Disintermediation in Europe: Causes in core and periphery economies

(1)

Disintermediation in Europe:

Causes in core and periphery economies

Sophie Meerstadt

MSc Finance Thesis

University of Groningen

Abstract

We divide the cyclical causes of the surge in disintermediation in Europe since the crisis into push factors, which are related to the reluctance or inability of banks to lend, and pull factors, which are related to the increased attractiveness of capital markets. Moreover, we propose a distinction between core and periphery countries within the euro area. Using three OLS models for the total euro, core, and periphery areas, we find that disintermediation has been triggered by pull factors in core countries, and by push factors in periphery countries. A panel model with interaction terms, while less robust, confirms these results. Our findings have implications for the access to finance of firms in Europe, especially for SMEs, and particularly in periphery countries.

(2)

1. INTRODUCTION

When companies require external financing, there are two sources that they can tap into. The first option is to apply for a bank loan. Alternatively, they can access the capital market by issuing debt. External financing in the US is predominantly capital market based, while European corporations remain largely dependent on bank loans.

One of the most important trends in non-financial corporate financing in Europe has been the shift away from bank-based financing. This phenomenon of disintermediation, which started in the US several decades ago, seems to have been triggered by the financial crisis in Europe. Increasingly, non-financial corporations (NFCs) have turned to capital markets for their financing needs.

This paper disentangles the possible causes of disintermediation, and divides them into two categories. The first category of possible causes consists of factors that have made banks more reluctant or less able to lend to NFCs since the crisis, thereby forcing these companies to access capital markets as they search for alternative sources of financing. These causes will be referred to as push factors. On the other hand, there are factors that have made capital markets a more attractive financing source relative to bank loans, which will be referred to as pull factors. Terms such as demand or supply have deliberately been avoided, since demand and supply exist in both bank loan and debt securities markets, so the use of these terms could be confusing.

There is a gap in the existing literature on disintermediation, as it tends to focus only on one of these two drivers of disintermediation. There are articles that discuss the more stringent capital requirements in place since the crisis, and how this has forced NFCs to issue debt securities to meet their financing requirements (Baker and Wurgler, 2013; Cohen, 2013; Matrynova, 2015). There are also articles that describe the appeal of capital markets as an alternative to bank loans, and the enormous growth in securities issuance as a result (Authers, 2014; Fancourt, 2015). Both of these strands of research only take into account one category of factors, and therefore both are incomplete. This paper contributes to the discussion on disintermediation by taking both

(3)

Moreover, existing empirical research on disintermediation ignores the heterogeneity within the euro area. Aggregating empirical analysis to the euro area level may obscure some important differences between countries. In particular, a theoretical distinction is often made between core economies, such as Germany, France and the Benelux, and periphery economies such as

Portugal, Ireland, Greece, Spain and Italy. Kaya and Meyer (2013) suggest that for NFCs in core countries, capital markets may primarily be an opportunity to lower funding costs (given the current low yields), while in the periphery increased capital market issuance may represent a search for alternatives to bank loans. This paper disaggregates euro area averages into core and periphery area data, thereby allowing for differentiation between causes of disintermediation in the two areas.

This paper will empirically examine the causes of the surge in disintermediation since the crisis. We expect disintermediation to be caused by both pull and push factors. Moreover, we expect that pull factors have been dominant in core countries, while push factors have been dominant in periphery countries.

Our findings could have important implications on the ability of SMEs to obtain financing. SMEs are less able to access capital markets, thus they are more dependent on banks. If disintermediation is caused by push factors, then the financing of SMEs is at risk, especially if disintermediation continues. Furthermore, even if disintermediation is caused by pull factors on average, it could still be the case that it is caused by push factors at the SME level. As SMEs constitute 99.7% of EU firms, their access to finance is high on the agenda of the European Commission (Mersch, 2014a). We review some of the policy initiatives that have been taken by the EC in light of our findings. We then determine whether they are adequate, and whether they target the most vulnerable companies.

(4)

The remainder of this paper is structured as follows. Section 2 reviews the existing research on disintermediation, and divides the causes found in the literature into push and pull factors. Push factors are related to the unwillingness or inability of banks to lend since the crisis, while pull factors have made the capital market a more attractive source of financing. Section 3 describes the data that is used in the analysis. A preliminary review of the data confirms the expectation that the causes of disintermediation are different for core and periphery countries. In section 4, the methodology is described. Three separate OLS regressions for the total euro, core and

periphery areas will test whether there is a significant difference between the types of factors that are dominant for core and periphery countries. A panel regression with interaction terms should yield similar results. Section 5 describes the results, which confirm that pull factors are the significant causes in core countries, while push factors have been significant in periphery countries.

These findings could have important implications for SMEs. In Section 6, the effectiveness of policy measures taken by the European Commission to facilitate access to finance of SMEs is discussed. While disintermediation is mostly push-driven in periphery countries, there are no initiatives that particularly target SMEs in the periphery.

2. LITERATURE REVIEW

In the US, disintermediation has been occurring since the 1980s, and some say even earlier. This was made possible by structural, long-term factors such as the advent of direct market financing, the development of technology, and the deregulation of banks (Pati and Schome, 2006). In contrast, up to the financial crisis, European NFCs were heavily dependent on banks for their financing, and capital markets were rather underdeveloped (IMF, 2014).

(5)

issuance has shown a strong and steady increase since 2009, only slowing slightly during the European sovereign debt crisis (S&P, 2012a).

Thus, a decrease in the reliance on banks loans can be observed since the financial crisis, as corporations increasingly turn to the capital markets for their financing needs (Authers, 2014). This could be a consequence of push factors, pull factors, or both.

2.1 Push factors: constraints on bank lending

Since Basel III, requirements concerning the quantity and quality of bank capital have become more stringent (Rosengren, 2013; Tanda, 2015), leading to a decrease in loan growth (Bridges et al, 2014). Also, securitization (the bundling and trading of bank debt) has decreased steeply,1 thereby further decreasing bank lending capacity (Maddaloni and Peydró, 2010).

Following the crisis, banks faced unfavourable economic conditions, rising funding costs and higher capital requirements (Kaya and Meyer, 2014). These conditions have made them less able and less willing to provide loans, leading to a decline in bank lending and a subsequent surge in capital market issuance as NFC searched for alternative sources of funds (Garralda, 2015; S&P, May 2012.; Riccio, 2016; S&P, July 2012). Research in this field generally mentions 3 possible channels of adjustment for banks when capital requirements are increased (Cohen, 2013;

Fidrmuc et al., 2014; Martynova, 2015; S&P, 2011):

- Issuing new equity is generally the least attractive way for banks to de-risk their balance sheets, as existing shares are diluted in this way. Not much adjustment has taken place via this channel (Cohen, 2013; Martynova, 2015).

- A significant share of the required adjustment for banks has occurred through an increase in retained earnings, for example, by reducing dividends or by boosting profits. The most

""""""""""""""""""""""""""""""""""""""""""""""""""""""""

1_{"Securitization"was"seen"as"one"of"the"major"culprits"of"the"financial"crisis"(Mersch"Speech,"7"April"2014)."Issuance"in"Europe"decreased"}

to"roughly"oneEeighth"of"2008"levels"(IMF,"2014,"Figure"1.19.4,"p29).""

2_"_{The"relationships"have"also"been"estimated"for"two"separate"periods:"up"until"the"crisis"(January"20014"to"March"2009)"and"}

(6)

direct way to increase profits is by raising spreads. Indeed, on average, banks have increased net margins since 2009 (Baker & Wurgler, 2013; Cohen, 2013).

- Lending growth usually tracks asset growth. However, from 2009 to 2012 it remained significantly below asset growth. This shows that banks have also responded to the stricter requirements by decreasing lending growth. (Aiyar et al., 2014; Cohen, 2013; IMF, 2014).

In summary, according to existing literature that covers the push side, banks have contracted their lending and raised margins on loans since the crisis.

2.2 Pull factors: demand for funds shifts towards capital markets

A separate strand of literature describes another category of factors that have pulled significant proportions of corporate financing away from banks and towards capital markets since the crisis:

- The crisis demonstrated the dangers of relying solely on banks for financing. The US actually recovered more quickly from the crisis than Europe due to a less bank-dependent system. As a consequence, since the crisis, company treasurers have aimed to diversify their external financing sources. (Fancourt, 2015; S&P, July 2012).

- A consequence of the steadily falling interest rates since the crisis has been the search for yield by investors (Authers, 2014). Extremely low deposit and government bond yields have pushed savers and investors up the risk curve (Billington et al., 2015; Fancourt, 2015; IMF, 2013; S&P, July 2012). Investor demand for corporate bonds in all risk segments has increased.

(7)

According to the literature, companies have been pulled towards capital markets by a preference for diversification and a lower cost of issuing bonds.

We have seen that push factors and pull factors are mentioned in the literature. The dominant causes may not be the same for different countries. Kaya and Meyer (2013) suggest that for NFCs in core countries capital markets are an opportunity to lower funding costs, while in the periphery increased capital market issuance is the result of a search for alternatives to bank loans.

These push and pull factors are all recent, cyclical causes that are related to the economic environment since the crisis. Structural causes of disintermediation were already present many years earlier. Examples are technological developments, the maturing of the European capital market, and the increased availability of information on the relative creditworthiness of potential borrowers, such as ratings from credit rating agencies (French and Leyshon, 2004). This paper focuses on the cyclical causes of disintermediation since the crisis, rather than on these structural causes.

2.3 Risks associated with disintermediation in terms of access to finance

Of course, there are benefits to moving away from the heavily bank-based financing system that was traditionally present in Europe. It reduces dependence on bank financing (which, as we have seen, can dry up quickly in a financial crisis), thereby increasing financial stability, and it

increases the available choice set for companies looking for financing. However,

disintermediation can also jeopardise the access to financing of certain (traditionally bank-dependent) companies.

Because they are much more dependent on bank credit than their American counterparts, European firms are a lot more vulnerable when the supply of bank credit decreases. Since the crisis, the increase in securities issuance has not always sufficiently been able to compensate for the drop in bank lending (IMF, 2014).

(8)

market involves fixed costs, such as commissioning an external rating, which are easier to absorb by large companies (Kaya & Meyer, 2013 and 2014). Also, the issuance needs to be large

enough to attract attention and to be sufficiently liquid (Kaya and Meyer, 2014). Thus, for the most part, euro area SMEs remain dependent on a small subset of banks that are large enough to diversify the idiosyncratic risks that are inherent to small enterprises (Giesecke et al., 2012; Mersch, 2014a; Martynova, 2015). Consequently, when bank credit contracts, capital markets are generally a poor substitute for SMEs (Garralda, 2015).

If capital market investors are indeed little inclined to lend to SMEs, then disintermediation could severely impact the ability of SMEs to obtain financing. The effects on access to finance as described by the literature are summarized quite nicely in the following quote (Riccio, 2016):

Credit rationing also affects deserving borrowers, who have great difficulty obtaining funds from banks to finance working capital or new investment projects, both in terms of availability of funds and their increased costs as well as demand for more guarantees. This has led to more stringent effects

on SMEs, who have less access to alternative channels to banks.

The ability of firms to access capital at a given point in time may also differ between countries. If disintermediation is driven by pull factors in core countries, and by push factors in periphery countries, as we expect, then it makes sense that periphery area companies will have more difficulty obtaining financing than core area companies.

3. DATA

(9)

All data should be of the same frequency of observation. The Bank Lending Survey (BLS) data is gathered on a quarterly basis, and all other data is collected on a monthly basis. Since the BLS is taken at the end of each quarter, and concerns the past three months, the quarterly data were transformed into monthly data by taking the same survey response for all three months in a quarter. The sample period starts in January 2004 and ends in December 2014, 2

so the total number of observations is 132.

These observations are on six variables (disintermediation and five explanatory variables) for eight different countries and for the entire euro area. By taking the averages of the data for the four core countries and the four periphery countries respectively, data for the core area and the periphery area are obtained. This means that there are data for eight countries (in a panel data set) and three areas (in three separate time series data sets).

3.1 Disintermediation in Europe

Disintermediation is defined using outstanding bank loan and capital market debt levels . For both these variables, total amounts outstanding at the end of each month to NFCs in millions of Euro are used. These can be found in the ECB Statistical Data Warehouse.

We define disintermediation as

!"# = !_!!!! , (1)

(10)

interpreted using the underlying trends in the right panel of Figure 1. The number of securities held has been increasing steadily since 1999. Before the crisis, bank loans increased, so

disintermediation decreased. In the years following the crisis, bank loans decreased, as banks de-risked their balance sheets in order to meet new capital requirements. Securities were still increasing at this point, so the combination of these two trends led to a sharp increase in disintermediation. Recently, the amount of bank loans held seems to have stabilised, but disintermediation is still increasing due to the continuing rise in securities.

To put these findings into perspective, a comparison should be made with the US, where disintermediation is a lot higher. In Europe, corporate bonds now represent 20%, while bank loans still represent 80%. In the US, this is 52% and 48% respectively (Moody’s, 2015).

For separate countries, and thus for the core and periphery areas, data on loans and securities held are available from January 2003. It is immediately noticeable from Figure 2 that

disintermediation is higher than the euro area average in core countries and lower in periphery countries. So NFCs in the periphery have been even more dependent on banks than the average European NFC.

The underlying variables for these two areas (Figure 3) show that in core countries bank loans recovered quite quickly following the crisis. In fact, the amount of bank loans is 6.3% higher today than it was at the peak just before the crisis. Disintermediation is driven by the rise in securities held, which increased by 92.2% between 2003 and 2015. In periphery countries, bank loans have not yet recovered. Here, disintermediation is driven by a very large increase in securities held (179.9%), but also by a large decrease in bank loans (27.5% since their highest point before the crisis).

To summarize, the data show an increase in disintermediation since the crisis across the entire euro area, in core countries and in periphery countries. Preliminary analysis of the variables that define disintermediation supports the hypothesis that the underlying causes are different for core and periphery countries in the EU.

(11)

(12)

3.2 Causes of disintermediation

To investigate the causes of disintermediation (DIS), the impact of five variables is examined. These are obtained for the euro area, the core countries and the periphery countries, for the 2004 to 2014 period. Capital market spreads and the risk-free rate are the same in all three areas. The descriptive statistics for the euro, core, and periphery area and the panel data are given in Appendix 1.

• For bank loan margins (MARGINS), the MFI (Monetary Financial Institutions) lending margins to NFCs from the ECB Statistical Warehouse are used. The loans are from MFIs except Money Market Funds and Central Banks. This variable is plotted in Figure 4 on page 14. While the core countries and total Euro area follow a similar path, the pattern is different for the periphery countries. The mean for the periphery area is higher than the euro average, while it is lower in the core area.

• Capital market spreads (CMSPREADS) are obtained from the ECB Statistical Warehouse for the euro area. The spreads are defined as the yield spread between A-rated NFCs and government bonds for a maturity of seven years (this is the only maturity available from the ECB). The spread ranges from 80 basis points to 1382 basis points, as can be seen in the descriptive statistics in the Appendix. Figure 5 shows that spreads increased

significantly during the financial crisis and during the Euro sovereign debt crisis.

(13)

(or decreases only very slightly). Correlations between CMSPREAD and RF are very low for all areas, so a lower risk-free rate translates into a lower cost of issuing capital market debt.

• The proportion of banks that report a net increase in enterprise demand for bank loans over the past 3 months (LOANDEMAND) is obtained from the ECB Bank Lending Survey (BLS). This quarterly survey is addressed to senior loan officers employed by a representative sample of banks in the euro area. Figure 7 shows that loan demand has followed the same general pattern for all three areas.

• The proportion of banks that report a net tightening of credit standards on loans to enterprises (CREDITST) is also obtained from the BLS. We can see from Figure 8 that credit standards are reported to have tightened more in periphery countries.

• A dummy is used to control for crisis periods (CRISIS). This dummy variable takes a value of 1 in crisis periods, and a value of 0 otherwise. Crisis periods are January 2008 to March 2009 and July 2011 to December 2011. These are the periods in which the credit market spread (an indicator of market stress) increased the most.

Each of the variables has a unit root (Appendix 15), indicating that they are non-stationary, as is quite common in financial data. Non-stationarity can lead to a spurious regression with a high R2 . In first differences, the unit roots are rejected (Appendix 16), so all variables are I(1) series. The model should therefore be estimated in first differences. Furthermore, there is evidence of non-normality in all variables except LOANDEMAND.

(14)

(15)

(16)

Of course, these five variables are not the only factors influencing disintermediation. Specifically, there are certain structural factors (such as technological development, the availability of more knowledge on creditworthiness for investors, and the maturing of the European capital market) that have been slowly triggering disintermediation for a longer time. This paper focuses only on the cyclical factors that have been present since the financial crisis. Thus the long-term factors are beyond the scope of this paper. They are controlled for by removing the trend (estimating the models in 1st

differences).

3.3 Implications of disintermediation in terms of access to finance

According to the literature, most of the decrease in bank lending is compensated by capital market issuance. However, not all companies can tap into this resource. It is relevant to determine which types of companies are most at risk of not being able to meet their financing needs as a result of disintermediation.

Data from the ECB Access to Finance Survey can provide information on this topic. This biannual survey, first held in 2009, provides evidence concerning the financing conditions faced by thousands of SMEs and large firms in the total euro area. Within the SME category, it further distinguishes between micro, small, and medium-sized firms. On a country level the data can only be found for SMEs. This means that different firm sizes can be compared at the euro level, while the core and periphery areas can be compared at the SME level.

The first survey item we look at is the percentage of firms that have reported using debt securities for financing in the past 6 months. As is evident from Figure 10 on page 18, this percentage is more than twice as large for large firms as for SMEs, but there is no clear distinction between micro, small and medium-sized firms. Figure 11 shows that SMEs in core countries use capital market financing less often than those in periphery countries.

(17)

areas is consistently below the euro area average, while the gap for SMEs in periphery countries is substantially higher.

A final survey item to take into account is the percentage of firms indicating that access to finance is their most pressing problem. For SMEs, this percentage is substantially higher than for large firms (Figure 14), and it consistently decreases with firm size. Figure 15 shows that

compared to the euro area average, the percentage is consistently higher in periphery countries, and lower in core countries.

(18)

(19)

(20)

4. METHODOLOGY

4.1 Euro, core, and periphery areas

We examine the effect of five explanatory variables on disintermediation. Because we expect the causes to be different in core and periphery countries, these relationships are examined for the euro, core, and periphery area separately. As is explained in the Section 3.3, all the variables contain a unit root. Therefore, the model is estimated in first differences:

!!"#! = !α + ! !!!!"#$%&'!+ ! !!!!"#$%&'(!+ ! !!!"#!+!

!!!!!!!!!!!!!!!!!!!!_!!"!"#$%&"#$_!+ !_!!"#$!%&'&_!!!+ !!_!!!!"!"_!+ ! !_! , (2)

where DIS is disintermediation, MARGINS are bank loan margins, CMSPREAD is the capital market spread, RF is the risk-free rate, LOANDEMAND is the increase in demand for bank loans,

CREDITST is the tightening of credit standards and CRISIS is a dummy that takes a value of 1 in

crisis months. All these variables were introduced in Section 3.3. D denotes that first differences are taken. As a bank loan is generally not approved immediately, changes in credit standards are not expected to have an effect until two periods after the change took place. Hence, this variable is lagged by two periods.

In the periphery area, markets are less liquid and the process of obtaining a bank loan is lengthier. For this reason, bank loan margins are also lagged by one period, so the model is !!"#! = α + !!!!"#$%&'!!!+ !!!!"#$%&'(!+ ! !!!"#!+

!!!!!!!!!!!!!!!!!!!!"!"#$%&"#$!+ !!!"#$!%&'&!!!+ !!!!"#$#$!+ ! !! (3)

As these are linear models for time series data, it is appropriate to use OLS to estimate the models. The relationships are estimated three times, using the data sets for the euro, core, and periphery areas respectively. This allows for distinction between the causes of disintermediation in the three areas.

4.2 Core and periphery countries

(21)

data. It increases the degrees of freedom (and thus the power) of hypothesis tests, and it helps mitigate problems of multicollinearity (Brooks). The panel is unbalanced, as some of the observations are missing for Ireland. The total number of observations is 984. The model to estimate is:

!!"#!" = !α + ! !!!"#$%&'(!"+ ! !!!"#$%&'(!!"+ ! !!!"#!"+ ! !!!"#$#$!"+ !!!"#$!"∗

!!!!!!!!!!!!!!!!!!!!!"#$%&'(!"+ !!!"#$!"∗ !"#$%&'(!!"+ !!!"#!!"∗ !"#!"! + ! !!" (4)

Here, DIS is disintermediation, MARGINS are bank loan margins, CMSPREAD is the capital market spread, RF is the risk-free rate and CRISIS is a dummy for the crisis months. The

variables vary over both countries (!) and time (!). LOANDEMAND and CREDITST are excluded because these measures are not very reliable at the country level.3

Interaction terms for CORE with the remaining variables are included. 4

CORE is a dummy variable that takes a value of 1 if the country is a core country, and a value of 0 if the country is a periphery country.

If the causes for disintermediation differ between core and periphery countries then the coefficients for the interaction terms should be significant.

This equation could be estimated in a pooled regression (using OLS). However, pooling assumes that the average values of the variables and the relationships between them are constant over time and across countries. For this reason, it is more appropriate to use panel techniques, which can capture the heterogeneity between countries by exploiting both the cross-sectional and the time series dimensions of the data. The most frequently used panel models are fixed effects models and random effects models.

Fixed effects models allow the intercept to vary cross-sectionally but not over time, while slope estimates are fixed. The error term !_! is decomposed into a country-specific effect !_! and a remaining error term !_!":

(22)

!!" = ! !!+ ! !!" (5)

"

Here, !! encapsulates all the variables that affect disintermediation cross-sectionally (across

countries), but that are time-invariant. The model can then be estimated using a least squares dummy variable (LSDV) approach:

!"# = !!!"#$%&'!"+ !!!"#$%&'(!"+ ! !!!"!" + ! !!!"#$#$!"+ !!!"#$!"∗

!!!!!!!!!!!!!!"#$%&'_!"+!!"#$_!"∗ !"#$%&'(_!"+ !_!!"#$_!"∗ !"_!"+ !_!!1_!+ !_!!2_!+ !_!!3_!+ !!!!!!!!!!!!!!_!!4_!+ !_!!5_!+ !_!!6_!+ !_!!7_!+ !_!!8_!+ !_!"!!!!!!!! (6) Here, the dummies 1 to 8 represent the eight countries. So !! takes a value of 1 for the first

country (Belgium), and a value of 0 for all other countries. The intercept α is excluded to avoid perfect multicollinearity between the dummy variables and the intercept.

An alternative model is the random effects model. As with the fixed effects model, different intercept terms that are constant over time are applied to each country. However, in a random effects model, the intercepts for each country can be decomposed into a common intercept α plus a random variable !! that varies across countries (but is constant over time). In this case, the

heterogeneity is not captured by dummies, but by the !! terms:

!!"#!" = α + !!!!"#$%&'!"+ ! !!!"#$%&'(!!"+ ! !!!!"!"+ ! !!!"#$#$!" + !!!"#$!"∗

!!!!!!!!!!!!!!!!!!!"#$%&'(!" + !!!!"#$!"∗ !!"#$%&'(!"+ !!!"#$!"∗ !"#!"+ !!+ !!+ !!+ !!+

!!!!!!!!!!!!!!!!!!!_!+ !_!+ !_!+ !_!+ !_!""" " " " " " " " (7)" The random effects model should produce more efficient estimation (Brooks, 2014). However, this model is only appropriate if the composite error term !!+ !!! is uncorrelated with all the

explanatory variables. We tested this using the Hausman test, which indicates that the random effects model is appropriate for our analysis (Appendix 6).

4.3 Hypotheses

(23)

The variables in the models can be divided along these lines. The expected relationship between bank loan margins and disintermediation is positive: when banks increase their margins, loans become more expensive. This is a push factor, since a higher price of loans indicates that banks are more reluctant to extend them. The relationship between capital market spreads and

disintermediation should be negative, as lower costs of capital market issuance should make this option more attractive. Lower capital market spreads are therefore a pull factor. The higher correlation between the risk-free rate and bank loan margins than between the risk-free rate and capital market spreads (Section 3.3) indicate that the risk-free rate is more related to

attractiveness of the capital market, and accordingly it is a pull factor. The relationship is expected to be negative, as a lower risk-free rate increases the attractiveness of the capital market. Also, disintermediation can increase as a result of a decrease in loan demand, so the relationship between these two variables should be negative. If loan demand decreases when disintermediation (securities as a proportion of the sum of bank loans and securities) increases, then it is a pull factor. A tightening of credit standards can force NFCs to seek alternative financing options. It is therefore a push factor, and should be positively related to

disintermediation.

In summary, the pull factors in our models are a lower capital market spread, a lower risk-free rate and lower loan demand. The push factors are higher bank loan margins and a tightening of credit standards on bank loans.

We formulate 3 hypotheses concerning the recent causes of disintermediation:

• H1: Both push and pull factors are significantly related to disintermediation in the euro area.

(24)

Table 1. Expected relationships between the 5 explanatory variables and disintermediation in the euro, core and periphery areas

This table shows the expected signs (and significance) of the 5 independent variables. Three separate OLS

regressions will be performed for the euro, core and periphery areas. In the euro area, all variables are expected to be significantly related to disintermediation. In the core area, the pull factors are expected to have a significant impact, while in the periphery area only push factors are expected to be significant. In this table, + denotes a significantly positive relationship, and - indicates a significantly negative expected relationship. If the cell is left blank, than no significant relationship is expected.

In the panel model, the CORE dummy differentiates between core and periphery countries. Interaction terms consisting of the core dummy with bank loan margins, capital market spreads and the risk-free rate are included to determine whether the impact of these variables is different between the core and periphery area. If this is the case, the interaction terms should be significant and have the correct signs.

5. RESULTS

5.1 OLS results for euro, core and periphery areas

Three separate OLS models are estimated for the euro, core and periphery areas. Equation (2) is estimated for the euro and core areas, and Equation (3) is estimated for the periphery area. Both can be found in Section 4.1.

The hypotheses were given in Section 4.3:

• H1: Both push and pull factors are significantly related to disintermediation in the euro area.

• H2: Only pull factors are significantly related to disintermediation in the core area. • H3: Only push factors are significantly related to disintermediation in the periphery area.

Consequently, we expect all variables to be significant for the euro area. Bank loan margins (MARGINS) and tightening of credit standards (CREDITST) are expected to have a positive sign, MARGINS CMSPREAD RF LOANDEMAND CREDITST

Euro + - - - +

Core - - -

Periphery + +

(25)

while capital market spreads (CMSPREADS),the risk free rate (RF) and the net increase in loan demand (LOANDEMAND) should have a negative sign. In the core area, only CMSPREADS, RF and LOANDEMAND should be significant and have the expected (negative) sign. In the

periphery area, only MARGINS and CREDITST are expected to be (positively) significant. The results are presented in the table below.

Table 2: Cyclical causes of disintermediation in the euro, core, and periphery areas

The table shows the results of the regression of disintermediation (defined as securities outstanding divided by the sum of securities and loans outstanding) on bank loan margins, capital market spreads, the risk-free rate, the net increase in loan demand and the net tightening of credit standards (lagged by two months). A crisis dummy is also included. Bank loan margins were lagged by one month for the periphery area, since markets are less liquid and the process of obtaining a bank loan is lengthier. The relationships in the three areas were estimated in three separate OLS models. The sample period is January 2004 to December 2014. The equations were estimated in first

differences, as all the variables contained a unit root. Standard errors are reported in parentheses. *, **, *** indicates significance at the 90%, 95%, and 99% level, respectively.

Euro Core Periphery

Constant 0.0003 0.0002 0.0004***

(0.0002) (0.0003) (0.0002)

Bank loan margins 0.2540** -0.0055 0.1092**

(0.1210) (0.2361) (0.0532) Capital market spreads -0.0306*** -0.0377** -0.0170**

(0.0109) (0.0153) (0.0079)

Risk-free rate -0.1963*** -0.3443*** -0.0171

(0.0754) (0.1112) (0.0424)

Loan demand -2.40E-5* -2.24E-05** 2.14E-05**

(0.0000) (0.0000) (0.0000) Credit standards -1.95E-05 -1.58E-05 -1.52E-06

(0.0000) (0.0000) (0.0000)

Crisis 0.0002 -0.0005 -0.0002

(0.0004) (0.0005) (0.0002)

R-squared 0.125 0.122 0.063

No. observations 131 131 131

Data sources: ECB Statistical Warehouse, Bloomberg Finance LP, ECB Bank Lending Survey

In the euro area, all variables except CREDITST are significant and have the expected signs.

(26)

LOANDEMAND at the 10% level. This demonstrates that both push and pull factors are cyclical

causes of disintermediation in the euro area, thereby confirming Hypothesis 1. The R2

is quite low at 0.125. There is no evidence for heteroskedasticity (Appendix 8A), however, there could be autocorrelation (Appendix 7A), and hence the model has been estimated in accordance with the Newey-West procedure, using HAC standard errors. The results can be found in Appendix 9A. There is no evidence for non-linearity (Appendix 10A). However, normality is rejected (Appendix 11A).

In the core area, RF is significant at the 1% level, and CMSPREAD and LOANDEMAND are significant at the 5% level. All three have the expected signs. The other variables are not

significant. The three significant variables are all pull factors, which supports Hypothesis 2. The R2

is similar to that of the euro area regression at 0.122. There is evidence of heteroskedasticity (Appendix 8B) and there could be autocorrelation (Appendix 7B), which is why HAC standard errors are used. The results are given in Appendix 9B. Although there is no evidence of non-linearity (Appendix 10B), normality is rejected (Appendix 11B).

(27)

capital market spreads decrease, more companies are able to tap into the capital market. This accelerates disintermediation. Higher bank loan margins therefore have a greater impact on disintermediation when they are accompanied by lower capital market spreads.

Disintermediation is driven by banks’ reluctance to lend rather than the attractiveness of capital markets, as is illustrated by the positive significance of demand for bank loans. This is in line with Hypothesis 3.

The R2

is even lower than for the other two areas at 0.063. There is no evidence of

heteroskedasticity (Appendix 8C), but autocorrelation could be present (Appendix 7C). For this reason, HAC standard errors were used, as is shown in the regression output in Appendix 9C. Linearity cannot be rejected (Appendix 10C), but there is evidence of non-normality (Appendix 11C). It is noticeable that CREDITST is not significant in any of the areas.5

In summary, the OLS results support all three hypotheses. Disintermediation is driven by pull factors in the core area, and by push factors in the periphery area. Both types of factors are present in the euro area. An important limitation of the results is that the R2

is quite low as a result of removing the trend (by taking first differences), which controls for the structural causes of disintermediation. The results are robust to statistical tests, although there is evidence of non-normality for all three areas.6

5.2 Panel results for core and periphery countries

To confirm the different cyclical causes of disintermediation in the core and periphery areas, the panel model given in Equation (4) in Section 4.2 is estimated. It includes interaction terms for a

CORE dummy with MARGINS, CMSPREAD and RF. If the coefficients for the interaction terms

(28)

• H2: Only pull factors are significantly related to disintermediation in the core area. • H3: Only push factors are significantly related to disintermediation in the periphery area.

Initially, in the panel regression, none of the terms were statistically significant (Appendix 14), which seemed odd considering the obvious differences between the two areas in the OLS results. Using another metric for bank loans margins (obtained by subtracting the risk-free rate from bank loan margins), the results were as presented in Table 3. The output can be found in Appendix 15.

Table 3: Cyclical causes of disintermediation in a panel model

The table shows the results of the regression of disintermediation (defined as securities outstanding divided by the sum of securities and loans outstanding) on a new definition of bank loan margins, capital market spreads, the risk-free rate, a crisis dummy, and interaction terms of margins, spreads and the risk-risk-free rate with a core dummy, which takes a value of 1 if the country is a core country. If these interaction terms are significant, then the effect of the variables differs between core and periphery countries. The sample period is January 2004 to December 2014. The equations were estimated in first differences, as all the variables contained a unit root. Standard errors are reported in parentheses. *, **, *** indicates significance at the 90%, 95%, and 99% level, respectively.

Constant 0.0002 (0.0002) Margins2 0.0806 (0.2144) CM Spreads -0.0167 (0.0201) Risk-free rate 0.0547 (0.2314) Crisis -0.0002 (0.0004) Margins2*CORE -0.6550** (0.3208) CM Spreads*CORE -0.0137 (0.0276) Risk-free rate * CORE -0.8000**

(0.3338) R-squared 0.014 No. observations 984

(29)

The first four terms are not significant. The CORE dummy is significant only when interacted with bank loan margins (MARGINS2) and with the risk-free rate (RF).

From the OLS tests, we know that the relationship between bank loan margins and

disintermediation is positive. The interaction term for MARGINS2 with CORE has a negative coefficient, meaning that the relationship is less positive in core countries. Thus, the relationship between this push factor and disintermediation is less strong in core countries. This fits

Hypothesis 3.

We know that the relationship between the risk-free rate and disintermediation is negative. The coefficient for RF interacted with CORE has a negative coefficient, so the relationship between these two variables is more negative (and thus stronger) in core countries. As the risk-free rate is a pull factor, this fits Hypothesis 2.

The OLS results showed that lower spreads always increase disintermediation, regardless of the area. Consequently, it is not surprising that the effect of capital market spreads is not

significantly different for core and periphery areas (the coefficient for CMSPREAD interacted with CORE is not significant).

Again, a limitation is that R2_{is very low. A panel model contains more heterogeneity, and} therefore the fit of the model is expected to decrease. Nevertheless, 0.01 is even lower than expected. Furthermore, a drawback of using RF in the definition of MARGINS2 is that there is near multicollinearity between the two, even in first differences (Appendix 5D). Another limitation is that the first three terms are not individually significant. There is also evidence of non-normality (Appendix 16). All in all, the panel results, although they do support the

hypotheses, are not very robust.

6. IMPLICATIONS

(30)

disintermediation, especially if it is caused by push factors, could cause significant problems for SMEs, particularly for the smallest firms and in periphery countries. This, in turn, could cause problems for the EU, since as 99,7% of firms in the EU are SMEs, they account for 60% of turnover and 70% of total employment (Mersch, 2014a), and 85% of new jobs are created in micro firms (Cats, 2014). If disintermediation continues, safeguarding access to finance will become one of the most important tasks for the European Commission (EC). To this end, this section will discuss developments in the market and review the policy measures taken by the EC so far.

During the crisis, banks reduced credit to their riskiest borrowers in a so-called ‘flight to quality’ (Martynova, 2015). Since SMEs are inherently more risky, banks were often reluctant to lend to them. Also, SMEs often have little assets of their own, so banks can be worried about the quality of the collateral in case the enterprise fails. This could reinforce push factors towards SMEs. Thankfully, bank lending has recovered in recent periods, accompanied by a decrease in interest rates and a widening of credit standards. However, most of the benefit from falling interest rates and easing terms flows to large firms, even more so in periphery countries. This even occurs to an extent that SMEs in Greece, Spain, Italy and Portugal consider interest rates too high to even apply for a loan (ECB Access to Finance results September 2015).

Of course, there are also non-bank ways to access finance. Due to the low prevailing interest rates on government and bank credit, corporate credit is one of the few areas that still offer an attractive yield. Increasingly, non-bank entities are competing with traditional banks to extend credit to corporations (Betlem and De Horde, 2015a). Also, the peer-to-peer lending market has grown substantially, although its size is still relatively small in Europe. The one exception in the UK, with a market size of roughly £2,3 billion at the end of 2014 (Betlem and De Horde, 2015b). Here, so-called ‘challenger banks’ have become serious competitors for the largest incumbent banks. However, they have exhibited little inclination to lend to SMEs, as these are generally more risky, and the challenger banks are unfamiliar with these risks. As a result, non-banks may not be able to compensate for the retrenchment of bank credit for SMEs (IMF, 2014).

(31)

(Dunkley, 2014). The EC would be well-advised to take similar measures, so that non-banks can start providing SMEs with the much-needed sources of funds, especially if disintermediation continues.

Total non-bank assets in the EU have more than doubled over the past decade (Constancia, 2014). This can bring many benefits, as it reduces the dependence on banks, and increases the financing options for companies (FD Editors in chief, 2015a). However, regulated and non-regulated providers of finance compete fiercely, and the borders between banks and non-banks are blurring (Constancia, 2014). Non-banks providing bank-like services without adequate regulation could pose risks to financial stability. This matter is already being discussed extensively in the UK (FD Editors in chief, 2015a), and the EC should do the same. The

challenge is to enjoy the benefits of non-bank banking, while minimizing their bank-like risks. If regulation is balanced with stimulating measures, as in the UK, then the ability of these

alternative banks to provide credit should not be affected.

In September 2015, European Commissioner for Financial Services Jonathan Hill proposed the Capital Markets Union (CMU), which will remove the barriers to investment flows between countries, thereby facilitating the flow of capital, especially to SMEs. This system is intended to complement banks as a source of financing, and to provide wider access to finance for start-ups and SMEs (ECB, Capital Markets Union Factsheet).

Another way to structurally increase the capacity of banks to lend to SMEs, initiated by the ECB as part of the CMU, is to reintroduce and promote securitization. Securitization was reduced to almost nothing after the crisis, despite the fact that securitization was a lot safer in Europe than in the US during the crisis (ECB, Simpler Securitization Factsheet). Nowadays, securitization even has stricter requirements, is more transparent, and brings very few defaults. Thus, the ECB is looking into measures to stimulate securitization (Mersch, 2014a). An increase in the use of ‘safe’ securitization brings two additional advantages: it could bridge the gap between

deleveraging banks and investors looking for yield, and it reduces the reliance on bank finance in future downturns (Mersch, 2014b).

(32)

there is a high degree of risk associated with these enterprises. Now the EaSI-guarantee has been established, which is intended for suppliers of microcredit loans. However, so far, the funds that were made available for this initiative have been limited. The European Commission has

communicated that it hopes to obtain more funds for this project in a negotiation with member states (Cats, 2015).

To summarize, the European Commission has taken considerable steps to secure access to finance for SMEs. It has initiated a Capital Markets Union, and is promoting ‘safe’

securitization. Moreover, the EaSI-guarantee targets those firms that are most credit-constrained (micro firms). However, the EC should also stimulate non-banks, so that these can become providers of credit for SMEs. This stimulation should be accompanied by regulation to minimize the associated risks. Furthermore, it is noticeable that there are no measures that focus

particularly on SMEs in periphery countries, although these are significantly more

credit-constrained than SMEs in core countries. Perhaps this is because existing research generally does not distinguish between these two areas. Additional measures should be taken that target

periphery area SMEs. One possibility is to stimulate the formation of credit unions in periphery countries. Credit unions are member-owned financial cooperatives that provide credit at

competitive rates.

7. CONCLUSION

Since the financial crisis, corporate financing in Europe, which was heavily bank-based before the crisis, has partly shifted towards capital markets. This phenomenon is known as

disintermediation. Understanding disintermediation is important because it can have implications on companies’ access to finance. This paper has examined the cyclical causes of

disintermediation in Europe since the crisis. The factors driving disintermediation were expected to be different for core and periphery countries in the euro area.

(33)

decreased cost of issuing bonds, both of which have made capital markets more attractive. In Section 3, basic analysis of the data shows that the trends underlying disintermediation are different for core and periphery countries. Section 4 outlines the methodology used in this paper. Section 5 presents the results. Three OLS regressions show that pull factors have been the principal causes in the core area, while push factors have been dominant in the periphery area. A panel regression with interaction terms, although not very robust, confirms that the cyclical causes of disintermediation have been significantly different for core and periphery countries, and that pull factors have had a greater effect in core countries, while push factors have had more impact in periphery countries.

Disintermediation has implications on the access to finance of SMEs, which often do not have the luxury of diversification. The Access to Finance data showed that the access to finance of the smallest firms is most at risk, especially in periphery countries. Section 6 discusses some

(34)

REFERENCES

Aiyar, S., Calomiris, C., Wiedalek, T., 2014. Does macropru leak? Evidence from a UK policy experiment, Journal of Money, Credit and Banking 46, 181-214.

Allianz Global Investors, Presentation Euro High Yield Bond, October 2015, p16.

Authers, J., 2014. Role of banks recedes in wake of crisis: banks’ centrality to the economy is being nibbled away. Financial Times,

http://www.ft.com/intl/cms/s/0/37a6f3ac-f9de-11e3-bb9d-00144feab7de.html#axzz3mg0gWxIF, accessed 09/09/15.

Baker, M. and Wurgler, J., 2013. Do strict capital requirements raise the cost of capital? Banking regulation and the low risk anomaly. NBER Working Paper No. 19018

Betlem, R., de Horde, C., 2015a. Britse crowdfunder gaat concurrentie aan met Nederlandse banken. Het Financieel Dagblad.

Betlem, R., de Horde, C., 2015b. Schaduwbanken treden in de openbaarheid. Het Financieel Dagblad.

Bermingham, C., Ubovic, S, 2015. Eurozone money and credit: lending standards. BNP Paribas Economic Desknote, p4.

Bridges, J., Gregory, D., Nielsen, M., Pezzini, S., Radia, A., Spaltro, M., 2014. The impact of capital requirements on bank lending. Vox EU, http://www.voxeu.org/article/impact-capital-requirements-bank-lending, accessed 10/09/15.

Brooks, C., 2014. Introductory econometrics for finance. Third edition, Cambridge University Press, Cambridge.

Cats, R., Start-upkrediet moet banen scheppen. FD, Brussels, 21 October 2015.

Cohen, B., 2013. How have banks adjusted to higher capital requirements? BIS Quarterly Review, September 2013, 25-39.

Constancia, V., October 2014. Beyond traditional banking: a new credit system coming out of the shadows. ECB speech, Frankfurt.

De Waard, P. EU begint met kapitaalunie. De Volkskrant, 30 September 2015.

Dunkley, E., 2014. UK to launch scheme to boost bank lending to small businesses. Financial Times.

ECB Access to Finance Survey,

(35)

ECB Statistical Warehouse Website, https://www.ecb.europa.eu/stats/html/index.en.html, accessed 11 November 2015.

ECB website, Capital Markets Union Factsheet, http://ec.europa.eu/finance/capital-markets-union/docs/cmu-factsheet_en.pdf, accessed 1 December 2015.

ECB website, Simpler securitization factsheet, http://ec.europa.eu/finance/capital-markets-union/docs/simpler-securitisation_en.pdf, accessed 1 December 2015.

Fancourt, D., 2015. European High Yield: Fewer covenants do not mean less value. M&G Fixed Income Perspective, September 2015.

FD Editors-in-chief, 2015. Speelkwartier over voor crowdfunding. Het Financieel Dagblad. Fidrmuc, J., Schreiber, P., Siddiqui, M., 2014. The transmission of bank funding to corporate loans: deleveraging and industry specific access to loans in Germany. Open Economies Review 26, 581-597.

French, S, Leyshon, A, 2004. The new, new financial system? Towards a reconceptualization of financial reintermediation. Review of International Political Economy 11:2, p263-288.

Furfine, C., 2000. Evidence on the response of US banks to changes in capital requirements. BIS Working Papers No 88.

Garralda, J., 2015. Financial disintermediation in international markets and global banks funding models. Estabilidad Financiera 26, 111-129.

Giesecke, K., Longsta, F., Schaefer, S., Strebulaev, I., 2012. Macroeconomic Effects of Corporate Default Crises: A Long-Term Perspective. NBER Working Paper Series - Working Paper 17854.

Hamilton, J., 2015. Cross-border boom creates global opportunities. M&G Investments institutional briefing.

Illes, A., Lombardi, M., 2013. Interest rate pass-through since the financial crisis. BIS Quarterly Review, September 2013.

International Monetary Fund, 2013. Shadow banking around the globe: how large, and how risky? IMF External Publications, Global Financial Stability Report 2013, Chapter 2,

https://www.imf.org/external/pubs/ft/gfsr/2014/02/pdf/c2.pdf, accessed 11/11/2015.

International Monetary Fund, 2014. Improving the balance between financial and economic risk taking. IMF External Publications, Global Financial Stability Report 2014, Chapter 1.

(36)

Kaya, O., Meyer, T., 2013. Corporate bond issuance in Europe: where do we stand and where are we heading? Deutsche Bank Research,

https://www.deutschebank.nl/nl/docs/Corporate_bond_issuance_in_Europe_Where_do_we_stan d_andwhere_are_we_heading.pdf, accessed 20/9/15.

Kaya, O., Meyer, T., 2014. Tight bank lending, lush bond market: new trends in European corporate bond issuance. Deutsche Bank Research,

http://www.dbresearch.com/PROD/DBR_INTERNET_EN-PROD/PROD0000000000333212/Tight+bank+lending%2C+lush+bond+market%3A+New+tren ds+in+European+corporate+bond+issuance.pdf, accessed 20/09/15.

Maddaloni, A., Peydró, J., Bank risk-taking, securitization, supervision and low interest rates: evidence from the Euro area and the U.S. lending standards. ECB Working Paper Series No 1248, October 2010, p7.

Martynova, N., 2015. Effect of bank capital requirements on economic growth: a survey. De Nederlansche Bank, http://www.dnb.nl/en/binaries/Working%20paper%20467_tcm47-319679.pdf, accessed 12/09/15.

Mersch, Y., 2014a. Banks, SMEs and securitization: Overcoming the challenges faced by small firms. ECB speech, London.

Mersch, Y, 2014b. Next steps for European securitisation markets. ECB speech, Barcelona. Moody’s Research Updates, September 2015, Moody’s.

Pati, A., Schome, D., 2006. Disintermediation in financial services: an empirical investigation of the Indian banking sector. The Indian Journal of Commerce 59, 153-163.

Popov, A. and Udell, G., 2012, Cross-border banking, credit access, and the financial crisis”, Journal of International Economics 87, 147-161.

Riccio, G., 2016. Direct access to the debt capital market by unlisted companies in Italy and the effect of changes in civil law: an empirical investigation. Financial Crisis, Bank Behaviour and Credit Crunch, edited by Rossi, S. and malavasi, R. Springer, 2016.

Rosengren, E., 2013. Bank capital: lessons from the U.S. financial crisis. Federal Reserve, Bank of Boston.

Standard & Poor’s, 2011. Why Basel III and Solvency II will hurt corporate borrowing in Europe more than in the U.S. S&P RatingsDirect,

(37)

Standard & Poor’s, 2012b. The credit overhang: bank financing for future corporate growth may be at risk. S&P RatingsDirect,

http://www.standardandpoors.com/spf/upload/Ratings_EMEA/2012-07-31_TheCreditOverhang_BankFinancingForFuture.pdf, accessed 10/09/15.

Tanda, A., 2015. The effects of bank regulation on the relationship between capital and risk. Comparative Economic Studies, 31-54.

Thakor, A. and Furlong Wilson, P., 1995. Capital requirements, loan renegotiation and the borrower’s choice of financing. Journal of Banking and Finance 19, 693-711.

(38)

(39)

(40)

Appendix!2:!Correlation!matrices! "

Appendix!2A:!Correlation!table!for!the!euro!area!time!series!data!set!!

MARGINS CMSPREAD RF LOANDEMAND CREDITST MARGINS 1.000000 -0.035106 -0.784532 -0.221416 -0.168867 CMSPREAD -0.035106 1.000000 0.012126 -0.339191 0.775208 RF -0.784532 0.012126 1.000000 0.417461 0.131599 LOANDEMAND -0.221416 -0.339191 0.417461 1.000000 -0.498175 CREDITST -0.168867 0.775208 0.131599 -0.498175 1.000000 ! Appendix!2B:!Correlation!table!for!the!core!area!time!series!data!set!

MARGINS CMSPREAD RF LOANDEMAND CREDITST MARGINS 1.000000 0.107803 -0.721211 -0.311942 -0.087198 CMSPREAD 0.107803 1.000000 0.012126 -0.030652 0.703977 RF -0.721211 0.012126 1.000000 0.430625 0.167804 LOANDEMAND -0.311942 -0.030652 0.430625 1.000000 -0.305512 CREDITST -0.087198 0.703977 0.167804 -0.305512 1.000000 ! Appendix!2C:!Correlation!table!for!the!periphery!area!time!series!data!set!!

(41)

Appendix!3:!Estimation!output!of!the!OLS!models!before!the!crisis! !

Appendix!3A:!Estimation!output!of!the!OLS!models!for!the!euro!area,!January!2004!to!March!2009!

Dependent Variable: DDIS Method: Least Squares Date: 12/20/15 Time: 11:01 Sample: 2004M02 2009M03 Included observations: 62

HAC standard errors & covariance (Bartlett kernel, Newey-West fixed bandwidth = 4.0000)

Variable Coefficient Std. Error t-Statistic Prob. C -0.000556 0.000226 -2.457755 0.0172 DMARGINS 0.007143 0.141516 0.050472 0.9599 DCMSPREAD -0.028878 0.014544 -1.985492 0.0521 DRF -0.097594 0.068262 -1.429706 0.1585 DLOAND -2.95E-05 1.49E-05 -1.976780 0.0531 DCREDITST(-2) -1.36E-05 2.29E-05 -0.592606 0.5559 CRISIS 0.000545 0.000472 1.155950 0.2527 R-squared 0.150658 Mean dependent var -0.000438 Adjusted R-squared 0.058003 S.D. dependent var 0.001430 S.E. of regression 0.001388 Akaike info criterion -10.21610 Sum squared resid 0.000106 Schwarz criterion -9.975935 Log likelihood 323.6990 Hannan-Quinn criter. -10.12180 F-statistic 1.626006 Durbin-Watson stat 1.753248 Prob(F-statistic) 0.157423 Wald F-statistic 3.017659 Prob(Wald F-statistic) 0.012716

!

Appendix!3B:!Estimation!output!of!the!OLS!models!for!the!core!area,!January!2004!to!March!2009!

(42)

Appendix!3C:!Estimation!output!of!the!OLS!models!for!the!periphery!area,!January!2004!to!March!2009!

Variable Coefficient Std. Error t-Statistic Prob. C -6.88E-05 0.000232 -0.297125 0.7675 DMARGINS(-1) 0.113050 0.131739 0.858138 0.3945 DCMSPREAD -0.016726 0.015192 -1.100997 0.2757 DRF 0.011093 0.052997 0.209314 0.8350 DLOAND 1.03E-05 1.50E-05 0.688918 0.4938 DCREDITST(-2) 2.38E-06 2.15E-05 0.110717 0.9122 CRISIS 0.000188 0.000229 0.819852 0.4158 R-squared 0.035023 Mean dependent var -6.19E-05 Adjusted R-squared -0.070247 S.D. dependent var 0.001229 S.E. of regression 0.001271 Akaike info criterion -10.39202 Sum squared resid 8.88E-05 Schwarz criterion -10.15186 Log likelihood 329.1525 Hannan-Quinn criter. -10.29772 F-statistic 0.332695 Durbin-Watson stat 1.836024 Prob(F-statistic) 0.916820 Wald F-statistic 2.191656 Prob(Wald F-statistic) 0.057511

" !

(43)

Appendix!4:!Estimation!output!of!the!OLS!models!after!the!crisis! "

Appendix!4A:!Estimation!output!of!the!OLS!models!for!the!euro!area,!April!2009!to!December!2014!

Variable Coefficient Std. Error t-Statistic Prob. C 0.000972 0.000235 4.128726 0.0001 DMARGINS 0.334727 0.144861 2.310675 0.0242 DCMSPREAD -0.015790 0.009517 -1.659208 0.1021 DRF -0.236770 0.141242 -1.676344 0.0987 DLOAND -1.54E-05 1.28E-05 -1.200698 0.2344 DCREDITST(-2) -7.83E-06 1.59E-05 -0.492858 0.6239 CRISIS -0.000769 0.000455 -1.691793 0.0957 R-squared 0.155161 Mean dependent var 0.001013 Adjusted R-squared 0.073403 S.D. dependent var 0.001690 S.E. of regression 0.001627 Akaike info criterion -9.907910 Sum squared resid 0.000164 Schwarz criterion -9.681261 Log likelihood 348.8229 Hannan-Quinn criter. -9.817991 F-statistic 1.897800 Durbin-Watson stat 1.917642 Prob(F-statistic) 0.095327 Wald F-statistic 3.414493 Prob(Wald F-statistic) 0.005630

!

Appendix!4B:!Estimation!output!of!the!OLS!models!for!the!core!area,!April!2009!to!December!2014!

(44)

Appendix!4C:!Estimation!output!of!the!OLS!models!for!the!periphery!area,!April!2009!to!December!2014!

(45)

Appendix!5:!Correlation!matrices!with!all!the!variables!in!first!differences!

! !

Appendix!5A:!Correlation!table!for!the!euro!area!time!series!data!set!(first!differences)!

DMARGINS DCMSPREAD DRF DLOANDEMAND DCREDITST DMARGINS 1.000000 -0.015701 -0.090438 0.068103 -0.026717 DCMSPREAD -0.015701 1.000000 -0.196348 -0.072520 0.036684 DRF -0.090438 -0.196348 1.000000 0.023367 -0.051752 DLOANDEMAND 0.068103 -0.072520 0.023367 1.000000 -0.070809 DCREDITST -0.026717 0.036684 -0.051752 -0.070809 1.000000 " Appendix!5B:!Correlation!table!for!the!core!area!time!series!data!set!(first!differences)"

DMARGINS DCMSPREAD DRF DLOANDEMAND DCREDITST DMARGINS 1.000000 0.106895 -0.160453 0.103099 -0.003712 DCMSPREAD 0.106895 1.000000 -0.196348 -0.031197 -0.024707 DRF -0.160453 -0.196348 1.000000 -0.078014 0.008466 DLOANDEMAND 0.103099 -0.031197 -0.078014 1.000000 -0.045238 DCREDITST -0.003712 -0.024707 0.008466 -0.045238 1.000000 " Appendix!5C:!Correlation!table!for!the!periphery!area!time!series!data!set!(first!differences)"

(46)

APPENDIX PART II: RESULTS OLS

"

Appendix!6:!Hausman!test!for!panel!model! !

Correlated Random Effects - Hausman Test Equation: EQ02

Test cross-section random effects

Test Summary Chi-Sq. Statistic Chi-Sq. d.f. Prob. Cross-section random 0.000000 7 1.0000 * Cross-section test variance is invalid. Hausman statistic set to zero.

Cross-section random effects test comparisons:

Variable Fixed Random Var(Diff.) Prob. DMARGINS2 0.086681 0.080568 0.000046 0.3679 DCMSPREAD -0.016493 -0.016675 0.000000 0.7207 DRF 0.070164 0.054712 0.000143 0.1964 CRISIS -0.000167 -0.000159 0.000000 0.5880 DMARGINS2*CORE -0.694560 -0.654970 0.000740 0.1457 DCMSPREAD*CORE -0.013655 -0.013668 0.000000 0.9811 DRF*CORE -0.856245 -0.800046 0.001559 0.1547 " " Appendix!7:!Estimation!output!of!the!OLS!models!for!the!time!series!data!sets! " " Appendix!7A:!Estimation!output!for!the!euro!area!model! " " ! !

(47)

Appendix!7B:!Estimation!output!for!the!core!area!model"

Variable Coefficient Std. Error t-Statistic Prob. C 0.000237 0.000221 1.075343 0.2843 DMARGINS -0.027559 0.201838 -0.136542 0.8916 DCMSPREAD -0.037272 0.019139 -1.947408 0.0537 DRF -0.349364 0.100005 -3.493458 0.0007 DLOAND -2.17E-05 1.22E-05 -1.781103 0.0773 DCREDITST(-1) -1.82E-05 1.53E-05 -1.185334 0.2382 CRISIS -0.000495 0.000625 -0.791878 0.4299 R-squared 0.121614 Mean dependent var 0.000249 Adjusted R-squared 0.079111 S.D. dependent var 0.002347 S.E. of regression 0.002253 Akaike info criterion -9.301503 Sum squared resid 0.000629 Schwarz criterion -9.147866 Log likelihood 616.2484 Hannan-Quinn criter. -9.239073 F-statistic 2.861321 Durbin-Watson stat 1.728420 Prob(F-statistic) 0.012069

"

Appendix!7C:!Estimation!output!for!the!periphery!area!model"

Dependent Variable: DDIS Method: Least Squares Date: 12/04/15 Time: 17:18

Sample (adjusted): 2004M02 2014M12 Included observations: 131 after adjustments

(48)

Appendix!8:!Heteroskedasticity!tests!(White),!test!equations!are!available!upon!request! "

Appendix!8A:!White’s!heteroskedasticity!test!for!the!euro!area!model!

Heteroskedasticity Test: White

F-statistic 1.151072 Prob. F(26,104) 0.3015 Obs*R-squared 29.27359 Prob. Chi-Square(26) 0.2988 Scaled explained SS 38.18431 Prob. Chi-Square(26) 0.0582

Appendix!8B:!White’s!heteroskedasticity!test!for!the!core!area!model"!

"

Appendix!8C:!White’s!heteroskedasticity!test!for!the!periphery!area!model!

" !

Appendix!9:!Estimation!output!of!the!OLS!models!for!the!time!series!data!sets!using!the! NeweyAWest!procedure!(Heteroskedasticity!and!Autocorrelation!Consistent!standard!errors)"

Appendix!9A:!Estimation!output!for!the!euro!area!model!using!HAC!standard!errors"

Variable Coefficient Std. Error t-Statistic Prob. C 0.000304 0.000225 1.347216 0.1804 DMARGINS 0.254012 0.121049 2.098426 0.0379 DCMSPREAD -0.030578 0.010868 -2.813538 0.0057 DRF -0.196305 0.075350 -2.605261 0.0103 DLOAND -2.40E-05 1.42E-05 -1.690213 0.0935 DCREDITST(-2) -1.95E-05 1.93E-05 -1.012954 0.3131 CRISIS -0.000223 0.000455 -0.489416 0.6254 R-squared 0.124523 Mean dependent var 0.000326 Adjusted R-squared 0.082161 S.D. dependent var 0.001727 S.E. of regression 0.001655 Akaike info criterion -9.918358 Sum squared resid 0.000340 Schwarz criterion -9.764722 Log likelihood 656.6525 Hannan-Quinn criter. -9.855929 F-statistic 2.939509 Durbin-Watson stat 1.533843 Prob(F-statistic) 0.010240 Wald F-statistic 4.976261 Prob(Wald F-statistic) 0.000134

"

!

(49)

"

Appendix!9C:!Estimation!output!for!the!periphery!area!model!using!HAC!standard!errors"

Variable Coefficient Std. Error t-Statistic Prob. C 0.000432 0.000156 2.765608 0.0066 DMARGINS(-1) 0.109158 0.053248 2.049993 0.0425 DCMSPREAD -0.016975 0.007919 -2.143753 0.0340 DRF -0.017072 0.042414 -0.402514 0.6880 DLOAND 2.14E-05 9.91E-06 2.158151 0.0328 DCREDITST(-2) -1.52E-06 9.09E-06 -0.167314 0.8674 CRISIS -0.000230 0.000185 -1.244636 0.2156 R-squared 0.063058 Mean dependent var 0.000409 Adjusted R-squared 0.017722 S.D. dependent var 0.001202 S.E. of regression 0.001191 Akaike info criterion -10.57609 Sum squared resid 0.000176 Schwarz criterion -10.42245 Log likelihood 699.7337 Hannan-Quinn criter. -10.51366 F-statistic 1.390911 Durbin-Watson stat 1.721549 Prob(F-statistic) 0.223431 Wald F-statistic 3.094910 Prob(Wald F-statistic) 0.007377

Appendix!10:!Ramsey!RESET!tests!of!the!OLS!models!for!the!time!series!data!sets,!unrestricted!test! equations!are!available!on!request!