University of Amsterdam
Amsterdam Business School
Master of Science, Business Economics
Finance Track
The Contribution of Financial Integration on Credit Growth in
the European Union
Pieter Willem Theodorus Huijink
10004648
06-07-15
Abstract
This thesis examines the relationship between financial integration and credit growth in the European Union, compares credit growth in the EU to other countries and analyses the association of credit growth with the crisis. We employ VECM techniques on an individual country level and on a panel level to examine the relationship and adjustment patterns between the two variables. The results show that there is no uniform relationship between financial integration and credit growth among EU members. The individual VECM analysis on the EU15 countries showed that there are differences across the countries when it comes to the existence of a cointegrating relationship and the adjustment patterns. In the majority of the cases, nine out of fifteen, credit adjusted after deviations from the long-run equilibrium between the two. The panel VECM indicates that financial integration responds to deviations from the long-run equilibrium. Using a propensity score matching analysis, we find that credit growth is significantly higher for EU members than for comparable non-EU member countries. These results suggest that higher credit growth may be related to membership of the European Union, or by common factors among the EU members. However, whether financial integration is the leading factor in that is doubtful, when we observe that there is no uniform relationship between these two variables across the EU15. With relation to the effect of credit growth on stock prices during the crisis, our methodology does not find any significant effect.
Statement of Originality
This document is written by Student Pieter Willem Theodorus Huijink who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of Contents
1. Introduction 4
2. Literature Review 7
2.1. International Financial Integration 7
2.2. Financial Integration in the European Union 9
2.3. Financial Integration and Credit Growth 11
3. Methodology and Data 13
3.1. Methodology 13
3.2. Financial Integration Measure 15
3.3. Propensity Score Matching 18
3.4. Data 21
3.5. The Financial Crisis 24
4. Results 26
4.1. Vector Error Correction Model Results 26
4.2. Robustness and Validity 36
4.3. Comparison of EU Members to Non-EU Members 39
5. Discussion 42
5.1. Economic Implications 42
5.2. The Financial Crisis 43
6. Conclusion 46
Reference List 48
1. Introduction
The past decades have seen a strong increase in globalization and financial integration across the world (Bai and Zhang, 2012). The European Union is a prime example of this. The 1 member states have gone through a process of financial liberalization as well as economic and monetary integration (Gehringer, 2013). The region is increasingly connected on various levels and an important aspect of this increased integration has been the integration of financial markets (Baele et al., 2004).
International financial integration, sometimes called financial globalization, is a process with benefits and risk. Financial integration allows investors to spread risks, it allows capital to be allocated to where it is most productive and it allows countries to exercise their competitive advantages (Stulz, 2005). However, we have witnessed the risks of financial integration in the financial crisis (Lane, 2013). The contagion due to the interdependence between markets has led to the spread of the financial crisis of 2008, which started as a subprime-mortgage crisis in the United States and resulted in a global financial crisis. Additionally, financial integration has led to asymmetries in credit growth and external positions between countries, which left some countries vulnerable when the crisis hit. These imbalances are mostly due to differences in financial market development (Mendoza, Quadrini and Ríos-Rull, 2009).
The financial crisis was preceded by a period of strong credit growth, as it has been the case for previous crises (Aikman, Haldane and Nelson, 2014; Reinhart and Rogoff, 2011). The relationship between financial integration and credit growth, and the possible influence of this relationship on the emergence of crises is shown schematically in Figure 1 (see Frost and Van Tilburg, 2014). This thesis focuses on the green line.
The relationship between financial integration and credit growth is not necessarily the same for members states of the European Union compared to the rest of the world. In the European Union, financial integration is part of a much broader integration process (Gehringer, 2013). This process can arguably amplify the relationship between financial integration and credit growth. The increased likelihood of cross-national operations of
Bai and Zhang (2012) measure financial integration using the world asset-output ratio. For their full
1
sample of 43 countries, this ratio increased from 8% in the 1970-1986 period to 18% in the 1987-2004 period.
!
companies in the region and the stronger links between regulatory institutions can further open channels through which financial integration facilitates the growth of credit. At the same time, the observable effect of financial integration may be marginalized by the other aspects of economic integration. If these other aspects of integration are stronger drivers of credit growth, financial integration may not play as big a role in determining credit growth as it does in other countries.
Understanding the relationship between financial integration and credit growth is important for (international) financial regulation, especially on the macro-prudential side. Looking at this from a European Union perspective provides valuable insight into the mechanisms of financial integration, due to the overall economic integration process. If we observe differences in credit growth between the European Union and other countries, we may need to pay more attention to the effects of integration in the European Union. For regulators and policy-makers alike, research into this topic helps decide on the need for measures and regulations focused on financial integration.
This thesis aims to understand how financial integration in the European Union has influenced credit growth within the European Union, and how this credit growth is different from other countries globally. The specific research question that is to be analyzed is: How did financial integration contribute to credit growth within the European Union?
The results of this thesis show that there is no uniform relationship between financial integration and credit growth among EU members. We use a vector error correction model on an individual country level and in a panel dimension, to analyze the relationship between the two variables and the adjustment patterns. The individual VECM analysis on the EU15 countries showed that there are differences across the countries when it comes to the existence of a cointegrating relationship and the adjustment patterns. The panel VECM indicates that financial integration responds to deviations from the long-run equilibrium
International Capital Flows
Domestic Factors
Private Credit Growth
Banking Crisis
Level of Private Credit Figure 1 - Overview of Channels between financial integration, credit growth and crises. (see Frost and Van Tilburg, 2014).
between Credit to GDP and financial integration, which leads to an adjustment back to the equilibrium. The propensity score matching analysis finds that credit growth is significantly higher for EU members than for comparable non-EU member countries. These results suggest that higher credit growth may be related to membership of the European Union, or by common factors among the EU members. However, whether financial integration is the leading factor in that is doubtful, when we observe that there is no uniform relationship between these two variables across the EU15. With relation to the effect of credit growth on the crisis, our methodology does not find any significant effect.
The thesis is structured as follows. In section 2, we analyze the existing literature. In section 3, we introduce the methodology and data sources. The results and robustness checks are found in section 4. In section 5, we will discuss the results and provide an extension to the recent financial crisis. The thesis concludes in section 6.
2. Literature Review
This thesis is related to various paths in the literature. First, it relates to literature on the consequences of financial integration and the links between financial integration and the credit cycle. Second, it is connected to the literature on the integration across the European Union. Third, it is part of the literature covering credit growth and its determinants.
2.1. International Financial Integration
There is extensive literature on the process and the consequences of financial integration, both for developed and developing countries. To start, financial integration can be defined across several markets, as Figure 2 shows. The general definition of financial integration is the liberalization of cross-border trade in financial assets (Stulz, 2005). However, there is a large variety of financial assets. This means that financial integration can imply liberalization of trade across the money, corporate-bond, government-bond, credit and equity markets (Baele et al., 2004).
Another problem concerns measuring financial trade liberalization. As Bai and Zhang (2012) argue, the removal of official trade barriers does not imply that no barriers remain. They study a dynamic stochastic general equilibrium model and find that implicit barriers still exist, focused mainly on the default risk on sovereign debt. This default risk prevents countries from sharing risks through endogenous constraints on borrowing. This argument is consistent with Kose et al. (2009), who describe the difference between the ‘de jure’ and the ‘de facto’ integration; actual integration may be different from the official lack of barriers.
Figure 2 - Definitions of Financial Integration (see Baele et al., 2004; Kose et al., 2009) Markets Money Corporate-bond Government-bond Credit Equity Type of Integration
De Jure / Official Financial Liberalization
The problem of consistent measures of financial integration is also discussed by Eichengreen (2001). He argues that even when we separate liberalization measures from actual capital flow measures, the used variables can have different interpretations. For instance, measures of financial liberalization can be continuous variables, or binary. The latter does not measure the intensity of controls, which is often a difficult task. Recent contributions that improve on measuring the intensity include the KAOPEN measure by Chinn and Ito (2006). These inconsistencies of measures exist across the field, and complicate a comparison of the literature.
Kose et al. (2009) composed a literature review on the topic, and do find evidence that financial integration can lead to higher growth and lower volatility. The results across the literature are mixed, which Kose et al. (2009) find is mainly due to the wide differences in the strength of supporting institutions (see also Edison et al., 2002). These institutions are organizations that enforce good governance and rule of law. Additionally, the macroeconomic data used in empirical literature makes it difficult to draw any definitive conclusions on the direction of the effect. The endogeneity problem between financial integration and growth can not be dealt with perfectly unless microeconomic data is used, in which case we would have more tools to address this problem. We will analyze a few empirical studies on the effect of financial integration on growth, look at the used methods and interpretations.
Quinn (1997) performs a multivariate regression analysis to show the relationships between financial integration and economic outcomes, using a de jure measure. He finds that financial integration is associated with economic growth. Bekaert, Harvey and Lundblad (2001) focus on emerging financial markets, using a de jure indicator for financial liberalization. They conclude that financial liberalizations are associated with increases in real economic growth in emerging markets. Bekaert, Harvey and Lundblad (2005) provide evidence that financial liberalization is associated with economic growth, now on a larger sample scale. Although their regressions have a predictive nature, they do not provide conclusive evidence on the direction of the effect.
A direct effect was established by Klein and Olivei (2008), who examine the effect of open capital accounts on financial depth and through that channel on economic growth. In their analysis, they use a simultaneous equation model. One interesting finding from their results is that the effect is mainly driven by developed countries in the sample. Gamra (2009)
provides a more extensive analysis on the empirical relation, focused on the East Asian region. His conclusion is that the effect of financial liberalization on growth is determined by the intensity of the liberalization. Specifically, partial liberalization is associated with higher economic growth, while full liberalization actually leads to lower economic growth.
There is a considerable group of literature that argues against an effect of financial integration on economic growth, for instance Rodrik and Subramanian (2009). Their main argument against a positive effect focuses on developing countries, arguing that these face investment constraints. This constraint leads to an appreciation of the real exchange rate when foreign finance flows in. This has adverse effects on economic growth, mainly through the traded goods sector. Although they do not provide substantial empirical evidence themselves, they argue that the evidence in favor of financial integration is unpersuasive and relies on indirect and speculative arguments. Arcand, Berkes and Panizza (2012) extend on the argument of Rodrik and Subramanian (2009) by showing that financial development can have negative contributions to economic growth above a certain threshold. Similarly, Kose, Prasad and Taylor (2011) show that the benefits of financial development can only be obtained above a threshold. They also argue that the thresholds are different across different types of asset classes and capital flows. This further complicates a conclusive discussion on the benefits and costs of financial integration on economic outcomes.
2.2. Financial Integration in the European Union
The increased financial integration within the European Union has been analyzed most intensively in the past 15 years, since the introduction of the Euro. Baele et al. (2004) examine the level of integration across several markets. Their basis of examining integration focuses on the law of one price, which argues that integrated markets with the same currency should provide the same price. They find that price differentials still exist. Of all the markets listed in Figure 2, Baele et al. (2004) find that the money market of the European Union is the most integrated and there is little integration in the credit market. Some of the main reasons for this include a home bias from investors and a fragmented consumer credit segment. It is important that the paper uses different measures for integration across various markets, ranging from price-based, quantity-based to news-based measures. The findings of Baele et
al. (2004) are consistent with the argument of Bai and Zhang (2012) on incomplete liberalization through implicit barriers. The inequality of integration across markets in the European Union is also a topic of discussion in the paper by De Guevara, Maudos and Pérez (2007), who find that retail markets were more integrated compared to wholesale markets.
Fratzscher (2002) focuses his analysis on the integration of stock markets in the European Union, comparing both time-invariant and time-varying integration using the uncovered interest rate parity as indicator for financial integration. He finds that stock markets are only highly integrated since 1996 and that this process has been led by the drive towards the monetary union. This drive has decreased exchange rate risk and led to monetary policy convergence. The conclusions are supported by Bartram, Taylor and Wang (2007). In their analysis, the increased integration in the stock markets is produced by the intention to join the monetary union. For this reason, they also find increased integration for the UK and Sweden, countries that were then expected to join the monetary union in due course. There is some discussion on whether the actual adoption of the Euro has contributed to increased financial integration. Bekaert et al. (2013) for instance find that this is not the case. They make the case that financial market integration was already at a point where the introduction of a single currency will have a limited contribution. Their assessment of the literature that does find an effect of the introduction of the euro is that those results most likely capture the effect of the European Union rather than the monetary union specifically.
We also examine literature on the consequences of this financial integration for the European Union specifically. The paper by Schmitz and Von Hagen (2011) provides a good starting point in this discussion relevant to this thesis. Specifically, they focus on current account balances of the members of the monetary union compared to non-members. They use trade balances against other members of the monetary union and external trade balances to the rest of the world. They find that capital flows follow differences in capital endowments, and that this effect is stronger for members of the monetary union. Schmitz and Von Hagen (2011) do not find an effect of the introduction of the monetary union on current account balances against the rest of the world. They conclude that the increased financial integration leads to a more efficient allocation of capital within the monetary union and see their result as a positive argument for increased financial integration.
Guiso et al. (2004) run several simulations to estimate the effects of financial integration on the European Union. In this analysis, they use several scenarios, ranging from integration comparable to the United States to a level of integration below the highest European Union level. They find that financial integration affects the European economy through financial development and the subsequent increase of sales of, especially, small and medium sized firms.
2.3. Financial Integration and Credit Growth
The literature on the credit cycle is relatively limited compared to research on the business cycle. Since the recent financial crisis, it is commonly part of a broader discussion on macro-prudential policy. The main conclusion on the workings of the credit cycle can be found in the paper by Kiyotaki and Moore (1997). Their model shows how credit limits amplify and lengthen the impact of a shock on asset prices. The mechanism essentially works as a pro-cyclical amplifier of the business cycle. With regard to the interaction between the credit cycle and crises, we have noted that financial crises are preceded by credit booms (Aikman, Haldane and Nelson, 2014; Reinhart and Rogoff, 2011). Jordà, Schularick and Taylor (2011) find that credit growth is the best predictor of financial instability, based on a study of 140 years across 14 developed countries. This analysis also looked at other factors, such as asset prices and imbalances. It argues that the imbalances and credit growth are linked too, but especially in the period before the second world war when the level of financialization was lower.
There are several mechanisms for the connection between financial integration and credit growth mentioned in the literature. Lane (2013) mentions this issue briefly, arguing that financial integration works on both supply and demand factors of credit growth. Credit supply is affected since banks, both domestic and foreign, can raise wholesale funding on international markets and increase bank equity through foreign investors. Credit demand is affected through the low interest rate environment created by capital inflows. This improves the net worth of domestic borrowers, as it increases domestic asset prices. Lane and McQuade (2012) establish that there is a correlation between capital flows and credit growth for the period 2003 to 2008. They do not conclude on the mechanism, but suggest that
international funding activities and the interplay of capital flows and macroeconomic variables can explain the results. They also emphasize that financial integration affects supply and demand factors, similar to Lane (2013). Using an analysis of 70 credit booms over the period 1960-2010, Mendoza and Terrones (2012) establish that credit booms often follow surges of capital inflows.
Integration within the European Union was not limited to financial markets only. The most visible aspect of integration has happened in international trade. Chen (2004) analyses this aspect and finds that technical barriers between countries matter, unlike non-tariff barriers, and that further market integration within the European Union leads to more trade between the countries. This paper is part of broader literature on so-called border effects. Another side of the economy where integration is expected to happen is on the labour market. However, Bartz and Fuchs-Schündeln (2012) find that although formal barriers have been lifted to a large extent, there are still significant limits to labour mobility. Their analysis is built on 21 years of data on the EU15 countries. They attribute the results to cultural and language aspects, rather than physical border controls.
Based on the previous paragraph, it is sensible to look further into international trade and specifically, whether it might have an effect on credit growth. There is literature that suggests an interaction, and most of the literature points to a direction from credit to international trade. Chor and Manova (2012) for instance, look at the effect of credit constraints on international trade during the recent financial crisis. They find that credit constraints were an important determinant of the collapse in international trade. Manova (2013) identifies three mechanisms through which credit constraints affect international trade, using a heterogeneous firm model. The identified mechanisms are the selection of firms into domestic production, the selection of domestic firms into exporting, and the level of firm exports. She establishes a direct effect by exploiting variation in financial development across countries and variation in financial vulnerability across sectors. There is no evidence in the related literature to suggest that international trade affects credit growth, domestically or internationally. This implies, in combination with the findings of Lane (2013), that financial integration may affect international trade through credit growth, or specifically by lowering credit constraints. Although this thesis does not test this aspect specifically, it is important to keep in mind different potential channels that could drive the result.
3. Methodology and Data
3.1. Methodology
Based on the literature, we have seen that financial integration may have an influence on economic outcomes, that financial integration is advanced in the European Union and that it influences the economies of the member states. We have also seen the consequences of credit growth, most importantly how it precedes financial crises. The question that has not been answered is how financial integration has affected credit growth in the European Union.
This thesis will answer the research question: How did financial integration contribute to credit growth within the European Union? The hypotheses of this thesis are that
1. Financial integration and credit growth are related.
2. Financial integration in the European Union resulted in higher credit growth in that region.
Answering this research question requires several steps. First, we need to estimate how financial integration and credit growth are related in the European Union. Second, we need to assess how the result for EU members differs from the result obtained in other countries, ceteris paribus except for the membership of the European Union.
We focus on the first hypothesis by using a vector error correction model for the EU15 countries separately and as a panel vector error correction model. Since both methods answer the research question from a different perspective, we implement both in this thesis.
Before we can estimate this model, we first have to see which of the time series of the various countries contain a unit root, using an Augmented Dickey-Fuller test. We will then perform the Johansen (1991) trace test to see which of the series are cointegrated. The test describes at what number of cointegrating relationships we can reject the null hypothesis that there is no cointegration. In our setup, the maximum number of cointegrating relationships is one. The VECM is given by equations (1) and (2).
(1) (2) ΔCt =βC+ βCC ,k k=1 p
∑
ΔCt−k βCF ,kΔFt−k k=1 p∑
+λC(
Ct−1−α0−α1Ft−1)
+ utC ΔFt =βF + βFC ,k k=1 p∑
ΔCt−k βFF ,kΔFt−k k=1 p∑
+λF(
Ct−1−α0−α1Ft−1)
+ utFThe system as provided in equations (1) and (2) is the representation of one cointegrating relationship with p lags. C is the Credit to GDP ratio. The change in the credit to GDP ratio can be regarded as credit growth. In this setup it is modeled as the response of the credit level on the short term, the lagged variables, and the long run, the error correction term. The financial integration measure F is defined following Bai and Zhang (2012) and Edison et al. (2002). The second equation thus models the response of financial integration to lagged changes in either of the two variables, and the error correction term. We will discuss the choices for the measure extensively in section 3.2. The VECM normally uses the number of lags suggested for the corresponding VAR by the AIC minus one. Due to the limited number of time periods in our sample, we will restrict the maximum number of lags to one when we estimate the VECM for separate countries. If the eigenvalue stability condition is not fulfilled with this lag, we remove the lag. We are then only able to estimate the coefficient for the error correction term, while we are unable to include lagged first differences. The stability condition says that the estimation has one unit modulus, imposed by the VECM, and all other moduli should be below one. We also estimate the model in a panel setting, in which case we can use more lags and will choose the number of lags according to the AIC. In the panel setting, we standardize the values of both variables by country.
In the VECM setting, we can identify short and long run responses. Specifically, the betas measure short run responses of the variable to the lagged changes. The lambda is the error correction parameter that measures the response to deviations from the long run equilibrium. The long-run cointegrating relationship is given by equation (3).
(3)
The estimated lambda should ensure a change in the variables to return to this equality. That means that past values of C and F and the equilibrium between the two help to determine the changes in these variables in the future.
The VECM analysis allows us to look at the relationship between financial integration and credit growth within the countries of the European Union, individually or as a group in the panel setting. However, we ideally would like to compare credit growth in the EU to credit growth outside of the EU too. To address this issue, we will use a second method,
being the estimation of the average treatment effect on the treated. Using propensity score matching, we can create a group of non-EU members with comparable characteristics concerning the determinants of credit growth. We can use this method to estimate the treatment effect of being a member of the European Union. The implementation of the propensity score matching technique will be explained in section 3.3., and the estimation of the treatment effect is given by equation (4).
(4) Note that this is the average treatment effect on the treated, or the treatment effect on the countries that actually received the treatment (in this case the EU members). This treatment effect is conditional on the propensity score! . The estimation in equation (4) balances
the control and treatment group. However, it balances on a finite number of chosen covariates, therefore almost certainly missing some information that creates a more accurate comparison group. This means we should always treat the results of this estimation with caution. We will perform two robustness checks on this estimation. First, we leave out one of the covariates sequentially. Second, we include the EU members that joined after 1995.
3.2. Financial Integration Measure
Bai and Zhang (2012) argue for the use of an explicit openness measure. The alternative would be a qualitative measure using financial regulation. However, this will measure official liberalization rather than actual integration, following the argument of Baele et al. (2004) on the limited financial integration across some markets due to non-regulatory factors. This argument is also in line with the evidence found in Lane and McQuade (2012). The openness measure used in the paper by Bai and Zhang (2012) is unsuitable for our analysis, since it measures the world asset to output ratio and as a result, aggregates financial integration for all countries. Using this measure on individual countries is not feasible.
To construct the measure for this thesis, we turn to the paper by Edison et al. (2002). They list six measures of integration, as well as the data associated with it. These measures are listed in Figure 3. The first two measures, IMF restrictions and the Quinn Measure, are
p( Xit)
qualitative measures and will not be used, following the previously mentioned arguments. The other four measures are considered as data is available. This is the case for the third and the fifth measure. The other two measures require access to the International Financial Statistics database. We do not examine measures that only capture FDI, following the argument of Quin, Schindler and Toyoda (2011) that FDI definitions are inconsistent. This creates a measurement bias, which becomes more problematic when the indicator relies solely on FDI. The two measures we will consider are described in equations (5) and (6).
(5)
(6)
Where SCF is Stock of Capital Flows, SCI is Stock of Capital Inflows. FDI and FDO represent foreign direct investment inflow and outflow respectively. PI and PO represent portfolio inflows and portfolio outflows respectively.
The reason why we use Stock of Capital Flows follows the argument of Edison et al. (2002, p.753) that openness includes the ability of foreigners to invest in a country and the ability of residents to invest abroad. The Stock of Capital Inflows variable can be used because capital inflow is deemed important for the development of countries. This means that the SCI indicator is different from the SCF indicator by focusing more on the development of the country. Edison et al. (2002) have examined the components of the indicators separately
SCFit = FDIit+ PIit+ FDOit+ POit
GDPit
SCIit = FDIit+ PIit
GDPit
Figure 3 - Financial Integration Measures (Edison et al., 2002) Financial Integration Measure
1. IMF Restrictions Qualitative measure of the existence of restriction on
capital account transactions.
2. Quinn Measure Qualitative measure of the intensity of restrictions on
capital inflows and outflows.
3. Stock of Capital Flows Quantitative measure of the sum of FDI and portfolio
inflows and outflows as a share of GDP.
4. Flow of Capital Quantitative measure of FDI and portfolio inflows
and outflows as a share of GDP.
5. Stock of Capital Inflows Quantitative measure of the sum of FDI and portfolio
inflows as a share of GDP.
6. Inflows of Capital Quantitative measure of FDI and portfolio inflows as
and find, for both indicators used in this study, that the components give similar results to the aggregated stock measure.
In this thesis, we will use the Stock of Capital Flows as primary measure, the measure described in equation (5). The Stock of Capital Flows measure allows for both domestic and foreign investments by residents and foreigners. Although Stock of Capital Inflows can be a useful measure when looking at developing or emerging countries, financial integration of EU member states is not likely to be focused on the prospect of incoming flows to stimulate economic development. The more likely argument is that financial integration for developed countries revolves around risk sharing through a combination of incoming and outgoing flows.
As a robustness check to this approach, we will analyze the results using Stock of Capital Inflows and an index of regulatory restrictions on capital movements. Several databases on this topic exist, the most appropriate for our analysis being the database described in Chinn and Ito (2006) (see also Chinn and Ito, 2008). It is an index based on restrictions on cross-border financial transactions as described by the IMF, as such related to the first measure described by Edison et al. (2002). This KAOPEN index covers 182 countries for the period of 1970 to 2013. The interpretation of this analysis will be slightly different compared to the original analysis. The KAOPEN index measures capital account liberalization, rather than actual financial globalization. The latter is reflected by our two original measures. A higher value on the KAOPEN index implies a higher level of financial integration for a country. Note that this is not the only indicator available, but there are some advantages for this indicator compared to others.
Satyanath and Berger (2007) note that this measure is preferable to many others, because the resulting continuous variable captures the intensity of capital controls, rather than just the presence or absence of them. Karcher and Steinberg (2013) discuss the index more
Table 1 - Correlation between the Financial Integration Indicators
SCF SCI KAOPEN
SCF 1
SCI 0.9933 1
KAOPEN 0.2104 0.1828 1
Note: Based on 2741 observations for SCF with SCI, 2571 observations for KAOPEN with SCF and 1578 observations for KAOPEN with SCI.
extensively and argue that KAOPEN is more suitable as a measure of long term capital account restrictions. The methodology used to construct the index fails to capture policy changes in a short period of time, and hence suffers from measurement error. This implies that it is difficult to draw definitive conclusions on the direction of the effect using this index. Quinn, Schindler and Toyoda (2011) assess a wide range of measures used for financial integration. Their discussion concludes that qualitative measures usually provide similar information, since they are often based on the same IMF statistics. The main advantage of KAOPEN according to them is the broad country and time coverage, public availability and the fact that it is a continuous variable.
To see how the three indicators are related to each other, the correlations between the three are given in Table 1. Stock of Capital Flows and Stock of Capital Inflows are highly correlated. This is not the case for the KAOPEN measure, although the correlation is positive. This does not mean that either of the measures is not valid, but it is remarkable and an indication that results may be different.
3.3. Propensity Score Matching
For the part in which we analyze the effect of EU membership on credit growth, we need to build a comparison group consisting of non-EU members. Stated differently, if the results end up being significant, this effect should be due to the membership of the European Union only and not due to a difference in the common trend between members and the control group. That means that the countries in the control group should be truly comparable in characteristics. The method of achieving this is to select countries in the dataset by using propensity score matching. This matching method is useful since this thesis concerns an observational study and the assignment to treatment and control group can not be made random in a similar way as an experimental situation. Propensity score matching provides a solution in such cases (Dehejia and Wahba, 2002). It essentially assigns a propensity score ! , which measures the probability of receiving treatment based on covariates
(Rosenbaum and Rubin, 1983). The propensity score is given by equation (7).
(7) p( Xit)
This propensity score can be used to match each treated observation to an observation that is (as close to) equally likely to have received treatment, conditional on the covariates used to calculate the propensity score. The covariates on which the observations will be 2 selected are listed in Figure 4. These variables define the economic state of the countries and their business cycle. The propensity is estimated using a logistics regression.
The covariates were chosen by following Djankov, McLiesh and Schleifer (2007) in their analysis on determinants of private credit. Since the EU countries are assumed to have an income above the worlds median level, the most relevant control variables are GDP and Inflation. To generate a variable that measures income development, we transform the GDP variable into a relative measure. We use a country’s GDP per capita divided by GDP per Capita in the United States. Because the EU has gone through a process of trade integration, it is also relevant to use a measure of trade as a control variable. This measure is imports and exports combined, as a share of GDP. Djankov, McLiesh and Schleifer (2007) did not specifically look at real estate prices, while a long-term relationship between the private credit and real estate prices has been established in the literature (see for instance Hofmann, 2001). This is probably due to a lack of reliable and consistent data on property prices across global samples. Data on comparing house prices across countries is complicated by the fact that types of houses and the region in which they are located are not directly comparable, especially on a national level. Some countries have relatively more urban areas, the houses can be of different sizes or quality or the rental market can be more (or less) important in the total housing market. In this thesis, we will use the real interest rate as a control variable
Adjusting for differences in the propensity score has been shown to remove biases from the
2
estimation. However, it is not always as efficient as adjusting for differences in the covariates. Hirano, Imbens and Ridder (2003) discuss this in more depth and propose using the inverse of a nonparametric estimate of the propensity score rather than the actual propensity score.
Figure 4 - Selection Covariates for Propensity Score Matching Covariates
Income Development (GDP per Capita relative to the
US) Measures country wealth
Inflation Measures changes in domestic price levels
Trade (Exports and Imports, Share of GDP) Measures trade openness
Real Interest Rate Indicator of lending restrictions
Table 2 - Estimation of the Propensity Score
This estimation denotes the results of the logistic likelihood regression used to calculate the propensity score. The sample is global, and the regression measures the likelihood of having a value of 1 in the EU dummy variable. A surprising outcome is the negative significant sign for Trade, which indicates that a higher sum of exports and imports relative to GDP is associated with a lower probability of being an EU member, keeping the other variables constant.
Dependent variable: EU Dummy Variable
Income Development 2.6363*** (0.3247) Inflation - 0.2207*** (0.0375) Trade - 0.0052*** (0.0018)
Real Interest Rate - 0.1589***
(0.0326) Capital Formation - 0.0668*** (0.0200) Constant 0.2036 (0.6105) Pseudo R2 0.3943 Observations 1935
Notes: Standard errors in parentheses. ***, ** and * denote significance levels of 1%, 5% and 10% respectively. Reported R2 is the Pseudo R2.
Figure 5 - Propensity Score Distribution
0 10 20 30 40 D e n si ty 0 .2 .4 .6 .8 Propensity Score
instead, following Hofmann (2001). This variable acts as an imperfect proxy for the 3 constraints on lending to households, which will be an indication of the booms and busts in the housing market. The interpretation of this variable has to be approached with caution, since the proxy is rough. Credit demand by companies is measured by gross capital formation, also called gross domestic investment. This variable measures additions to fixed assets and inventories as a share of GDP. Other, more direct indicators would be preferable, but data availability limits this. For example, trade credit applications are likely to be in line with overall credit demand by firms. However, the University of Amsterdam does not have a subscription to such data. A new database by the World Bank on the ease of doing business contains data on information provision about credit and the quality of institutions, which would be preferable. Unfortunately, the dataset does not cover even half the sample period of this thesis.
First, we calculate the propensity score for all countries in the sample using a logistic likelihood regression. The results of this regression are given in Table 2. All coefficients in this analysis are significant. The distribution of the propensity scores across EU and non-EU members is shown in Figure 5. Note that this represents the unmatched sample for the non-EU members. We see that the mean of the distribution is higher for non-EU members, which is what we would expect. We also see that most observations show very low similarity to the EU observations. Based on this, we expect a big difference between the unmatched and matched estimations of the treatment effect.
3.4. Data
Credit data is widely available and in various definitions, in terms of sources of credit (all financial institutions or specifically banks) and destination of credit (governments or private institutions). For this analysis, the credit data will be taken from the World Bank database, following the definition of credit to the private sector by banks and other financial institutions as a share of GDP. For the VECM analysis, it was necessary to interpolate the data for the Credit to GDP ratio. In this thesis, we use linear interpolation.
The real interest rate is the lending rate corrected for inflation, which is especially important since
3
The data needed to construct the financial integration measure is obtained from the dataset constructed by Lane and Milesi-Ferretti (2007). The dataset was constructed for the paper by Lane and Milesi-Ferretti (2001). This dataset is continuously being updated, also since the paper of 2007 was published. It now covers the years 1970 to 2011 for 188 countries. All data is denoted in millions of US dollars. 4
The covariates are taken from the World Bank database. GDP data, inflation, exports, imports, the real interest rate and gross capital formation can be found in the World Bank Development Indicators. Inflation is defined as the annual GDP deflator, a more comprehensive inflation indicator than purely inflation of consumer prices. Export and imports are defined as the exports and imports of goods and services as a share of GDP. These include exports and imports to all other countries in the world and as a result, measure total trade integration rather than trade integration within the EU alone. This consistent definition does allow for comparison with the global sample. Gross capital formation is defined as a share of GDP.
The European Union consisted of 15 members from 1995 until 2004, as listed in Appendix 1. This thesis will analyze data ranging from 1995 to 2011. This period includes the credit crisis that started in 2008, which provides a new dimension on the analysis by looking at the effect in an unstable crisis situation too. The analysis will use all the EU15 member countries and compare them with other developed and emerging countries that have experience financial globalization, but were not part of the European Union. Comparable countries can be found in the group of countries that later joined the European Union (see Appendix 1 for a list). However, their entry in the European Union might be endogenous. It can be argued that they joined the European Union as a result of increased integration with the other countries. To prevent distortion in the estimate, all of the new members on top of the EU15 will be excluded from the non-European Union group. Similarly, candidates or 5
The original dataset for the paper by Milesi-Ferretti (2001) covered 67 countries for the period of
4
1970-1998. In addition to increasing the number of countries, the revised dataset also uses more data sources and an improved methodology. The second update in 2007 contained 145 countries.
We will use the sample including new members for a robustness check.
potential candidates to EU membership are excluded (see Appendix 1 for a list). In the EU15 6 group, we do not take account of whether the members are part of the Eurozone (the EMU). The evidence suggests that the Euro in itself has not led to any further financial integration. After these restrictions, we are left with a global sample of 143 countries, of which 15 countries are from the European Union, covering the years 1995 to 2011. The sample has a total of 2788 observations.
For the robustness check using the regulatory measure of restrictions on financial openness, the KAOPEN index is used. This database was constructed by Chinn and Ito (2006) (see also Chinn and Ito, 2008) and covers 182 countries for the period of 1970 to 2013. The data is selected according to the same restrictions as for the original analysis. We have discussed the limitations and considerations with this dataset in section 3.2. on the financial indicators.
The countries were completely removed from the dataset. Slovakia was already absent in the dataset
6
originally. Kosovo was in the dataset as an independent country, even though the country is not officially recognized by the UN. The EU explicitly does not take an opinion on this issue by declaring it as a potential candidate for membership.
Table 3 - Mean values of variables across EU and non-EU members EU Members (9.1%) Non-EU Members (90.9%) Global Sample (100%) Credit to GDP ratio 110.3283 (44.0651) 38.3720 (38.0396) 45.2898 (44.0901) SCF 8.6733 (27.2364) 0.7257 (2.3861) 1.4651 (8.9033) SCI 5.1461 (16.4935) 0.4834 (1.0823) 0.9161 (5.2960) Income Development 0.8582 (0.3370) 0.1546 (0.2711) 0.2206 (0.3454) Inflation 2.1929 (1.5876) 16.6683 (143.9304) 15.3068 (137.0576) Trade 93.8422 (60.4769) 87.0654 (53.8565) 87.7113 (54.5471)
Real Interest Rate 4.3086
(3.0137) 8.0809 (15.3232) 7.8070 (14.8109) Gross Capital Formation 23.8861 (12.2636) 22.3467 (2.9065) 23.7327 (11.6816) KAOPEN 2.3110 (0.3133) 0.0901 (1.5277) 0.2948 (1.5939) Notes: Standard errors in parentheses.
Table 3 shows the summary statistics for all variables used in the analysis, including the three constructed measures of financial integration, the KAOPEN index and all control variables. We list the statistics for three separate groups: EU members, non-EU members and the global sample. The statistics follow an expected pattern. The financial integration measures are higher for EU members than for non-EU members. GDP and Export levels are similarly higher in the EU, as is the level of Credit to GDP.
3.5. The Financial Crisis
One of the major reasons for the relevance of the research in this thesis is derived from the connection between credit booms and financial crises (see for instance Aikman et al., 2014). Although the theoretical foundation on the influence of credit growth on the occurrence and severity of financial crises is solid, empirical evidence is limited. To see whether the results of this thesis are relevant beyond explaining just credit growth, we will construct an empirical methodology to test whether credit growth is associated with stock returns during the recent financial crisis.
To do this, we use research data by Kenneth French. He has constructed country portfolios, and calculated market returns for those country portfolios. We will also use the global variant of the Fama and French (1993) factors. We can use this data to construct the excess returns for a country and correct for the three factors that are modeled to explain stock returns. By adding the proposed explanatory variable of the change in credit growth, we can see whether it has any significant additional influence. In some sense, we treat the countries as individual stocks and the global portfolio as the market. Differences between countries in the credit to GDP ratio will provide insight. We will approach this in three different ways: (1) the change of the credit to GDP ratio in the same year, (2) a one year lag of the change in the credit to GDP ratio and (3) the average of the change in the credit to GDP ratio over the past 5 years. This will allow for an estimation of how credit booms are associated with the stock markets during a crisis. The regression is given by equation (8).
(8)
In this regression, the dependent variable is the excess return in country i. The three factors are global excess returns, SMB (excess returns of small firms over big firms in market capitalization) and HML (excess returns of firm with a high book to market ratio over firms with a low ratio). We add the change in the credit to GDP ratio, including an interaction effect with the crisis dummy variable equal to one for 2008. Data for this analysis is restricted to those countries of which Fama and French have constructed a portfolio in their international returns data. Those countries are: Austria, Australia, Belgium, Canada, Denmark, Finland, France, Germany, Hong Kong, Ireland, Italy, Japan, Malaysia, The Netherlands, New Zealand, Norway, Singapore, Spain, Sweden, Switzerland and the United Kingdom. The period is restricted to the period used in this thesis, 1995 to 2011. After this data restriction, we end up with 347 observations. This is relatively small, which we should keep in mind when analyzing the results. Robust standard errors clustered on country level will be used, following Petersen (2009), who argues in favor of clustering in a finance setting. The reason is that residuals may be correlated within clusters or across time. His arguments apply to this methodology too, even though it uses country data rather than firm data. The clustered standard errors correct for serial correlation within a cluster, on a country level. Note that we still make the assumption that observations are independent between clusters, meaning that we do not correct for serial correlation between countries.
We are especially interested in the significance of ! , which measures whether the change in the Credit to GDP ratio is associated with stock markets during the crisis. Establishing a direct effect will be difficult in this methodology.
4. Results
4.1. Vector Error Correction Model Results
The first section of the results focuses on the vector error correction model in which we analyze the long run relationship between financial integration and credit for countries in the European Union. The used sample is the EU15 countries for the years 1995 to 2011. We will present the results in several tables and in several steps. The first step is testing whether the variables are integrated of order one. The second step is to test the number of cointegrating relations. The third and final step is the estimation of the VECM. We will use the Stock of Capital Flows measure to estimate the relationship.
We start with the unit root analysis. The results are given in Table 4. All of the variables have a unit root in their levels. Stock of Capital Flows does not have a unit root in the first difference for any country. The Credit to GDP ratio has a unit root in the first difference for Belgium, Finland, France, Germany, Ireland, Luxembourg, Portugal, Spain and the United Kingdom.
The second step is to test for cointegration. For cointegration, it is required that both variables are stationary in their first difference. We consider all countries of the EU15. We test for cointegration using the Johansen (1991) test. As noted in the methodology section, we restrict the model to the number of lags under which the eigenvalue stability condition is met. Since we have 2 variables, we can have at most 1 cointegrating relation. The results 7 show both the trace statistic and the maximum eigenvalue. If either of the two rejects the null hypothesis that the number of cointegrating relations is equal to zero, we will conclude that there is cointegration between the two variables. We see that this is the case for Austria, Denmark, Finland, France, Portugal and the United Kingdom. Note that these cointegrating relationships are estimated under the assumption that the individual variables are integrated of order one, even though we failed to reject the null hypothesis that this is the case for many countries in Table 4.
The final step is to calculate the vector error correction model. Again, we estimate the model for all EU15 countries, even though only Austria, Denmark, Finland, France, Portugal
The estimation should have one unit modulus and all other moduli less than one.
and the United Kingdom showed cointegration. We estimate two versions: on an individual level and a panel data version. In the individual country model we allow for a maximum number of lags under which the eigenvalue stability condition is met. For Greece and Spain, the estimation is unstable if we include more than 1 lag on the VAR level, which means we can only estimate the lambda for the error correction term for those two countries. For the other countries, we include 2 lags on the VAR level, which implies 1 lagged first difference in the VECM setting of equation (1) and (2). We focus first on the interpretation of the lambdas, the error correction coefficients. These measure the responses of the respective variables to deviations from the long-run equilibrium, which is described in the error correction term.
Table 6 provides the results of the individual model estimations. The results show that for Austria, Belgium, Denmark, Germany, Ireland, Luxembourg, the Netherlands, Portugal and the United Kingdom, Credit to GDP adjusts after a deviation from the long run equilibrium. This is observed with the negative significant lambda for the error correction term in equation (1). For France, Credit to GDP adjusts, but the observed positive coefficient indicates that the adjustment moves away from the equilibrium. In the estimations for equation (2), we see that for Finland, Greece, Ireland and Italy, financial integration adjusts after a deviation from the equilibrium. Spain and Sweden do not show any significant coefficients for the error correction term.
The cointegrating relationship is provided in Table 7, again calculated for all EU15 countries. This follows the setup as described in section 3: C - α1 × F - α0. The negative sign
in this equation is included in the table, such that a negative sign implies a positive α1 and vice versa. We see from the results that the coefficient for α1 is significant in all cases. The
coefficient is negative for Finland, France, Germany and Italy. We see that the magnitude of the coefficients varies a lot between the countries, reflecting the differences in values of financial integration and Credit to GDP, relative to each other.
Combining Tables 6 and 7 allows us to reflect upon the stability of the adjustment. In the equation for the Credit to GDP adjustment, we always expect to see a negative significant sign if we estimate a significant adjustment. We see that this is not the case for France, indicating that the adjustment does not imply a return towards equilibrium. In the equation for financial integration, we require a positive sign if α1 in the cointegrating relationshipis
Table 4 - Augmented Dickey Fuller Tests for Unit Root
This table shows the ADF statistics for the EU 15 countries for the period 1995 to 2011. The first difference of the Credit to GDP ratio is not stationary for Belgium, Finland, France, Germany, Ireland, Luxembourg, Portugal, Spain and the United Kingdom.
Country Variable Level ADF Test Statistic First Difference ADF
Test Statistic Austria Credit to GDP - 1.128 - 3.464** SCF - 0.926 - 3.978*** Belgium Credit to GDP - 0.561 - 1.843 SCF - 0.707 - 4.953*** Denmark Credit to GDP - 1.097 - 3.794*** SCF - 1.853 - 4.299*** Finland Credit to GDP 1.363 - 2.037 SCF - 2.379 - 3.233** France Credit to GDP 1.507 - 2.757* SCF - 2.539 - 3..719** Germany Credit to GDP - 1.859 - 2.921* SCF - 2.246 - 4.055*** Greece Credit to GDP 1.086 - 4.544*** SCF - 1.836 - 4.238*** Ireland Credit to GDP - 0.807 - 1.364 SCF - 0.444 - 4.159*** Italy Credit to GDP 0.603 - 3.613** SCF - 2.399 - 3.815*** Luxembourg Credit to GDP - 0.788 - 2.803* SCF - 0.738 - 6.198*** Netherlands Credit to GDP - 1.104 - 4.781*** SCF - 2.324 - 4.232*** Portugal Credit to GDP - 1.269 - 1.627 SCF - 1.522 - 5.063*** Spain Credit to GDP 0.334 - 0.834 SCF - 1.747 - 4.549*** Sweden Credit to GDP - 1.677 - 6.002*** SCF - 1.907 - 4.738***
United Kingdom Credit to GDP - 0.839 - 0.886
SCF - 2.246 - 4.546***
Table 5 - Johansen Test for Cointegration between Credit to GDP and SCF
This table shows the trace statistic and maximum eigenvalue test for the EU15 countries over 1995 to 2011, using the number of lags under which the estimation meets the eigenvalue stability condition with a maximum of two. The results show cointegration for Austria, Denmark, Finland, France, Portugal and the United Kingdom.
Country Number of
Cointegrating Relations Trace Statistic Maximum Eigenvalue
Austria (lags: 2) r = 0 16.8003* 15.5752* r ≤ 1 1.2251 1.2251 Belgium (lags: 2) r = 0 9.2379 8.0335 r ≤ 1 1.2044 1.2044 Denmark (lags: 2) r = 0 18.2982* 14.8162* r ≤ 1 3.4819 3.4819 Finland (lags: 2) r = 0 19.0490* 18.4103* r ≤ 1 0.6388 0.6388 France (lags: 2) r = 0 15.1122 15.1047* r ≤ 1 0.0075 0.0075 Germany (lags: 2) r = 0 12.9780 10.8725 r ≤ 1 2.1055 2.1055 Greece (lags: 1) r = 0 5.6362 5.0259 r ≤ 1 0.6103 0.6103 Ireland (lags: 2) r = 0 9.7185 9.1996 r ≤ 1 0.5189 0.5189 Italy (lags: 2) r = 0 9.5809 9.3583 r ≤ 1 0.2226 0.2226 Luxembourg (lags: 2) r = 0 6.2875 6.2694 r ≤ 1 0.0181 0.0181 Netherlands (lags: 2) r = 0 14.2267 8.5275 r ≤ 1 5.6993 5.6993 Portugal (lags: 2) r = 0 20.2320* 17.1369* r ≤ 1 3.0951 3.0951 Spain (lags: 1) r = 0 8.2008 7.8886 r ≤ 1 0.3122 0.3122 Sweden (lags: 2) r = 0 8.4023 7.7034 r ≤ 1 0.6989 0.6989 United Kingdom (lags: 2) r = 0 16.7747* 9.8293 r ≤ 1 6.9454 6.9454
Table 6 - Individual Country VECM Results - Relationship between Credit to GDP and Financial Integration
This table shows the VECM results for all the EU15 countries for 1995 to 2011. For Greece and Spain, we do not observe the coefficients for the lagged variables, since the eigenvalue stability condition is not met. For Austria, Belgium, Denmark, Germany, Ireland, Luxembourg, the Netherlands, Portugal and the United Kingdom, we see that Credit to GDP adjusts after deviations from the long-run equilibrium. For France, the variable adjusts too, but moves away from the equilibrium. For Finland, Greece, Ireland and Italy, financial integration adjusts after a deviation from the long-run equilibrium.
Country Equation
Variables
Error Correction Term ∆Ct-1 ∆Ft-1 Constant
Austria ∆C - 1.1349*** (0.3045) 0.6063** (0.2602) - 10.2498** (3.5159) 0.0001 (0.8915) ∆F 0.0020 (0.0311) 0.0280 (0.0266) - 0.2406 (0.3593) 0.0729 (0.0911) Belgium ∆C - 0.3515*** (0.1289) 0.8944*** (0.2119) - 5.3224* (2.7215) - 0.0038 (1.0102) ∆F - 0.0044 (0.0144) 0.0387 (0.0237) - 0.6151** (0.3046) 0.3080*** (0.1130) Denmark ∆C - 0.5845** (0.2875) - 0.1284 (0.2784) - 34.5808 (42.6958) 0.0008 (8.9096) ∆F 0.0028 (0.0028) - 0.0011 (0.0028) 0.1121 (0.4227) 0.1545* (0.0882) Finland ∆C 0.0132 (0.0084) 0.3070 (0.2368) - 2.9993** (- 1.3446) 0.2192 (1.0651) ∆F - 0.0050*** (0.0014) 0.0667 (0.0407) 0.3506 (0.2312) 0.5828*** (0.1825) France ∆C 0.0151** (0.0062) - 0.5141 (0.4264) - 1.6383 (1.9023) 0.0074 (0.7914) ∆F - 0.0005 (0.0009) - 0.0393 (0.0625) - 0.0257 (0.2788) 0.2149* (0.1160) Germany ∆C - 0.0841*** (0.0312) - 0.1774 (0.2655) - 6.7471 (5.0247) - 0.0009 (0.7722) ∆F - 0.0016 (0.0015) 0.0155 (0.0130) - 0.2042 (0.2467) 0.0473 (0.0379)
Table 6 - Individual Country VECM Results (continued)
Country Equation Variables
Error Correction Term ∆Ct-1 ∆Ft-1 Constant
Greece ∆C - 0.0179 (0.0198) 0.0070 (6.6487) ∆F 0.0006** (0.0003) 0.2084** (0.1012) Ireland ∆C - 0.2658*** (0.0892) 0.7781*** (0.2144) - 6.7360*** (1.7089) 0.1593 (4.0040) ∆F 0.0315** (0.0143) - 0.0077 (0.0345) 0.0892 (0.2746) 1.3645** (0.6434) Italy ∆C 0.0022 (0.0051) - 0.1115 (0.3449) 8.5404 (8.7406) 0.1140 (9.4640) ∆F - 0.0004*** (0.0001) 0.0049 (0.0094) - 0.0145 (0.2376) 0.7168*** (0.2573) Luxembourg ∆C - 0.4607** (0.2324) 0.2994 (0.2861) - 0.3559 (0.3610) 2.8474 (4.8561) ∆F 0.1217 (0.1867) - 0.3190 (0.2300) - 0.3774 (0.2901) 10.7752*** (3.9026) Netherlands ∆C - 0.1726** (0.0843) - 0.1819 (0.2375) - 10.9633** (4.4542) 0.0179 (4.6839) ∆F 0.0062 (0.0063) 0.0112 (0.0177) 0.0047 (0.3326) 0.4991 (0.3497) Portugal ∆C - 0.3057*** (0.0664) 0.7029*** (0.1368) - 23.5541*** (6.6111) 0.0002 (1.5217) ∆F 0.0006 (0.0036) - 0.0012 (0.0073) - 0.3523 (0.3547) 0.0910 (0.0816)
Table 6 - Individual Country VECM Results (continued)
Country Equation Variables
Error Correction Term ∆Ct-1 ∆Ft-1 Constant
Spain ∆C - 0.0555* (0.0330) 0.0040 (5.2448) ∆F 0.0011* (0.0006) 0.2070** (0.0949) Sweden ∆C - 0.3530* (0.2101) - 0.3313 (0.2467) - 20.8689 (14.1635) 0.0069 (5.9263) ∆F 0.0085 (0.0041) - 0.0047 (0.0048) - 0.2431 (0.2779) 0.2856** (0.1163) United Kingdom ∆C - 0.2117*** (0.0705) 1.0550*** (0.2421) 0.8542 (6.0405) 0.0006 (2.3051) ∆F 0.0060 (0.0039) 0.0003 (0.0134) - 0.0359 (0.3347) 0.0215 (0.1277)
Table 7 - Cointegrating relationship
This table shows the cointegrating relationship between Credit Growth and Financial Integration for the EU15 countries for the years 1995 to 2011. The general Structure is C - α1F - α0 in which the coefficient for the Credit to GDP ratio is always 1. The coefficient for financial integration is significant in all cases. Note that a negative sign indicates a positive α1 in the equation.
Country Credit to GDP Financial Integration Constant
Austria 1 - 12.8950*** (0.7629) - 94.6985 Belgium 1 - 5.3946*** (1.6741) - 67.5944 Denmark 1 - 185.9097*** (13.1715) 60.1828 Finland 1 159.5233*** (30.9399) - 183.0235 France 1 550.9047*** (127.8507) - 643.1643 Germany 1 117.2025*** (35.1899) - 215.6971 Greece 1 - 707.8622*** (251.1609) - 173.4885 Ireland 1 - 17.0837*** (2.6522) - 53.3557 Italy 1 1488.5*** (505.2332) 1003.94 Luxembourg 1 - 1.4033*** (0.3036) 12.5683 Netherlands 1 - 48.2407*** (10.1862) - 56.0509 Portugal 1 - 93.2166 ( 7.6799) - 65.3677 Spain 1 - 291.3844*** (59.8355) - 33.7382 Sweden 1 - 48.2884*** (14.6745) - 27.6704 United Kingdom 1 - 70.1223*** (19.1018) - 9.2239
Notes: Standard errors in the parentheses. ***, ** and * denote significance levels of 1%, 5% and 10% respectively.
Table 8 - Panel Vector Error Correction Model - Relationship between Credit to GDP and Financial Integration
This table shows the result of the panel vector error correction model using the EU15 countries over the period 1995 to 2011. The estimation includes two lags on VAR level, which means that one lag is included in the VECM. This resulted in the lowest AIC. We also add a constant in the error correction term and regression, analogous to equations 1 and 2. The results show that the error correction term has a significant coefficient in the equation of financial integration. Credit to GDP responds to the lagged changed in financial integration with a negative adjustment.
Equation Variable Coefficient
∆C EC - 0.0347 (0.0199) ∆Ct-1 0.0027 (0.0676) ∆Ft-1 - 0.0987** (0.0489) Constant 0.1668*** (0.0291) ∆F EC 0.2157*** (0.0260) ∆Ct-1 - 0.0942 (0.0883) ∆Ft-1 - 0.0035 (0.0639) Constant 0.1666*** (0.0380)
Notes: Standard errors in the parentheses. ***, ** and * denote significance levels of 1%, 5% and 10% respectively.
Table 9 - Panel Cointegrating Relationship
This table shows the cointegrating relationship of the panel vector error correction model using the EU15 countries over the period 1995 to 2011. The general Structure is C - α1 × F - α0 in which the coefficient for the Credit to GDP ratio is always 1. This cointegrating relationship is the error correction term used in Table 8.
Credit to GDP Financial Integration Constant
1 - 2.0010*** (0.1671) 0.1379
Notes: Standard errors in the parentheses. ***, ** and * denote significance levels of 1%, 5% and 10% respectively.
