Lending Conditions, Macroeconomic Fluctuations, and the Impact of Bank Ownership

Transcription

1 Lending Conditions, Macroeconomic Fluctuations, and the Impact of Bank Ownership Abstract This paper provides evidence on the drivers and consequences of fluctuations in bank behavior along the economic cycle. Using bank micro data from Germany, we find that commercial and cooperative banks adjust their lending volume and conditions cyclically, whereas that link is much less pronounced for public sector savings banks. This result is stable regarding different measures of business activity, including regional economic conditions. However, the impact of a bank s size and capitalization on the cyclicality of its lending policy remains ambiguous. We also find that higher fluctuations in loan rates and a more stable loan growth policy lead to higher individual bank profitability. These results provide valuable implications for bank management, borrowers, and supervisors, and enrich the debate on the optimal design of financial systems. JEL classification: G20, G21 Keywords: Bank lending, Bank ownership, Business cycles

2 1. Introduction The common observation that bank behavior (lending policy, financing, riskiness) fluctuates over time, and particularly along the macroeconomic cycle, has inspired recent contributions of academics, regulators, and practitioners alike (see e.g. Rajan, 1994; Lown and Morgan, 2006; Quagliariello, 2007). However, research could not yet provide clear evidence on the question which banks behave more or less cyclical compared to others under similar circumstances stated more generally, on the drivers of cyclical behavior. Furthermore, the consequences of more or less cyclical bank behavior regarding profitability and riskiness are still unclear. This paper provides robust and surprising evidence on these questions using a nearly 20 year-long bank micro data set from Germany. Financial systems are usually classified into market-based types, like the United States or the United Kingdom, and bank-based systems, like Germany, France, or Italy (Levine, 2002). The latter category is characterized by a high dependency of company and private borrower financing on stable bank-customer relationships rather than on quick transactions with changing creditors. Notice however that the lending policy of individual banks differs considerably even within these systems, enabling us to analyze determinants of bank behavior under relatively homogeneous systemic conditions for Germany. In the German financial system, which is composed of three distinct pillars of public sector savings banks, cooperative banks (credit unions) and private commercial banks (see Hackethal, 2004), both savings and cooperative banks claim that they commit to their clients in good times as in bad, which distinguishes them from other banks. 1 As they are also supposed to represent the main financing source for small and medium enterprizes, one might expect these banking sectors being important drivers of economic stability. It is a timeless debate how financial systems should be shaped in general (see for an overview 1 Extracts from image brochures: Savings banks take responsibility for sustainable growth of Small and Medium Enterprizes (SME), which they rank higher than fast transactions. Even in times when large private banks reduce SME lending, savings banks and Landesbanks reliably provide financing for these companies., or stated by cooperative banks: The principles of helping our members and clients to help themselves and supporting their entrepreneurial activities are in the foreground [of our behavior]. 2

3 Allen and Gale, 1999), and for Germany, reports like the Financial Systems Stability Assessment of the International Monetary Fund (IMF, 2003) recommend to reduce barriers to consolidation, within or across pillars, and thereby facilitate market-oriented restructuring. On the other hand, the three-pillar structure of the German financial system reliably provided financial services and was a driver of permanent economic stability in the last 50 years. Notice however that, as the IMF annotates, average bank profitability has decreased in recent time, questioning the sustainable stability of the German system unless intensified structural changes take place. Taking an individual bank s perspective, we analyze how a bank s ownership affects its tendency to cyclically adjust lending conditions, potentially providing us insights concerning the assets and drawbacks of Germany s three-pillar banking system. In this paper, we test the following three hypotheses (H1 H3) regarding cyclical bank behavior. Firstly, we take up the claim of savings banks and cooperatives to commit to their clients in good times as in bad and hypothesize that a) savings banks and b) cooperative banks follow a less cyclical lending policy than private commercial banks (H1). This may also be related to savings banks public mission of economic support and the commitment of credit unions to support their cooperative members. Secondly, an important precondition of constant bank behavior is a sufficient bank capitalization that allows for equity reductions in recessions, and for equity increases as economic conditions improve. Additionally, we believe that smaller banks with their more direct bank-customer relationship are more likely to maintain their lending conditions throughout the cycle. Therefore, we hypothesize that a) better capitalized banks and b) smaller banks follow a less cyclical lending policy (H2). Thirdly, we analyze the consequences of cyclical behavior and investigate whether banks which cyclically adjust lending conditions exhibit higher profitability irrespective of the economic situation (H3). If this were the case, a switch from a stable to a cyclical lending policy could, despite the destabilizing effect of cyclical bank behavior, positively influence economic stability due to higher bank solvency. 3

4 The contribution of our paper to the existent literature is twofold. First, we address the literature on bank ownership by analyzing how public or cooperative bank ownership is related to the cyclicality of bank behavior. Second, we contribute to the mainly macrooriented literature on lending and business cycles, which is closely linked to stability aspects, by providing new evidence from micro level data. In doing so, we also consider the significance of regional economic conditions for bank lending. Additionally, besides the drivers of more or less cyclical bank behavior, we analyze its consequences as well. Our results have implications for bank owners and management, borrowers, and supervisors. The literature on state-owned enterprizes justifies their existence on one hand with social welfare maximization through the prevention of market failure, e.g. the provision of financial services in regions or industries where credit rationing occurs. On the other hand, politicians as the creators of government-owned enterprizes may be self-interested and run these firms so as to increase their own wealth, maximize voting support, create jobs for party members etc. La Porta et al. (2002) are among the first to extend this general literature to an analysis of the government ownership of banks. They document that public banking is prevalent all over the world, and an important determinant of financial development. In another seminal paper, Sapienza (2004) finds that state-owned banks set lower loan rates and favor firms in depressed areas as well as large firms. Notice that the latter finding, which is based on Italian individual loan data from the period , contradicts German savings banks claim to particularly support small and medium-sized enterprizes. Contrarian evidence regarding the support of these companies is provided by Delgado et al. (2007), who show that indeed Spanish savings banks and cooperatives specialize in relational lending and small firm funding. Taking a more systemic perspective, Bichsel (2006) analyzes whether state-owned banks in Switzerland enhance banking competition. He finds that public banks margins are not relatively low, nor are their interest rates particularly borrower-friendly or relatively costsensitive. Hence, a disciplinary effect of state-owned banks on competitors prices cannot 4

5 be detected. Using a large data set comprising 179 countries, Micco et al. (2007) detect cost and profitability disadvantages of state-owned banks in industrial countries. More specifically, analyzing large banks in 14 European countries, Iannotta et al. (2007) find that both cooperatives and government-owned banks face lower costs, but nevertheless exhibit a lower profitability than privately owned banks. While public sector banks in their sample are characterized by poor loan quality, the opposite is true for cooperative banks. Theoretical work regarding mutual lending institutions has been established by Smith et al. (1981), focusing on collective decision making in credit unions, and O Hara (1981), taking a property rights perspective. In a recent comprehensive paper, Fonteyne (2007) claims that as being owned by their consumers, cooperative banks success can be explained by lower informational asymmetries. Most importantly for our analysis, Hesse and Cihak (2007) find that cooperative banks returns are indeed more stable than those of commercial banks, however not directly relating this finding to the business cycle. The macro-oriented literature relevant for our analyses is based on the theory of a supply-driven bank lending channel and a demand-driven balance sheet channel (Bernanke and Gertler, 1995). Recent empirical research has been conducted by Bikker and Hu (2002), who find in a large, international sample, that bank profits and lending volume fluctuate procyclically, and that loan growth seems to be rather demand- than supply-driven. Demirguc-Kunt and Huizinga (1999) document that net interest margins and the profitability of banks in 80 countries during are positively related to the respective interest rates. Finally, papers like Quagliariello (2007) show how banks riskiness fluctuates over the economic cycle, also taking into account the possible procyclical consequences of banking regulation. Combining this literature with bank ownership aspects, the paper most closely related to ours is by Micco and Panizza (2006), who analyze international bank-level data for the period stemming from Bankscope. They find that loan growth is 5

6 positively related to GDP growth, with this procyclical link to be less pronounced for state-owned banks. When restricting the sample to industrial countries, however, the significance of this result is rather low. Furthermore, their finding does not prove true when analyzing the growth of other earning assets (except loans), indicating that public banks intentionally smooth credit and do not just behave lazy. The remainder of this paper is structured as follows. Section 2 describes the macro and micro data used, reports summary statistics, and explains peculiarities of the German banking system. Section 3 presents the results from multivariate tests of our three hypotheses and reports robustness checks. Section 4 provides a summary and concludes. 2. Description of the data We analyze yearly balance sheet and income statement data for individual banks in Germany during the period provided by the German private data supplier Hoppenstedt. Our sample is restricted to public sector savings banks, cooperative banks (credit unions) and commercial banks (credit banks). Specifically, development banks, investment advisory firms, thrift institutions, branches of foreign banks, and other specialized banks (including Landesbanks and head institutions of cooperatives) are excluded. As we focus our analyses on the lending activity to customers, banks are as well excluded if the fraction of total customer loans over total assets is below 25%. The raw data set is an unbalanced panel, and to be able to analyze bank behavior over the business cycle, banks are only considered if they exhibit at least 5 bank-year observations, respectively. In total, our sample thus consists of 483 savings banks, 382 cooperatives, and 91 commercial banks, summing up to 13,665 bank-year observations from 956 institutions. Table 1 reports summary statistics. Insert Table 1 here 6

7 It can be seen in Panel A that our final sample covers more than 80% of the the savings banks in Germany (as reported in Deutsche Bundesbank statistics for 2004), whereas the coverage for cooperatives and commercial banks is below 25%. Notice however that a comparison of the accumulated bank assets in our sample and in Germany, both for the year 2004, shows that for all banking sectors, our data set covers at least 70% of total assets. 2 This means that the banks in our sample account for approximately 80% of commercial banking activity, underlining their representativeness from a systemic perspective. Panel B summarizes the main variables employed in our empirical analyses for the full sample as well as differentiated by banking sectors. All statistics are referring to bank-year observations. It is evident that savings banks are on average almost twice as large as cooperatives in terms of total assets or total customer loans. The mean size of commercial banks is again more than 6 times higher than that of savings banks. Also, there is a lot of variation within sectors, and we therefore carefully control for bank size in the subsequent analyses. Our main measure of lending behavior is loan growth, defined as the percentage change of total customer loans: LG t = total customer loans (t) total customer loans (t 1) 1. We intentionally exclude lending to other banks, as this is a separate business activity with a fundamentally different risk-return structure. Amounting to 7.17%, the mean of commercial banks loan growth is 2 percentage points higher than average loan growth of savings banks and cooperatives (5.20%). Notice that values for loan growth below 14.99% (0.5%-quantile) and above +37, 45% (99.5%-quantile) are considered as outliers, possibly due to data errors, and excluded from our analyses. Since we cannot directly observe loan rates, we have to rely on an indirect measure of the average loan rate, the relative interest income RII t = interest income from loans (t) average total loans (t 1,t), which is on average slightly higher for commercial banks. 2 The coverage ratio of 78.4% for commercial banks is slightly upwards biased. Bundesbank statistics are based on individual financial statements, and we generally also use individual statements, however replace them with consolidated statements whenever they are not available. 7

8 As loans that are additionally granted throughout the year t do not pay full interest for that year, interest income is scaled with average total lending in the 2 preceding years in the denominator. Outliers below the 0.5%-quantile (4.09%) and above the 99.5%-quantile (14.02%) are truncated as mentioned above. As bank-specific control variables serve the relative net interest result (RNIR t ), which is defined accordingly to the relative interest income (RII t ), except that in the numerator of the fraction, interest expenses as the bank s refinancing costs are subtracted. Panel B of Table 1 documents that this bank profitability measure averages 3.32% for commercial banks and is much lower for cooperatives (2.19%) and savings banks (1.62%). The equity-to-total assets ratio (ET A t ) as a key measure of bank solvency amounts on average to 8.40% for commercial banks, 5.01% for cooperatives, and 4.36% for savings banks, possibly due to the government s liability for savings banks in Germany that still existed during most years in our sample period. Additionally, we control for the maturity structure of a bank s loan portfolio by defining the long term loan ratio LT LR t = customer loans with maturity > 5 years total customer loans, which is on average relatively similar for cooperatives (61.1%) and savings banks (69.7%), however much lower for commercial banks (39.9%). The interbank loan ratio IBLR = interbank loans total lending reveals that commercial banks (22.0%) are on average more active in lending to other banks than are cooperatives (17.5%) and savings banks (12.5%). Summarizing, our descriptive statistics document cooperatives and savings banks being relatively similar in many bank characteristics, allowing us for an immediate comparison of our results at least for these groups. It is well known that there has been a substantial mergers and acquisitions activity of German banks in recent years (see for possible reasons and consequences Elsas, 2007). When the assets and the income of two merging banks are pooled and consolidated, it is evident that structural changes in their growth and profitability measures take place that cannot be explained by our empirical models of ordinary bank behavior. As the data do not allow us to directly control for M&A activity, we introduce a double criterion 8

9 to detect the respective bank-year observations, which proved to work quite accurately in a small control sample. Observations are dropped if both the equity rises by more than 26.9% (95%-quantile) and the number of employees increases by more than 17.4% (95%-quantile) for the same year and bank. The overlap of the two sub-criteria is quite high (72.4%), and so a total of 513 observations are dropped. Insert Figure 1 here Macro data for Germany, which are our key explanatory variables, are taken from OECD statistics and illustrated in Figure 1. We concentrate on the real GDP growth rate, representing the state of the German economy in general and loan demand in particular. During the 19 years of our sample period, there have been at least two phases of economic boom ( and ) as well as two major recessions ( and ). Macroeconomic interest rates are important determinants of loan rates and bank profitability, and we therefore also include the 3-month interest rate as well as the 10-year government bond yield for Germany. As Figure 1 shows, there is a clear co-movement of the two variables, with higher fluctuations of the short-term interest rate. Having a peak in the early 1990s, interest rates have steadily decreased thereafter. 3 With its distinct separation of public sector savings banks, cooperatives, and private commercial banks, the German banking system is particularly well suited for an analysis of bank ownership s consequences on lending behavior. However, as there are several peculiarities of German banking sectors, some additional explanations are in order. First, recall that the German financial system is characterized as rather bank-based with a relative importance of stable bank-customer relationships and house banks. Second, the lending activity of savings banks and cooperatives is generally restricted to exactly one distinct geographic market. Germany is partitioned into these regions, so that there is hardly any competition within the savings banks or the cooperative banks sector. Hackethal (2004) gives an overview over other institutional characteristics. 3 An exception is the increase of interest rates at the end of the economic upturn in the year

10 3. Empirical analysis 3.1. Bank ownership and the cyclicality of lending behavior In this section we test the impact of bank ownership on the volatility of bank lending, in particular on the link between the economic cycle and lending behavior. We start with a graphical analysis of the volatility of bank behavior over the cycle that is displayed in Figure 2. Insert Figure 2 here In a comparison of average volatility of bank behavior between commercial banks, cooperatives, and savings banks, the unbalanced nature of our panel may lead to distortions. We therefore restrict this analysis to banks for which at least 17 observations are available, which reduces the number of banks by 48.9%. The standard deviation of loan growth (Panel A) and of the relative interest income (Panel B) is calculated from the time-series for each of the 500 remaining banks. Box-plots show how these standard deviations are distributed across banking sectors. It turns out in Panel A of Figure 2 that, although mean loan growth rates are similar for cooperatives and savings banks, banks in the latter group exhibit significantly less variation in loan growth. As expected in H1, both cooperatives and savings banks lending activity fluctuates less than that of commercial banks. Evidence for the volatility of the relative interest income in Panel B is less clear, but again, average loan rates of commercial banks fluctuate significantly more than those of cooperatives and savings banks. This could however be related to the fact that the fraction of long-term loans is lowest for commercial banks, allowing them to adjust loan rates at relatively short notice. Please notice that the preceding analysis can only reveal how bank lending behavior fluctuates in general, which may well be due to managerial decisions not systematically related to the business cycle. Therefore, we propose the following multivariate model to 10

11 explain changes in loan growth of bank i in year t: LG i,t = α + β 1 GDP t + β 2 LG i,t 1 + β 3 RII i,t + β 4 ET A i,t + β 5 LT LR t + ɛ i,t We believe that it is rather the change in the loan growth rate from the year t 1 to t than the actual growth rate that is determined by macroeconomic fluctuations and therefore define our entire model in first differences. Thereby, we also implicitly control for possible bank-specific fixed effects in our panel data set. According to theory and the empirical literature, we expect a procyclical lending policy and thus a positive coefficient for our main explanatory variable GDP t. Additionally, we include the lagged dependent variable LG i,t 1, as it might be that loan growth exhibited an unsystematic (positive or negative) shock in the preceding year, possibly due to the introduction or the abolishment of new lending products or because of external bank growth activity that our M&A criterion fails to detect. As we expect this unsystematic effect to be reversed in the subsequent year, the coefficient β 2 should be negative. The change in the relative interest income RII i,t is included as control variable. On one hand, it is more likely that a bank expands its lending when new, profitable lending segments yielding relatively high loan rates are discovered. On the other hand, Foos et al. (2007) report that loan growth could also have a negative impact on the relative interest income, as banks may be underpricing their competitors to attract new customers. Foos et al. (2007) also show that the equity-to-total assets ratio is negatively related to loan growth, and we thus include changes in that ratio ( ET A i,t ) as additional control variable. Finally, a high long term loan ratio (LT LR i,t ) is obstructive for a bank to adjust its lending cyclically. We control for changes in that ratio and expect it to be negatively related with loan growth. Results from OLS regressions with Huber-White robust standard errors, controlling for clustering at the individual bank-level, are reported in Table 2. 11

12 Insert Table 2 here Panel A compares regression results for our full sample with findings for each bank group. First, as we expected, a positive and significant impact of GDP growth on loan growth is confirmed. Our control variables also show their expected sign. Second, a comparison of the estimated coefficients for commercial banks, cooperatives, and savings banks leads to interesting results. It turns out that the link between GDP growth and loan growth for savings banks is much less pronounced than for commercial banks or cooperatives. Notice that the coefficient for cooperatives is 6 times higher than that for savings banks, 4 which is even more surprising as these banks are similar in many respect. Furthermore, the loan growth of cooperatives seems to be more sensitive to the relative interest income and to the long term loan ratio, whereas savings banks growth depends to a higher degree on changes of the equity-to-total assets ratio. These results prove true if we exclude banks for which less than 17 observations are available (quasi-balanced panel), however for brevity, results are not presented here. As we believe that a bank s reaction to positive macroeconomic shocks may differ from that to negative shocks, we restrict our analysis in Panel B of Table 2 to years with accelerated economic growth ( GDP t > 0) and in Panel C to years with weakened economic growth ( GDP t < 0). To obtain a homogeneous sample in these two panels, only banks with at least 17 bank-year observations are analyzed. Our empirical results show that the reaction on an economic improvement is substantially stronger than on an economic slowdown. Aditionally, the pattern that savings banks tend to adjust their lending less cyclically than cooperatives is confirmed for both differentiations. Hence, our analyses until now strongly support Hypothesis H1 regarding savings banks cyclicality. One problem with the linear regression model presented so far is that there may be unobserved macroeconomic effects in some years which are not captured by GDP growth. Therefore, we also test the link between fluctuations in lending and the economic cycle 4 Tests as proposed by Chow (1960) confirm that coefficients are significantly distinct. 12

13 using the following fixed effects model for loan growth (LG i,t ), which is similar to that proposed by Micco and Panizza (2006): LG i,t = α + β 1 (SAV i MACRO t ) + β 2 (COOP i MACRO t ) + β 3 (SIZE i,t MACRO t ) + β 4 (ET A i,t MACRO t ) + γ i + δ t + ɛ i,t Indicator dummy variables for savings banks (SAV i ) and cooperatives (COOP i ) are interacted with our main explanatory variable MACRO t. Furthermore, we control for bank size using the natural logarithm of total assets (SIZE i,t ) as well as for bank capitalization using the equity-to-total assets ratio (ET A i,t ) and interact both variables with MACRO t. Fixed effects are included on a bank-specific (γ i ) and year-specific (δ t ) level. Table 3 presents the empirical results. Insert Table 3 here For an interpretation of the coefficients from our fixed effects models, please notice that the year-specific fixed effect captures the entire macroeconomic impact on the loan growth rate in each year. We learned from our OLS regressions that loan growth generally shows a procyclical behavior. Thus, if some explanatory variable causes the extent of cyclical bank behavior to diminish, the coefficient of the interaction term should show a negative sign, and vice versa. All regressions in Table 3 are estimated for the quasi-balanced panel including only banks with at least 17 observations each. Panel A uses the GDP growth rate as macroeconomic explanatory variable for loan growth. The interaction terms with SAV i have a highly significant and negative coefficient, meaning that savings banks indeed behave less cyclically. Furthermore, large banks and banks with a lower equity-to-total assets ratio exhibit a slightly more cyclical lending policy. These findings are robust for different specifications of control variables in regression models 2 and 3. 13

14 In Panel B of table 3, we apply another macroeconomic variable, namely the business climate index of the economic situation in Germany provided by the Ifo Institute for Economic Research at the University of Munich, as an early indicator of economic activity in Germany. It could well be that the GDP growth rate used so far is a poor proxy for the current state in the economic cycle due to its merely backward-looking definition. Most importantly, all findings from Panel A are confirmed with this alternative measure of business activity. Notice however that the explanatory power of the GDP growth rate is substantially higher, as R-squares of the models in Panel A are almost 8 times larger. Therefore, in our subsequent analyses, we prefer using the GDP growth rate as as explanatory macro variable. 5 Another important aspect of lending behavior is the cyclical adjustment of loan rates. If it is indeed the case that savings banks and cooperatives commit to their customers in good times as in bad, this should imply relatively constant loan rates over the macroeconomic interest rate cycle. We model adjustments in the average loan rate RII i,t as follows: RII i,t = α + β 1 INT EREST t + β 2 RII i,t 1 + β 3 SIZE i,t + β 4 LG i,t + β 5 ET A i,t + β 6 IBLR i,t + β 7 LT LR i,t + ɛ i,t As are our regressions for the cyclicality of loan growth, this model is defined in first differences. The main explanatory variable is the macroeconomic interest rate INT EREST t, for which we employ the 3-month interest rate (INT 3M t ), or alternatively the 10-year government bond yield (INT 10 t ). Since a bank s refinancing costs as well as the competitive loan rate are strongly positively associated with the key interest rate, we expect a positive impact on RII i,t. The change in the lagged dependent variable RII i,t 1 is included as it is not unlikely that there is some serial correlation between changes in loan rates. We further control for bank size using the natural logarithm of total assets 5 One exception is the robustmess check using regional economic conditions in section

15 (SIZE i,t ) as well as for loan growth (LG i,t ) and for the equity-to-total assets ratio (ET A i,t ). More importantly, the interbank loan ratio (IBLR i,t ) should have a negative impact on the relative interest income, as loan rates in interbank lending are typically lower than those in customer lending. Moreover, a high long term loan ratio (LT LR i,t ) positively influences the average loan rate as long as there is a normal term structure of interest rates, and we therefore expect a negative coefficient β 7. Results from OLS regressions are reported in Table 4 Insert Table 4 here In Panel A, the 3-month interest rate is used as main explanatory variable. The expected positive impact of the macroeconomic interest rate is confirmed for all banking groups, and it turns out that the link between INT 3M t and the relative interest income RII i,t is significantly less pronounced for cooperatives and for savings banks than for commercial banks. 6 This is not exclusively driven by the higher average long term loan ratio of the latter groups, as we firstly control for changes in that ratio in our regressions and secondly obtain the same pattern when considering only commercial banks with high LT LR i,t as reference group. Notice that there is indeed a positive serial correlation for changes in the relative interest income. Furthermore, larger banks tend to have higher average loan rates than smaller institutions. The coefficients for loan growth and the equity-to-total assets ratio have very small values and are thus economically not significant. As expected, a high fraction of interbank loans over total lending influences the relative interest income negatively. Finally, only for cooperative banks, a positive impact of the long term loan ratio on RII i,t can be observed, which is consistent with our prediction. Summarizing, in accordance with Hypothesis H1, cooperatives and savings banks adjust their loan rates less cyclically. Panel B uses 10-year government bond yields instead of short-term interest rates to explain fluctuations in the average loan rate. The findings of Panel A are confirmed 6 Tests as proposed by Chow (1960) confirm that coefficients are significantly distinct. 15

16 with two exceptions: Bank size is not significant any more, and the long term loan ratio has a negative impact on the relative interest income, possibly because adjustments of loan rates are not feasible at short notice if loan maturities are on average long. Altogether, our analyses reveal that cooperative banks indeed adjust their loan rates less cyclically than commercial banks, and public sector savings banks follow an even less cyclical lending policy with regard to interest income as well as lending volume. Hence, Hypothesis H1 is mainly supported by our empirical results Bank capitalization or size and the cyclicality of lending behavior In the analyses in the preceding section, it remains open whether our findings are exclusively driven by differences in ownership. It could well be that the on average smaller size or better capitalization of cooperatives relative to savings banks is actually responsible for their more cyclical behavior. Therefore, we reconsider our baseline model from section 3.1 to explain changes in loan growth and classify banks in each group into three size or capitalization terciles. Recall that we hypothesized in H2 that better capitalized banks and smaller banks follow a less cyclical lending policy. Insert Table 5 here Our analysis starts in Table 5 with a differentiation by size terciles for the full sample (Panel A), cooperative banks (Panel B) and savings banks (Panel C). We calculate the average size (total assets) for each bank over time and classify it into one of the models 1 3. No separate breakdown of commercial banks is conducted as their quite small sub-samples (terciles) fail to provide significant results. For savings banks, it turns out that the smallest tercile exhibits the strongest impact of GDP growth on loan growth, which contradicts our proposition of less cyclicality due to a more direct bank-customer relationship for these banks. Among cooperatives, the differentiation by bank size does not lead to significantly different coefficients, and these are still four times higher than 16

17 for the smallest savings banks tercile. Thus, it is unlikely that the observed stronger cyclicality of cooperative banks lending is solely driven by their on average smaller size. Also, the size pattern in our full sample analysis (Panel A) mainly represents the unobserved ownership characteristics, in particular the less pronounced cyclical behavior of savings banks which are mostly classified into the mid oder even upper size tercile. This is even more striking as our fixed effects models in Table 3 reveal a small, however significantly positive impact of bank size on the cyclicality of lending, which is also in line with our prediction in Hypothesis H2. Insert Table 6 here Table 6 reports the results regarding the impact of bank capitalization on the cyclicality of loan growth. Again, banks are classified into terciles according to their average equity-to-total assets ratio. Panel C shows that only savings banks with a relatively good capitalization exhibit a significant link between GDP growth and loan growth. For cooperatives (Panel B), which have on average a higher equity-to-total assets ratio, the cyclicality of lending behavior is relatively homogeneous across capitalization terciles. Notice again that the seemingly clear pattern in Panel A s full sample analysis seems to be mainly driven by ownership characteristics. Our findings from the fixed effects models in Table 3 suggest that bank capitalization influences the cyclicality of lending negatively, which is again consistent with Hypothesis H2. Notice also that Stolz and Wedow (2005) observe the capital buffer of German savings and cooperative banks to fluctuate anticyclically, which is more pronounced for savings banks. This could be a consequence of their relatively stable lending behavior, inducing capital ratios to diminish in bad times and to recover in economic upturns. Like for bank size, however, we cannot find clear evidence for or against Hypothesis H2, which implies that bank ownership turns out to be the dominant driver of more or less cyclical lending behavior. 17

18 3.3. Consequences of cyclical bank behavior So far, we focused on the drivers of cyclicality in loan growth and loan rate setting. However, from a systemic point of view, it is also interesting whether banks which behave more cyclically gain profitability advantages. This proposition is not trivial as a stable or even countercyclical lending policy could be helpful to establish durable bank-customer relationships. These are very important in Germany s (house) bank-based financial system and may lead to economic rents, e.g. due to informational advantages. Following this reasoning, it is clear that we should define bank profitability as an average over the economic cycle, since the consequences of cyclicality in bank behavior may only appear in the long run. Therefore, we propose the following regression model for average profitability P ROF i of bank i: P ROF i = α + β 1 SD LG i + β 2 SD RII i + β 3 SIZE i + β 4 ET A i + β 5 SAV i + β 6 COOP i + ɛ i The main explanatory variables are the standard deviations for each bank s loan growth SD LG i and relative interest income SD RII t. Although these variables do not measure whether fluctuations in lending behavior are linked to the economic cycle, they should be suitable proxies for cyclicality in bank behavior. As control variables, the natural logarithm of total bank assets (SIZE i ) and the equity-to-total assets ratio (ET A i ) are added. Since existing papers report profitability (dis)advantages of savings banks and cooperatives, we additionally include indicator dummy variables for these banking sectors (SAV i and COOP i ). Models are estimated as OLS regressions with robust standard errors, coefficient estimates are displayed in Table 7. Insert Table 7 here A total of three alternative measures of average bank profitability P ROF i are employed. We use the relative net interest result RNIR i for our analyses in Panel A, the average 18

19 return on equity ROE t = on assets ROE = operating result (t) average book equity (t 1,t) operating result (t) average total assets (t 1,t) in Panel B, and the average return in Panel C. The operating result is defined as earnings from ordinary operations before taxes. It turns out that the standard deviation of loan growth shows a negative impact on the bank s return on equity, whereas its coefficients in Panels A and C are statistically and/or economically not significant. This means that a stable lending lending policy may also lead to profitability advantages. However, we observe a positive and significant impact of the standard deviation of average loan rates SD RII i on the relative net interest result, indicating that it pays for a bank to adjust loan rates cyclically. These findings are confirmed when controlling for the possible multicollinearity between SD LG i and SD RII i in models 2 and 3. Coefficients for SIZE i are always significantly negative, which means that smaller banks have profitability advantages over larger ones. Furthermore, banks with a higher equity-to-total assets ratio ET A i exhibit on average a higher net interest result and return on assets. This could however be due to the fact that these banks obtained a better capitalization because of their higher profitability. Interestingly, our ownership dummy variables both for savings and cooperative banks show insignificant coefficients (with two exceptions), indicating that fundamental profitability (dis)advantages for these banking groups do not exist. These results have important implications from a systemic point of view. First and most importantly, it is not bank ownership per se that drives its profitability. Also, a relatively stable policy with regard to loan growth does not necessarily reduce bank profitability, but influences a bank s return on equity positively. However, our finding that the average interest income fluctuates less for cooperatives and least for savings banks could have a detrimental impact on interest margins. Thus, bank supervisors and politicians have to weigh the provision of relatively stable lending conditions against the advantage of more profitable savings banks. Notice that in a sense, the latter would also improve the stability of the German financial system. 19

20 3.4. Robustness Checks We complete the empirical analysis by testing our results for robustness with regard to two aspects. First, as the activity of savings and cooperatives banks is restricted to distinct geographic regions, it could be that they adjust their lending to regional economic conditions rather than to the economic cycle for Germany. Second, savings banks and cooperatives might behave less cyclically just because their management does not face strong incentives to regularly adjust the lending policy, i.e. they do not intentionally exhibit stable bank behavior but are rather lazy. This would contradict our proposition that savings banks public mission of economic support and the commitment of credit unions to support their cooperative members motivates their less cyclical lending policy. Germany consists of 16 federal states, and most states are further divided into administrative regions, so that we obtain 40 distinct regions for which economic data are available for the period We reconsider our baseline model from section 3.1 to explain loan growth and substitute the GDP growth rate for Germany by the growth rate of regional per-capita GDP (RGDP t ). Regression results are reported in Table 8. Insert Table 8 here It turns out that for all three specifications, the entire sample period in Panel A, and years with accelerated (weakened) regional economic growth in Panel B (C), the link of loan growth to regional economic conditions is less pronounced than to the German GDP growth, as reported in Table 2. This could be due to higher unsystematic fluctuations in regional economic activity adding too much noise in our main explanatory variable which is hence unable to explain lending behavior properly. In any case, our concern that savings and cooperative banks lending behavior could in fact be driven by regional economic conditions is eliminated. The issue of possibly lazy savings or cooperative banks is addressed by analyzing other aspects of bank behavior except lending to customers. Specifically, we investigate 20

21 the deposit taking activity and the evolution of provision income over the economic cycle. The first difference of the percentage change in total deposits from year t 1 to t, as well as the first difference in the growth of provision income are to be explained in OLS regressions, where we apply again our baseline models from section 3.1. Coefficient estimates are reported in Table 9. Insert Table 9 here Our empirical results suggest that the less cyclical behavior of savings banks and cooperatives is restricted to their lending activity. As Panel A documents, deposit taking of banks shows a countercyclical pattern with negative coefficients. These are slightly higher for commercial banks, however fundamental differences between cooperatives and savings banks, that were prevalent in previous analyses, cannot be detected. Even more striking are our results for the growth rate of provision income. Panel B shows that this measure is by far more clearly associated with the economic cycle for savings and cooperative banks. To sum up, these findings indicate that our observation of savings banks and cooperatives smoothing credit is essentially driven by the public mandate of economic support and the commitment to cooperative members. This task affects other bank activities to a lesser extent, so that a higher cyclicality of deposit taking and provision income can well be observed. 4. Conclusions This paper investigates the drivers and consequences of cyclical bank behavior. Using data from more than 950 German banks over the period , three hypotheses on the relation between bank ownership, bank size, bank capitalization, and the cyclicality of lending behavior are tested. First, as we hypothesize in H1, loan growth and the average loan rate of savings and cooperative banks fluctuates less. However, when analyzing the link between these fluctuations and the business cycle, it turns out that the impact of 21

22 GDP growth on loan growth is similar in size for commercial banks and cooperatives, but significantly less pronounced for public sector savings banks. This finding proves true using different sample specifications and methodological approaches. Furthermore, we find that savings banks and cooperatives indeed adjust their loan rates less cyclically than commercial banks, which is consistent with savings banks public mission of economic support and cooperative banks claim to commit to their clients in good times as in bad. Second, we cannot detect a clear relation between a bank s size or capitalization and the cyclicality of its behavior, as H2 suggests. This finding supports our view that it is indeed a bank s ownership and not other possibly correlated characteristics that is the driver of the less cyclical lending policy of savings banks and cooperatives. Third, our test of H3 reveals that banks which show relatively stable loan growth rates over the economic cycle exhibit higher returns on equity, whereas higher fluctuations in relative interest income increase bank profitability via higher interest margins. Finally, robustness checks verify that regional economic conditions do not drive savings and cooperative banks lending policy to a higher degree than does GDP growth for Germany, and that these banks intentionally smooth credit, which is not the case for other banking activities. This paper has several implications. First, the ongoing debate about a necessary restructuring in the German financial system is enriched by our finding that public sector savings banks behave significantly less cyclically, which does not necessarily reduce their profitability. Thus, they are an important factor for sustainable financial stability and economic growth as they reliably provide financing for small and medium enterprizes. Second, borrowers deciding about their financing sources should bear in mind our finding that they can count on savings and cooperative banks rather than on private commercial banks when economic conditions deteriorate. Finally, bank managers in Germany s bank (relationship)-based financial system should note that a more stable lending policy does not essentially affect bank profitability negatively, however allows for the establishment of durable bank-borrower relationships which may be beneficial for both sides. 22

23 References Allen, Franklin, and Douglas Gale, 1999, Comparing Financial Systems (MIT Press). Bernanke, Ben S., and Mark Gertler, 1995, Inside the black box: The credit channel of monetary policy transmission., Journal of Economic Perspectives 9, Bichsel, Robert, 2006, State-Owned Banks as Competition Enhancers, or the Grand Illusion, Journal of Financial Services Research 30, Bikker, Jacob A., and Haixia Hu, 2002, Cyclical Patterns in Profits, Provisioning, and Lending of Banks, De Nederlandsche Bank Staff Report No. 86. Chow, Gregory C., 1960, Tests of Equality Between Sets of Coefficients in Two Linear Regressions, Econometrica 28, Delgado, Javier, Vicente Salas, and Jesus Saurina, 2007, Joint size and ownership specialization in bank lending, Journal of Banking & Finance 31, Demirguc-Kunt, Asli, and Harry Huizinga, 1999, Determinants of Commercial Bank Interest Margins and Profitability: Some International Evidence, World Bank Economic Review 13, Elsas, Ralf, 2007, Preemptive distress resolution through bank mergers, Working Paper, LMU Munich. Fonteyne, Wim, 2007, Cooperative Banks in Europe - Policy Issues, IMF Working Paper No. 07/159. Foos, Daniel, Lars Norden, and Martin Weber, 2007, Loan Growth and Riskiness of Banks, Working Paper, University of Mannheim. Hackethal, Andreas, 2004, German banks and banking structure, in Jan P. Krahnen, and Reinhard H. Schmidt, ed.: The German Financial System (Oxford University Press). 23

24 Hesse, Heiko, and Martin Cihak, 2007, Cooperative Banks and Financial Stability, IMF Working Paper No. 07/02. Iannotta, Giuliano, Giacomo Nocera, and Andrea Sironi, 2007, Ownership structure, risk and performance in the European banking industry, Journal of Banking & Finance 33, International Monetary Fund, 2003, Germany: Financial System Stability Assessment, IMF Country Report 03/343. La Porta, Rafael, Florencio Lopez-de-Silanes, and Andrei Shleifer, 2002, Government Ownership of Banks, Jounal of Finance 57, Levine, Ross, 2002, Bank-Based or Market-Based Financial Systems: Which Is Better?, Journal of Financial Intermediation 11, Lown, Cara, and Donald P. Morgan, 2006, The Credit Cycle and the Business Cycle: New Findings Using the Loan Officer Opinion Survey, Journal of Money, Credit, and Banking 38, Micco, Alejandro, and Ugo Panizza, 2006, Bank ownership and lending behavior, Economics Letters 93, , and Monica Yanez, 2007, Bank ownership and performance. Does politics matter?, Journal of Banking & Finance 31, O Hara, Maureen, 1981, Property rights and the financial firm, Journal of Law & Economics 24, Quagliariello, Mario, 2007, Banks riskiness over the business cycle: a panel analysis on Italian intermediaries, Applied Financial Economics 17, Rajan, Raghuram G., 1994, Why Bank Credit Policies Fluctuate: A Theory and Some Evidence, Quarterly Journal of Economics 109,

25 Sapienza, Paola, 2004, The effects of government ownership on bank lending, Journal of Financial Economics 72, Smith, Donald J., Thomas F. Cargill, and Robert A. Meyer, 1981, An Economic Theory of a Credit Union, Journal of Finance 36, Stolz, Stephanie, and Michael Wedow, 2005, Banks regulatory capital buffer and the business cycle: evidence for German savings and cooperative banks, Deutsche Bundesbank Banking and Financial Studies Discussion Paper No. 07/

26 Figure 1: Macroeconomic variables during This graph displays the development of our main macroeconomic variables during Data on the real GDP growth rate, the 3-month interest rate, and the 10-year government bond yield for Germany are taken from OECD statistics. GDP growth years interest rates GDP growth 3 month interest 10 year interest 26