Small Business Economics

, Volume 41, Issue 3, pp 717–732 | Cite as

Bank loan terms and conditions for Eurozone SMEs

Article

Abstract

The evolution of bank loan price and non-price terms and conditions (T&Cs) for the 2009–2011 period are investigated using firm-level survey data for a sample of Eurozone small and medium-sized enterprises. The raw firm responses, which are of a discrete nature denoting tightening or easing of the T&Cs, or no change at all, are modeled by a bivariate ordered probit model. According to the results obtained, there are sizeable differences between countries, with the protagonists of the sovereign debt crisis (Greece, Ireland, Portugal, Spain) exhibiting predicted probabilities of tightening that are considerably above the Eurozone average. In addition, price T&Cs exhibit a substantial tightening over time. Finally, firms’ net interest expenses and profitability emerge as important determinants in explaining the cross-sectional variation in bank loan T&Cs that they face.

Keywords

Bank loan Bivariate ordered probit Terms and conditions Small and medium enterprises 

JEL Classifications

C25 G21 L26 

1 Introduction

The outbreak of the financial crisis in the fall of 2008 produced a wave of bank failures and subsequent massive bailout schemes on both sides of the Atlantic. The Eurozone in particular has witnessed further challenges since most of its economies have also plunged into recession and, moreover, some of its member states (mainly Greece, Ireland, Portugal, Spain; GIPS hereafter) have been the subjects of an ongoing sovereign debt crisis. These adverse developments have reinforced the initial crisis shock, and there is no doubt that they are bound to affect several aspects of European banks’ behavior. One of the most important aspects is how bank lending behavior is affected since bank lending behavior directly relates to the real economy. In particular, financial distress in the banking industry can have a negative impact on the supply and cost of lending, as banks reduce their credit risk exposure and attempt to recapitalize their positions. These developments are expected to affect more harshly small and informationally opaque businesses due to the amplified adverse selection and moral hazard problems that typically characterize the lender (bank)–borrower (firm) relationship.

Banks typically attempt to overcome these informational problems by engaging in screening, contracting, and monitoring activities (Diamond 1984; Berger and Udell 1990; Diamond 1991; Berger et al. 2001). Loan contracts offered by banks are multidimensional and specify a set of terms and conditions (T&Cs hereafter) that have to the double aim to mitigate credit risk stemming from a lending relationship and to enhance their ability to monitor borrowers over the time span of the loan contract (Strahan 1999; Watanabe 2005). T&Cs fall into two categories: (1) price T&Cs, such as the interest rate and the cost of financing (any other charges, fees, and commissions), and (2) non-price T&Cs, such as collateral requirements, the loan size, the loan maturity, and loan covenants. Banks use non-price T&Cs such as maturity and debt covenants to facilitate monitoring (Berlin and Loeys 1988; Berlin and Mester 1992), and they use other non-price T&Cs such as loan size and collateral to limit credit losses. Restricting loan size limits the bank’s potential exposure, while higher collateralization reduces loss due to default by potentially enhancing recovery rates. In a similar vein, shortening the contractual maturity of loans limits risk since the lender is exposed to the underlying credit risk for a shorter horizon. However, non-price T&Cs’ elements are not sufficient on their own to fully eliminate risk, and therefore banks price the residual risk through the interest rate and other types of charges and fees (Strahan 1999).

In this study I investigate how Eurozone banks specified T&Cs in offered loan contracts during the period 2009–2011, with a special focus on small and medium-sized enterprises (SMEs hereafter). The SMEs are of particular interest for three main reasons. First, SMEs rely heavily on external financing, of which the lion share is bank loans (Berger and Udell 1995). Secondly, the lender–borrower informational asymmetry of SMEs is exacerbated due to their informational opacity (Hu and Schiantarelli 1994; Jensen and McGuckin 1997; Audretsch and Elston 2002). Thirdly, SMEs represent the overwhelming majority (99.8 %) of enterprises active within the EU-27’s non-financial business economy, some 20.9 million, which together accounted for two out of every three jobs (66.7 %) and for 58.6 % of value added within the non-financial business economy during the study period (Eurostat Pocketbooks 2011).

The empirical analysis presented here relies on the European Commission and European Central Bank Survey on the access to finance of SMEs. The survey was launched in 2009 and is conducted twice a year, therefore providing five cross-sectional waves of firm-level survey data for representative Eurozone SMEs for use in this analysis. Firm responses for each T&Cs’ element take the form of tightening, being unchanged, or easing, having as a reference period the previous 6 months. Based on the raw responses I create a discrete ordered variable for the evolution of each T&Cs’ element and set out to address three research questions:
  1. (1)

    Are country effects mapped into T&Cs confronting SMEs? One would expect country heterogeneity to play an important role in any cross-country dataset. For the particular dataset under scrutiny, such heterogeneity might be even more pronounced, given the apparent cross-country variation in terms of the degree of involvement with the sovereign debt crisis. Hence, it is fruitful to investigate whether bank T&Cs pick up on such an effect and, if so, do they quantify it.

     
  2. (2)

    Are there any discernible patterns in T&Cs over time? The fact that the particular dataset covers the current crisis almost from its outset provides the opportunity to explore how the T&Cs evolved. Moreover, during the study period, the sovereign debt crisis, which started in Greece, has spread to other peripheral Eurozone countries. Hence, it is interesting to investigate whether these developments have resulted in any tendencies as far as the evolution of T&Cs is concerned. Clearly, given that only a short period of time is analyzed, it would be farfetched to talk about a time trend in the standard time-series terminology. Still, if any time patterns were to be uncovered they would be indicative of the evolution of conditions faced by SMEs.

     
  3. (3)

    Which are the firm-specific determinants of T&Cs? Obviously, after controlling for country heterogeneity, the cross-sectional variation in firms’ characteristics would be expected to be reflected on the cross-sectional variation of T&Cs. Economic theory suggests that banks would apply tighter loan T&Cs on cross-sectionally riskier firms.

     

The present study makes a threefold contribution to the literature: (1) it is the first empirical study that considers a detailed breakdown of six T&Cs’ elements, in contrast to previous studies that usually consider a subset of T&Cs due to data unavailability; (2) it is the first study that analyzes T&Cs for the Eurozone’s banking sectors; (3) it provides evidence for the behavior of T&Cs during the sovereign debt crisis period.

According to the empirical results presented here, there is strong evidence for an increased tightening over time in the price T&Cs’ elements (interest rate and cost of finance). The predicted joint probability for interest rate and cost of finance tightening has exhibited a steep increase of about 144 % between the first semester of 2009 and the first semester of 2011. This finding gives a vivid picture of the evolution of price T&Cs faced by SMEs in the Eurozone, i.e., they have been steadily deteriorating. In contrast, non-price T&Cs do not exhibit any significant time variation during this period. Another important finding is the strong between-country heterogeneity, implying that the T&Cs faced by SMEs are to a large extent determined by the country in which they operate. Moreover, the country variation exhibits a very clear and strong pattern, with GIPS, which are subject to the sovereign debt crisis, showing a substantially higher likelihood of experiencing tightening of the T&Cs. For example, while the average predicted joint probability of interest rate and cost of finance tightening is about 32 %, the corresponding probability reaches values of 54 % for Spain, 49 % for Portugal, 47 % for Ireland, and 45 % for Greece. In comparison, for the Eurozone’s three largest economies, these are 15 % (Germany), 22 % (France), and 34 % (Italy). A similar picture emerges when the predicted probabilities for the rest of the T&Cs’ pairs are compared, with GIPS showing values substantially above the Eurozone average and France and Germany showing predicted probabilities far below the Eurozone mean. An examination of firm-specific characteristics reveals that the probability of T&Cs being tightened is higher for firms with increased net interest rate expenses and firms with decreased profits.

The remainder of the paper is structured as follows: Sect. 2 offers a detailed description of the data, Section 3 presents the econometric methodology adopted, Sect. 4 discusses the empirical findings, and Sect. 5 presents the conclusion.

2 Data issues

2.1 Sample properties of T&Cs

We use the European Commission and European Central Bank Survey on the access of SMEs to finance. The survey was launched in 2009 and is conducted twice a year, therefore providing five waves of data for this analysis (until the first semester of 2011). The survey predominantly covers micro, small, and medium-sized firms, but also includes a small percentage of large firms. Firms covered in the survey belong to the Manufacturing, Construction, Wholesale-Retail, Services, or Mining sectors. The following countries are continuously covered: Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Portugal, and Spain.

Waves are essentially populated by random sampling, leading to pooling of data across them. However, the data provider recently released information that enables the construction of a panel dataset since a small number of firms have been surveyed in more than one survey. The analysis reported here exploits this development and used two datasets: a pooled dataset and a panel dataset.

The part of the survey of particular interest here relates to firms’ responses to the evolution of bank T&Cs on the financing available to them. The exact question that is posed to the firms is as follows:Open image in new window

Depending on the specific elements of the T&Cs an increase or a decrease indicates tightening or easing. In particular, for price T&Cs, tightening (easing) is denoted whenever they are increased (decreased). Turning to the non-price T&Cs, tightening (easing) is denoted by an increase (decrease) in the requirement(s) for collateral and others (which is referred to as loan covenants in the remainder of the paper). In contrast, for the size of loan and maturity of the loan, tightening (easing) is denoted by a decrease (increase).

In this analysis, the raw data are recoded in the following manner in order to obtain an operational and uniform measurement. Let (I), (j) denote the firm (respondent) and specific T&Cs’ element, respectively. Then a set of new variables is generated that attains the following values:
$$ {\text{TC}}_{i,j} = \left\{ {\begin{array}{lll} {0,\quad {\text{ if respondent }}i{\text{ mentioned tightening of the }}j{\text{ T\& Cs element }}} \\ {1,\quad {\text{if respondent }}i{\text{ mentioned no change of the }}j{\text{ T\& Cs element}}} \\ {2,\quad {\text{if respondent }}i{\text{ mentioned easing of the }}j{\text{ T\& Cs element }}} \\ \end{array} } \right\} $$
(1)

It becomes apparent that the new variables retain the information embodied in the original responses. In addition, the ordering denotes movement towards easing as the variable attains higher values. Note that for simplification purposes, the fact that the responses are pooled from waves surveyed at different time periods is not explicitly stated.

Table 1 reports the evolution of T&Cs by element and countries. For space conservation reasons, only the pooled sample (all firms across all countries and time periods) will be discussed here. Given that the data cover the period 2009–2011 one would expect bank loan T&Cs to exhibit a tendency towards tightening. Such a tendency is evident in the interest rates charged to SMEs, where 52 % of the firms responded that banks increased the interest rate on offered loans. The remaining 31 % responded that banks did not change their interest rates, while 16 %, a considerable portion, mentioned that interest rates had decreased. A similar pattern emerges for the cost of financing (charges, fees and commissions), with almost 52 % of the respondents mentioning that the cost of financing had increased, 44 % reporting that it was unchanged, and only a small fraction of just below 5 % mentioning that financing cost was lower. A rather different picture emerges for non-price T&Cs. In particular, for each non-price T&Cs’ element the status quo dominates, i.e., most firms responded that there was no change in the T&Cs of the loan contracts offered by banks. For instance, loan maturity was reported as unchanged by 81 % of firms, while approximately 60 % of respondents reported the status of the remaining T&Cs’ elements (loan size, collateral requirements, loan covenants) as ‘unchanged’. However, for each of these T&Cs’ elements, tightening accounted for the second largest portion of responses. However, there are apparent differences between the elements of T&Cs. For example, for collateral requirements and loan covenants, tightening was much higher than easing (37 vs. 3 % for the former and 35 vs. 3 % for the latter). In contrast, for loan size and loan maturity, responses were almost evenly split between tightening and easing.
Table 1

Evolution of terms and conditions by element and country (1st semester 2009 to 1st semester 2011)a,b

 

Price T&Cs

Non-price T&Cs

Interest rates

Cost of finance (charges, fees, commissions)

Size of loan or credit line

Maturity of loan

Collateral requirements

Covenants

Pooled sample

  Tightened

52.90

51.70

20.47

10.08

37.29

35.60

  Unchanged

31.02

43.69

59.74

80.84

59.57

61.64

  Eased

16.08

4.62

19.79

9.08

3.14

2.76

Austria

  Tightened

45.39

34.39

7.73

5.91

35.26

33.04

  Unchanged

38.05

60.35

67.01

84.00

62.35

66.09

  Eased

16.55

5.26

25.26

10.09

2.39

0.87

Belgium

  Tightened

38.76

39.36

14.44

6.38

33.02

31.10

  Unchanged

38.95

52.73

63.62

86.49

61.64

63.58

  Eased

22.28

7.91

21.94

7.13

5.34

5.31

Finland

  Tightened

37.93

30.31

6.83

6.34

32.87

29.29

  Unchanged

51.72

68.29

78.42

81.34

62.28

69.29

  Eased

10.34

1.39

14.75

12.32

4.84

1.43

France

  Tightened

38.11

45.93

10.99

6.27

35.96

26.07

  Unchanged

39.24

48.09

67.26

85.60

60.75

71.14

  Eased

22.65

5.98

21.75

8.13

3.29

2.79

Germany

  Tightened

30.37

23.37

9.46

5.01

29.03

27.10

  Unchanged

45.71

72.18

67.19

86.44

67.57

70.33

  Eased

23.93

4.45

23.36

8.55

3.40

2.57

Greece

  Tightened

74.16

58.86

38.68

21.65

46.64

47.90

  Unchanged

17.98

36.29

45.42

66.52

49.56

48.65

  Eased

7.87

4.86

15.90

11.83

3.80

3.44

Ireland

  Tightened

70.09

61.50

44.62

25.13

50.12

55.27

  Unchanged

22.65

36.95

41.98

69.95

46.85

40.87

  Eased

7.26

1.55

13.41

4.92

3.03

3.86

Italy

  Tightened

54.27

56.10

15.75

6.38

27.62

26.71

  Unchanged

29.11

39.28

66.51

86.11

70.01

71.10

  Eased

16.62

4.62

17.74

7.51

2.37

2.19

Netherlands

  Tightened

47.61

34.51

25.32

13.44

31.42

28.24

  Unchanged

38.56

60.33

50.38

78.49

62.57

66.18

  Eased

13.83

5.16

24.30

8.06

6.01

5.59

Spain

  Tightened

74.20

73.69

35.44

16.55

54.14

56.22

  Unchanged

15.56

22.59

45.04

71.68

43.32

41.26

  Eased

10.24

3.72

19.52

11.77

2.54

2.53

Portugal

  Tightened

66.40

71.08

24.04

8.78

24.66

29.60

  Unchanged

25.80

25.53

62.44

82.94

72.60

66.95

  Eased

7.80

3.39

13.52

8.28

2.74

3.44

T&Cs, Terms and conditions

Data in table are presented as the proportion (percentage) of respondents giving a specific response

aThe sum of tightened, unchanged, and eased may differ from 100 due to rounding-off errors

bResponses are aggregated across all firm sizes and sectors

Thus, a tentative conclusion is that during the sample period bank loan price T&Cs were tightened, which comes as no surprise since this is a normal reaction by banks in a highly volatile financial environment coupled with recession in real economic activity. For non-price T&Cs the same can be said for collateral requirements and loan covenants, which probably reflects the fact that the average expected default probability has increased due to the financial crisis and the resulting recession in the Eurozone. Thus, banks have put stricter conditions in place to minimize loss due to default via higher collateralization and guarantees. In addition, the increased information requirements, more complex procedures, and increased length of time required for loan approval reflect the more cautious behavior of banks in screening potential borrowers more thoroughly. Loan size and loan maturity behave rather differently from the other T&Cs since they show no evidence of either tightening or easing. In terms of loan maturity, one possible explanation for such behavior might be provided by banking practice that usually offers standardized loan maturities which are to some extent indivisible (i.e., 1, 3, 6 months, etc.). Similarly, the predominance of status quo for loan size may be explained by the possibility that the loan is covered through higher collateral requirements. It should be borne in mind that the apparent low variation in loan maturity and loan size can be expected to adversely affect the performance of any empirical model.

2.2 Firm characteristics

The inherently latent nature of a potential borrower’s riskiness leads banks to base their assessment on observed information. Unfortunately, the survey contains limited information on firm characteristics that can serve as possible determinants of the bank loan T&Cs it faces. The first group of covariates includes firm-specific indicators that reflect the evolution of financial conditions and performance. The second group captures firm characteristics, such as size and age. Table 2 shows these covariates in detail.
Table 2

Definition of Covariates

Notation

Definition

\( \Updelta \left( S \right)^{ + } \)

\( \begin{array}{*{20}c} { 1 , {\text{ if turnover increased during past 6 months}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

\( \Updelta \left( S \right)^{0} \)

\( \begin{array}{*{20}c} { 1 , {\text{ if turnover was unchanged during past 6 months}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

\( \Updelta \left( S \right)^{ - } \)

\( \begin{array}{*{20}c} { 1 , {\text{ if turnover decreased during past 6 months}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

\( \Updelta \left( {\text{NIE}} \right)^{ + } \)

\( \begin{array}{*{20}c} { 1 , {\text{ if net interest expenses increased during past 6 months}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

\( \Updelta \left( {\text{NIE}} \right)^{0} \)

\( \begin{array}{*{20}c} { 1 , {\text{ if net interest expenses were unchanged during past 6 months}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

\( \Updelta \left( {\text{NIE}} \right)^{ - } \)

\( \begin{array}{*{20}c} { 1 , {\text{ if net interest expenses decreased during past 6 months}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

\( \Updelta \left( \pi \right)^{ + } \)

\( \begin{array}{*{20}c} { 1 , {\text{ if profits increased during past 6 months}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

\( \Updelta \left( \pi \right)^{0} \)

\( \begin{array}{*{20}c} { 1 , {\text{ if profits were unchanged during past 6 months}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

\( \Updelta \left( \pi \right)^{ - } \)

\( \begin{array}{*{20}c} { 1 , {\text{ if profits decreased during past 6 months}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

\( {{\Updelta}}\left( {\text{D/A}} \right)^{ + } \)

\( \begin{array}{*{20}c} { 1 , {\text{ if debt related to assets increased during past 6 months}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

\( {{\Updelta}}\left( {\text{D/A}} \right)^{ 0} \)

\( \begin{array}{*{20}c} { 1 , {\text{ if debt related to assets was unchanged during past 6 months}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

\( {{\Updelta}}\left( {\text{D/A}} \right)^{ - } \)

\( \begin{array}{*{20}c} { 1 , {\text{ if debt related to assets was decreased during past 6 months}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

Micro

\( \begin{array}{ll} 1 , & {\text{ if}}\, \#\, {\text{of employees }} \le 9 \\ 0 ,& {\text{ otherwise}} \\ \end{array} \)

Small

\( \begin{array}{ll} 1 ,& {\text{if}}\;10 \le \#\;{\text{of employees }} \le 4 9 \\ 0 ,& {\text{ otherwise}} \\ \end{array} \)

Medium

\( \begin{array}{ll} 1 ,& {\text{ if}}\; 50 \le \#\,{\text{of employees }} \le 2 4 9 \\ 0 ,& {\text{ otherwise}} \\ \end{array} \)

Large

\( \begin{array}{ll} 1 ,& {\text{ if}} \# {\text{of employees}} > 2 4 9 \\ 0 ,& {\text{ otherwise}} \\ \end{array} \)

\( {\text{Age}}2^{ - } \)

\( \begin{array}{*{20}c} {1 , {\text{ if firm age < }}2{\text{ years}}} \\ {0 , {\text{ otherwise}}} \\ \end{array} \)

\( {\text{Age}}2 - 4 \)

\( \begin{array}{*{20}c} {1 , {\text{ if firm age between }}2{\text{ and 4 years}}} \\ {0 , {\text{ otherwise}}} \\ \end{array} \)

\( {\text{Age}}5 - 9 \)

\( \begin{array}{*{20}c} { 1 , {\text{ if firm age between 5 and 9 years}}} \\ { 0 , {\text{ otherwise}}} \\ \end{array} \)

\( {\text{Age}}10^{ + } \)

\( \begin{array}{*{20}c} {1 , {\text{ if firm age > 10 years}}} \\ {0 , {\text{ otherwise}}} \\ \end{array} \)

At this point it is interesting to briefly discuss the direction in which these firm characteristics are expected to affect T&Cs according to economic theory. Firm age is usually viewed as an indicator of a firm’s quality, since longevity may be a signal for survival ability and quality of management, as well as of the accumulation of reputational capital (Diamond 1991; Oliner and Rudebusch 1992). Moreover, the information gap is relatively narrower for older firms given their longer track record (Petersen and Rajan 1994; Cressy 1996). Another dimension that may be related to the degree of asymmetric information is firm size. Larger firms are more likely to have developed a reputation over time that lessens their incentive to behave in ways that could increase the probability of experiencing financial distress. In contrast, smaller firms face a higher relative probability of failure (Jensen and McGuckin 1997) and proportionately higher monitoring costs (Boocock and Woods 1997). In addition, smaller firms may have lower collateral relative to their liabilities than larger ones, and unit bankruptcy costs are likely to decrease with size (Gertler and Gilchrist 1994; Hu and Schiantarelli 1994; Gilchrist and Himmelberg 1995; Audretsch and Elston 2002; Vijverberg 2004). In the context of the Merton (1974) option-pricing model, leverage is used as an inverse proxy of firm credit quality because more levered firms, ceteris paribus, face a greater likelihood of insolvency. Leverage could also exacerbate moral hazard problems since highly levered borrowers may have a greater incentive to substitute high-risk assets for low-risk ones after a loan. Also, more profitable firms or firms with a higher cash flow are expected to be able to borrow more from banks since they are more likely to have the means to service their debt.

3 Econometric methodology

Given the ordered nature of firm responses to the evolution of each element of T&Cs, they must be modeled by an Ordered Probit. However, it is standard practice for banks to use various T&Cs’ elements in conjunction (Brick and Palia 2007), suggesting that the elements of T&Cs are not independent of each other. Thus, each T&Cs’ model should condition other T&Cs, otherwise the former would suffer from omitted variable bias. However, the inclusion of T&Cs in the vector of regressors would introduce endogeneity bias that could be treated by resorting to an instrumental variable (IV) estimation. A number of studies have attempted to tackle this issue. For example, Berger and Udell (1995), who explored the determinants of interest rates charged to small businesses, also included a dummy variable indicating whether the loan was collateralized or not. In a similar vein, Strahan (1999) controlled for loan maturity and collateralization in his equation for the lending rate. The primary aim of these two studies was to explore the determinants of lending rate and thus they did not explicitly model the remaining T&Cs.

Three other studies moved the literature forward by attempting to deal with the endogeneity of T&Cs’ elements using an IV approach. For example, in their analysis of loans to SMEs in the UK, Cressy and Toivanen (2001) considered three T&Cs (the interest rate, the size of the loan, and collateral). In their empirical set-up the interest rate is assumed to depend on the remaining two T&Cs and, therefore, these are instrumented. According to the results of these authors, endogeneity is not rejected, and the interest rate charged on loans is positively related to loan size, while negatively related to pledging collateral. Another important study was performed by Brick and Palia (2007) who were primarily concerned about the effects of relationship lending on T&Cs. These authors, focused on the interrelationship of loan rate premium, firm, and personal collateral, finding evidence in favor of jointness of these T&Cs, driven by the presence of the relationship lending. In a recent study, Posey and Reichert (2011) extend the work of Brick and Palia (2007) by considering an extra element of T&Cs. They too employ an IV approach to take into account the potential endogeneity of T&Cs’ elements, reporting evidence in favor of simultaneity among terms of lending.

In the analysis presented here, however, the implementation of IV is inapplicable given the lack of sufficient firm information that could be used as instruments. However, as shown in Table 3, there are specific pairs of T&Cs in the dataset that are highly correlated: (1) the two price elements (interest rate and cost of finance), (2) the size and maturity of loan, and (3) collateral requirements and covenants. Thus, both the omitted variable and the endogeneity biases can be handled by modeling each of these pairs using a bivariate ordered probit.
Table 3

Sample correlation between elements of T&Cs

Elements

Interest rates

Cost of finance

Size of loan

Maturity of loan

Collateral requirements

Covenants

Interest rates

Cost of finance

0.428a

Size of loan

0.083

0.091

Maturity of loan

0.045

0.047

0.334a

Collateral requirements

0.204

0.270

0.151

0.046

Covenants

0.206

0.275

0.129

0.054

0.574a

aSpecific pairs of T&Cs in the dataset that are highly correlated

Thus, the aim is to estimate the joint probability distribution of these two ordered categorical variables. In particular, the assumption is made that for each of the pairs, the involved latent variables (denoted by 1 and 2) are determined as follows (for a detailed exposition see Sajaia 2008):
$$ \begin{gathered} {\text{TC}}_{i,1}^{*} = z^{\prime}_{i,1} \beta_{1} + \varepsilon_{i,1} \hfill \\ {\text{TC}}_{i,2}^{*} = z^{\prime}_{i,2} \beta_{2} + \gamma {\text{TC}}_{i,1}^{*} + \varepsilon_{i,2} \hfill \\ \end{gathered} $$
(2)
where \( \left( {\beta_{1} ,\beta_{2} } \right) \) are vectors of unknown parameters, \( \gamma \) is an unknown scalar, and \( \left( {\varepsilon_{i,1} ,\varepsilon_{i,2} } \right) \) are the error terms. The \( \left( {z^{\prime}_{i,1} ,z^{\prime}_{i,2} } \right) \) are vectors of explanatory variables satisfying \( E\left( {z^{\prime}_{i,1} \varepsilon_{i,1} } \right) = 0 \) and \( E\left( {z^{\prime}_{i,2} \varepsilon_{i,2} } \right) = 0 \).
However, instead of observing the latent variables we only observe the categorical variables \( \left( {{\text{TC}}_{i,1} ,{\text{TC}}_{i,2} } \right) \)such that:
$$ {\text{TC}}_{i,1} = \left\{ {\begin{array}{lll} {0,\quad{\text{ if TC}}_{i,1}^{*} \le c_{11} \, } \\ {1,\quad{\text{if }}c_{11} < {\text{TC}}_{i,1}^{*} \le c_{21} } \\ {2,\quad{\text{if }}c_{21} < {\text{TC}}_{i,1}^{*} \, } \\ \end{array} } \right\},{\text{TC}}_{i,2} = \left\{ {\begin{array}{lll} {0,\quad{\text{ if TC}}_{i,2}^{*} \le c_{12} \, } \\ {1,\quad{\text{if }}c_{12} < {\text{TC}}_{i,2}^{*} \le c_{22} } \\ {2,\quad{\text{if }}c_{22} < {\text{TC}}_{i,2}^{*} \, } \\ \end{array} } \right\} $$
(3)

The \( \left( {c_{11} ,c_{21} ,c_{12} ,c_{22} } \right) \) denote unknown cutoff points satisfying the conditions that \( \left( {c_{11} < c_{21} } \right) \) and \( \left( {c_{12} < c_{22} } \right) \).

The probability that \( \left( {{\text{TC}}_{i,1} = k} \right) \) and \( \left( {{\text{TC}}_{i,2} = l} \right) \) is:
$$ \begin{aligned} \Pr \left( {{\text{TC}}_{i,1} = k,{\text{TC}}_{i,2} = l} \right) & = \Pr \left( {c_{k - 1,1} < {\text{TC}}_{i,1}^{*} \le c_{k,1} ,c_{l - 1,2} < {\text{TC}}_{i,2}^{*} \le c_{l,2} } \right) = \\ & = \Pr \left( {{\text{TC}}_{i,1}^{*} \le c_{k,1} ,{\text{TC}}_{i,2}^{*} \le c_{l,2} } \right) - \Pr \left( {{\text{TC}}_{i,1}^{*} \le c_{k - 1,1} ,{\text{TC}}_{i,2}^{*} \le c_{l,2} } \right) \\ & - \Pr \left( {{\text{TC}}_{i,1}^{*} \le c_{k,1} ,{\text{TC}}_{i,2}^{*} \le c_{l - 1,2} } \right) + \Pr \left( {{\text{TC}}_{i,1}^{*} \le c_{k - 1,1} ,{\text{TC}}_{i,2}^{*} \le c_{l - 1,2} } \right) \\ \end{aligned} $$
(4)
If \( \left( {\varepsilon_{i,1} ,\varepsilon_{i,2} } \right) \) are distributed as bivariate standard normal with correlation \( \rho \) the individual contribution to the likelihood function could be expressed as:
$$ \begin{aligned} \Pr \left( {{\text{TC}}_{i,1} = k,{\text{TC}}_{i,2} = l} \right) & = \Upphi \left( {c_{k,1} - z^{\prime}_{i,1} \beta_{1} ,\left( {c_{l,2} - \gamma z^{\prime}_{i,1} \beta_{1} - z^{\prime}_{i,2} \beta_{2} } \right)\zeta ,\tilde{\rho }} \right) \\ & - \Upphi \left( {c_{k - 1,1} - z^{\prime}_{i,1} \beta_{1} ,\left( {c_{l,2} - \gamma z^{\prime}_{i,1} \beta_{1} - z^{\prime}_{i,2} \beta_{2} } \right)\zeta ,\tilde{\rho }} \right) \\ & - \Upphi \left( {c_{k,1} - z^{\prime}_{i,1} \beta_{1} ,\left( {c_{l - 1,2} - \gamma z^{\prime}_{i,1} \beta_{1} - z^{\prime}_{i,2} \beta_{2} } \right)\zeta ,\tilde{\rho }} \right) \\ & + \Upphi \left( {c_{k - 1,1} - z^{\prime}_{i,1} \beta_{1} ,\left( {c_{l - 1,2} - \gamma z^{\prime}_{i,1} \beta_{1} - z^{\prime}_{i,2} \beta_{2} } \right)\zeta ,\tilde{\rho }} \right) \\ \end{aligned} $$
(5)
where \( \Upphi \) is the bivariate standard normal cumulative distribution function, \( \zeta = \frac{1}{{\sqrt {1 + 2\gamma \rho + \gamma^{2} } }} \) and \( \tilde{\rho } = \zeta \left( {\gamma + \rho } \right) \).

The parameters in the system of equations are identified only by imposing a minimum of an exclusion restriction on vectors \( \left( {z^{\prime}_{i,1} ,z^{\prime}_{i,2} } \right) \).

4 Empirical results

Table 4 shows the estimation results from the bivariate-ordered probit models for the pairs of T&Cs. The independence of equations is rejected for all three pairs analyzed, suggesting that the bivariate model is appropriate. Another point worth noting is that the explanatory power (Wald test for overall significance) is clearly higher for the price T&Cs, while the worst fit is encountered for loan size and loan maturity. The latter reflects the sample properties of the dependent variables that exhibit very low variation.
Table 4

Pooled bivariate ordered probit modela

Variables

Pair A

Pair B

Pair C

Dependent variable

Interest rates

Cost of finance

Loan maturity

Loan size

Covenants

Collateral requirements

\( \Updelta \left( S \right)^{ + } \)

0.023

(0.024)

0.016

(0.025)

0.028

(0.021)

\( \Updelta \left( S \right)^{0} \)

0.037

(0.031)

−0.028

(0.026)

0.044*

(0.022)

\( \Updelta \left( {\text{NIE}} \right)^{ + } \)

−0.801***

(0.032)

0.650***

(0.185)

−0.116***

(0.035)

0.039

(0.037)

−0.228***

(0.032)

0.085

(0.057)

\( \Updelta \left( {\text{NIE}} \right)^{0} \)

−0.351***

(0.031)

0.348***

(0.043)

−0.049

(0.034)

0.020

(0.036)

0.051

(0.032)

−0.082**

(0.038)

\( \Updelta \left( \pi \right)^{ + } \)

0.104***

(0.030)

−0.052

(0.066)

0.089***

(0.033)

0.099**

(0.041)

0.129***

(0.030)

\( \Updelta \left( \pi \right)^{0} \)

0.080***

(0.029)

−0.058

(0.045)

0.103***

(0.029)

0.016

(0.035)

0.081***

(0.024)

\( \Updelta \left( {\text{D/A}} \right)^{ + } \)

−0.030

(0.030)

−0.010

(0.048)

0.173***

(0.031)

−0.082***

(0.031)

−0.011

(0.047)

\( {{\Updelta}}\left( {\text{D/A}} \right)^{ 0} \)

0.044

(0.027)

−0.057*

(0.030)

0.136***

(0.027)

0.037

(0.028)

−0.057

(0.035)

Micro

−0.004

(0.029)

0.044

(0.045)

−0.015

(0.061)

0.022

(0.035)

−0.129**

(0.058)

0.059

(0.075)

Small

−0.016

(0.032)

0.075

(0.059)

0.033

(0.058)

−0.050

(0.041)

−0.168***

(0.053)

0.085

(0.071)

Medium

−0.110*

(0.057)

0.198**

(0.089)

0.094

(0.058)

0.043

(0.037)

−0.121**

(0.054)

0.079

(0.068)

\( {\text{Age}}10^{ + } \)

0.091

(0.074)

−0.050

(0.091)

−0.129

(0.081)

0.114

(0.089)

0.130*

(0.071)

−0.210**

(0.093)

\( {\text{Age}}5 - 9 \)

0.157*

(0.080)

−0.078

(0.117)

−0.056

(0.087)

0.066

(0.094)

0.120

(0.077)

−0.199**

(0.099)

\( {\text{Age}}2 - 4 \)

0.135

(0.082)

−0.082

(0.106)

−0.113

(0.091)

0.167*

(0.097)

0.143*

(0.083)

−0.303***

(0.108)

\( {\text{Age}}2^{ - } \)

0.008

(0.098)

0.112

(0.146)

−0.101

(0.108)

−0.062

(0.124)

0.298***

(0.107)

−0.550***

(0.137)

Trend

−0.266***

(0.008)

−0.260***

(0.025)

0.008

(0.009)

0.023**

(0.010)

0.011

(0.008)

−0.014

(0.010)

Sector dummies

Included

Included

Included

Included

Included

Included

Country dummies

Included

Included

Included

Included

Included

Included

Observations

11,907

11,614

11,413

Wald test (overall significance)

2922.11***

222.62

703.08

Rhob

−0.916

−0.605

−0.767

Gammac

1.124***

(0.028)

0.987***

(0.078)

1.369***

(0.021)

Log pseudo-likelihood

−18618.41

−17187.08

−15132.49

Wald test of indep. equations

3.06*

13.30***

16.69***

Country effects (p value)

0.000

0.000

0.000

*, **, *** denote significance at the 10, 5, and 1 % level, respectively

aSome covariates have been excluded for identification purposes

bDenotes the cross-equation error correlation

cDenotes the coefficient of the endogenous element of the T&Cs

Starting with the pair of price T&Cs (‘interest rates/cost of finance’), it is evident that the time trend enters the models with significantly negative coefficients, indicating the increasing (decreasing) likelihood that the probability of price T&Cs tightens (eases) as time passes. Uniformity of country effects was emphatically rejected (p = 0.000), implying that there is substantial country heterogeneity. Hence, controlling for firm-specific attributes and sector effects, there is a significant portion of variation accounted for by the country within which a firm operates.

For the firm-specific characteristics, firms whose net interest rate expenses either increased or remained unchanged (compared with those whose net interest rate expenses decreased) are found to be less likely to enjoy an interest rate easing. One way to gauge this is that everything else equal, firms with an increased burden of interest expenses are viewed by banks as being riskier. Given their lower likelihood for interest rate easing, this creates a vicious circle, with perhaps higher interest expenses in the future. Another possibility is simply that of a reverse causality, where the higher likelihood of interest rate tightening leads to higher interest expenses. The pooled nature of the dataset does not allow further exploration of this issue. In addition, firms whose profits increased or remained unchanged (compared with those whose profits decreased) are less likely to witness an interest rate tightening. With regards to the cost of finance, the result is counter-intuitive, i.e., firms whose net interest rate expenses either increased or remained unchanged are more likely to face an easing. For the pair ‘loan maturity/loan size’, the time trend enters with a significantly positive coefficient in the latter equation, suggesting that the likelihood of banks granting larger loans increases over time. ‘Loan maturity’ easing is more likely for firms whose profits increased or remained unchanged. In addition, there is a higher likelihood for easing ‘loan Maturity’ for firms whose net interest rate expenses increased, and firms whose leverage has increased. In the ‘loan size’ equation, increased profitability is associated with a higher probability of easing. With respect to the third pair, we find that the probability of ‘covenants’ being tightened is higher for firms with increased net interest expenditures, firms whose profits decreased, and firms whose leverage increased. There is also an increased easing likelihood of ‘covenants’ for younger firms, while the opposite holds for ‘collateral requirements’.

This discussion leads to a consideration of various SME subgroups in order to investigate the impact on the (predicted) probability of joint tightening for each T&Cs pair. Two groups are formed for the country-level scenarios: one group that includes firms operating in countries that have been hit by the sovereign debt crisis (Greece, Ireland, Portugal, Spain), and a second group representing the largest economies in the Eurozone (France, Germany, Italy), with the Netherlands for comparison purposes. The predicted probabilities by wave are also calculated in order to shed light on the time path of T&Cs. Finally, in terms of firm profile, the following subgroups are considered: (1) firms whose net interest expenses increased; (2) firms whose profits increased. Table 5 presents the predicted probabilities across the different subgroups. For comparison purposes, the unconditional (sample) probabilities and the predicted probabilities for the “average firm” in the sample are reported.
Table 5

Predicted probabilities of tightening across hypothetical small and medium-sized enterprise subgroups (pooled bivariate ordered probit model)

Variables

Interest rates and cost of finance

Loan size and loan maturity

Collateral requirements and covenants

Unconditional probability

0.383

0.068

0.276

Average predicted probability

0.319

0.050

0.228

SME subgroups

Predicted probability

\( \Updelta \left( {\text{NIE}} \right)^{ + } = 1 \)

0.504

0.058

0.303

\( \Updelta \left( \pi \right)^{ + } = 1 \)

0.253

0.036

0.177

First semester 2009

0.169

0.052

0.234

Second semester 2009

0.215

0.049

0.228

First semester 2010

0.278

0.047

0.222

Second semester 2010

0.345

0.051

0.227

First semester 2011

0.413

0.049

0.228

Greece

0.449

0.091

0.307

Ireland

0.474

0.140

0.354

Portugal

0.490

0.052

0.160

Spain

0.547

0.068

0.391

France

0.223

0.035

0.189

Germany

0.151

0.029

0.175

Italy

0.346

0.040

0.176

Netherlands

0.222

0.055

0.152

SME, Small and medium-sized enterprise

Comparison of the predicted probabilities across countries produces interesting results. Recall that the predicted joint probability of interest rate and cost of finance tightening for the average firm is about 32 %. For countries that are involved in the sovereign debt crisis the corresponding probability is substantially higher, reaching values of 54 % (Spain), 49 % (Portugal), 47 % (Ireland), and 45 % (Greece). The corresponding probabilities for the Eurozone’s three largest economies are 15 % (Germany), 22 % (France), and 34 % (Italy). While Germany and France are considerably below the mean predicted probability, the same does not hold for Italy. Graph 1 depicts these predicted probabilities.
Graph 1

Predicted probability of interest rate and cost of finance tightening across (selected) countries

A similar picture emerges when the predicted probabilities for the remaining T&Cs pairs are considered. Portugal, Spain, Ireland, and Greece exhibit values substantially above the Eurozone average, while France and Germany show predicted probabilities far below the Eurozone mean. Again Italy is very close to the Eurozone average. A plausible explanation is that Italy has been increasingly involved in the debt crisis over time.

Equally interesting results are drawn when the time path of predicted joint probabilities are tracked in terms of the price T&Cs. The predicted joint probability for interest rate and cost of finance tightening exhibited a steep increase of about 144 % between the first semester of 2009 and the first semester of 2011, reaching values of 16 and 41 %, respectively. These findings give a vivid picture of the evolution of price T&Cs faced by SMEs in the Eurozone, namely, the price T&Cs have been steadily deteriorating. In contrast, non-price T&Cs did not exhibit any significant time variation during this period. The predicted probabilities for interest rate and cost of finance tightening are shown in Graph 2.
Graph 2

Predicted probability of interest rate and cost of finance tightening across time

With respect to the firm characteristics, those firms with increased net interest expenses tend to exhibit a higher predicted probability of T&Cs’ tightening, while the opposite holds for firms with increased profits. These are shown in Graph 3.
Graph 3

Predicted probability of interest rate & cost of finance tightening across (selected) firm types

4.1 Exploiting the panel dimension

A limited number of firms were surveyed in more than one wave, thus providing a panel dimension. Consequently, using this dataset it is possible to determine whether the results of the preceding analyses are still valid.

In this analysis, essentially pooled bivariate ordered probit models are investigated, since firm (random) effects will not be accounted for. Even though the sample size of the panel is considerably smaller, it allows two adjustments that provide some advantages compared to the pooled dataset. First, the panel structure is exploited by lagging all regressors into one period, which avoids any potential endogeneity of firm characteristics and also permits a causal analysis. Second, the time-means of all time-varying regressors can be included: \( \bar{z}^{\prime}_{i} \), in the spirit of a correlated random effects ordered probit (Mundlack 1978; Chamberlain 1984). The underlying rationale for this tactic is that the (between-) variation of time means would capture a portion of unobserved heterogeneity. The full specification is as follows:
$${\text{TC}}_{{i,1,t}}^{*} = z_{{i,1,t - 1}}^{\prime } \delta _{1} + \bar{z}_{{i,1}}^{\prime } \lambda _{1} + u_{{i,1,t}} {\text{TC}}_{{i,2,t}}^{*} = z_{{i,2,t - 1}}^{\prime } \delta _{2} + \gamma {\text{TC}}_{{i,1,t}}^{*} + \bar{z}_{{i,2}}^{\prime } \lambda _{2} + u_{{i,2,t}} $$
(6)
where \( \left( {\delta_{1} ,\delta_{2} ,\lambda_{1} ,\lambda_{2} } \right) \) are vectors of unknown parameters, \( \gamma \) is an unknown scalar, and \( \left( {u_{i,1,t} ,u_{i,2,t} } \right) \) are the error terms.
Table 6 shows the estimation results from the bivariate ordered probit models for each T&Cs’ pair. The findings can be summarized as follows. Starting with the price T&Cs, both show a significantly negative time trend, suggesting that the joint probability of tightening for the ‘interest rate/cost of finance’ are increasing over time. In all other T&Cs, the time trend plays no role. As far as ‘country heterogeneity’ is concerned, the results suggest that it accounts for a significant portion of the observed variation in T&Cs. Again, firms located in GIPS are found to have the highest likelihood of encountering T&Cs’ tightening, a finding which is more pronounced for the price T&Cs.
Table 6

Bivariate ordered probit model (based on panel dataset)a

Variables

Pair A

Pair B

Pair C

Dependent variable

Interest rates

Cost of finance

Loan maturity

Loan size

Covenants

Collateral requirements

\( \Updelta \left( S \right)_{i,t - 1}^{ + } \)

−0.018

(0.036)

0.046

(0.048)

−0.067

(0.056)

\( \Updelta \left( S \right)_{i,t - 1}^{0} \)

−0.096

(0.095)

−0.038

(0.061)

0.061

(0.083)

\( \Updelta \left( {\text{NIE}} \right)_{i,t - 1}^{ + } \)

0.427***

(0.106)

0.370***

(0.105)

0.022

(0.098)

−0.041

(0.092)

0.087

(0.100)

0.013

(0.150)

\( \Updelta \left( {\text{NIE}} \right)_{i,t - 1}^{0} \)

0.210**

(0.103)

0.254***

(0.096)

0.0007

(0.089)

0.002

(0.085)

−0.062

(0.098)

0.139

(0.126)

\( \Updelta \left( \pi \right)_{i,t - 1}^{ + } \)

−0.091

(0.093)

−0.066

(0.096)

−0.150

(0.099)

0.098

(0.093)

0.048

(0.101)

−0.023

(0.118)

\( \Updelta \left( \pi \right)_{i,t - 1}^{0} \)

−0.123

(0.081)

−0.101

(0.087)

−0.098

(0.091)

0.065

(0.082)

−0.038

(0.090)

0.055

(0.101)

\( \Updelta \left( {\text{D/A}} \right)_{i,t - 1}^{ + } \)

0.057

(0.110)

−0.008

(0.097)

0.056

(0.112)

−0.095

(0.093)

−0.009

(0.104)

−0.043

(0.125)

\( \Updelta \left( {\text{D/A}} \right)_{i,t - 1}^{0} \)

0.016

(0.107)

−0.049

(0.088)

−0.064

(0.090)

0.042

(0.082)

0.017

(0.095)

−0.040

(0.111)

Micro

0.106

(0.107)

−0.130

(0.326)

0.344

(0.357)

−0.325

(0.340)

−0.123

(0.330)

0.237

(0.395)

Small

0.216

(0.247)

−0.066

(0.282)

0.365

(0.336)

−0.288

(0.314)

0.181

(0.299)

−0.035

(0.380)

Medium

0.123

(0.320)

0.048

(0.239)

0.270

(0.285)

−0.249

(0.269)

−0.092

(0.251)

0.286

(0.340)

\( {\text{Age}}10^{ + } \)

−0.179

(0.277)

−0.050

(0.270)

0.185

(0.283)

−0.227

(0.272)

0.110

(0.264)

0.114

(0.379)

\( {\text{Age}}5 - 9 \)

0.066

(0.312)

0.195

(0.304)

0.171

(0.336)

−0.258

(0.307)

0.108

(0.303)

0.075

(0.388)

\( {\text{Age}}2 - 4 \)

−0.272

(0.352)

−0.067

(0.324)

0.193

(0.334)

−0.245

(0.316)

−0.142

(0.316)

0.321

(0.398)

\( {\text{Age}}2^{ - } \)

−0.176

(0.420)

0.047

(0.391)

−0.358

(0.375)

0.317

(0.359)

−0.187

(0.389)

0.622

(0.594)

Trend

−0.219**

(0.106)

−0.101***

(0.027)

−0.047

(0.030)

0.032

(0.026)

−0.022

(0.029)

0.037

(0.036)

Sector dummies

Included

Included

Included

Included

Included

Included

Country dummies

Included

Included

Included

Included

Included

Included

Time-series means

 \( \Updelta \left( S \right)^{ + } \)

−0.076

(0.106)

−0.054

(0.117)

−0.121

(0.114)

0.043

(0.113)

0.129

(0.123)

−0.032

(0.133)

 \( \Updelta \left( S \right)^{0} \)

0.155

(0.111)

0.212

(0.142)

−0.035

(0.125)

0.012

(0.116)

0.165

(0.138)

−0.150

(0.150)

 \( \Updelta \left( {\text{NIE}} \right)^{ + } \)

−1.427***

(0.234)

−1.168***

(0.146)

−0.0062

(0.139)

0.012

(0.133)

−0.509***

(0.146)

0.229

(0.341)

 \( \Updelta \left( {\text{NIE}} \right)^{0} \)

−0.428***

(0.170)

−0.302***

(0.139)

−0.041

(0.135)

0.046

(0.129)

0.176

(0.143)

−0.313

(0.195)

 \( \Updelta \left( \pi \right)^{ + } \)

0.454***

(0.143)

0.429***

(0.152)

−0.113

(0.172)

0.204

(0.143)

0.089

(0.155)

−0.019

(0.214)

 \( \Updelta \left( \pi \right)^{0} \)

0.285**

(0.134)

0.302***

(0.134)

−0.012

(0.164)

0.096

(0.124)

−0.074

(0.133)

0.088

(0.151)

 \( \Updelta \left( {\text{D/A}} \right)^{ + } \)

−0.175

(0.135)

−0.149

(0.139)

−0.247

(0.159)

0.293**

(0.134)

−0.203

(0.145)

0.173

(0.173)

 \( \Updelta \left( {\text{D/A}} \right)^{0} \)

−0.100

(0.126)

−0.132

(0.125)

−0.085

(0.121)

0.088

(0.115)

−0.030

(0.137)

0.104

(0.168)

 Micro

0.486

(0.484)

0.367

(0.494)

−0.369

(0.583)

0.342

(0.555)

−0.001

(0.514)

−0.339

(0.689)

 Small

0.534

(0.446)

0.490

(0.461)

−0.555

(0.574)

0.446

(0.535)

−0.332

(0.488)

−0.028

(0.686)

 Medium

0.387

(0.420)

0.376

(0.431)

−0.356

(0.534)

0.346

(0.507)

−0.041

(0.454)

−0.355

(0.660)

 \( {\text{Age}}10^{ + } \)

0.528

(0.399)

0.422

(0.399)

0.376

(0.477)

−0.391

(0.435)

0.053

(0.405)

−0.321

(0.610)

 \( {\text{Age}}5 - 9 \)

0.573

(0.431)

0.458

(0.431)

0.381

(0.507)

−0.328

(0.466)

0.029

(0.439)

−0.318

(0.641)

 \( {\text{Age}}2 - 4 \)

0.630

(0.495)

0.373

(0.455)

0.575

(0.514)

−0.588

(0.476)

0.310

(0.460)

−0.608

(0.59)

 \( {\text{Age}}2^{ - } \)

0.451

(0.563)

0.414

(0.563)

0.964

(0.579)

−0.896*

(0.526)

0.610

(0.580)

−1.293

(0.929)

Diagnostics

 Observations

2,392

2,325

2,290

 Wald test (overall significance)

412.20***

156.25***

211.96***

 Rhob

0.948***

(0.099)

−0.967***

(0.078)

−0.906***

(0.178)

 Gammac

−0.789***

(0.250)

1.091**

(0.068)

1.298***

(0.177)

 Log pseudo-likelihood

−3590.47

−3330.35

−2953.39

 Wald test of independent equations

3.35*

2.81*

2.28

*, **, *** denote significance at the 10, 5, and 1 % level, respectively

aSome covariates have been excluded for identification purposes

bDenotes the cross-equation error correlation

cDenotes the coefficient of the endogenous T&Cs’ element

In terms of firm characteristics, for the price T&Cs the (lagged) increased or unchanged net interest expenses tend to increase the likelihood of easing, which if taken at face value is at odds with economic theory. However, these effects are overturned when the impact of the net interest expenses time means are taken into account, which are significantly negative, implying an increased probability of tightening. In addition, the time means of increased and unchanged profits are significantly positive, leading to a higher likelihood of easing. For the non-price T&Cs there are no noteworthy impacts with respect to firm characteristics.

The predicted joint probabilities of tightening for T&Cs pairs across various SME subgroups are given in Table 7. Similar to the pooled dataset, there is a clear dichotomy in terms of the country effect, where GIPS exhibit a probability of tightening that is well above the Eurozone average. This pattern is encountered across all T&Cs, but it is more pronounced for the price T&Cs. In particular, the average predicted joint probability of tightening for price T&Cs is 32 %, while for Greece, Ireland, Spain and Portugal these are 45, 46, 53 and 54 %, respectively. In contrast, the corresponding probabilities for France, Germany, Italy and Netherlands are 21, 15, 34 and 19, respectively.
Table 7

Predicted probabilities of tightening across hypothetical SME subgroups (bivariate ordered probit model based on the panel dataset)

 

Interest rates and cost of finance

Loan size and loan maturity

Collateral requirements and covenants

Unconditional probability

0.365

0.070

0.277

Average predicted probability

0.321

0.044

0.210

SMEs subgroups

Predicted probability

\( \Updelta \left( {\text{NIE}} \right)_{i,t - 1}^{ + } = 1 \)

0.439

0.052

0.282

\( \Updelta \left( \pi \right)_{i,t - 1}^{ + } = 1 \)

0.266

0.030

0.161

First semester 2009

Second semester 2009

0.191

0.038

0.217

First semester 2010

0.259

0.041

0.212

Second semester 2010

0.338

0.043

0.203

First semester 2011

0.417

0.049

0.212

Greece

0.452

0.085

0.363

Ireland

0.462

0.117

0.309

Portugal

0.530

0.075

0.206

Spain

0.542

0.053

0.356

France

0.217

0.033

0.171

Germany

0.155

0.027

0.151

Italy

0.344

0.033

0.159

Netherlands

0.199

0.039

0.134

There is no time effect for the non-price T&Cs since for the pair ‘loan size/loan maturity’ the probability of tightening shows a rather weak tendency to increase, while for the pair ‘collateral requirements/covenants’ it is clearly flat. The elements of the price T&Cs show a totally different behavior. The predicted probability of joint tightening started at 19 % in the second semester of 2009 and has moved continuously upwards to 25 % (first semester 2010), 33 % (second semester 2010), and 41 % (first semester 2011).

Finally, firms with increased net interest expenses face a 43 % predicted probability of encountering a tightening of the joint price T&Cs, while the corresponding probability for firms with increased profits is just 26 %. These particular groups of firms show a small difference in predicted probabilities of about two 2 percentage points for ‘loan size/loan maturity’, while for ‘collateral requirements/covenants’ their difference is about 12 percentage points (28 % for firms with increased net interest expenses vs. 16 % for firms with increased profits).

All in all, the findings based on the panel dataset are in line with those obtained from the pooled dataset, indicating a robustness of the results.

5 Conclusions

The empirical analysis reported here utilized firm-level data from the European Commission and European Central Bank Survey on the access to finance of SMEs. Firms state their credit experiences with respect to the evolution of a variety of T&Cs’ elements, taking the form of tightening, being unchanged, or easing. In this analysis, I empirically explored the role of country, time, and firm profile to explain the observed variation in T&Cs.

According to these empirical results, country heterogeneity is an important factor in explaining T&Cs. In particular, a clear pattern is uncovered, with firms located in GIPS facing a substantially higher likelihood of encountering a tightening of T&Cs, a finding which could be attributed to the sovereign debt crisis. In addition, price T&Cs exhibit a significant tendency towards an increased tightening likelihood over time, strongly suggesting that the SMEs in the Eurozone face a steadily deteriorating environment. In terms of firm profile is concerned, net interest expenses and profitability contain important information for the observed variation in the evolution of T&Cs across SMEs. Essentially, the probability of T&Cs’ tightening is higher for firms whose interest expenses burden increases and lower for firms whose profitability increases.

The analysis suffers from several caveats that must be borne in mind. First, there is an intrinsic limitation to the dataset in that it is a narrow set of available firm-specific financial variables. Secondly, the firm responses are of a discrete and qualitative nature, which classifies them in the same group based only on direction (tightening etc.) without any reference to value.

Future research could be advanced in two ways: first, with the availability of better quality and more extensive datasets along the lines mentioned in the previous paragraph and second, with the availability of data directly related to banks’ decision-making. At the moment the only relevant available datasets correspond to the banking sector as a whole and consequently do not offer the required level of disaggregation.

Notes

Acknowledgments

I would like to thank two anonymous referees for their insightful comments. Any remaining errors and ambiguities remain my responsibility.

Financial disclosure

Financial support was provided by the Athens University of Economics and Business Research Center (ELKE).

References

  1. Audretsch, D., & Elston, J. (2002). Does firm size matters?: Evidence on the impact of liquidity constraints on firm investment. International Journal of Industrial Organization, 20, 1–16.CrossRefGoogle Scholar
  2. Berger, A., Klapper, L., & Udell, G. (2001). The ability of banks to lend to informationally opaque small businesses. Journal of Banking and Finance, 25, 2127–2167.CrossRefGoogle Scholar
  3. Berger, A., & Udell, G. (1990). Collateral, loan quality, and bank risk. Journal of Money, Credit and Banking, 8, 839–856.Google Scholar
  4. Berger, A., & Udell, G. (1995). Relationship lending and lines of credit in small business finance. Journal of Business, 68(3), 351–381.CrossRefGoogle Scholar
  5. Berlin, M., & Loeys, J. (1988). Bond covenants and delegated monitoring. Journal of Finance, 43(2), 397–412.CrossRefGoogle Scholar
  6. Berlin, M., & Mester, L. (1992). Debt covenants and renegotiation. Journal of Financial Intermediation, 2, 95–133.CrossRefGoogle Scholar
  7. Boocock, G., & Woods, M. (1997). The evaluation criteria used by venture capitalists: Evidence from a UK venture fund. International Small Business Journal, 16, 36–57.CrossRefGoogle Scholar
  8. Brick, I., & Palia, D. (2007). Evidence of jointness in the terms of relationship lending. Journal of Financial Intermediation, 16, 452–476.CrossRefGoogle Scholar
  9. Chamberlain, G. (1984). Panel data. In S. Griliches & M. Intriligator (Eds.), Handbook of econometrics (pp. 1247–1318). Amsterdam: North Holland.Google Scholar
  10. Cressy, R. (1996). Are business startups credit rationed? Economic Journal, 106, 1253–1270.CrossRefGoogle Scholar
  11. Cressy, R., & Toivanen, O. (2001). Is there adverse selection in the credit market? Venture Capital, 3(3), 215–238.CrossRefGoogle Scholar
  12. Diamond, D. (1984). Financial intermediation and delegated monitoring. Review of Economic Studies, 51, 393–414.CrossRefGoogle Scholar
  13. Diamond, D. (1991). Monitoring and reputation: The choice between bank loans and directly placed debt. Journal of Political Economy, 99, 688–721.Google Scholar
  14. Eurostat Pocketbooks. (2011). Key figures on European business with a special feature on SMEs. Eurostat, LuxembourgGoogle Scholar
  15. Gertler, M., & Gilchrist, S. (1994). Monetary policy, business cycles and the behavior of small manufacturing firms. Quarterly Journal of Economics, 109, 309–340.CrossRefGoogle Scholar
  16. Gilchrist, S., & Himmelberg, D. (1995). Evidence on the role of cash flow for investment. Journal of Monetary Economics, 36, 541–572.CrossRefGoogle Scholar
  17. Hu, X., & Schiantarelli, F. (1994). Investment and financing constraints: A switching regression approach using US firm panel data. Review of Economics and Statistics, 80, 466–479.CrossRefGoogle Scholar
  18. Jensen, J., & McGuckin, R. (1997). Firm performance and evolution: Empirical regularities in the US microdata. Industrial and Corporate Change, 6, 25–47.CrossRefGoogle Scholar
  19. Merton, R. (1974). On the pricing of corporate debt: The risk structure of interest rates. Journal of Finance, 29(2), 449–470.Google Scholar
  20. Mundlack, Y. (1978). On the pooling of time series and cross-section data. Econometrica, 46(1), 69–85.CrossRefGoogle Scholar
  21. Oliner, S., & Rudebusch, G. (1992). Sources of the financing hierarchy for business investment. Review of Economics and Statistics, 74(2), 643–654.CrossRefGoogle Scholar
  22. Petersen, M., & Rajan, R. (1994). The benefits of lending relationships: Evidence from small business data. Journal of Finance, 49, 3–37.CrossRefGoogle Scholar
  23. Posey, R., & Reichert, A. (2011). Terms of lending for small business lines of credit: The role of loan guarantees. International Journal of Business and Finance Research, 5(1), 91–102.Google Scholar
  24. Sajaia, Z. (2008). Maximum likelihood estimation of a bivariate ordered probit model: Implementation and Monte Carlo simulations. The Stata Journal, 2,1–18.Google Scholar
  25. Strahan, P. (1999). “Borrower risk and the price and nonprice terms of bank loans”, Staff Report 90. New York: Federal Reserve Bank of New York.Google Scholar
  26. Vijverberg, C. (2004). An empirical financial accelerator model: Small firms’ investment and credit rationing. Journal of Macroeconomics, 26, 101–129.CrossRefGoogle Scholar
  27. Watanabe, W. (2005). How are loans by their main bank priced? Bank effects, information and non-price terms of contract. The Research Institute of Economy, Trade and Industry (RIETI) Discussion Paper Series 05-E-028. RIETI, Tokyo. Available at: http://www.rieti.go.jp/en

Copyright information

© Springer Science+Business Media New York 2012

Authors and Affiliations

  1. 1.Department of Accounting and FinanceAthens University of Economics and BusinessAthensGreece

Personalised recommendations