, I also examine the causal relationship between banks and stock markets, for which conflicting views also exist. On the one hand, since bank and stock markets provide substitute sources of finance, the explosive growth of stock markets may lead to a decline of the banking sector. On the other hand, it may be possible that the development of stock markets occurs simultaneously with that of the banking system since the latter can offer complementary services to issuers of equity. The results appear to be in line with the second view: in five of the eight countries analysed, stock market development appears to cause, in the long-run, banking system development.

This research brings some important contributions to the finance-growth literature. First, it extends the existing time series evidence by looking at cointegration and patterns of causality in five new countries, which, to my knowledge, was not previously attempted within this particular framework or time period. Although my research is limited to a number of developed European countries, its results are of equal interest to developing countries, since it helps them understand how the two types of financial systems affect economic growth, which may have important policy implications. Second, my results bring clear evidence that patterns of causality are sensitive to country specific characteristics like the dominance of a particular type of financial sector, i.e. banks or financial markets. Moreover, they show that while the causality between banks and economic growth may differ among countries, there is consistent long tern causality between stock market development and output growth in almost all countries considered. Finally, I also look at the relationship between the two types of financial sectors and shade some light on the patterns of causality between the two.

(6)

6 2. Literature review

2.1 Theoretical considerations

The role of the financial sector has been well recognized in the growth literature. The theoretical underpinnings of this role can be traced back to the work of Patrick (1966) who points out two possible relationships between financial development and economic growth. First, he considers a “demand-following” phenomenon in which the creation of modern financial institutions and services stems from them demand for these services from the economy. According to this view, the more an economy grows, the greater will be the enterprises’ demand for external funding and consequently the greater the need for financial intermediation. Patrick (1966, p. 175) recognizes that : “ [A]s a consequence of real economic growth, financial markets develop, widen and become more perfect thus increasing the opportunities for acquiring liquidity and for reducing risk, which in turn feeds back as a stimulant to real growth”. Second, a “supply-leading” hypothesis is considered, where the creation of financial institutions occurs before there is a demand for it, as a way of transferring resources from traditional sectors to modern, high-growth sectors. Furthermore, the author recognizes a possible interaction between the two phenomena, where the supply-leading impetus may initially promote innovation-driven development, but, as the process of real growth occurs, the demand-following financial response takes over and becomes the more dominant force.

Since Patrick (1966), numerous theoretical models have been developed to support both his supply and demand hypotheses, among others by Greenwood and Jovanovic (1990), Levine (1991), Bencivenaga and Smith (1991), Pagano (1993) and Blackburn and Hung (1998). The demand-following line of argument is shared by Greenwood and Jovanovic (1990) who build a model in which financial institutions arise endogenously to facilitate trade in the economy. In their model the role of financial institutions in improving the allocation of resources is achieved by collecting and analysing information which allows individuals to earn a higher expected return on their investments. Furthermore, they also consider financial institutions to provide means through which agents can pool and diversify risk which allows for safer returns. Therefore, the model assumes a two-way causality: first, economic growth provides the avenue for developing a financial structure, then, this financial superstructure which enables a more efficient investment will, in turn, further promote economic growth.

(7)

7

investment opportunities due to liquidity constraints. Through these functions, banks favour capital accumulation by altering the fraction of an economy's savings that is held in unproductive liquid assets. If we then consider that savings behaviour influences growth rates, it would result that financial intermediaries will tend to promote growth.

Blackburn and Hung (1998) also propose a theoretical analysis of the linkages between economic growth and financial development. Their model assumes a positive, two-way casual relationship between the two indicators, while focusing on another channel through which financial intermediation can enhance growth, namely, the monitoring of risky investment projects in the presence of asymmetric information. As the number of investment projects increases in a growing economy, financial institutions, as delegated monitoring agencies, will emerge endogenously to provide the most efficient means of channeling savings into investments. This will be achieved through diminishing the agency costs that must be paid by companies to secure loans.

Overall, the theoretical models presented above do not distinguish between the separate roles of equity markets and other financial intermediaries. Boyd and Smith (1998) formally characterize the complementarities between the two sectors of the financial system by developing a theoretical model in which firms through the issue of some equity find it cheaper to issue debt. Their analysis also infers that debt and equity markets are likely to become more complementary as economies become more highly developed.

The main insight that can be drawn from the studies presented above is that the causality between financial development and economic growth is a controversial issue which could be potentially solved only by resorting to empirical research.

2.2 Empirical evidence

(8)

8

Earlier empirical literature has solely focused on the role of the banking system as a promoter of economic activity. King and Levine (1993) show that bank development helps explain economic growth in a sample of 80 countries. However, by overlooking the stock market, it is difficult to assess whether the positive relationship between bank development and economic growth holds when controlling for stock market development or whether banks and equity markets each have an independent impact on economic growth. Rousseau and Wachtel (2000) consider several reasons why stock markets are an important component of the financial sector that may promote economic development. First, equity markets are a liquid market mechanism that makes venture capital investments more attractive by offering an exit option. This also enhances entrepreneurial activity in countries with well functioning equity markets. Second, the existence of stock markets facilitates capital inflows – both foreign direct investments and portfolio investments. Thirdly, stock markets provide information that improves the efficiency of financial intermediation in general.

Levine and Zervos (1998) bring evidence in this direction by empirically investigating whether measures of stock market liquidity, size, volatility, and integration with world capital markets are robustly correlated with current and future rates of economic growth using data on 47 countries from 1976 through 1993. They find that initial levels of both stock market liquidity -measured by the value of stock trading relative to the size of the market or the economy- and banking development -measured by bank loans to private enterprises divided by GDP- are significantly correlated with current and future rates of economic growth. Their results confirm a strong, positive link between the initial level of financial development and the current level of economic growth. However, based on their findings nothing can be inferred about the potential relationship between the contemporaneous levels of financial sector development and economic growth or the direction of causality between the two.

Beck and Levine (2004) largely use the same variables and data as Levine and Zervos (1998) but try to reduce some statistical short-comings of simultaneity and unobserved country-specific effects by using improved econometric techniques and data averaged over five-years. Their results strongly reject the hypothesis that financial development is unrelated to growth. Although their findings suggest that both markets and banks independently spur economic growth, their results are inconsistent when different estimation techniques or control variables are used. More specifically, either the bank development measure or the stock market variable losses its statistical significance, which leads the authors to conclude that it is difficult to identify which specific financial institution is more closely associated with economic success.

(9)

-9

measured through value traded- rather than size is the key channel through which stock markets enhance growth.

Although the empirical research presented so far proves a strong, positive connection between financial development and growth, the use of cross-sectional techniques cannot identify dynamics or differences among countries. It is likely that in some countries finance is a leading sector whilst in others it lags behind the real sector. This means that the causality of the relationship derived from these studies is only valid on average, if at all (Demetriades & Hussein, 1998: 391). Among the first papers to explore causality in the VAR framework is Demetriades and Hussein (1998). They provide evidence on the existence of a stable long-run linear relationship between economic and financial development by carrying out cointegration tests based on both the Engle and Granger (1987) two-step procedure and the Johansen (1988) maximum likelihood method. Their approach is also novel with regards to the data set used which is comprised exclusively of 16 not very developed countries. They focus mainly on the development of the banking sector for which they find a stable relationship with real GDP per capita in 13 of the countries considered. Their causality tests seem to favour a bi-directional relationship between finance and growth, which is very much tied to country specific elements. This leads the authors to raise a question about the dangers of lumping together in cross-section equations, countries with very different experiences in relation to financial development (Demetriades & Hussein, 1998: 406).

(10)

10 3. Variables, Data and Hypotheses

3.1 Variable definitions

The variables used to capture the finance-growth nexus are consistent with theoretical specifications and previous studies, and are measured quarterly. It is widely accepted that economic growth is best captured by the logarithm of real GDP (Demetriades & Hussein, 1996; Levine & Zervos, 1998; Rousseau & Wachtel, 2000; Arestis et al., 2001; Beck & Levine, 2004).

Stock market development is measured by the logarithm of the ratio of stock market value to nominal GDP. Although previous cross-country studies have found other measures of market liquidity, such as the ratio of the value traded to GDP, to be more closely linked to economic growth, Arestis et al. (2001) argue that market capitalization is more likely to have time series properties making it more suitable for a cointegration analysis. Furthermore, although market capitalization is a widely used indicator of stock market development, its main weakness is that it may fluctuate excessively over time, reflecting any excess volatility in stock prices. However, if the stock prices are non-stationary and exhibit close to unit root behaviour (Baltagi, Demetriades & Law, 2009), this makes the methodological framework used in this paper the more appropriate.

The development of the banking system is measured by the logarithm of the ratio of domestic bank credit to nominal GDP. Finally, stock market volatility- measured by an eight-quarter moving standard deviation of the end-of-quarter change of the stock market price index- has also been shown to be an important characteristic of stock markets that undermines their ability to promote efficient allocation of investments (Arestis et al., 2001). Whether volatility has a positive or negative influence on economic growth is, however, still a matter of debate. Although some volatility of prices is desirable, many have argued that the presence of excess volatility results in an inefficient allocation of resources (Federer, 1993).

3.2 Data sources

(11)

11

1973- Q4 2009; Sweden during Q1 1982- Q4 2009; the United Kingdom during Q1 1973- Q4 2009; Italy during Q1 1973- Q4 2009; Belgium during Q1 1980-Q4 2009. The focus of the research is on European countries, which to my knowledge was never attempted so far. It looks at the most developed European nations, measured by nominal GDP, for which the stock market capitalization variable is available from the early 1980’s.

The four variables are obtained as follows. Output, measured by the logarithm of real GDP, is obtained from the IMF International Financial Statistics database and Datastream. Stock market capitalization and end-of-quarter price indexes are extracted from Datastream. Finally, domestic bank credit is obtained from IMF International Financial Statistics database1. For brevity, the descriptive statistics and correlations between the variables are presented in Appendix A.

3.3 Hypotheses

This section considers some popular hypotheses about the relative importance of the capital markets and banking systems in the eight countries studied. Firstly, the legal origin argument put forward by La Porta et al. (1997) offers the venue for a preliminary analysis. Based on the idea that common law systems, originating from English law, are better suited for the development of capital markets than civil law based systems, primarily rooted in French law, I would expect a greater development of stock markets in the UK, which is the single common law country in the sample considered. However, similar to Rajan and Zingales (2003), the evidence presented in Table 1 does not support the legal origin argument.

Table 1 presents the evolution of the ratios of stock market capitalization to GDP and domestic credit to GDP over the sample period. Here, The Netherlands, a country with a civil law system of French origin, has the highest ratio of stock market capitalization to GDP, in the 1973 to 1980 period. Similar evidence is brought by Rajan and Zingales (2003) who report a ratio of stock market value to GDP of 0.56 for The Netherlands in 1913, much higher, even at that time, than some common law countries such as the US. Sweden, a country of Scandinavian civil law origin, offers the same argument, by having a stock market capitalization ratio higher than that of UK in the 2001 to 2009 period. Demetriades (2008) brings a similar argument in dismissing the power of the legal hypothesis to explain the link between financial development and economic growth across countries. As a result, I will not further test this hypothesis.

1

(12)

12

TABLE 1

Evolution of stock market value and domestic claims over GDP (period averages)

Stock Market Value over GDP

Years Country 1973-1980 1981-1990 1991-2000 2001-2009 Germany 0.48 0.59 1.15 1.63 The Netherlands 0.74 1.28 3.82 3.61 France 0.11 0.32 1.58 2.87 Austria 0.04 0.11 0.46 1.18 Sweden - 0.38 2.34 3.49 UK 0.56 1.13 3.12 3.41 Italy 0.05 0.25 0.98 1.62 Belgium 0.22 0.52 1.54 2.36

Domestic credit over GDP

Years Country 1973-1980 1981-1990 1991-2000 2001-2009 Germany 3.32 4.26 4.94 5.39 The Netherlands 2.80 3.42 4.03 6.85 France 2.30 3.43 3.83 4.47 Austria 2.11 3.68 4.76 5.02 Sweden - 2.38 2.01 4.04 UK 1.69 2.71 4.68 6.57 Italy 3.69 3.26 3.70 4.33 Belgium 2.03 2.47 5.15 4.42

More fruitful avenues are offered by the popular distinction made between the “bank-centered” and “market-centered” financial systems in promoting economic growth. Many scholars have argued that bank-based systems are better at mobilizing savings, identifying good investments, and exerting sound corporate control, particularly during the early stages of economic development and in weak institutional environments. Others emphasize the advantages of stock markets in allocating capital, providing risk-management tools, and mitigating the problems associated with excessively powerful banks (Levine, 2000:398). Panel data evidence (Levine, 2000) has failed to find a strong support towards either system in better promoting economic growth. However, time series research such as that of Arestis et al. (2001) largely confirms that the relative importance of the two financial systems in explaining economic growth is consistent with the bank-based versus market-based hypothesis. Thus, I consider this framework as a useful starting point in assessing the relative importance of the two financial systems in explaining the economic development of the eight countries studied. It should be noted here that, from the sample of countries analysed in this study, Germany, Italy, Austria, Belgium, France are commonly recognized as bank-based, while UK, Sweden and The Netherlands as market-based.

(13)

13

implications over the patterns of causality between finance and growth. Arestis and Demetriades (1997) argue that certain types of institutional structures of financial systems may be in a better position to enhance economic growth than others. They argue that in the case of bank-based systems the causality between financial intermediation and output growth is likely to be either from finance to growth or bi-directional. This is due to the fact that bank-based financial systems are arguably better at solving principal-agent problems and promote longer time horizons, which encourage financial stability and foster a framework for the implementation of successful economic policies. By contrast, market-based systems are more concerned with short-term performance and thus the causality in this case is expected to be from growth to finance, although a bi-directional relationship cannot be ruled out.

Finally, turning to the relationship and causality between banks and stock markets, two competing hypotheses can also be formulated. First, considering the theoretical model proposed by Boyd and Smith (1998), the development of stock markets complements that of debt markets, making them operate more efficiently. Their analysis also implies that the two types of financial systems become more complementary as economies become more and more developed. Considering the level of economic development of the countries analysed in this paper, I expect a causal link going from stock markets to the banking system in all of them. An alternative hypothesis is offered by Demirgüç-Kunt and Maksimovic (1996) who empirically support the view of a casual link between stock markets and banks in only in developing economies, while in countries with already highly developed stock markets further development leads to a substitution of equity for debt financing.

4. Methodology

A popular strategy used for data that is thought to be non-stationary and cointegrated is the Johansen (1988) technique based on VARs. Arestis et al. (2001) argue that for sample sizes of around one hundred data points (which is the case for most countries analysed here), the maximum likelihood approach of Johansen generally performs better than other estimators of long-term relationships. They therefore use it to determine the number of cointegrated vectors amongst the variables specified in the VAR models and to examine the direction of their causality.

(14)

14 , (1)

then this can be converted into a vector error correction model (VECM) in the form:

, (2)

where y is a (nX1) vector of I(1) variables (in this case n=4), Γi= -( I - β1 – βi), (i=1,…, k-1), Π=-( I - β1 - …- βk), D is a set of I(0) deterministic variables such as a constant, trend

2

and dummies, and u is the vector of normally distributed errors with zero mean and constant variance.

The test of cointegration centers on the rank r of the Π matrix via its eigenvalues, denoted by λi. The rank of the matrix is equal to the number of its characteristic roots that are different from zero. Thus, to test the null hypothesis that there are at most r cointegration vectors amounts to the following:

H

0

:

λi =

0

i= r+1,…, n

where only the first r eigenvalues are non-zero. Johansen (1988) provides two statistics for testing this null hypothesis, i.e. the trace statistic:

(3)

and the maximal-eigenvalue statistic:

) (4)

where r is the number of cointegrating vectors and is the estimated value for the ith ordered eigenvalue for the Π matrix. If is significantly different than zero, then this indicates a cointegrating vector. The test is conducted in a sequence until the null hypothesis that there are r cointegrating vectors in favor of the alternative that there are r+1 (for λtrace)or more than r (for λmax)

cannot be rejected. Critical values are provided by Osterwald-Lenum (1992).

Since there are four variables which constitute the vector

y

t, the dimension of Π is 4x4 and its rank can

be at most equal to four. If the rank of Π is zero, then the variables are not cointegrated, whilst a rank of four implies that the vector process

y

t is stationary.

If r<4, Π can be defined as the product of two matrixes, α and β’, of dimensions (nXr) and (rXn), respectively:

2

(15)

15

Π=αβ’ (5) The columns of the β’ matrix have an economic interpretation as the cointegrating vectors. That is, after normalization, they may be interpreted as long run parameters (Charemza & Deadman, 1992). The coefficients of α represent the speed of adjustment of a particular variable with respect to a disturbance in the equilibrium relation. Furthermore, for r cointegrating vectors in β, there exist r columns in α which contain at least one non-zero element. This implies that each of the r non-zero columns of α contains information on which cointegrating vector enters which short-run equation, as well as on the speed of the short-run response to disequilibrium. Thus, a test for the presence of all zeros in a row i of α, i.e. αij=0, (i=1, … , n; j=1, … , r), indicates that the cointegrating vectors in β do not enter the equation determining Δyit and, as a result, this variable is weakly exogenous to the system (Harris, 1995). Arestis et al. (2001) interpret this weak exogeneity in a cointegrated system as evidence of long-run causality going from the exogenous variables to the endogenous ones.

In empirical applications, the Johansen method faces with several caveats. First, the results are highly sensitive to the lag length selection. Charemza and Deadman (1992) argue that when empirical analysis is concerned exclusively with the estimation and identification of the cointegrating vectors, the lag length is set such that the VAR residuals are free of autocorrelation, even if this implies relatively long lags. Another, more general approach is the use of a multivariate information criterion (Brooks, 2008) such as Akaike (1974) or Schwarz (1978). This paper will follow both approaches in attempting to estimate the lag length optimally. However, if they happen to yield conflicting results, the lag length will be set such that the VECM residuals are empirically Gaussian as the Johansen method proposes.

Second, within the framework of error correction models, the question of whether an intercept or trend ought to enter the short and/or the long-run model is frequently raised (Harris, 1995). Several models can be considered:

1. The level data have no deterministic trends and the cointegrating equations do not have intercepts;

2. The level data have no deterministic trends and the cointegrating equations have intercepts; 3. The level data have linear trends but the cointegrating equations have only intercepts; 4. The level data and the cointegrating equations have linear trends;

5. The level data have quadratic trends and the cointegrating equations have linear trends.

(16)

16

implies estimating the cointegrating vector under all models and presenting them from the most restrictive (i.e., r=0 and Model 1) to the least restrictive alternative (i.e., r=n-1 and Model 5). The test procedure is then to move through from the most restrictive model to the least and at each stage to compare the trace or maximal-eigenvalue statistic to its critical value and only stop the first time the null hypothesis is not rejected. Arestis et al. (2001), among others, implement cointegration tests along these lines, and thus, this paper will also follow this approach.

Finally, the Johansen reduced rank regression procedure only determines how many unique cointegration vectors span the cointegrating space. Therefore, it is necessary to impose restrictions motivated by economic arguments and then test whether the columns of β are identified.

5. Results

The first step in performing cointegration tests involves testing the stationary of the variables. This is achieved using the ADF (Augmented Dickey-Fuller) procedure referred to earlier. The results are presented in Appendix B and show that all variables are in I(1). I then perform the cointegration analysis for each of the eight countries, the results of which are presented in Tables 2-10. Since cointegration tests are sensitive the lag length of the VECM, in Panel A of each table, I present the criteria I used for indentifying the optimal lag length. These include the Akaike and Schwarz information criteria and the autocorrelation tests of the VAR residuals. When the two approaches suggest different lag lengths, the number of lags is set such that there is no autocorrelation in the VAR residuals (Charemza & Deadman, 1992). Next, the Pantula principle for identifying the appropriate restrictions on the intercept and trends in the short- and long-run models is presented in Panel B of each table. Panel C reports the trace and maximal-eigenvalue statistics of the Johansen cointegration tests as well as the normalized cointegrating relationships. The normalization of the cointegrating vectors is done on the variable for which evidence of error correction is found, that is, a negative α coefficient. Finally, Panel D of each table reports the weak exogeneity tests for each variable.

5.1 Germany

(17)

17

Table 2 GERMANY

PANEL A Test statistics and the choice criteria for selecting the lag order

Variables included in the unrestricted VAR: gdp is the logarithm of real GDP, mkt is the logarithm of the ratio of stock

market capitalization to nominal GDP, bnk is computed as the logarithm of the ratio of domestic credit to nominal GDP, vol is the eight quarter moving standard deviation of the price index

Lag AIC SC LM-Stat Prob

0 -4.74939 -4.57970 1 -15.9009 -15.3918 34.7296 0.0043 2 -15.9780 -15.1296 37.8229 0.0016 3 -16.0818 -14.8939 34.4639 0.0047 4 -17.4289 -15.9017 16.4498 0.4220 5 -17.8004 -15.9337* 11.1429 0.8006 6 -17.8472 -15.6417 24.1975 0.0853 7 -17.8941 -15.3486 18.0337 0.3219 8 -18.0038 -15.1190 43.5065 0.0002 9 -18.0374* -14.8132 15.4794 0.4898 * indicates lag order selected by the criterion; AIC: Akaike information criterion; SC: Schwarz information criterion LM test - Null Hypothesis: no serial correlation at lag order h;

Diagnostic tests VECM residuals: χ2

Normality(2): 5.3692 (gdp); 2.6257 (mkt); 2.0493 (bnk); 140.2663 (vol);

χ2Heteroskedasticity(810): 839.3865

PANEL B The Pantula Principle ( 5 lags)

H0 R n-r Model 2 Model 3 Model 4

λ max test 0 3 54.5726 54.4892 54.7088 1 2 36.1756 34.9016 34.9139 2 1 17.1713* 14.6938 22.7225 3 0 9.8386 6.6152 6.6185 λ trace test 0 3 124.0506 115.6130 123.9431 1 2 69.4780 61.1238 69.2343 2 1 33.3024* 26.2223 34.3204 3 0 16.1311 11.5285 11.5979 * Denotes the first time when the null hypothesis is not rejected at the 10% significance level

PANEL C Johansen Cointegration Test

Null Hypothesis Eigenvalue Trace Statistic Maximal Eigenvalue Statistic None ** 0.3191 124.0506 54.5726

At most 1 ** 0.2249 69.4780 36.1756 At most 2 0.1139 33.3024 17.1713 At most 3 0.0669 16.1311 9.8386 At most 4 0.0433 6.2926 6.2926 Normalized vector 1: bnk = 0.2215 x mkt + 1.6601 x vol + 2.9384 x dummy + 1.5273 x c [1.9113]* [0.9171] [6.7301]** [7.4684]** Normalized vector 2: gdp = 0.6257 x bnk + 0.2337 x mkt - 2.6136 x vol + 12.3530 x c [7.0426]** [4.6690]** [-4.8665] ** [86.7027] **

*(**) Denotes rejection of the hypothesis at the 5%(1%) level; dummy represents a dummy variable for Q4 1987 and Q4 1989- Q1 1991; Critical values are provided by Osterwald-Lenum(1992);

t-statistics of the normalized coefficients under the null that the parameter is zero are in parenthesis [ ]

PANEL D Weak Exogeneity

gdp bnk mkt vol

α (vector 1) 0.0042 -0.0031 -0.0032 0.0041 α (vector 2) -0.0140 0.0623 0.0478 -0.0914

p-value 0.0000 0.0032 0.7687 0.0000

(18)

18

In the case of Germany, both SC and the LM autocorrelation tests suggest a lag length of 5. Apart from no serial correlation, the diagnostics tests on the VECM residuals (presented below panel A) show that they also follow a normal distribution (with the exception of the error-correction equation for vol) and that there is no heteroskedasticity. Thus, the VECM residuals are empirically Gaussian as the Johansen method proposes.

The Pantula principle in Panel B shows that the first time the null hypothesis of no cointegration is not rejected is under model 2. Thus, cointegration tests are performed by allowing no deterministic trends in the level data and an intercept in the cointegrating space. Both trace and maximal-eigenvalue statistics in panel C show two cointegrating vectors among the financial development and economic growth variables for the period 1973 to 2009. The first vector suggests a positive and significant relationship between bank and stock market development. In the second vector, both bnk and mkt are positively related to gdp and the relationships are significant at a 1% level. Stock market volatility has a negative and significant impact on GDP suggesting that the presence of excess volatility may be detrimental to efficient resource allocation and, therefore, reduce growth. It is worthwhile noticing that while both stock market capitalization and domestic credit are positively related to output growth, the coefficient of bank development is almost three times bigger than the one of stock market development. This result is similar to the one obtained by Arestis et al. (2001) for the 1973-1997 period and confirms the “bank-based” view of Germany.

The results of the weak exogeneity tests are presented in Panel D. The null hypothesis of weak exogeneity is rejected in the case of real GDP and the ratio of domestic credit to GDP. Hence, similar to Arestis et al. (2001), in Germany, there appears to be bi-directional causality between output and banking system development, while stock market development is weakly exogenous to the system. Also, as hypothesized, the relationship between banks and stock markets is complementary, with the development of capital markets causing, in the long run, the expansion of the banking system.

(19)

19

Table 3 THE NETHERLANDS PANEL A Test statistics and the choice criteria for selecting the lag order

0 -4.9715 -4.7837 1 -17.7014* -17.1379* 14.7984 0.5395 2 -17.5541 -16.6149 15.4520 0.4918 3 -17.4696 -16.1547 17.7073 0.3413 4 -17.4172 -15.7266 46.5422 0.0001 5 -17.4752 -15.4089 14.6903 0.5474 * indicates lag order selected by the criterion; AIC: Akaike information criterion; SC: Schwarz information criterion LM test - Null Hypothesis: no serial correlation at lag order h

PANEL B The Pantula Principle ( 1 lag)

H0 r n-r Model 2 Model 3 Model 4

λ max test 0 3 73.4813 70.0207 72.8692 1 2 40.8293 16.8129 17.2841 2 1 13.4982* 9.6952 13.6093 3 0 9.3154 6.5509 9.5550 λ trace test 0 3 142.3291 103.2382 118.7554 1 2 68.8477 33.2175 45.8863 2 1 28.0184* 16.4046 28.6022 3 0 14.5202 6.7094 14.9930

* Denotes the first time when the null hypothesis is not rejected at the 10% significance level

At most 1 ** 0.2806 68.8477 40.8293 At most 2 0.1031 28.0184 13.4982 At most 3 0.0724 14.5202 9.3154 At most 4 0.0411 5.2048 5.2048 Normalized vector 1: gdp= 0.4120 x bnk + 0.4884 x vol – 2.2680 x dummy + 10.5994 x c [6.6935]** [0.7316] [-9.2579]** [199.496]** Normalized vector 2: mkt= 9.0478 x gdp - 21.7349 x vol + 46.5518 x dummy + 100.5843 x c [3.9859] ** [-1.5124] [8.6729]** [3.9649]**

*(**) Denotes rejection of the hypothesis at the 5%(1%) level; dummy represents a dummy variable for Q1 1979 and Q4 1987; Critical values are provided by Osterwald-Lenum(1992);

t-statistics of the normalized coefficients under the null hypothesis that the parameter is zero are in parenthesis []

α (vector 1) -0.0035 -0.0067 -0.0003 -0.0015 α (vector 2) 0.0010 0.0081 -0.0037 -0.0004

p-value 0.0000 0.0000 0.9509 0.8370

(20)

20

5.2 The Netherlands

The results for The Netherlands are reported in Table 3. Here, I introduce a dummy variable in Q1 1979 and Q4 1987 to capture blips in the data due to the second oil crisis in 1979 and the stock market crash in 1987. After the introduction of the dummy variable, both Akaike and Schwartz information criteria suggest an optimal lag length of 1. Furthermore, there appears to be no autocorrelation in the residuals at lag order 1, although the VECM residuals fail the heteroskedasticity test and only the residuals of the ECM for gdp and bnk are normally distributed.

Similar to Germany, the cointegration test, performed under Model 2, suggests two cointegrating vectors. The identifying normalizations show a positive and significant relationship between GDP and bank development in the first vector. Here, volatility also appears to be positively related to the real gross domestic product although the relationship is not significant. In the second vector, the positive and highly significant relationship between gdp and mkt is consistent with the market-based view. The volatility of the price index is again not statistically significant. The weak exogeneity tests show that the variables capturing stock market development are exogenous to the system. This implies that in the Netherlands causality flows from stock market development to output growth. There is also evidence that, in the long run, stock markets determine the development of the banking system. Both, gdp and bnk are endogenous, so a bi-directional relationship exists between them. The causality between the stock market and GDP is different than what Arestis and Demetriades (1997) expect in a market-based economy. Nevertheless, the results confirm the well-known importance of the stock market in the development of the Dutch economy.

5.3 Belgium

Turning to Belgium, a break in the data is observed during Q1 to Q4 1992, when the country experienced a deep recession. I therefore introduce a dummy variable to account for this period. In Panel A, the criteria for choosing the optimal lag length shows contradicting results: AIC suggests 8 lags, while SC only 1. Since there appears to be no autocorrelation in the residuals, following Charemza and Deadman (1992), I set number of lags to the largest value, i.e. 8.

(21)

21

Table 4 BELGIUM

0 -3.6595 -3.5316 1 -12.8716 -12.1042 33.1890 0.1264 2 -14.5450 -13.13812* 24.5303 0.4889 3 -14.4780 -12.4316 22.3482 0.6156 4 -15.2006 -12.5148 32.9380 0.1326 5 -15.4888 -12.1635 16.6983 0.8924 6 -15.2871 -11.3222 22.4242 0.6111 7 -15.5572 -10.9528 20.1277 0.7401 8 -15.7114* -10.4675 32.8358 0.1352 * indicates lag order selected by the criterion; AIC: Akaike information criterion; SC: Schwarz information criterion LM test - Null Hypothesis: no serial correlation at lag order h

Null Hypothesis Eigenvalue Trace Statistic Maximal Eigenvalue Statistic None * 0.39911 104.2144 51.4442

At most 1 0.2252 52.7702 25.7702 At most 2 0.1139 26.9999 12.2095 At most 3 0.0811 14.7904 8.54223 At most 4 0.0510 6.2482 6.2482

Normalized vector: gdp = 0.4663 x bnk + 0.1168 x mkt + 2.0940 x vol - 3.7053 x dummy + 10.3058 x c [2.8018]** [2.0955]** [1.1696] [-6.0039]** [30.8576]**

*(**) Denotes rejection of the hypothesis at the 5%(1%) level; dummy represents a dummy variable for Q1 1992-Q4 1992 Critical values are provided by Osterwald-Lenum(1992);

t-statistics of the normalized coefficient under the null that the parameter is zero are in parenthesis []

α -0.0017 -0.0179 0.0047 0.0035

p-value 0.0481 0.0000 0.6867 0.0902

(22)

22

again not statistically significant. Regarding the weak exogeneity tests, a surprising finding is that stock market development seems to cause GDP. Although this is not what one might expect in a country with a continental bank-based financial system, Degryse and Ongena (2005) argue that when it comes to general economic, financial and technological development, Belgium is very similar to the US, i.e. a market-based system.

Furthermore, since both gdp and bnk are endogenous to the system, a bi-directional relationship exists between output growth and the banking system in the case of Belgium. Again, there is evidence of causality going from capital markets to banking system development in accordance with the hypothesis put forward by Boyd and Smith (1998).

5.4 France

The financial and macroeconomic data for France suggests a break in the mid 1980’s and in 1987 relating to the deregulation of the French banking system and the stock market crash in 1987. Consequently, a dummy variable is introduced to account for these periods. The cointegration results for France, which allow for the above breaks and an intercept in the cointegrating space, are presented in Table 5, Panel C. The trace statistic shows evidence of two cointegrating vectors after allowing for two lags in the VECM. On the basis of the signs and significance of the adjustment parameters in the α matrix, normalization is performed with respect to gdp and mkt. The identifying restrictions which are accepted by the data are the exclusion of mkt in the first vector and of gdp in the second. Thus, the first cointegrating relationship suggests a positive link between output growth and the ratio of domestic credit to nominal GDP, as well as a positive impact of stock market volatility on gross domestic product. All identified coefficients are significant at a 1% level.

(23)

23

Table 5 FRANCE

Lag AIC SC LM-Stat Probability

0 -9.5584 -9.3819 1 -21.7816 -21.2522* 14.5328 0.5591 2 -21.8723* -20.9899 13.2713 0.6528 3 -21.8573 -20.6220 20.4135 0.2022 4 -21.7133 -20.1252 17.7326 0.3398 5 -21.6084 -19.6673 17.4219 0.3588 6 -21.4693 -19.1753 17.1234 0.3777 7 -21.4050 -18.7580 17.7187 0.3406 8 -21.4963 -18.4964 22.8687 0.1173

* indicates lag order selected by the criterion; AIC: Akaike information criterion; SC: Schwarz information criterion LM test - Null Hypothesis: no serial correlation at lag order h

Diagnostic tests VECM residuals: χ2Normality(2): 1.0458 (gdp); 2.0220 (mkt); 4567.364 (bnk); 106.3682 (vol);

λ max test 0 3 51.2620 51.2527 53.3454 1 2 26.9244 24.1633 24.4308 2 1 14.5797* 14.2168 16.3528 3 0 13.0874 13.0493 13.0632 λ trace test 0 3 111.8960 103.8099 117.4086 1 2 60.6340* 52.5571 64.0631 2 1 33.7095 28.3938 39.6324 3 0 19.1298 14.1771 23.2796

Null Hypothesis Eigenvalue Trace Statistic Maximal Eigenvalue Statistic

None ** 0.3066 112.7354 51.2620

At most 1 **(λ trace) 0.1749 69.0629 26.9244

At most 2 0.0989 30.8282 14.5797

At most 3 0.0892 18.1804 13.0874

At most 4 0.0422 8.0119 6.0424

Normalized vector 1: gdp = 0.8170 x bnk + 2.3171 x vol - 1.3410 x dummy + 4.8744 x c [4.8139]** [3.0163]** [-7.2989]** [33.5430]** Normalized vector 2: mkt = 5.5438 x bnk + 12.2969 x vol - 5.5006 x dummy - 4.4710 x c [6.1430]** [3.0103]** [-5.6302] ** [-5.7858] **

*(**) Denotes rejection of the hypothesis at the 5%(1%) level; dummy represents a dummy variable for Q3 1982, Q2 1983 and Q4 1987; Critical values are provided by Osterwald-Lenum(1992);

t-statistics of the normalized coefficient under the null that the parameter is zero are in parenthesis [ ]

gdp Bnk mkt Vol

α (vector 1) -0.0331 0.0003 0.1399 -0.0283 α (vector 2) 0.0078 0.0032 -0.0324 0.0087

p-value 0.0024 0.5171 0.3295 0.4952

(24)

24

Turning to the weak exogeneity tests, I find that only the GDP vector is endogenous to the system, implying that, in France, causality goes from finance to growth. This result is in line with Arestis and Demetriades’s (1997) expectations in a bank-based financial system. Furthermore, since both measures of financial development are weakly exogenous to the system, nothing can be said regarding the patterns of causality between the two.

5.5 Italy

Table 6 presents the results for Italy. In this case, a dummy variable is introduced to account for the two economic crises Italy passed during the sample period, i.e. during Q1 to Q4 1974 and Q1 1992 to Q4 1993. After the introduction of this dummy, both information criteria suggest a lag length of 1. However, since there is still autocorrelation in the VAR residuals, the lag length is set to 2. The Pantula principle depicted in Panel B shows that the first time the null hypothesis of no cointegrating vector is not rejected is under model 4 for at most one cointegrating vector in case of the maximal-eigenvalue test. Thus, the cointegration test is performed by allowing a trend in the level data and in the cointegrating space.

Panel C shows evidence of a single cointegrating vector at a 5% level. This vector is normalised on gdp, for which a negative and significant error correction coefficient is found. The cointegration relationship shows a positive and highly significant link between both measures of financial development and real gross domestic product. However, it is important to note that the magnitude of the banking system development is six times bigger that the parameter of stock market capitalization. This is consistent with the bank-centred financial system of Italy. Furthermore, volatility appears to exhibit a negative, albeit very small and not significant effect, on economic growth. This result is not surprising in the traditionally considered high-volatility stock market of Italy (Bilio & Pelizzon, 2003).

(25)

25

Table 6 ITALY

0 0.7780 0.9469 1 -10.4379* -9.9312* 34.5992 0.0045 2 -10.3802 -9.5358 19.1263 0.2622 3 -10.3066 -9.1244 38.8141 0.0012 4 -10.2672 -8.7472 29.7186 0.0195 5 -10.2684 -8.4106 22.8654 0.1174 6 -10.3564 -8.1608 20.2478 0.2093 7 -10.2875 -7.7541 30.6350 0.0150 8 -10.3116 -7.4404 10.2872 0.8512

* indicates lag order selected by the criterion; AIC: Akaike information criterion; SC: Schwarz information criterion LM test - Null Hypothesis: no serial correlation at lag order h

λ max test 0 3 29.4602 28.7352 41.9677 1 2 23.8850 23.6547 23.6596* 2 1 14.4528 14.3198 17.9028 3 0 9.8354 9.1983 10.7411 λ trace test 0 3 81.1081 77.4261 100.8538 1 2 51.6479* 48.6909* 58.8861* 2 1 27.7629 25.0362 35.2265 3 0 13.3101 10.7164 17.3237

Null Hypothesis Eigenvalue Trace Statistic Maximal Eigenvalue Statistic None * 0.25282 100.8538 41.9677

At most 1 0.15152 58.8861 23.6596 At most 2 0.11691 35.2265 17.9028 At most 3 0.07188 17.3237 10.7411 At most 4 0.04468 6.5826 6.5826 Normalized vector: gdp = 1.5136 x bnk + 0.2708 x mkt - 0.0012 x vol - 0.6408 x dummy - [4.3130]** [3.8042]** [-0.0172] [-5.4220]** - 0.0123 x trend – 12.0048 x c

[-4.3833]** N.A.

*(**) Denotes rejection of the hypothesis at the 5%(1%) level; dummy represents a dummy variable for 1974Q1-Q4 and 1992Q1-1993Q4;Critical values are provided by Osterwald-Lenum(1992);

α -0.0316 0.1065 0.0936 -0.0054

p-value 0.0130 0.0001 0.2040 0.9658

(26)

26 5.6 United Kingdom

The cointegration analysis and causality tests between financial development and economic growth in the case of UK are reported in Table 7. The LM test in Panel A shows that the first time the null hypothesis of no serial correlation is not rejected is at a lag length of 5, while the information criteria suggest a lag length of 1. In order to accommodate for the condition of empirically Gaussian residuals, I allow for 5 lags in the VECM.

The cointegration test is performed by allowing an intercept in the VECM equations, as well as a dummy variable corresponding to the 1974 stock market crash and the 1987 reforms of the British financial system. Once these structural breaks are taken into account, there is evidence of two cointegrating vectors. On the basis of the sign of the error correction coefficient and the data-identifying restrictions, the vectors are normalized on stock market capitalization and banking system development. Both cointegrating relationships suggest a straightforward positive relationship between banks and stock markets in the UK. Moreover, in the first cointegrating vector, a strong and significant relationship also exists between gdp and mkt. The weak exogeneity tests in Panel D confirm the positive influence that stock market development has on output growth in the UK. Bank development is also weakly exogenous to the system suggesting that, causality also flows from the banking system to economic growth. Thus, for the 1973 to 2009 period, in UK, causality goes from the financial system development to real GDP. The evidence provided here is much stronger than the one found by Arestis et al. (2001), who conclude that for the 1968 to 1997 time frame, the causal link between finance and growth in the UK “is, at best, weak”.

Furthermore, in the UK, volatility is endogenous to the cointegrating system, although it appears to exhibit a positive and significant link to the stock market capitalization ratio in the first vector, and a negative effect on banking system development, in the second.

(27)

27

Table 7 UK

Lag AIC SC LM-Stat Probability

0 -2.5515 -2.3834 1 -13.9116* -13.4073* 30.1208 0.0174 2 -13.7879 -12.9474 33.5035 0.0063 3 -13.7721 -12.5954 53.6016 0.0000 4 -13.7238 -12.2109 42.4061 0.0003 5 -13.6568 -11.8078 11.8467 0.7545 6 -13.5634 -11.3781 12.1802 0.7315 7 -13.4897 -10.9683 6.4741 0.9821 8 -13.3462 -10.4886 29.9050 0.0185 * indicates lag order selected by the criterion; AIC: Akaike information criterion; SC: Schwarz information criterion LM test - Null Hypothesis: no serial correlation at lag order h

None ** 0.2885 110.6323 48.3332

At most 1 * 0.1847 62.2991 28.9898

At most 2 0.1028 33.3093 15.3980

At most 3 0.0723 17.9113 10.6600

At most 4 0.0498 7.2514 7.2514

Normalized vector 1: mkt = 1.6693 x gdp + 1.5429 x bnk + 18.9892 x vol - 24.5599 x c [1.7099]* [4.0615]** [4.7297]** [-2.0161]** Normalized vector 2: bnk= 0.7491 x mkt - 16.6923 x vol + 8.9251 x dummy + 1.9567 x c [4.5736]** [-4.7720]** [6.5973] ** [5.2862] **

*(**) Denotes rejection of the hypothesis at the 5%(1%) level; dummy represents a dummy variable for Q2 1972- Q1 1975 and Q1 1987; Critical values are provided by Osterwald-Lenum(1992);

t-statistics of the normalized coefficients under the null hypothesis that the parameter is zero are in parenthesis [ ]

gdp bnk mkt Vol

α ( vector 1) 0.0083 -0.0000 -0.0087 -0.0184 α ( vector 2) 0.0007 -0.0077 0.0118 -0.0023

p-value 0.0526 0.1724 0.2697 0.0000

(28)

28 5.7 Austria

In the case of Austria both the Akaike and Schwarz information criteria suggest a lag length of 5. Moreover, at this lag length there appears to be no serial correlation or heteroskedasticity in the residuals and the normality test is passed in all error-correction equations, apart from the one corresponding to volatility. The Pantula principle presented in Panel B of Table 8 shows that the first time the null hypothesis is not rejected at a 10% level is under Model 2 for at most 2 cointegrating vectors. Based on these results, the cointegration test is performed with 5 lags in the VECM and an intercept in the cointegrating equation. To accommodate for the residual outliers, I introduce a dummy variable in Q1 1974 and Q1 1987 corresponding to the international oil crisis and stock market crash, respectively. The results, presented in Panel D, suggest two cointegrating vectors in between 1973 and 2009 in Austria. The data-identifying restrictions suggest the exclusion of the stock market variable from the first cointegrating relationship and of volatility from the second one. Hence, the first vector depicts a positive and highly significant relationship between bnk and gdp, while volatility is inversely related to bank development. Similarly, the second vector also shows a positive link between finance and growth and, in line with the bank-based view of the Austrian economy, the impact of bank development on real GDP is significantly higher than the one of stock markets. Furthermore, volatility appears to be negatively linked to output growth in this case.

The weak exogeneity tests in Panel D show that GDP is weakly exogenous to the system, while bank development is not. This implies that, in Austria, economic growth causes in the long run the development of the banking system. However, nothing can be said about the causality between output growth and stock markets since both variable appear to be weakly exogenous to the system. Also, as hypothesized, stock markets have a causal influence on the development of banks in the long run.

5.8. Sweden

(29)

29

Table 8 AUSTRIA

0 -2.2481 -2.1603 1 -13.9151 -13.4761 18.5201 0.2943 2 -14.2039 -13.4138 19.0622 0.2655 3 -14.6973 -13.5560 13.5534 0.6319 4 -16.0844 -14.5919 37.7133 0.0017 5 -16.6269* -14.7833* 21.1523 0.1727 6 -16.5764 -14.3816 20.4767 0.1995 7 -16.4927 -13.9467 25.3782 0.0634 8 -16.5592 -13.6620 21.6591 0.1545 * indicates lag order selected by the criterion; AIC: Akaike information criterion; SC: Schwarz information criterion LM test - Null Hypothesis: no serial correlation at lag order h

None * 0.2320 93.9743 35.6266

At most 1 * 0.1884 58.3477 28.1772 At most 2 0.0995 30.1705 14.1492 At most 3 0.0771 16.0213 10.8243 At most 4 0.0378 5.1970 5.1970 Normalized vector 1: bnk = 1.9561 x gdp - 1.2498 x vol + 15.2734 x dummy - 19.4993 x c [4.2232]** [-2.5152]** [5.5933]** [-3.8254]** Normalized vector 2: gdp= 0.0428 x mkt + 0.5623 x bnk - 5.8594 x dummy + 9.9630 x c [2.0797]** [6.8493]** [-5.3231] ** [59.6925] **

*(**) Denotes rejection of the hypothesis at the 5%(1%) level; dummy represents a dummy variable for Q1 1974, and Q1 1987; Critical values are provided by Osterwald-Lenum(1992);

gdp Bnk Mkt Vol

α ( vector 1) -0.0009 -0.0018 0.0032 0.0016 α ( vector 2) -0.0008 0.0054 0.0126 -0.0010

p-value 0.6476 0.0029 0.5814 0.7822

(30)

30

stock market capitalization to GDP and output growth, while volatility also influences positively the development of financial markets. The causality tests, in Panel D also confirm the market-based view, since stock market capitalization is endogenous to the system and real GDP is exogenous. Thus, in Sweden causality goes from growth to finance. It is interesting to notice that banking system development is also weakly exogenous to the system, which implies that there appears to be a long run causality going from banks to stock markets.

6. Robustness checks

This section presents some robustness checks performed with alternative measures of financial development, as well as alternative estimation techniques. Firstly, I check the sensitivity of my results to alternative measures of banking system and stock market development. King and Levine (1993) and Rousseau and Wachtel (2000), among others, consider the ratios of broad money (M2) and stock of liquid liabilities (M3) to GDP as measures of intermediary activity able to capture the notion of “financial depth”. Rousseau and Wachtel (2000) also find that, in cross-sectional specifications, the ratio of value traded to GDP has a better explanatory power than stock market capitalization. As a result, I construct a dataset for a sub-sample of countries for which data on these alternative measures is available for sufficiently long periods of time and repeat the empirical analysis in sections 4 and 5. The results are presented in Appendix C.1 for M2/M3 and C.2 for value traded.

Overall, the results are broadly similar to the ones obtained with the original variables. However, in some cases, due to shorter samples of data, the results can only be viewed as indicative. For The Netherlands, data on M2 was obtained for the full period Q1 1977 to Q4 2009. Panel A in Appendix C.1 shows that there is still evidence of two cointegrating vectors, with the link between GDP and the alternative bank development variable being quantitatively larger than the one obtained in section 5.2. More importantly, the causality test shows identical patterns stressing once again the finance to growth causality in the case of The Netherlands.

(31)

31

Table 9 SWEDEN

0 -1.9267 -1.7059 1 -11.8570 -11.1948 18.2282 0.3107 2 -11.9070 -10.8033 19.6445 0.2367 3 -12.1529 -10.6078 21.5070 0.1598 4 -13.9392 -11.9526 19.3733 0.2498 5 -14.4775 -12.0494* 19.7430 0.2320 6 -14.5204* -11.6509 7.0650 0.9720 7 -14.3405 -11.0295 12.9369 0.6774 8 -14.3120 -10.5595 18.3096 0.3061 * indicates lag order selected by the criterion; AIC: Akaike information criterion; SC: Schwarz information criterion LM test - Null Hypothesis: no serial correlation at lag order h

H0 R n-r Model 2 Model 3 Model 4

λ max test 0 3 35.1082 33.5138 53.3932 1 2 30.3744 21.7776 23.2230 2 1 17.0963* 16.4702 19.0421 3 0 9.1283 5.5140 16.4636 λ trace test 0 3 92.4128 77.3925 114.8613 1 2 57.3046 43.8788 61.4681 2 1 26.9302* 22.1011 38.2451 3 0 9.8339 5.6309 19.2030

At most 1 * 0.2737 57.3046 30.3744 At most 2 0.1647 26.9302 17.0963 At most 3 0.0916 9.8339 9.1283 At most 4 0.0074 0.7055 0.7055 Normalized vector 1: gdp = 0.0922 x bnk + 0.1594 x mkt - 0.8135 x dummy + 13.1094 x c [4.0929]** [15.4748]** [-5.7286]** [-347.889]** Normalized vector 2: mkt= 5.7540 x gdp + 0.7301 x vol + 6.4826 x dummy + 76.2367 x c [9.5001]** [1.7261]* [6.0288] ** [9.5316] **

*(**) Denotes rejection of the hypothesis at the 5%(1%) level; dummy represents a dummy variable for Q4 1987, Q4 1989, Q2-Q3 1996, Q2-Q3 1999; Critical values are provided by Osterwald-Lenum(1992);

gdp bnk Mkt vol

α ( vector 1) -0.0555 -0.0058 0.8794 0.2048 α ( vector 2) 0.0189 0.0124 -0.2992 -0.0636

p-value 0.3440 0.6528 0.0325 0.0183

STOCK MARKETS, BANKS AND ECONOMIC GROWTH

UNIVERSITY OF GRONINGEN

Faculty of Economics and Business

MSc Business Administration, Specialization Finance

STOCK MARKETS, BANKS AND ECONOMIC

GROWTH

Supervising professor:

Prof. Dr. Kasper Roszbach

Author:

Oana Peia (s1942743)

2

Abstract:

This paper investigates the empirical relationship between financial development and

economic growth for eight developed European countries. Using time series data within a

VECM framework, I find evidence of a robust long-term link between both stock markets and

banks and real GDP growth in all the countries analysed. The causality of this relationship is

examined through weak exogeneity tests. The results suggest that causality patterns differ

with the type of financial institution analysed: in most countries, stock market development

causes, in the long run, economic growth, while the relationship between banking sector

development and real GDP growth appears to be bi-directional.

3

Table of contents:

1. Introduction

2. Literature review

2.1 Theoretical considerations

2.2 Empirical evidence

3. Variables, Data and Hypotheses

3.1 Variable definitions

3.2 Data sources

3.3 Hypotheses

4. Methodology

5. Results

5.1 Germany

5.2 The Netherlands

5.3 Belgium

5.4 France

5.5 Italy

4

1. Introduction

5

6

2. Literature review

2.1 Theoretical considerations

7

2.2 Empirical evidence

8

-9

10

3. Variables, Data and Hypotheses

3.1 Variable definitions

3.2 Data sources

11

3.3 Hypotheses

12

13

4. Methodology

14

, (1)

, (2)

H

:

0

(3)

) (4)

y

y

15

16

5.

Results

5.1 Germany

17

18

19

20

5.3 Belgium

21

22

5.4 France

23