The Relationships between Profitability, Capital, and Risk: Commercial vs. Saving/Mortgage Banks

Jacob Paroush* and Ben Z. Schreiber**

December 2008

Keywords: Profitability, Capital, Risk, Commercial Banks, Mortgage Banks JEL: G21; G32; G12

* Bar-Ilan University, Economics Department, Ramat-Gan and Ashkelon Academic College ** Corresponding author: Bank of Israel, Box 780, Jerusalem 91007, ISRAEL Phone: (972) 2-6552595, E-mail: [email protected] We thank A. Berger, E. Elyasiani, Y. Landskroner, and the participants of seminars at the Bank of Israel and Bar Ilan University for helpful comments. We also aknowledge Aharon Meir Center for Banking of the Department of Economics at Bar-Ilan University for financial support.

Electronic copy available at: http://ssrn.com/abstract=1312402

The Relationships between Profitability, Capital, and Risk: Commercial vs. Saving/Mortgage Banks

December 2008

Abstract This study examines the relationships between capital, profitability, and risk in U.S. commercial banks versus saving/mortgage banks for the period q1/1995-q4/2006. Following Froot and Stein (1988), we distinguish between these two banking sectors by their loan portfolios: commercial banks extend larger and unique loans with unstable probability of default rates relatively to saving/mortgage banks. Based on this analysis, we hypothesize a positive relation between risk and profitability in commercial banks and a positive relation between risk and capital in saving/mortgage banks. Contrary to the current literature, which analyzes only two out of the three variables: capital, profitability, and credit risk, our approach considers simultaneous analysis of the above three variables. The simultaneous analysis as well as the empirical results, provides an answer to the question, why in the sample period commercial banks were riskier than saving/mortgage banks while having a similar capital level.

-2Electronic copy available at: http://ssrn.com/abstract=1312402

1. INTRODUCTION

The three main characteristics of a bank are capital, profitability, and risk. The relationships between these variables are usually examined in pairs i.e., capital and profitability, capital and risk, and risk and profitability. Such analyses are only partial and the results - whether on a theoretical or empirical basis - are equivocal. For example, regulatory bodies tend to focus on risk and the capital held against unexpected losses. Thus, neglecting profitability may cause an inaccurate assessment of the overall conditions of examined banks if some of them are characterized by high profitability while others not (all other things being equal). On the other hand, investors sometimes consider profitability and risk but not capital. Omitting capital may also cause biased assessments if there are banks whose capital cushion is large versus banks with low capital levels (all other things being equal). In addition, if one of the three variables is omitted, econometric estimates of regression coefficients may suffer from specification errors. The present study depicts the relations among all the three variables simultaneously by comparing two different banking sectors: commercial versus saving/mortgage banks. One may expect a different relation between profitability and credit risk in the two sectors even if banks maintain the same capital level. During the sample period, commercial banks were more profitable and riskier than saving/mortgage banks while capital level was similar in both banking sectors. This raises the question why capital level is similar, given that the required capital by Basel I for loans secured by residential property was 50% lower than that of a commercial loan. We attempt to explain why the relations between profitability, capital, and risk are different in these two banking sectors. The average size of a loan and its variance in commercial banks is larger than in saving/mortgage banks. Thus, the costs per one dollar of assessing the riskiness of borrowers and monitoring their financial conditions in commercial banks are smaller than the respective costs in saving/mortgage banks. Following Froot and Stein (1998), we conjecture that commercial banks will increase the required yield and interest margin (profitability) for riskier borrowers while saving/mortgage banks will raise capital levels for riskier mortgage

-3-

takers. During the sample period, commercial banks were riskier than saving/mortgage banks while their capital level was similar. The simultaneous analysis of all three variables: profitability, capital, and risk, enables us to explain these different relations that occurred in saving/mortgage banks versus commercial banks during the sample period. As expected, we found that profitability and credit risk were positively related in commercial banks while capital and credit risk were positively related in saving/mortgage banks. In contrast, the relations between both capital and credit risk in commercial banks and profitability and risk in saving/mortgage banks, are insignificant or less robust. The rest of the paper is organized as follows. Section 2 surveys the current literature. Section 3 links profitability, capital, and risk and presents relevant hypotheses concerning these relationships. Section 4 discusses the methodology, Section 5 displays results, Section 6 elaborates on the implications of the results, and section 7 concludes.

2. LITERATURE SURVEY In what follows we survey the current literature on the relations between each pair of the variables: profitability, capital, and risk.

2.1 The relationship between profitability and capital It is generally accepted (see Berger, 1995; Barth et al., 1998) that the Capital Asset Ratio (hereafter CAR) is negatively correlated with Return On Capital (hereafter ROC). According to this hypothesis, the negative relationship is obtained, ceteris paribus, in a one-period model where deposit rates are not influenced by bank risks. However, assuming information symmetry between the depositors and the bank i.e., ‘market discipline` exists and deposit and stock markets are perfect, a rise in CAR due, for example, to the substitution of equity and debt, should entail a reduction of the bank's risk to fail. In such a case, risk-averse depositors who regard capital as a cushion against unexpected losses will be satisfied with a lower interest rate on deposits. This in

-4-

turn, ceteris paribus, should increase Net Interest Margin (hereafter NIM) and thus ROC. On the other hand, a rise in CAR increases capital, and therefore may reduce profitability either due to the increase in the denominator of ROC or due to the perception that the bank is safer. Thus, an increase in CAR might have an ambiguous effect on ROC. According to the Expected Bankruptcy Costs Hypothesis (henceforth EBCH), if a bank’s capital is below its optimum level, a rise in capital should reduce the yield required on deposits. Consequently, the increase in net income (the numerator in ROC) will have a greater effect than the rise in capital (the denominator in ROC), ceteris paribus, and altogether one can expect a positive relationship between capital and profitability. On the other hand, if capital is above its optimum level as perceived by depositors, the increase in capital reduces the interest rate required on deposits, so that the relationship between capital and profitability is expected to be negative. In general, EBCH assumes ‘market discipline’ either for well-capitalized or under-capitalized banks and ROC influenced by loans and deposits but not by operational activity e.g., commisions. According to the Signaling Hypothesis (see Acharya, 1988), managers have ‘inside information’ regarding future performance. If their compensation packages include stocks and/or stock options it will be cheaper for a safe bank than for a risky bank to signal expected improved performance in the future by increasing capital today. Therefore, capital entails profitability. Stiroh (2000) gives another argument for this causation. When banks overcome high entry barriers by increasing their capital levels, they gain access to profitable activities such as issuing guarantees and subordinated notes, and acting as intermediators in derivative markets.

2.2 The relationship between capital and risk A negative relationship between capital and risk is expected when all deposits are insured with a flat premium rate i.e., there is no ‘market discipline’. In this case, the marginal cost of increasing bank risk and/or reducing the level of capital is zero. This is because in the view of the regulators,

-5-

the insurance premium does not change with risk or capital, and for the insured depositors the interest demanded on their deposits is the same as that on a riskless asset. On the other hand, when the insurance premium is adjusted to risk, e.g., including the level of financial leverage, there is less incentive to change the financial leverage (Osterberg and Thomson, 1989). The "optimum capital buffer theory" suggests that banks have an incentive to hold more capital than required as an insurance against a violation of the regulatory minimum capital requirements (Heid et al., 2004). Hence, banks with relatively large capital buffers expected to maintain their capital buffers (increase both capital and risk) while banks with small capital buffers aim at rebuilding an appropriate capital buffer (increase capital and decrease risk). Alfon et al. (2004), who found a negative relationship between capital and risk in U.K. banks and building societies, mention several explanations for the actual capital levels, which are substantially higher than required. The parameters mentioned are: the distance from minimum capital requirements, the internal risk assessments by bank managers and their sophistication in managing risk, the level perceived as appropriate by rating agencies and depositors (market discipline), and the costs of raising extra capital. Flannery and Rangan (2004) explain the capital build-up of US banks during the 90s by increased capital requirements such as the FDIC Improvement Act (1991), high profitability of the banking industry along with higher risk levels, and the withdrawal of implicit government guaranties (increased market discipline). Cebenoyan and Strahan (2004) found that banks which used the loan sales market for risk management purposes held less capital and were more profitable but riskier than other banks. This evidence is in line with the Froot and Stein (1998) model that active risk management can allow banks to hold less capital and to invest in riskier assets.

2.3 The relationship between risk and profitability Stone (1974), Booth and Officer (1985) and Flannery and James (1984) applied a Two-Index Model in banking. They found a positive correlation between the yield on bank shares and changes in stock

-6-

and bond indices (reflecting risks). In a competitive business environment where symmetrical information between the bank and its borrowers prevails, one can expect positive relationships between profitability and risk. This should be the result of risk premium demanded by the bank from its borrowers and by the bank stakeholders (See also Saunders et al., 1990; Shrieves and Dahl, 1992). The trade-offs between pricing credit risk and setting capital aside are mainly related to parameters such as regulation, competition, sophistication in risk management, and the type of credit portfolios. In particular, a bank might not fully price its loan portfolio for the following reasons: (a) Cost of data collection for each borrower or project is usually greater than the benefit. A case in point is mortgages or standard loans, (b) The population of borrowers is relatively homogeneous but not correlated, the amount of the particular loan is not significant, and the distribution of loan repayments is relatively known, (c) The risk is not directly connected to the borrower e.g., management or operational risks. In practice, banks price credit risk and simultaneously set aside capital so the differences between various banking sectors e.g., commercial and saving/mortgage banks are related maily to the dosages of capital and profitability. This is true despite the blurring of the distinction between commercial and saving/mortgage banks during the past few years. Below we link profitability, capital, and credit risk based on Froot and Stein (1998) and EBCH adjusted to credit risk.

3. LINKING BANK PROFITABILITY, CAPITAL, AND RISK

Froot and stein (1998) distinguish between priced risk and non-priced risk. They argue, within a stylized model, that the risks of a bank's loan portfolio which consists of nontradable assets should not be fully priced. This is because the portfolio cannot be fully hedged, especially if it is composed of many small loans. On the other hand, a loan portfolio with many tradable assets can be hedged in the market, thus, risks should be priced. In the first case the cost per dollar loan is higher than in the second case. More specifically, banks usually price the loans and simultaneously set aside capital

-7-

against unexpected losses. However, one can distiguish between two types of loans or projects. The first loan type includes loans that are characterized by a relatively small size, are more homogenous, have low correlation with other loans in the portfolio, and have a stable default rate. Examples of such projects are credit extended to households or mortgages. Hence, the costs per dollar of credit of assessing the risk profile of the borrowers and monitoring them is relatively large. As a result of these costs the banks are reluctant to charge an average interest rate, based on history or similar projects, but instead set aside more capital to limit their exposure to the borrowers' credit risk (all other things being equal). We call these projects "non-priced projects" ala Froot and stein (1998). The other loan type is characterized by loans of a large size and relatively small costs per dollar of credit of assessing and monitoring the credit risk. Such projects can be large commercial loans with unfamiliar default rates where the banks carefully consider and monitor the projects as well as the borrowers' financial conditions. We call these projects "priced projects". In these projects the banks will price the loans more accurately, so the provisioned capital should be smaller compared to the loans in the first category (all other things being equal). Based on this distinction between priced and non-priced projects we conjecture that the relationship between profitability and credit risk in commercial banks is more positive than in saving/mortgage banks while the relationship between capital and credit risk in saving/mortgage banks is more positive than in commercial banks. This conjecture is a consequence of the different mode of operation: commercial banks are characterized by more priced projects such as C&I (Commercial and Industrial) loans, therefore one should expect a positive relationship between profitability and credit risk. Contrarily, saving/mortgage banks extend more non-priced projects such as mortgages, so the relationship between capital and credit risk should be positive. Of course, the blurring of the disttinction between commercial and saving/mortgage banks especially during the last years of the sample period makes it harder to distinguish between the two banking sectors. The different relationships of profitability, capital, and risk between commercial and

-8-

saving/mortgage banks within the EBCH framework are presented in Figure 1. The figure presents the EBCH adjusted to credit risk and our hypotheses about the distinction between commercial and saving/mortgage banks. [Enter Figure 1 here] Under EBCH, each bank either commercial or saving/mortgage should be located at the top of the hyperbola. That point reflects the highest profitability, given capital and credit risk as percieved by bank stakeholders i.e, depositors and rating agencies (see Berger, 1995 and Alfon et al., 2004). However, the different profiles of the various banking sectors are reflected by their location on the graph. In the beginning all banks are located on the lower curve and exposed to the very same risk level. Commercial banks are located on the left top side as they hold relatively less capital and their profitability is higher while saving/mortgage banks are characterized by the opposite: they hold more capital but their profitability is lower. In between there are unspecialized banks whose profile is the combination of the two. EBCH refers to CAR as a cushion against unexpected losses but does not take into account different credit risk levels. In the figure, the upper curve represents higher credit risk levels. As a result, we expect that all banks may adjust their capital/profitability in order to return to the new optimal positions1. Commercial banks increase the yield on loans, hence, the NIM and profitability to compensate for the higher risks rather than capital. This is shown by the bold up arrow – a movement from (CapC, ProfC) to (Cap*C, Prof*C) where, Prof and Cap are profitability and capital, respectively and asterisks represent the new optimal positions. Saving/Mortgage banks increase capital rather than pricing the new higher risks, thus, they should move from (CapSM, ProfSM) to the right (Cap*SM, Prof*SM). Finally, Typical unspecialized banks, the largest group, move up and to the right as they increase both profitability and capital. This movement from (CapT, ProfT) to (Cap*T, Prof*T) is symbolized by the up and to the right bold arrow.

-9-

4. THE METHODOLOGY Equations (1a) and (1b) summarize our hypotheses concerning commercial and saving/mortgage banks, respectively as follows: Prof = f C (Cap, +/−

(1a)

Cap = g C (Prof,

Risk, A) ⊕ Risk, B)

+ /−

+

Risk = h C ( Prof, Cap, C) ⊕ Prof = f SM (Cap, +/−

(1b)

Cap = g SM (Prof, + /−

+ Risk, A) + Risk, B) ⊕

Risk = h SM ( Prof, Cap, C) +

⊕

Where, A, B, and C are vectors of explanatory variables, and subscripts C and SM refer to commercial and saving/mortgage banks, respectively. The circles around the plus signs represent our main hypotheses regarding the relations between capital, profitability, and risks in both commercial (equation 1a) and saving/mortgage (equation 1b) banks. Prof and Risk are expected to positively relate for commercial banks while Cap and Risk should be positively related for saving/mortgage banks. In what follows we measure capital as bank accounting capital to total assets (CAR), profitability as Net Interest Margin (NIM), and risk as Provisions for loan losses over total loans. We focus on NIM rather than the familiar Return On Capital (ROC), in order to exclude operational and other profits from profitability. Hence, our conjectures are within the framework of EBCH except the adjustment to credit risk. As exogenous variables (A, B, and C in equations (1a),(1b)) we chose the log of total asset (Lasset) as the size control variable and another two macro variables: log of GDP and short term interest rates (Tbills). Both were found as important variables explaining the –––––––––––––––––––––––––––

1

The new risks level reflects a movement of the curve rather than a movement along the curve.

- 10 -

profitability, capital and credit risk of banks (see for examle Demirguc-Kunt and Huizinga, 1999). We also added a fourth explanatory variable relating to the composition of investors: Ind2C&I. This variable which is the ratio of loans extended to individual divided by commercial and Industrial (C&I) loans reflects the relative amount of mortgages versus commercial loans. As we shall see later, this variable characterizes the two banking sectors. The four variables and a dummy – Dum2 representing the second period after the slowdown of 2001-2002, were ran in differences (∆) in order to avoid different degrees of integration (see Table 3 hereafter). Following DemirgucKunt and Huizinga (1999), we expect that size (Lasset) should negatively affect NIM while GDP should positively affect NIM in both banking sectors. In boom periods, for example, we expect more profitability or less realized credit losses or both. In contrast, interest rates should negatively influence CAR as high interest rate levels increase the opportunity cost of capital holdings. Our expectation concerning the fourth explanatory variable chosen, Ind2C&I, is that it will negatively affect PROV in both banking sectors. This is because mortgages or individual loans are perceived as more diversified and less risky than C&I loans. After adding the explanatory variables to the various equations, there are two different variables in each equation – a necessary condition for identification of the three equations system. Following EBCH adjusted to risk, we assume that capital, profitability, and credit risk are determined simultaneously. However, we do not conjecture any ex ante relationships between CAR and NIM i.e., it can be either positive or negative. Within the context of EBCH, a positive relationship means that an increase in capital is perceived as good news to the bank stakeholders, so, profitability should increase more than capital, while the opposite holds for a negative relationship. In what follows we test these hypotheses within a context of three simultaneous equations using first differences of the variable chosen.

- 11 -

5. THE DATA The database consists of two sectors of U.S. banks for the period q1/1995–q4/2006 (48 quarters). The two sectors are commercial banks (7,402 institutions in q4/2006) and mortgage/saving banks (1,279 institutions in q4/2006). All banks are insured by the Federal Deposit Insurance Corporation (FDIC) either through the Bank Insurance Fund (BIF) or through the Savings Association Insurance Fund (SAIF). The commercial banking sector includes national banks and depository trust companies while the saving/mortgage banking category includes savings banks and savings and loan institutions supervised by the Office of Thrift Supervision (OTS). The quarterly data on banks are obtained from the formal FDIC's web site, based on Quarterly Call Reports. The advantage of the database is its reliability, consistency, and the lack of sample biases as it contains all insured US banks. Table 1 depicts some basic information on the two banking sectors: commercial versus saving/mortgage during the sample period. [Enter Table 1 about here] In addition to NIM, other measures of profitability in the table are Return On Capital (ROC) and Yield On Earning Assets (YOEA). Another measure for capital is CAD (Capital Adequacy of the Basel I directive) while alternate measures of credit risk are net charge-offs to total loans (CHRG_OF), Past Due loans to total loans (PastD), and Non-Performing loans to total loans (NONPER)2. Other variables presented in the table are the log of bank assets (Lasset) and the ratio of gross 1-4 family mortgages to total assets (MORTG). Table 1 shows that saving/mortgage banks are less profitable, exposed to smaller credit risk, and hold slightly more capital than commercial banks. For example, the ROC of commercial banks is three percentage points higher than that of saving/mortgage banks while NIM of commercial banks is almost one percentage point higher than that of the saving/mortgage banks. YOEA of 2

We examined more risk variables than the ones presented here but the correlation coefficients between them and the ones chosen were very high so they were excluded, in order to avoid multicolinearity.

- 12 -

commercial banks is also higher than that of saving/mortgage banks. Credit risk, such as loan-loss provision (PROV) and net charge-offs of commercial banks (CHRG_OF) are more than twice those that prevail in saving/mortgage banks. This is partially explained by regulatory requirements that dictate much lower provision levels for mortgages compared to other loans.3 Contrarily, the capital level represented by CAR is quite similar between these two banking sectors. Because of the similar capital and lower risk levels of saving/mortgage banks it is reasonable that the CAD of the latter (15.3%) is higher than that of commercial banks (12.6%). Credit risks in commercial banks are also more volatile than in saving/mortgage banks. For example, the standard deviation of CHRG_OF in saving/mortgage banks is less than one third of that of commercial banks (0.06 versus 0.21). This is consistent with our assumption that credit risks in saving/mortgage banks are relatively known and stable. The differences between the two banking sectors are reflected by the ratio of family mortgages to gross assets (MORTG). This figure in commercial banks is more than three times that of commercial banks (47.2 versus 15.4). Splitting the sample period into two sub periods reveals that in most cases there are no significant differences between the first period (q1/1995 – q4/2000) and the second period (q1/2001 – q4/2006) except for CAR. In both banking sectors, CAR during the second period was higher than in the first period. However, the growth rate of CAR in saving/mortgage banks was steeper than that of commercial banks. Figure 2 depicts the development of the main three variables representing capital, profitability, and risk. [Enter Figure 2 about here] It can be seen that profitability (NIM) and risk (PROV) of commercial banks are larger than those of saving/mortgage banks during the sample period. This is true both in boom periods such as 1999-2000 and 2005-2006 and in periods of a slowdown such as 2001-2002. The third variable in the triangle is capital (CAR) which appeared to be the same for both banking sectors during the first period. The substantial development after the slowdown period of 2001-2002 was an increase 3

Under BASEL I guidelines, (iv) collateral and guarantees: "loans to private individuals for residential house purchases have a very low record of loss" therefore, the weight of such loans secured by the property determined at 50

- 13 -

in CAR in both banking sectors along with a decrease in PROV and NIM. This can be explained by a perception of banks' stakeholders that the business environment is riskier given there is market discipline. Yet, as mentioned above, the increase in CAR was more prominent in saving/mortgage banks compared with commercial banks, perhaps due to the former's difficulties in raising equity in the capital markets4. In addition, the blurring of the distinctions between commercial and saving/mortgage banks enabled the latter to increase their C&I loans; which are usually riskier than loans extended to individuals (see Figure 3 hereafter). This and the large losses of the saving and loans industry during the 80s and the 90s may have caused saving/mortgage banks to raise their capital levels even more than commercial banks did. In order to test the similarity of the two banking sectors: commercial vs. saving/mortgage banks regarding the variables in Table 1, we ran three parametric tests (mean, median, and variance) and another three empirical tests: Kolmogorov-Smirnov, Kuiper, and Cramer-Von Mises. The latter compare the entire empirical distribution of any two series. [Enter Table 2 about here] By these tests we could reject the null hypothesis that the variables are similar at 99% confidence level, except for CAR5. This development raises the question: why saving/mortgage banks held capital similar to that of commercial banks while their risk levels were much lower especially given that capital requirements regarding mortgages secured by the purchased houses were half of that required on commercial loans?. Our hypothesis addresses this issue by introducing both the three simultaneously determined variables – profitability, capital, and risk and the different selection of these variables by the two banking sectors. –––––––––––––––––––––––––––

percent rather than 100 percent of a regular commercial loan. 4 There are more (large) commercial bank equities which are traded in capital markets than saving/mortgage banks. This can partially explain why the latter increase their CAR more than the former (see Heid et al., 2004). 5

We test the similarity of mean, median, and variance of the variables by using several tests, e.g., Anova. Usually they

provided the same results. We also divided the sample into two sub-periods and re-examined the similarity of the variables in both banking sectors. The results were quite similar to those accepted for the entire sample period.

- 14 -

If the above variables are not of the same integration level, we could get spurious relationships between them. In order to test for the integration level of these variables we ran three kinds of Augmented Dickey Fuller (ADF) tests: ADF with no constant and trend, ADF with a constant but without a trend and ADF with both a constant and a trend6. [Enter Table 3 about here] According to Table 3 (upper panel) most variables are I(1) except some related to saving/mortgage banks. However, when we take the first differences of the variables rather than the variables themselves (lower panel of Table 3) almost all variables turned out to be stationary. The next step was to test our main hypothesis in (1a) and (1b) regarding commercial and saving/mortgage banks by using the first differences of the variables and running Three-Stage LS regressions (3SLS)7, as follows: ( 2.1) ∆NIM = α1 + β1, 2 ∆CAR + β1,3 ∆PROV + χ1∆Lasset + δ 1∆Lgdp + φ1 Dum 2 + ε 1 ( 2.2) ∆CAR = α 2 + β 2,1∆NIM + β 2,3∆PROV + χ 2 ∆Lasset + δ 2 ∆IR + φ2 Dum 2 + ε 2 ( 2.3) ∆PROV = α 3 + β 3,1∆NIM + χ 3, 2 ∆CAR + χ 3∆Lasset + δ 3∆Ind 2C & I + φ3 Dum 2 + ε 3 Where, NIM, CAR, and PROV represent profitability, capital, and credit risk, respectively, in the two banking sectors, Lasset is the log of total asset, Lgdp is the log of US GDP, IR is the short term interest rate (Tbills), and Ind2C&I is the ratio of loans extended to individuals to Commercial and Industrial (C&I) loans. Dum2 represents the second period when the banks increased their capital levels after the slowdown of 2001-2002 (it takes on the value 1 for the period 2003-2006 and 0 otherwise). The data were seasonally adjusted using X-12 RegARIMA methodology and the variables regressed in differences − ∆ (=log(vart/vart-1)) so as avoiding different levels of integration. We also added to equations (2.1) - (2.3) an autoregressive term (AR) of one quarter

6

Using other tests such as the Phillip Pheron test did not change the results substantially. Based on Table 1 and Figure 2, there is heteroskedacticity in some variables, so this procedure, which also accounts for correlations between the exogenous variables and the residuals and between the residuals themselves, is suitable. 7

- 15 -

lag, when needed, in order to correct for serial correlation. Table 4 depicts the results of the three equations system regarding commercial and saving/mortgage banks8 for the entire period. [Enter Table 4 about here] The table is divided into three vertical panels: the left panel presents the regression results of the three equations related to commercial banks, the center panel presents the results related to saving/mortgage banks, and the right panel presents the gap between the coefficients of two sectors and their significance level. Almost all of the coefficients are in line with our hypotheses. Particularly, CAR and NIM are negatively related in saving/mortgage banks but positively related in commercial banks yet, in commercial banks they are not significant. Negative relations mean that increasing capital is perceived as bad news, and therefore profitability declines. By the Signaling Hypothesis described earlier, CAR should result in NIM, as was found in Demirguc-Kunt and Huizingat (1999). In Berger (1995) it is determined simultaneously and expected to be negative, while under EBCH it depends on the perception of the bank's stakeholders. If they consider the current capital level as higher than the optimal level, any addition of bank capital should result in a negative return. The opposite will occur if the bank's stakeholders consider the capital as smaller than the optimal level. The results are consistent with Berger and EBCH only for saving/mortgage banks, whereas, the current CAR level is perceived as higher than the optimal. Our main testable hypothesis concerns the relationships between credit risk and both profitability and capital. As expected, profitability (NIM) and credit risk (PROV) are positively related in commercial banks while capital (CAR) and credit risk (PROV) are positively related in saving/mortgage banks. The influence of PROV on NIM in commercial banks is significantly stronger than that of saving/mortgage banks (0.238 vs. 0.126 both at a significance level of 99%). In contrast, the influence of PROV on CAR in saving/mortgage banks is significantly stronger than that for commercial banks (0.211 vs. 0.041 where only the former is significant, at a significance level of 99%). The positive association between capital and risk is in line with Shrives and Dahl 8

Very small banks – below USD100 million in assets were excluded from the sample.

- 16 -

(1992) yet, it is significant only for saving/mortgage banks while the positive relation between profitability and credit risk is consistent with Flannery and Rangan (2004) for both banking sectors. The RHS of the table shows the gaps between the coefficients of the two banking sectors and their significance levels. As conjectured, the gaps of the impact PROV has on both NIM in commercial banks and CAR in saving/mortgage banks are significant. The gap in the NIM equations is 0.112 with a significance level at 99% (T-stat = 2.59) while in the CAR equation the gap is -0.171 with a significance level of 99% (T-stat = -2.65). The influence of NIM and CAR on PROV, which is usually not included in the literature, but appears significant in our three equations system, points to an interesting phenomenon. NIM and CAR positively affect PROV significantly except for the influence of CAR in commercial banks. The effect of NIM on PROV in commercial banks is stronger compared with saving/mortgage banks (3.008 vs. 2.244 although this gap is insignificant) while the impact CAR has on PROV in saving/mortgage banks is larger than that of commercial banks (1.580 vs. 0.164; a gap whose significance level is 95% - T-stat = -2.26). These are consistent with the above findings that (1) the influence of PROV on NIM in commercial banks is stronger than the influence of PROV on NIM in saving/mortgage banks and (2) the influence of PROV on CAR in saving/mortgage banks is stronger than in commercial banks. The results confirm our conjectures of simultaneity in banks' profitability (NIM), capital (CAR), and risks (PROV). They are also in line with our conjecture that commercial banks tend to price their loans such that the larger the credit risk, the higher the demanded yield/risk premium and the NIM (all other thing being equal). In contrast, saving/mortgage banks prefer setting aside capital rather than increasing NIM. The latter is due to relatively stable and known credit risk, small loans per each borrower, diversification among borrowers, and regulatory guidelines that dictate setting aside a certain amount of capital against mortgages regardless of the borrower risk level. Finally, the increase in CAR in both banking sectors also reflects the fact that the risk environment was perceived as more dangerous after the

- 17 -

2001-2002 slowdown of the US economy. That perception caused banks to raise their CAR in order to absorb unexpected losses especially saving/mortgage banks. Concerning the explanatory variables in Table 4, Lasset positively influences NIM and negatively affects PROV in saving/mortgage banks while the opposite is true with commercial banks. In both banking sectors, Lasset positively influences CAR. The unexpected positive influence of size (Lasset) on NIM in saving/mortgage banks together with the negative impact of the ratio Ind2C&I on PROV can be explained by the high growth rate of C&I loans during the sample period in saving/mortgage banks. This long-term development of blurring the distinctions between saving/mortgage and commercial banks is shown in Figure 3. [Enter Figure 3 about here] It can be seen that the ratio of loans to individuals over C&I loans decreased dramatically for saving/mortgage banks – from almost 4 at the beginning of the sample period to 1.5 at the period end. This development was the result of a substantial growth (90%) in C&I loans as shown in the figure (the dashed line – right axis). Thus, during the sample period saving/mortgage banks increased their assets by extending more C&I loans. As the NIM of C&I loans is usually larger than that relating to loans to individuals, both Lasset and NIM grew.9 The influence of Lasset on PROV in commercial banks is significantly positive while in saving/mortgage banks it is significantly negative. This may also reflect the differences between these two banking sectors during the sample period; controlling for the changes in Ind2C&I and the other endogenous variables. The dummy variable - Dum2 that represents the second period positively influenced NIM and negatively influenced PROV in commercial banks while in saving/mortgage banks it was insignificant. The other exogenous variables were also found to be insignificant. As a robustness check, we re-ran the three equations system with a window size of 36 quarters (3 years). In most cases, the relations remained as in Table 4 although the significance levels were 9

In Table 1, NIM of saving/mortgage banks already increased from the first to second period (an increase from 3.1% to 3.2%) while the opposite occcured to the NIM of commercial banks (from 4.2% to 3.8%).

- 18 -

often lower. We also re-ran the three equations system using other endogenous and exogenous variables. For example, the alternative variables for profitability (NIM) were ROC and YOEA and for risk (PROV) were PastD and NONPER (see Table 1). In many cases, the coefficients of the three endogenous variables changed signs or were not significant as these presented in Table 4. This means that our results are sensitive to timing or are related to other activities of the bank rather than loan extending. The above alternatives such as ROC and YOEA reflect other activities such as operational profits or profits on bank security portfolios while PastD and NONPER represent risk variables that determined several quarters after NIM.10 In order to assess the benefit of our three simultaneous equations system versus other models we re-ran the three equations in pairs i.e., profitability and capital, capital and risk, and profitability and risk. [Enter Table 5 about here] In Table 5, the coefficients in panels A and C are similar to those of the three equations system presented in Table 4. For example, in panel C, the influence of PROV on NIM is 0.239 (T-stat = 7.74) while the respective figure in Table 4 is 0.238 (T-stat = 7.65). As in Table 4, Lasset affects the two banking sectors differently and this influence is usually significant. The main difference between Tables 4 and 5 relates to the influence of PROV on both NIM and CAR in saving/mortgage banks (panel B). The coefficients of PROV in the CAR equation and CAR in the PROV equation are much smaller than their respective figures presented in Table 4 (0.090 versus 0.211 and 1.079 versus 1.580, respectively). Moreover, the significance levels of the former are lower (95% versus 99%). This evidence points to the stability of the relations between all three variables examined on the one hand, but also to the importance of our simultaneous equations approach on the other hand. Finally, we re-ran the three equations system with the variables themselves rather than with the differences and the results were very similar to those presented in Table 4. This also confirms the stability of the regression results of Table 4. 10

We added lags to PasD, NONPER, ROC, and YOEA but the results were unsatisfied. We also ran Granger's causality tests on the above variables but we could not find any causality up to 4 lags.

- 19 -

6. A DISCUSSION

We asked above, why are commercial and saving/mortgage banks dissimilar with regard to profitability and risks, but alike in terms of capital. Alfon et al. (2004) found that commercial banks in the UK held more extra capital (regarding capital requirements) than building societies. Their findings are consistent with banking theory and practices, as commercial banks are exposed to higher levels of risks than saving/mortgage banks. Moreover, according to the model developed by Froot and Stein (1998) and the evidence of Cebenoyan and Strahan (2004), active risk management allows banks to hold less capital and to invest in riskier assets. However, the literature usually neglects the third rib of the triangle namely profitability. Froot and stein (1998) connect between theoretical pricing models such as the CAPM and practical pricing models such as Risk Adjusted Return On Capital (RAROC). The former assumes costless capital adjustments; thus, it is suitable for banks whose asset portfolios are traded (risks are priced), so risks can be hedged. The latter considers capital adjustment costs, hence, it is suitable for banks whose assets are not traded (nonpriced risks) thus risks cannot be hedged. Under the Basel I Accord (1988), mortgage banks are required to set aside smaller capital levels against loans they extend and have less judgment ability regarding the required level compared to commercial banks11. Contrary to the above, our findings are different in the sense that saving/mortgage banks held slightly more capital than commercial banks; although their exposure to risk was much lower (see Table 1). The differences can be explained by our triangle approach emphasizing the simultaneous relations and the contribution of profitability to the two other variables: capital and risk, and the structural differences between the two banking sectors. Following Froot and stein (1998) we relate more priced projects/risks to commercial banks and more non-priced projects/risks to saving/mortgage banks. By taking into consideration the three variables simultaneously, the differences between these two banking sectors are structural in the sense that we expect a higher capital level per one dollar of credit risk and

- 20 -

lower profitability in saving/mortgage banks compared to commercial banks. Additionally, the negative relation found between profitability and capital in saving/mortgage banks suggests (according EBCH) that stakeholders perceive the capital level as more than the optimal one. Our inference suggests that these differences will not fade away in the future as long as saving/mortgage banks extend more small loans to households with low and stable default rates compared to commercial banks. The distinction between the two banking sectors and the simultaneity of the three endogenous variables are of particular interest to investors, policy makers, researchers, and bank regulators. For the latter, on the one hand, the current regulatory practice of taking into account only risks and capital levels is incomplete, whilst the regulator's distinction between mortgage and commercial banks is justified, as the characteristics of the two are quite different. For investors, especially in saving/mortgage banks with many non-traded loans, the focus only on risk and profitability is also inappropriate if the access of the bank to capital markets is limited i.e., there are capital adjustment costs (see Froot and Stein, 1998).

7. CONCLUSIONS

This study examines the relationships between capital, profitability, and risk of US commercial banks compared to saving/mortgage banks for the period q1/1995 – q4/2006. The evidence reveals substantial differences between these two banking sectors during the sample period: commercial banks are characterized by higher credit risk and profitability compared to saving/mortgage banks. As capital is approximately equal in both sectors, risk adjusted capital is lower and profitability is higher in commercial banks compared to saving/mortgage banks. Following Froot and stein (1998), who distinguish between priced and non-priced risks, we explain the differences between these two sectors by discriminating between two types of loans/projects: –––––––––––––––––––––––––––

11

This is also the Basel II proposal regarding mortgages (see BIS, November 2005, "International Convergence of Capital Measurment and Capital Standards"). Compared with a weight of 50% on mortgages to individuals secured by

- 21 -

(1) Common projects such as mortgages characterized by high pricing costs per dollar of credit and known/stable probability of default, and (2) Unique projects such as large-scale loans characterized by small pricing costs per dollar of credit and unknown/unstable probability of default. Based on this analysis, we hypothesize that banks with many homogenous but diversified loan portfolios, such as saving/mortgage banks, set aside more capital instead of fully pricing the loan. In contrast, banks whose loans are unique and large, such as commercial banks, tend to price the loans instead of setting capital aside. As a result of this analysis, profitability should be simultaneously incorporated in a triangle of capital, profitability, and risk, in order to explain the structural differences between these two banking sectors. We test our hypothesis on the US commercial banking sector versus saving/mortgage banks, using a 3SLS regression. The results are consistent with our conjectures: the relationships between both credit risk and profitability for commercial banks and between credit risk and capital for saving/mortgage banks are positive and robust. In contrast, the relations between, credit risk and profitability, for saving/mortgage banks and between, credit risk and capital, for commercial banks are either not significant or less robust. The relation between profitability and capital is negative for saving/mortgage banks. By the Expected Bankruptcy Cost Hypothesis (EBCH) adjusted to credit risk, this means that capital held by saving/mortgage banks is perceived by the banks' stakeholders as above the optimal level. Future research is required in the following topics: (1) Introducing a formal model that explains the differences between various types of banks, (2) Incorporating other banks' activities rather than loan extending. For example, services that also yield fees, and (3) Testing the hypothesis/model inferences by using particular banks within the context of panel data. –––––––––––––––––––––––––––

property under Basel I guidelines, the respective figure by Basel II recommendations is 35%.

- 22 -

REFERENCES Alfon I., I Argimon, and P. Bascunana-Ambros (2004), What Determines how Much Capital is Held by UK Banks and Building Societies, Occasional Paper Series 22, FSA, London. Acharya, B. (1988). A Generalized Econometric Model and Test of a Signaling Hypothesis with Two Discreet Signals, Journal of Finance 43, 413–429. Barth, J., D. Groper, and J. Jahera (1998), A Multi-country Analysis of Bank Capital and Earnings, Review of Pacific Basin Financial Markets and Policies 1(2), 123-55. Cebenoyan, A.S. and P.E. Strahan (2004), Risk Management, Capital Structure, and Lending at Banks, Journal of Banking and Finance 28, 19-43. Berger, A.N. (1995), The Relationship between Capital and Earnings in Banking, Journal of Money, Credit, and Banking 27(2), 432–456. Berger, A.N., Bonime, S.D., Covitz, D.M., and D. Hancock, (2000), Why are Bank Profits so Persistent? The Roles of Product Market Competition, Informational Opacity, and Regional/Macroeconomic Shocks, Journal of Banking and Finance 24, 1203-1235. Booth, J.R., and D.T. Officer (1985), Commercial Bank Stocks, Interest Rate, and Systematic Risk, Journal of Economics and Business 37(4), 303–310. Demirguc-Kunt, A. and H., Huizinga (1999), Determinants of Commercial Bank Interest Margins and Profitability: Some International Evidence, the World Bank Economy Review, 13 (2), 379-408. Flannery, M., and J.C. James (1984), The Effect of Interest Rate Changes on the Common Stock Returns of Financial Institutions, Journal of Finance 39(4), 1141–1153.

- 23 -

Flannery, M. and K. Rangan (2004), What caused the capital build-up of the 1990s? FDIC Center for Financial Research Working Paper No. 2004-03, August.

Froot K.A. and J.C. Stein (1998), Risk Management, Capital Budgeting, and Capital Structure Policy for Financial Institutions: An Integrated Approach, Journal of Financial Economics 47, 5582.

Heid F., Porath D., and S. Stolz (2004), Does Capital Regulation Matter for Bank Behavior? Evidence for German Savings Banks, Deutsche Bundesbank discussion paper Series 2, 03/2004.

Osterberg, W.P., and J.B. Thomson (1989), Bank Capital Requirements and the Riskiness of Banks: A Review, Federal Reserve Bank of Cleveland, 10-17. Saunders, A., E.Strock, and N.G. Travlos (1990), Ownership Structure, Deregulation, and Bank Risk Taking, Journal of Finance 45, 643–654.

Shrieves, R.E., and D. Dahl (1992), Relationship between Risk and Capital in Commercial Banks, Journal of Banking and Finance 16(2), 439–457. Stiroh, K.J., (2000), How did bank holding companies prosper in the 1990s?, Journal of Banking and Finance 24, 1703-1745. Stone, B.K. (1974), Systematic Interest Rate Risk in a Two-Index Model of Returns, Journal of Financial and Quantitative Analysis 9(5), 709–21.

- 24 -

Figure 1

Expected Bankruptcy Cost Hypothesis (EBCH) adjusted to risk: Commercial versus Saving/Mortgage banks Prof

Commercial banks Prof*C Typical unspecialized banks ProfC Prof*T ProfT

Saving/Mortgage banks

Prof*SM/ProfSM

CapC Cap*C

CapT

Cap*T

CapSM

Cap*SM

Cap

According to the EBCH, a typical unspecialized bank sets aside capital and prices loans simultaneously, so, it will maximize profitability (top of the hyperbola) given bank risks. We symbolize the optimal point as (Cap, Prof) where Cap is bank capital, Prof is bank profitability, and subscripts C, T, and SM refer to Commercial, Typical, and Saving/Mortgage banks A right hand movement from the optimal point (the top of the hyperbola) should result in a negative relation between Cap and Prof. This means that adding more capital is not optimal. In contrast, a left hand movement should result in a positive relation between Cap and Prof which means that adding capital is indeed optimal. The three types of banks are located along the lower line by their characteristics. Given the same risk levels commercial banks will price the risks while saving/mortgage banks will tend to set aside capital instead. This is reflected by their optimal initial position along the lower line (Cap C CAR (%) - Capital Asset Ratio: Commercial versus Saving/Mortgage banks 13 12 11 c_CAR

10

m_CAR

9 8 7 Jul-04

Mar-04

Nov-03

Jul-03

Mar-03

Nov-02

Jul-02

Mar-02

Nov-01

Jul-01

Mar-01

Nov-00

Jul-00

Mar-00

Nov-99

Jul-99

Mar-99

Nov-98

Jul-98

Mar-98

Nov-97

Jul-97

Mar-97

Nov-96

Jul-96

Mar-96

Nov-95

Jul-95

Mar-95

Risk => PROV (%) - Provision to loan and lease losses to total loans: Commercial versus Saving/Mortgage banks 1.4 1.2 1.0 c_PROV

0.8 0.6

m_PROV

0.4 0.2 0.0 Jul-05

Mar-05

Nov-04

Jul-04

Mar-04

Nov-03

Jul-03

Mar-03

Nov-02

Jul-02

Mar-02

Nov-01

Jul-01

Mar-01

Nov-00

Jul-00

Mar-00

Nov-99

Jul-99

Mar-99

Nov-98

Jul-98

Mar-98

Nov-97

Jul-97

Mar-97

Nov-96

Jul-96

Mar-96

Nov-95

Jul-95

Mar-95

This figure depicts the development of the three main variables across the sample period. It can be seen that profitability and risk were quickly influenced by the slowdown period of 2001-2002, especially, commercial bankswhile capital levels were raised only two years later. The letters c_ and m_ represent commercial and saving/mortgage banks, respectively.

Figure 3

The ratio of loans to individuals over C&I loans (Ind2C&I) Commercial versus saving/mortgage banks 90%

4

80%

3.5

70% 3 60% 2.5

50%

2

40% 30%

1.5 20% 1

10% 0%

0.5 -9 ar

06 pSe 6 -0 ar M 05 pSe 5 -0 ar M 04 pSe 4 -0 ar M 03 pSe 3 -0 ar M 02 pSe 2 -0 ar M 01 pSe 1 -0 ar M 00 pSe 0 -0 ar M 99 pSe 9 -9 ar M 98 pSe 8 -9 ar M 97 pSe 7 -9 ar M 96 pSe 6 -9 ar M 95 pSe 5

M

c_ind2C&I

m_Ind2C&I

m_C&I

c_C&I

This figure presents the development of the ratio: loans to individuals over Commercial and Industrial loans (C&I) along the sample period (left axis) and the accumulated growth rate of the latter (right axis). It can be seen that the dramatic decline in the ratio of Mortgage/Savings banks is the result of the substantial growth rate of the denominator - C&I loans. The letters c_ and m_ represent commercial and saving/mortgage banks, respectivel

Table 1

Main Characteristics of Commercial vs. Saving/Mortgage banks (q1/1995 - q4/2006, %)

Commercial Banks

Saving/Mortgage Banks

(All period) Avg Median Std Min Max Skewness Kurtosis

Lasset YOEA 6.8 7.1 6.8 7.7 0.11 1.20 6.6 4.9 7.0 8.4 0.1 -0.5 -1.1 -1.4

NIM 4.0 4.1 0.26 3.4 4.4 -0.5 -0.9

ROC CAR 14.4 9.0 14.5 8.7 0.76 0.71 12.9 7.8 16.0 10.3 -0.3 0.7 -0.7 -0.7

Avg Median Std Min Max Skewness Kurtosis

Lasset YOEA 6.9 6.1 6.9 6.0 0.1 0.9 6.8 4.9 7.0 8.1 0.1 0.8 -1.2 -0.1

NIM 3.8 3.8 0.2 3.4 4.2 0.4 -0.7

ROC CAR 14.2 9.5 14.1 9.2 0.9 0.6 12.9 8.7 15.5 10.3 0.1 0.2 -1.5 -1.7

Avg Median Std Min Max Skewness Kurtosis

Lasset YOEA 6.7 8.0 6.7 8.2 0.1 0.4 6.6 6.3 6.8 8.4 0.1 -3.0 -0.9 11.5

NIM 4.2 4.1 0.1 4.0 4.4 0.0 -1.4

ROC CAR 14.7 8.4 14.7 8.4 0.6 0.3 13.7 7.8 16.0 9.2 0.3 0.5 0.3 3.1

PastD PROV CHRG_OF NONPER 1.1 0.7 0.7 0.7 1.2 0.6 0.6 0.7 0.18 0.22 0.21 0.16 0.8 0.4 0.4 0.5 1.4 1.2 1.1 1.0 -0.6 0.9 1.0 0.0 -0.9 0.1 0.3 -1.0

(All period) CAD MORTG 12.6 15.4 12.5 14.6 0.26 1.57 12.1 13.7 13.0 18.8 0.1 0.8 -1.2 -1.0

Lasset 6.1 6.1 0.09 6.0 6.3 0.5 -1.0

YOEA 6.8 7.3 1.03 5.0 7.9 -0.5 -1.4

NIM ROC CAR 3.2 11.3 9.3 3.2 11.4 8.7 0.14 1.48 1.15 2.9 7.8 7.9 3.5 13.9 12.3 0.6 -0.4 1.0 0.4 -0.1 -0.4

CAD MORTG 12.7 16.6 12.7 17.0 0.2 1.4 12.3 14.1 13.0 18.8 0.1 -0.4 -1.3 -1.2

Lasset 6.2 6.2 0.1 6.1 6.3 0.2 -1.4

YOEA 6.0 5.8 0.9 5.0 7.9 0.9 -0.1

NIM ROC CAR 3.2 11.7 10.1 3.2 11.6 10.2 0.2 1.5 1.1 2.9 8.7 8.5 3.5 13.9 12.3 0.4 0.0 0.0 -0.2 -0.9 -1.4

CAD MORTG 12.5 14.3 12.4 14.3 0.3 0.3 12.1 13.7 13.0 15.0 0.5 0.3 -1.0 0.2

Lasset 6.0 6.0 0.0 6.0 6.2 1.8 4.0

YOEA 7.5 7.7 0.5 5.6 7.8 -3.6 14.9

NIM ROC CAR 3.1 11.0 8.5 3.1 11.4 8.4 0.1 1.5 0.3 3.0 7.8 7.9 3.3 13.9 9.4 0.2 -0.7 1.0 -0.7 0.3 1.8

(q1/2001 - q4/2006) PastD PROV CHRG_OF NONPER 1.0 0.7 0.7 0.7 1.0 0.7 0.7 0.7 0.2 0.3 0.3 0.2 0.8 0.4 0.4 0.5 1.4 1.2 1.1 1.0 0.5 0.3 0.1 0.1 -1.1 -1.3 -1.1 -1.7

The definition of the above variables are as follows: Lasset - Log of total assets (in US dollar terms) YOEA - Yield On Earning Assets CAR - Equity capital to total assets

PROV - Loan-Loss Provisions to Gross Loans NIM - Net Interest Margin PastD - Past Due loans to total loans

MORTG 47.2 46.8 1.88 43.5 50.6 0.3 -0.8

(q1/2001 - q4/2006)

(q1/1995 - q4/2000) PastD PROV CHRG_OF NONPER 1.2 0.7 0.6 0.7 1.2 0.6 0.6 0.7 0.1 0.1 0.1 0.1 1.1 0.5 0.4 0.6 1.4 1.1 1.1 1.0 0.1 2.0 2.9 1.0 -0.4 7.5 12.7 -0.2

PastD PROV CHRG_OF NONPER CAD 1.0 0.3 0.3 0.7 15.3 1.0 0.3 0.3 0.6 15.2 0.20 0.06 0.06 0.26 1.05 0.7 0.2 0.2 0.4 13.7 1.4 0.4 0.5 1.4 17.2 0.3 0.1 1.1 1.0 0.5 -0.8 -1.3 3.7 -0.2 -0.9

PastD PROV CHRG_OF NONPER CAD 0.9 0.3 0.3 0.6 15.6 0.9 0.3 0.3 0.6 15.5 0.1 0.1 0.0 0.1 1.4 0.7 0.2 0.2 0.4 13.7 1.1 0.4 0.3 0.7 17.2 -0.1 0.1 -0.1 -0.8 -0.1 -1.3 -1.4 -0.5 -0.3 -1.9

MORTG 47.2 46.4 2.4 43.5 50.6 0.2 -1.5

(q1/1995 - q4/2000) PastD PROV CHRG_OF NONPER CAD 1.1 0.3 0.3 0.9 15.0 1.1 0.3 0.3 0.9 15.2 0.2 0.1 0.1 0.3 0.5 0.8 0.2 0.2 0.5 14.0 1.4 0.4 0.5 1.4 15.7 -0.2 0.2 1.1 0.1 -0.8 -1.3 -1.2 2.0 -1.5 -0.2

CHRG-OF - Net charge off to gross loans NONPER - Non Performing loans to total loans ROC - Net income to equity capital CAD - Total capital to risk-weighted assets (Capital Adequacy) MORTG - Gross 1-4 family mortgages to gross assets

MORTG 47.2 47.1 1.2 45.1 49.2 0.2 -1.0

Table 2

Empirical Distribution Function (EDF) and parametric similarity tests: Commercial Vs. Saving/Mortgage banks

Variable name

Kolmogorov Smirnov

EDF tests Cramer Von Mises

Mean Kuiper

T test

Parametric tests Median Variance KruskalVan der Wallis Waerden Bartlet Levene

YOEA

0.222 0.00

0.980 0.01

0.489 0.00

1.32 0.19

5.15 0.02

6.08 0.01

1.07 0.30

3.43 0.07

NIM

0.500 0.00

7.519 0.08

1.000 0.00

18.74 0.00

72.27 0.00

63.08 0.00

16.92 0.00

23.14 0.00

ROC

0.422 0.00

6.637 0.07

0.844 0.00

13.02 0.00

64.86 0.00

57.82 0.00

20.12 0.00

12.08 0.00

CAR

0.089 0.48

0.127 0.00

0.222 0.75

1.70 0.09

0.94 0.33

1.86 0.17

10.66 0.00

10.97 0.00

PastD

0.204 0.00

1.079 0.01

0.429 0.00

3.14 0.00

8.38 0.00

7.71 0.01

0.73 0.39

0.59 0.44

PROV

0.500 0.00

7.508 0.08

1.000 0.00

12.23 0.00

72.52 0.00

63.26 0.00

65.31 0.00

33.89 0.00

CHRG_OF

0.489 0.00

7.525 0.08

0.978 0.00

13.20 0.00

71.67 0.00

62.81 0.00

55.12 0.00

31.63 0.00

NONPER

0.144 0.05

0.622 0.01

0.533 0.00

0.30 0.77

1.05 0.30

0.25 0.62

11.06 0.00

10.63 0.00

CAD

0.500 0.00

7.534 0.08

1.000 0.00

17.69 0.00

72.76 0.00

63.48 0.00

72.46 0.00

51.09 0.00

MORTG

0.500 0.00

7.472 0.08

1.000 0.00

90.80 0.00

72.76 0.00

63.39 0.00

1.54 0.21

0.91 0.34

This table presents EDF and parametric similarity tests of selected variables from the two bank types. The null hypothesis is that the variable in question is similar in both bank types. For EDF, we implement Kolmogorov-Smirnov (KS), Cramer - Von Mises (CM), and Kuiper tests. For parametric similarity tests we conduct T test for the mean, Kruksal-Wallis and Van der Warden tests for the median, and Bartlet and Levene tests for the variance. For each variable the upper figure is the test figure while the lower figure is the probability for rejecting the null which is: the banking sectors are similar (for a description of the variables see Table 1). The results usualy reject the null hypothesis, i.e. the two banking sectors are not similar except for CAR.

Table 3

Unit Root Tests by Augmented Dickey Fuller (ADF) Commercial Banks No constant & Trend Variables in Levels CAR ROC CHRG-OF PROV PastD NONPER NIM YOEA CAD IND2C&I Lgdp Int (Tbill)

Variables in 1st Difference CAR ROC CHRG-OF PROV Lasset NONPER NIM YOEA CAD IND2C&I Lgdp Int (Tbill)

Constant, No Trend

Saving/Mortgage Banks

Constant & Trend

No constant & Trend

Constant, No Trend

Constant & Trend

1.007 (0.92) -0.489 (0.50) -0.798 (0.37) -0.753 (0.39) -0.768 (0.38) -0.671 (0.42) -1.834 (0.06) -0.145 (0.63) -0.759 (0.38) 1.140 (0.93) 11.062 (1.00) -1.214 (0.20)

-1.488 (0.53) -2.795 (0.07) -2.554 (0.11) -2.023 (0.28) -1.959 (0.30) -2.800 (0.07) -0.405 (0.90) -0.866 (0.79) -2.109 (0.24) -0.867 (0.79) -0.722 (0.83) -2.394 (0.15)

-3.694 (0.03) -3.257 (0.09) -2.465 (0.34) -2.459 (0.35) -2.897 (0.17) -3.026 (0.14) -2.473 (0.34) -2.007 (0.58) -2.087 (0.54) -3.071 (0.13) -2.322 (0.41) -2.182 (0.49)

1.448 (0.96) -0.409 (0.53) -1.779 (0.07) -0.833 (0.35) -1.266 (0.19) -1.623 (0.10) -0.961 (0.30) -0.012 (0.67) 0.465 (0.81) 5.974 (1.00) 11.062 (1.00) -1.214 (0.20)

-0.171 (0.93) -4.189 (0.00) -4.810 (0.00) -2.990 (0.04) -2.212 (0.21) -2.941 (0.05) -1.473 (0.54) -1.017 (0.74) -0.979 (0.75) -1.143 (0.69) -0.722 (0.83) -2.394 (0.15)

-1.796 (0.69) -3.998 (0.02) -4.254 (0.01) -2.197 (0.48) -1.925 (0.62) -3.919 (0.02) -1.502 (0.82) -3.786 (0.03) -1.550 (0.80) -1.107 (0.92) -2.322 (0.41) -2.182 (0.49)

-14.729 (0.00) -9.460 (0.00) -13.166 (0.00) -13.403 (0.00) -0.979 (0.29) -1.837 (0.06) -7.448 (0.00) -0.263 (0.59) -8.805 (0.00) -1.131 (0.23) -0.979 (0.29) -2.909 (0.00)

-14.932 (0.00) -9.427 (0.00) -13.182 (0.00) -13.404 (0.00) -3.232 (0.02) -1.748 (0.40) -7.730 (0.00) -1.728 (0.41) -8.746 (0.00) -1.128 (0.70) -3.232 (0.02) -2.879 (0.06)

-15.534 (0.00) -9.354 (0.00) -12.953 (0.00) -13.159 (0.00) -3.351 (0.07) -1.727 (0.72) -7.862 (0.00) -5.528 (0.00) -8.681 (0.00) -0.995 (0.93) -3.351 (0.07) -3.042 (0.13)

-8.900 (0.00) -11.646 (0.00) -10.202 (0.00) -6.268 (0.00) -0.979 (0.29) -19.200 (0.00) -5.540 (0.00) -2.318 (0.02) -11.675 (0.00) -8.255 (0.00) -0.979 (0.29) -2.909 (0.00)

-9.132 (0.00) -11.547 (0.00) -9.993 (0.00) -6.147 (0.00) -3.232 (0.02) -19.050 (0.00) -5.564 (0.00) -2.435 (0.14) -11.574 (0.00) -7.235 (0.00) -3.232 (0.02) -2.879 (0.06)

-9.666 (0.00) -7.575 (0.00) -9.993 (0.00) -6.114 (0.00) -3.351 (0.07) -18.714 (0.00) -5.520 (0.00) -1.883 (0.65) -7.546 (0.00) -7.438 (0.00) -3.351 (0.07) -3.042 (0.13)

This table presents the integration level of the variables included in the regressions (for a description of the variables see Table 1). The figures in parentheses are based on MacKinnon (1996) one-sided p-values. Number of lags are determined by Schwartz Information Criterion (SIC). It can be seen that almost all the variables in 1st differece are stationary by the ADF's unit root test.

Table 4 3SLS regression results of three equations system: Changes in Profitability (NIM), Capital (CAR), and Risk (PROV) Commercial Banks

Saving/Mortgage Banks

Differences between banks (Commercial - Saving/Mortgage)

∆ΝΙΜ Coefficient t Statistic -0.002 -0.52 0.045 0.34 0.238 7.65** -0.014 -4.63** 0.279 0.84 0.019 3.96** 0.097 0.73

Endogenous Variable: Constant ∆CAR ∆PROV ∆Lasset ∆Lgdp Dum2 AR Adj. R-Square D.W.

0.33 1.85

Constant ∆NIM ∆PROV ∆Lasset ∆IR Dum2 Adj. R-Square D.W.

0.69 1.82

Constant ∆ΝΙΜ ∆CAR ∆Lasset ∆Ind2C&I Dum2 AR Adj. R-Square D.W.

0.46 1.77

Endogenous Variable: Constant ∆CAR ∆PROV ∆Lasset ∆Lgdp Dum2 AR

∆CAR Coefficient t Statistic -0.003 -0.54 -0.351 -1.50 0.211 4.62** 0.009 4.33** 0.011 0.49 0.015 1.56

Constant ∆NIM ∆PROV ∆Lasset ∆IR Dum2

Endogenous Variable: Constant ∆NIM ∆PROV ∆Lasset ∆IR Dum2

∆CAR Gap T-Stat. 0.019 1.69 0.386 1.35 -0.171 -2.65** -0.004 -1.36 -0.005 -0.18 -0.020 -1.66

0.21 1.71 ∆PROV t Statistic 0.011 0.49 2.244 3.77** 1.580 3.60** -0.035 -2.80** -0.014 -0.14 -0.008 -0.28 0.109 0.85

Endogenous Variable:

Endogenous Variable:

Coefficient Constant ∆ΝΙΜ ∆CAR ∆Lasset ∆Ind2C&I Dum2 AR Adj. R-Square D.W.

∆ΝΙΜ Gap T-Stat. -0.439 -1.88 0.236 1.45 0.112 2.59** -0.019 -4.64** 0.375 1.12 0.008 0.66 -0.044 -0.23

0.03 1.98

Endogenous Variable:

Adj. R-Square D.W. ∆PROV Coefficient t Statistic 0.006 0.44 3.008 8.43** 0.164 0.37 0.045 6.34** -0.040 -0.17 -0.079 -4.69** 0.239 1.65

Endogenous Variable:

Constant ∆CAR ∆PROV ∆Lasset ∆Lgdp Dum2 AR Adj. R-Square D.W.

∆CAR Coefficient t Statistic 0.016 1.67 0.035 0.21 0.041 0.90 0.005 2.29* 0.006 0.44 -0.005 -0.67

Endogenous Variable:

∆ΝΙΜ Coefficient t Statistic 0.436 1.87 -0.192 -1.96* 0.126 4.19** 0.005 1.78 -0.096 -1.87 0.011 1.05 0.141 1.06

Endogenous Variable:

Constant ∆ΝΙΜ ∆CAR ∆Lasset ∆Ind2C&I Dum2 AR

∆PROV Gap T-Stat. -0.005 -0.18 0.764 1.10 -1.416 -2.26* 0.080 5.58** -0.026 -0.10 -0.071 -2.08* 0.130 0.67

0.04 1.93

The definitions of the variables are: NIM - Net Interest Margin, CAR - Book equity to total assets, PROV- provisions for loan and lease losses, Lasset - total assets in logs, Lgdp - the US GDP in logs, IR - Tbill rate (daily means of the quarter), Ind2C&I - the ratio of loans extended to individuals to Commercial and Industrial loans, Dum2 takes the value 1 for the period 2003-06 and 0 otherwise, AR is an autoregressive term of one lag. The differences between banks are calculated as: (1) For the mean - coefficients of commercial banks minus those of saving/mortgage banks, (2) For the significance we use a test of differences between where, µC, µSM is the coefficient of commercial and saving/mortgage banks, respectively, and σC, σSM are the respective standard deviations. two unrelated series as follow: * Represents a confidence level at 95% and ** represents confidence level at 99%. Instrumental variables are the lags of the exogenous ones. The data used for the regressions are seasonally adjusted using X-12 Reg ARIMA model and small banks (less than $100m in assets) were excluded from the sampl

Table 5 3SLS Regression results of two equations systems: Changes in NIM and CAR, CAR and PROV, and NIM and PROV Commercial Banks

Saving/Mortgage Banks

Differences between banks (Commercial - Saving/Mortgage)

A) NIM & CAR system Endogenous Variable: Constant ∆CAR ∆NIM ∆Lasset ∆Lgdp ∆IR Dum2 Adj. R-Square D.W.

∆ΝΙΜ Coefficient T-Statistic -0.004 -0.76 0.317 1.96 -0.004 0.150

-2.96** 0.30

-0.002

-0.45 0.04 1.90

∆CAR Coefficient T-Statistic 0.000 -0.07 0.241 0.007

1.91 10.61**

-0.006 0.002

-1.12 0.41 0.68 1.79

∆ΝΙΜ Coefficient T-Statistic 0.006 0.89 0.061 0.67 0.005 -0.775

3.64** -1.23

-0.009

-1.48 0.23 1.85

∆CAR Coefficient T-Statistic -0.004 -0.55 0.116 0.005

0.48 2.54**

-0.008 0.020

-0.61 1.88

∆ΝΙΜ Gap T-Stat. -0.010 -1.17 0.256 1.38 -0.008 0.924

-4.67** 1.14

0.007

0.90

∆CAR Gap T-Stat. 0.003 0.48 0.125 0.001

0.46 0.61

0.002

0.14

0.19 1.78

B) CAR & PROV system Endogenous Variable: Constant ∆CAR ∆PROV ∆Lasset ∆IR ∆Ind2C&I Dum2 AR Adj. R-Square D.W.

∆CAR ∆PROV Coefficient T-Statistic Coefficient T-Statistic -0.001 -0.24 0.041 2.03* -0.629 -1.21 0.027 0.74 0.005 3.36** -0.014 -1.13 -0.005 -0.91 0.222 0.78 0.003 0.65 -0.067 -2.83** 0.410 2.97** 0.68 1.74

0.40 2.15

∆CAR ∆PROV Coefficient T-Statistic Coefficient T-Statistic 0.004 0.61 0.011 0.49 1.079 2.22* 0.090 2.01* -0.006 -1.43 -0.028 -1.96* -0.009 -0.83 -0.020 -0.17 0.020 2.35** -0.015 -0.50 -0.040 -0.45 0.062 0.47 0.09 1.94

∆CAR Gap T-Stat. -0.004 -0.65

∆PROV Gap T-Stat. 0.031 1.02 -1.709 -2.40**

-0.063 0.012 0.003

-1.08 2.52** 0.29

0.014

0.73

-0.017

-1.65

0.242 -0.052 0.348

0.79 -1.37 1.82

0.08 1.99

C) NIM & PROV system Endogenous Variable: Constant ∆NIM ∆PROV ∆Lasset ∆Lgdp ∆Ind2C&I Dum2 AR Adj. R-Square D.W.

∆ΝΙΜ ∆PROV Coefficient T-Statistic Coefficient T-Statistic -0.002 -0.50 0.006 0.45 3.017 8.48** 0.239 7.74** -0.014 -4.65** 0.046 6.88** 0.274 0.83 -0.031 -0.14 0.019 3.97** -0.080 -4.68** 0.102 0.80 0.245 1.71 0.35 1.85

0.48 1.78

∆ΝΙΜ ∆PROV Coefficient T-Statistic Coefficient T-Statistic 0.004 0.60 0.016 0.81 2.380 3.97** 0.113 3.92** 0.005 5.10** -0.042 -3.41** -0.504 -0.84 0.010 0.10 -0.008 -1.42 0.022 0.78 0.031 0.29 0.25 1.82

∆ΝΙΜ Gap T-Stat. -0.006 -0.78 0.126 -0.019 0.778

2.99** -6.08** 1.14

0.088

6.28**

0.027

3.60**

-0.041 -0.101 0.214

-0.17 -3.13** 1.20

0.07 2.02

The definitions of the variables are: NIM - Net Interest Margin, CAR - Book equity to total assets, PROV- provisions for loan and lease losses, Lasset - total assets in logs, Lgdp - the US GDP in logs, IR - Tbill rate (daily means of the quarter), Ind2C&I is the ratio of loans extended to individuals divided by Industrial and Commercial loans, and AR is a one period autoregressive term. The differences between banks are calculated as: (1) For the mean - coefficients of commercial banks minus those of saving/mortgage banks, (2) For the significance we use a test of differences between two unrelated series as follow: where, µC, µSM is the coefficient of commercial and saving/mortgage banks, respectively, and σC, σSM are the respective standard deviations. * Represents a confidence level at 95% and ** represents a confidence level at 99%. Instrumnetal variables are the lags of the exogenous ones. The data used for the regressions are seasonally adjusted using X-12 Reg ARIMA model. Small banks (less than $100m in assets) were excluded from the sample.

∆PROV Gap T-Stat. -0.010 -0.42 0.637 0.91