Modern Banking

[ 498 ]

M O D E R N B A N K I N G

efficiency model holds, β ≥ 0. Berger and Hannan (1989) collected quarterly data from 470 banks in 195 local banking markets over a 2.5-year period, from 1983 to 1985, with 3500 – 4000 observations in six deposit categories. The dependent variables were retail deposit rates paid by commercial banks, as reported in the Federal Reserve’s monthly survey of selected deposits and other accounts.29 Banks in the sample were assigned to local markets, which were defined as metropolitan statistical areas (MSAs) or non-MSA counties. Banks with less than 75% of their deposits in one local market were deleted from the sample.

Berger and Hannan used two concentration ratios to measure the degree of firm concentration in the banking market. The ‘‘three firm’’ concentration ratio, CR3 is defined as the proportion of output attributed to the top three firms in the industry. More generally, this ratio is written as CRn, where n is the output share produced by the top n firms in the industry. The Herfindahl index30 was also used, defined as H = s2i , where si is the market share of the ith firm. These measures were constructed both with and without the inclusion of saving and loans firms.

The vector x included a number of additional explanatory variables:

žThe growth rate of deposits in the bank’s market, which may reflect local supply and demand conditions, and could have either sign.

žThe number of bank branches divided by total bank branches plus savings and loan

branches in the local market – it should have a negative coefficient if costs rise with the number of branches. Local per capita income was included to control for factors affecting the supply of funds to banks – in a non-competitive market, it may reflect a greater or lesser elasticity of deposit supply. The local bank wage, reflecting a cost factor, was another explanatory variable. Its sign is not predicted, because bank wages could also reflect local income differences.

žWhether a state in which a given bank operates prohibits (UNIT) or limits (LIM) branch banking. To the extent that such regulations limit entry, and therefore raise costs, one would expect to observe a negative coefficient.

The different concentration measures yielded similar results, so only the results using CR3 were reported. The β coefficient on the concentration variable was found to be negative and significant at the 1% level – that is, the more concentrated the market, the lower the deposit rate, a finding which is consistent with the SCP hypothesis but not the relative efficiency model. For example, ceteris paribus, banks in the most concentrated markets were found to pay money market deposit rates which were 25 – 100 basis points less than what was paid on the less concentrated markets. Similar findings were obtained for all but some certificate of deposit (CD) rates. For the regressions using the short-term CD rates, there were some large and significantly negative rates; a few of the coefficients were insignificant. But for

29The six rates were: MMDA – money market deposit account, 10 quarters, September 1983–December 1985; SNOW, super now* account, 10 quarters, September 1983–December 1985; CD rates – certificate of deposit rates for 3, 6, 12 and 30 months, nine quarters from January 1983–December 1985 (CD rates had not been deregulated in September 1983).

30A more general measure of concentration which does not rely on a single arbitrary cut-off point.

[ 500 ]

M O D E R N B A N K I N G

share of the market, its total asset size, and variables capturing market risk and state ownership of banks (if any) in the local market. Schematically:

where

i : bank i’s profit, measured by return on assets

Note the dependent variable is now a measure of profit rather than price. Molyneux and Forbes pool data from a number of European countries32 for 1986 (756 banks), 1987 (1217 banks), 1988 (1538 banks) and 1989 (1265 banks). Each European country is treated as a separate local market.

The SCP hypothesis would predict α1 > 0 = α2. The relative efficiency hypothesis implies α1 = 0 < α2. When the data were pooled and the measure of market share is given in either deposits or assets, the SCP hypothesis is supported. In the regressions using yearly data, the coefficient on α2 is negative and insignificant, a rejection of the relative efficiency hypothesis. The government dummy coefficient is positive and significant – state owned banks are more profitable. The coefficient on asset (size) is negative and insignificant, suggesting that size does not influence profitability. The significant, positive coefficient on (K/A) indicates that the higher the capital adequacy, the more profitable the bank.

Altunbas and Molyneux (1994) used a three-stage least squares estimator to estimate the structural equations as well as the reduced form equations [like (9.12) and (9.13) above] to test the profits – concentration relationship. The results of OLS regressions favour SCP, as in Molyneux and Forbes. However, the three-stage least squares test results lend some support to both the SCP and the relative efficiency paradigms, and cast doubt on the validity of the OLS reduced form equations. These ambiguous results indicated the need for more sophisticated models and econometric techniques, and the inclusion of direct measures of efficiency in the model. A key issue is simultaneity: there may be more than one link between the coefficients, implying that the regression coefficients are biased.

Berger (1995) introduces two efficiency measures, and tested four hypotheses:

žHypothesis 1: the traditional SCP model.

žHypothesis 2: the Relative Market Power Hypothesis. Firms with a higher market share can exert more market power and earn higher profits, independent of how concentrated the market is.

The relative efficiency model is divided into two, to allow for either X-efficiency and/or scale economies.

32 Accounting data from IBCA is used for 18 European countries. The countries were Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Norway, Portugal, Spain, Sweden. Switzerland, Turkey and the UK.

33 Berger (1995) also uses price measures.

[ 502 ]

M O D E R N B A N K I N G

negative relationship between profits and concentration. Just one CONC coefficient was found to be positive and significant. He claims that evidence supporting SCP (higher concentration yields higher profit) in earlier papers is likely due to correlations between other variables, such as concentration and market share.

žThere is some evidence for the relative market power hypothesis. In estimations of equation (9.16), 45 out of 60 MS coefficients are positive, 22 of these are significant. Only four are significantly negative. Since the efficiency measures are included, it suggests that larger firms have gained market power through advertising, networks, and so on.

žThe relative efficiency school argues that it is the efficiency of banks that allows them to capture a higher market share, and therefore perform better. There is some support for the X-efficiency version of this hypothesis – X-efficiency has a significant and positive influence on profits. However, there is little evidence of a significantly positive coefficient for X-efficiency in the market share or concentration equations, which is needed to explain the higher profitability, i.e. greater efficiency (through better management of resources) should increase market share or concentration, which in turn increases profits. While there is evidence of a link between X-efficiency and profits, there is none to support the idea that X-efficiency raises market share or concentration. So while one necessary condition is satisfied, the other is not.

žBerger finds no evidence to support the scale efficiency version of the relative efficiency hypothesis.

žPerhaps the most important conclusion is how small changes in these variables affect a bank’s profitability. Based on the size of the coefficients, Berger reports that ROA would rise by 0.142% and ROE by 1.9% if a bank increased its market share, X-efficiency and scale efficiency by 10%, respectively. It is unlikely a bank could achieve these very large increases simultaneously, unless, Berger argues, it is done through a merger. Since most of the rise in profitability comes from an increase in X-efficiency, the acquiring firm should be looking for an inefficient target, which is likely to be able to be made as efficient as the acquiring firm.

žBerger also notes that the R2 on most of the regressions are low, under 10%. Including the market and efficiency variables raises them to about 13%. With such a low explanatory power, it suggests profitability sources come from elsewhere, such as portfolio choices or other factors not considered here.

Goldberg and Rai (1996) conduct a Berger type exercise for 11 European countries34 between 1988 and 1991. They use the large banks from each of these countries, most of which have extensive branch networks, where, unlike the USA, deposit and loan rate decisions are taken by head office and are quoted by all the national branches. There are 79 banks, ranging from 1 in Belgium to 15 in Italy. Following Jackson (1992), the authors divide the sample into high and low concentration countries, depending on their Herfindahl index and the three bank concentration ratio scores. The UK, Belgium, Finland, Sweden and Denmark were classified as high concentration countries. Data were pooled over the four years, and dummy variables used for the first three years.

34 Austria, Belgium, Denmark, Finland, France, Germany, Italy, Spain, Sweden, Switzerland, United Kingdom.

[ 503 ]

C O M P E T I T I V E I S S U E S I N B A N K I N G

The variables tested are similar to Berger (1993), except for the performance measures. In addition to ROA and ROE, they use net interest margin/total assets as a proxy for pricing by banks. They also try NIR: non-interest returns. The control variables include per capita income [(wages + salaries)/number of employees], size (log of total assets) and a measure for risk: total liabilities/total assets. Another difference from the Berger study is the use of stochastic frontier analysis to obtain the efficiency measures.35 The estimating equations are similar to the three shown above for the Berger study. The main findings from the Goldberg and Rai (1996) study are:

žThe results are highly sensitive to the performance measure used.

žThere is only evidence to support the scale efficiencies version of the relative efficiency hypothesis in low concentration countries.

žThere is no support for the relative market power hypothesis.

žThe two concentration measures do very little to explain bank performance.

žThe results are at odds with those of Molyneux and Forbes (1996), which support SCP, and Altunbas and Molyneux (1994), though the latter paper finds some evidence to support SCP and relative efficiency. The authors note that based on their findings, there is no reason for regulators to restrict bank mergers or cross-border acquisitions in Europe.

Also similar to Berger (1995), Goddard, Molyneux and Wilson (2001) investigate the SCP hypothesis for 15 European countries from 1980 to 1996. Though the explanatory power is very low, they report evidence to support the SCP hypothesis. Mendes and Rebelo (2003) use Portuguese banking data from 1990 to 1999. For the first half of the 1990s there is evidence of SCP, but in the later half, following regulatory reforms, their results support the relative market power hypothesis, that is, firms with a greater market share can earn higher profits, independent of concentration.

Corvoisier and Gropp (2002) employ a Cournot model of loan pricing, where banks are price makers in the loan market but face a given deposit rate, and show that differences in the deposit rate and loan rate will occur in markets with a low number of banks, n. As the number of banks rises to ∞, the deposit rate will approach the loan rate, making the loan market perfectly competitive. In the theoretical model, they also show the loan rate depends on aggregate loan demand, the elasticity of aggregate loan demand, the probability of default by borrowers and the ban’s operating costs.

They go on to test an empirical version of the model, where:

MARGINic = β0 + β1iCONCic + β2iRISKc + β3NRISKc + β4i(C/I)c + β5iDLc

35 Berger (1993) uses deviations from an average residual over a time horizon to obtain efficiency measures. Goldberg and Rai (1996) estimate the deviations from the stochastic cost frontier, then use the error terms to obtain measures of X-efficiency for each bank. Scale efficiency measures are also obtained using the SCF model.

[ 504 ]

M O D E R N B A N K I N G

where

MARGINic : the difference between a bank retail interest rate and a money market rate for product i and country c

CONCic : the Herfindahl index for product i in country c – it declines as n, the number of banks, and also as their shares become more similar

RISKc : a proxy for the probability of default by borrowers, which is the share of problem loans in country c. If that is not available, NRISKc, a dummy – valued at 1 if there is no measure for problem loans, 0 otherwise

C/I : the average cost to income ratio in country c, a proxy for operating costs

DL : the consumer and producer confidence indices for each country, which act as a proxy for the aggregate demand for loans

SMc : the extent of stock market capitalisation in a country c – proxies for the elasticity of aggregate loan demand; also uses the ratio to total assets in a country’s banking system to GDP

I : indicator dummy – 1 if the Herfindahl index describes concentration in the product market I, 0 otherwise

They test equation (9.17) for a number of loan and deposit categories:

1.Overall, short-term and long-term loans;

2.Mortgage loans;

3.Demand, fixed term and savings deposits.

Corvoisier and Gropp make interesting use of the Herfindahl index. They compute Herfindahl indices (recalibrated) for several bank products in each country: customer loans, short-term loans, long-term loans, mortgages and demand, savings and time deposits. For example, using consumer loans of bank k and the total number of banks in the country, the Herfindahl index is defined as:

k=1

where

Lk : consumer loans of bank k

K : total number of banks in the country

They also use the more conventional Herfindahl computation based on total assets. By this measure they find that concentration increases in the period 1995 – 99, but at a slower rate than in earlier years. On average, it grew by about 10%. The largest countries (e.g. Germany) show the least concentration, while Finland and the Netherlands show the greatest concentration. The product specific indices tend to follow the same pattern. However, the differences between products is notable. In Italy, the Herfindahl indices vary from 25 to 160 for deposits and loans. In Germany, they vary from 5 to 30; in Finland, from 350 to 500. Concentration in deposit markets tends to be higher than in loan markets.

[ 505 ]

C O M P E T I T I V E I S S U E S I N B A N K I N G

The main objective of the study was to look at the relationship between concentration and margins. The authors estimate equation (9.17) using three models, each of which eases the restrictions on slopes across different products and markets. The results show:

žConcentration has different effects depending on the type of product under consideration.

žThe more concentrated the market for loans and demand deposits, the higher the margins, which supports the structure – conduct – performance hypothesis, and supports the presence of collusion.

žFor savings and time deposits, the more concentrated the market, the lower the margins, which does not support the SCP model. The authors try to explain the result by arguing that proximity to the bank is important for demand deposits, but not for savings and time deposits. They suggest this difference could make the market more contestable.36 However, if contestability was determining interest margins, then there should be no relationship between changes in concentration and price, since the number of firms does not matter in a contestable market. Furthermore, during the period studied, the development of technology was such that bank location grew increasingly less important for all products, except for customers who insist on using a branch. A more likely explanation is that current accounts (demand deposits) lack close substitutes, whereas for savings and time deposits, there are alternatives (e.g. sweep accounts, low risk mutual funds, government bonds). This could make the demand for savings and time deposits more price elastic, especially if the amount saved is quite high, say, in excess of $1500, and customers are prepared to lock away their money for a period of time. The presence of close non-bank financial substitutes makes the market less concentrated than it appears.

A related problem may be the use of the interest rate in the computation of margins. Corvoisier and Gropp are somewhat vague on the rates used in the study: the data are reported to come from national central banks of reporting countries. However, it is well known these data tend to be highly aggregated, for example, an average interest rate for the big four or five banks in a country. A more precise measure would be constructed from each bank’s deposit or loan rate corresponding to each product, for varying amounts, at different maturities.

Angelini and Cetorelli (2003) look at competition in the Italian banking market from 1984 to 1997. They find evidence of an increase in competition post-1992, the year the European Union’s second Banking Directive (see Chapter 5) came into effect, which introduced a single passport for European banks, so banks could branch more easily across Europe. Using firm based balance sheet data on approximately 900 Italian banks over the 14-year period, they compute Lerner indices. A Lerner index is defined as:

36 See the next subsection for more detail on the meaning of contestable banking markets.

[ 506 ]

M O D E R N B A N K I N G

where

n : number of firms

ε: the elasticity of demand for the industry product, defined as a positive number

ν: the conjectural variation for output, that is, the representative firm’s belief about how industry output responds to its own output

Thus, the Lerner index measures the relative mark-up of price over marginal cost. The higher the index, the less competitive the market is. Looking at the results for commercial banks, they find Lerner indices remain largely unchanged in the first part of the period 1984 – 92, but drop after 1993, suggesting competitive conditions increased post-1992. The authors also compared a group of banks that were involved in a merger/acquisition and those that were not. However, they can find no discernible differences in the two grouped Lerner indices, suggesting consolidation had little effect on market power.

With the Lerner index as the dependent variable, regression analysis is used to test a number of explanatory variables which could have affected mark-ups. Different measures of market structure included the number of banks and the Herfindahl index. As expected, the coefficient on number of firms is positive, that is, as the number of firms increases, the Lerner index falls. A negative relationship is found between the level of concentration and the relative price mark-up, that is, as concentration rises, the mark-up falls – contrary to the prediction of the SCP hypothesis. The finding is, the authors claim, likely due to the fact that the increased consolidation was a strategic reaction by banks anticipating increased contestability. Consolidation and restructuring increased efficiency, which was passed on to consumers, hence the fall in the index.

There are two problems with these arguments. First, if, as the authors claim, contestability was a driving force, then the number of firms in the market should have no effect on pricing. Second, using Herfindahl or the number of firms as explanatory variables creates a potential simultaneity problem, because the Lerner index is derived from the number of firms [see equation (9.19)]. Thus, these results must be treated with extreme caution. Overall, their results show increased consolidation in the Italian banking sector coincided with a fall in the Lerner index, suggesting bank mergers have increased bank efficiency, which in part has been passed on to consumers. This could be part of a defensive strategy – becoming more competitive in the face of anticipated entry by banks headquartered in other states.

9.4.4. The Panzer–Rosse Statistic and Contestable

Banking Markets

Some empirical studies consider the question of whether banking markets are contestable. A contestable market is one in which incumbent firms are vulnerable to ‘‘hit and run’’ entry and exit, and given this threat, behave as though they are price takers, pricing products at average cost (equal to marginal cost with a horizontal cost curve), thereby maximising consumer surplus. This type of entry is possible if the market is one where customers can switch suppliers faster than the suppliers can reprice, if incumbents and newcomers have access to similar technology and factor prices and there are no sunk or irrecoverable costs.

[ 507 ]

C O M P E T I T I V E I S S U E S I N B A N K I N G

Sunk costs are fixed costs which cannot be recovered when a firm leaves the market/industry. Not all fixed costs are sunk costs. For example, if used machinery has a secondary market value, it can be sold, making the costs fixed but not sunk. In the banking industry, some experts argue that most of the costs are fixed but not sunk, making it contestable. New firms enter if incumbent firms are acting as ‘‘price-makers’’, that is, deposit or loan rates are lower (higher) than perfectly competitive rates, ‘‘hit’’ the market and capture market share with lower prices. They remain in the market until profit margins begin to fall when existing firms react by lowering their prices. Having made a quick profit, these firms, with virtually no sunk costs, ‘‘run’’ or exit the market when increased competition narrows profit margins.

Under these assumptions, new entrants capture market share by offering lower ‘‘prices’’. This type of market is known as contestable; the mere threat of entry keeps existing banks pricing their products at marginal cost. There are important policy implications if a market is found to be contestable. It will not matter if there are only a few firms in the industry, for example, a banking oligopoly. The mere threat of entry will mean incumbent banks price their products at marginal cost, and consumer surplus is maximised. Hence, there is no need for governments to implement policies to encourage greater entry into the market.

Shaffer (1982) and Nathan and Neave (1989) argue the Panzer – Rosse (1987)37 statistic (PR) can be used to test for contestability and other forms of competition in, respectively, the US and Canadian banking markets. The technique involves measuring market power by looking at how changes in factor prices affect firms’ revenues, quantifying the firms’ total revenue reaction to a change in factor input prices. For example, for a given change in factor prices, revenues rise less than proportionately; in the case of monopoly, there should be no response, while in perfect competition, there will be an equiproportionate increase in gross revenues.

In Shaffer (1982) and Nathan and Neave (1989), input prices consisted of the unit price of labour, the unit price of premises, and the ratio of interest expenses to total deposits for banks. PR, the Panzer – Rosse statistic, is defined as the numerical value of the elasticity of total revenue with respect to a chosen vector of input prices.

Shaffer (1982) used data for unit banks in New York, and estimated the PR statistic to be 0.318. He concluded that banks in the sample behave neither as monopolists (their conduct was inconsistent with joint monopoly) nor as perfect competitors in the long run. In Nathan and Neave (1989) a similar methodology was applied, using cross-section data (1982 – 84) from the Canadian banking system, PR values for 1983 and 1984 were found to be positive but significantly different from both zero and unity. These PR values, they argued, confirmed the absence of monopoly power among Canadian banks and trust companies. Nathan and Neave concluded their results were consistent with a banking structure exhibiting features of monopolistic, contestable competition.

Molyneux, Lloyd-Williams and Thornton (1994) tested for contestability in German, British, French, Italian and Spanish markets, using a sample of banks from these countries, for the period 1985 – 89. The authors found the PR for Germany (except 1987), the UK,

37 Originally known as the Rosse – Panzer statistic after Rosse and Panzer (1977) used it to test for competition in the newspaper industry. Following Panzer and Rosse (1987), it has come to be known as the Panzer – Rosse statistic.