CONSUMER SWITCHING COSTS AND FIRM PRICING: EVIDENCE FROM BANK PRICING OF DEPOSIT ACCOUNTS

Transcription

1 THE JOURNAL OF INDUSTRIAL ECONOMICS Volume LIX June 2011 No. 2 CONSUMER SWITCHING COSTS AND FIRM PRICING: EVIDENCE FROM BANK PRICING OF DEPOSIT ACCOUNTS Timothy H. Hannan w Robert M. Adams z We employ extensive information on bank deposit rates and area migration patterns to examine pricing relationships implied by switching costs. We argue that, because of the trade-off between attracting new customers and exploiting old ones, banks offer higher deposit rates in areas experiencing more in-migration. Further, because greater outmigration implies that a locked-in customer will not be with the bank for as many periods, banks will offer lower deposit rates in areas exhibiting greater out-migration. Also, because this effect of out-migration logically depends on the extent of in-migration, an interaction effect exists. We find evidence strongly supporting these relationships. I. INTRODUCTION FORMANYDIFFERENTPRODUCTSANDSERVICES, consumers who have purchased from one firm incur (or perceive they incur) costs if they switch to a competitor s offering. Examples are the costs associated with learning to use a new brand, the need for compatibility with existing equipment, the costs of overcoming uncertainty about the quality of unfamiliar brands, the psychological switching cost associated with brand loyalty, and, of course, the more direct costs that are sometimes incurred to change brands. An example is the transaction costs that a depositor incurs when changing banks. This requires not only the closing of one account and the opening of another, but in recent years it has come to mean also the notification of employers regarding automatic deposit of wages and notification of potentially numerous commercial enterprises regarding authorized electronic withdrawals (automatic or otherwise). It has long been recognized that the existence of such switching costs can have significant implications for pricing decisions. The exact implications The views expressed herein are those of the authors and do not necessarily reflect the views of the Board of Governors of the Federal Reserve System or its staff. The authors would like to thank Elizabeth Kiser, June K. Lee and Ron Borzekowski for helpful comments and Miranda Mei, Stephanie Ramirez and Jason Scott for excellent research assistance. w Authors affiliations: Federal Reserve Board, Washington, DC 20551, U.S.A (retired). hannan.timothy@gmail.com. z Federal Reserve Board, Washington, DC 20551, U.S.A. robert.m.adams@frb.gov. The JournalofIndustrialEconomicsr2011Blackwell PublishingLtdand the EditorialBoardof TheJournalofIndustrialEconomics. Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX42DQ, UK, and 350 Main Street, Malden, MA 02148, USA. 296

2 CONSUMER SWITCHING COSTS AND FIRM PRICING 297 depend on the underlying theoretical model employed. The simple (and naı ve) one-period model, wherein all customers are locked in and there are no new customers to attract, yields monopoly pricing when switching costs are high enough. The more commonly presented two-period model, where customers purchase a service or commodity in the first period and are locked in during the second period, yields high prices in the second period but can produce prices below those associated with short-run profit maximization in the first period. The intuition, of course, is that first period prices are lower because firms compete for customers that they can exploit in the second period. A two-period model is, however, not very useful for analyzing pricing in the general case in which new customers enter the market each period, some customers leave the market each period, and firms do not find it profitable to discriminate fully between new and old customers. Beggs and Klemperer [1992] have developed a model designed to address the pricing implications of switching costs in this context; they find that under plausible assumptions, prices and profits are higher in the presence of switching costs. 1 They also show that prices rise as firms discount the future more, fall as consumers discount the future more, fall as the turnover of consumers increases, and fall as the rate of market growth increases. 2 In this paper, we employ data from the banking industry and information on banking-market migration patterns to test several pricing implications of switching costs that can be derived from the Beggs and Klemperer [1992] model. As noted above, substantial switching costs should be relevant to customer behavior in the banking industry a point well recognized by participants in the industry. Features such as direct deposit, automated debits, and electronic bill payments in particular are often cited as roadblocks to switching accounts. 3 Also, it appears that banks cannot fully discriminate between new and old customers. While they are observed to offer free checks and other fairly minor benefits to new customers, 1 These assumptions are (1) discounting of future profits, (2) the recognition by firms that charging a higher price today will cause rivals to have more locked in customers and therefore charge higher prices tomorrow, and (3) the recognition by customers that responding to a low price today will mean a higher price tomorrow. 2 They also find that larger firms or firms with larger market shares charge higher prices than smaller firms or firms with smaller market shares, but these implications derive from a particular assumption in their model that is not likely to apply in the case of the banking industry. See Hannan [2008]. 3 Typical is a comment by Susan E. C. Riley, a senior vice president at 1 st Commonwealth Bank of Virginia, who notes that People think those links are really difficult to break. They will stay in a relationship because they think they have to. cited in Here s a Switch Easing the Customer Migration Path, American Banker, March 13, In response, some banks are investing in switch kits to make the switching process easier, but, as noted by Michael Dobbins, a senior vice president at Charter One, switching banks isn t easy. This [a switching kit] just makes it easier, cited in Do Account-Switching Kits Help Banks Draw Clients? American Banker, April 15, See also Kiser [2002] for a more detailed discussion of the evidence. TheJournalofIndustrialEconomicsr 2011BlackwellPublishingLtd and theeditorialboardof TheJournalofIndustrialEconomics.

3 298 HANNAN AND ADAMS differences in deposit rates between new and old customers are not observed for most account categories, perhaps because of the arbitrage opportunities or customer bitterness that would result from observably different rates offered to different customers for the same account type. Several features of the banking industry allow identification of the impact of switching costs on prices. The most important of these features is the local nature of bank retail deposit pricing. Despite changes that may have broadened the geographic scope of deposit markets in recent years, competition for many different types of deposit funds is still regarded as local in nature, meaning that pricing of deposit accounts can vary by area (see Amel and Starr-McCluer [2002], and Amel, Kennickell, and Moore [2008]). We shall argue that migration into and out of local banking markets is relevant to bank pricing when switching costs exist. Importantly, migration both into and out of these markets vary cross-sectionally and over time, making possible tests of the implications of switching costs not possible for many other industries. Another relevant characteristic of the industry is that switching costs may have been increasing over time as a result of the increasing use of direct deposits and arranged withdrawals. A final relevant characteristic concerns the difference in deposit products offered by banks. Of the several different types of deposit accounts that banks typically offer, it is possible that some types of accounts entail higher switching costs than others. Subject to qualifications discussed below, these differences also offer the potential to identify the role of switching costs in firm pricing. Results show a strong and robust relationship between bank deposit rates and the rates of both market in-migration and out-migration. As suggested by an application of the Beggs and Klemperer model to the banking industry, deposit rates are found to increase with the rate of migration into a market and decline with the rate of migration out of the market, with the rate of in-migration having a less positive effect on deposit rates, the higher is the rate of out-migration. These results imply that switching costs are very much a factor in explaining bank deposit rates and that banks consider the future profitability of locked-in depositors in choosing current deposit rates. Expectations based on the presumed differences in switching costs across four different types of deposit accounts receive only weak support, and possible reasons for this are discussed below. As indicated in the literature review below, this paper is the first to investigate the role of switching costs in bank deposit-rate setting since the introduction of the numerous deposit-related innovations, and it is the first to investigate the relationship between bank deposit rates and migration measures using a large sample of banks and deposit rates. Also of importance, it is the first to investigate the role of out-migration in bank deposit-rate setting, allowing for a test of intertemporal decision making by banks. The JournalofIndustrialEconomicsr2011BlackwellPublishingLtdandthe EditorialBoardof The JournalofIndustrialEconomics.

4 CONSUMER SWITCHING COSTS AND FIRM PRICING 299 The paper is organized as follows: Section II reviews the relevant literature, and section III discusses predictions regarding pricing implications of switching costs. Section IV outlines the tests to be conducted, and section V describes the data used. Section VI presents results, and a final section concludes. II. THE LITERATURE The theoretical literature on switching costs is vast, and we refer the reader to Klemperer [1995] and Farrell and Klemperer [2006] for extensive reviews. We simply note some of its salient aspects. A common fixture in this literature has been the two-period model, where firms cannot commit to future prices. Consumers that choose to purchase from a given firm in the first period are locked in for the second period. This causes firms to exploit their market power in the second period and price aggressively in the first. This bargainsthen-rip-off pattern is a main theme of many two period models. Such models can explain some instances of pricing behavior when cohorts can be identified by the firm. For example, banks offer college students gifts and free services to induce them to open accounts, followed in later years by highly profitable pricing. A two-period model is, however, less useful for analyzing competition over many periods when new customers enter the market in every period, some old customers leave, and firms are unable to profitably charge new and old customers different prices. Because the empirical environment that we wish to explore contains all three of these elements, the tests proposed and conducted in the paper are derived from the multi-period model of competition and switching costs presented in Beggs and Klemperer [1992]. In this model, the prices chosen by firms under these circumstances reflect a tradeoff between attracting new customers and exploiting old, locked-in customers. Empirical models of switching costs are far less numerous. Two studies of the banking industry actually report estimates of switching costs. Using highly aggregated data, Kim et al. [2003] estimate the magnitude of switching costs by deriving and then estimating a first-order condition, a market-share equation and a supply equation under the assumption of Bertrand behavior. Using data on Norwegian bank loans, they estimate switching costs of 4.12 per cent for the typical customer s loan, which seems quite substantial. Shy [2002], using data on prices and market shares, finds that the costs of switching deposits ranges from 0 to 11 per cent for deposit customers of Finnish banks. Some other empirical studies relevant to the banking industry have sought to test the implications of switching costs for pricing behavior. Two previous studies of bank pricing have noted that newly arrived customers in an area in essence do not face switching costs if their move required that they leave their previous bank, while existing residents could be subject to substantial additional costs if they were to switch their accounts to a different TheJournalofIndustrialEconomicsr 2011BlackwellPublishingLtd and theeditorialboardof TheJournalofIndustrialEconomics.

5 300 HANNAN AND ADAMS institution. If bank pricing reflects a compromise between exploiting old customers (made possible by switching costs) and attracting new ones, then it follows that banks located in markets with greater in-migration would optimally offer higher deposit rates (charge lower prices), all else equal. Sharpe [1997], the first to test this implication of switching costs, used monthly data on six-month CD rates and on money market deposit account (MMDA) rates offered by 222 banks from October 1983 to November Defining local banking markets as Metropolitan Statistical Areas (MSA s) or non-msa counties, Sharpe estimated the proportion of new movers into each market during this period by extrapolating from Census data applying to years , substantially earlier than the years examined. Using a pooled time series of 5 annual cross sections (and adjusting for time effects), Sharpe found that, consistent with predictions, the proportion of household migration into a market has a generally positive (pro-competitive) effect on deposit rates, all else equal. Because of the very limited sample of banks employed by Sharpe for a time period now 25 years old, a reexamination of this issue, we believe, is in order. Hannan et al. [2003] employed a measure of migration in their investigation of ATM surcharges imposed by banks on non-depositors. They report that banks in local markets with higher levels of in-migration are more likely to impose a surcharge a finding consistent with the hypothesis that ATM surcharges, because they can attract rather than repel new depositors, 4 are more likely to be imposed in markets where a greater proportion of the population can be more readily attracted. Because of the availability of annual IRS data on household migration for years more recent than those examined by Sharpe [1997], Hannan et al. [2003] used a more direct measure of market in-migration that does not involve a cumbersome extrapolation from migration data applying to earlier periods. An issue that arises in the context of these analyses is whether decision making is intertemporal. If banks do not consider the implications for future periods of attracting new customers in the current one, then new customers influence prices simply because they represent a source of more elastic demand in the current period. This current-period analysis is essentially the one modeled and presented by Sharpe [1997]. If, however, banks consider the gains obtainable in future periods from attracting a new customer in the current one, then the new customer is worth considerably more to the bank, and the extent to which new customers are present in a market should make a greater difference to bank behavior than if only the current period is considered. Both imply a negative relationship between in-migration and price (a positive relationship between in-migration and deposit rates). 4 The reason is that depositors typically do not pay surcharges for the use of their own bank s ATM s, making it more desirable to open an account at a bank with many ATM s if it is surcharging. The JournalofIndustrialEconomicsr2011BlackwellPublishingLtdandthe EditorialBoardof The JournalofIndustrialEconomics.

6 CONSUMER SWITCHING COSTS AND FIRM PRICING 301 This observational equivalence, however, does not apply to a measure of out-migration. The extent of out-migration from a market may influence the prices that banks charge or offer, but only if banks look to future periods and realize that a newly attracted customer is less valuable, the greater is the likelihood that that customer will migrate out of the market. Thus, outmigration measures should be positively related to price (negatively related to deposit rates) only if the bank considers the impact of attracting new customers during the current period on the profits obtainable in future periods. Accordingly, measures of market out-migration will be employed in the analysis reported below. III. SOME PRICING IMPLICATIONS OF SWITCHING COSTS Switching costs may be a factor in many industries characterized by the entry of new consumers over different time periods, the exit of old consumers over different time periods, and the inability of firms to distinguish between new and old customers in their pricing decisions. Under these circumstances, the following relationships may be derived for the more typical case in which the price (unlike the case of bank deposit rates) is paid by the consumer to the firm: First, an increase in the rate of in-migration results in a decrease in the prices that firms charge their customers. The reason is that the firm finds it optimal to charge lower prices, the larger the proportion of customers that are new to the market and that therefore can be readily attracted. Second, an increase in the rate of out-migration causes firms to increase the prices that they charge their customers, all else equal. The reason for this effect is only slightly less intuitive. It occurs only if the firm considers new customers as exploitable in future periods after they are locked in, rather than just a more elastic source of product demand in the current period. Greater out-migration deters firms from offering low prices to attract new customers, because new customers (on average) will not remain with the firm for as long (and therefore be exploitable) as would otherwise be the case. A final implication is that the effect of in-migration should depend on the level of out-migration, and vice versa. We discuss this interaction in more detail in the context of bank deposit rates. The proposed test of these predictions will focus on retail deposit rates of banks. Because deposit rates are prices that are paid to the customer by the firm, rather than paid to the firm by the customers, the predicted effects are the opposite of those that obtain in the more common case discussed above. In the presence of switching costs, greater in-migration into retail banking markets should cause banks to offer higher deposit rates because of the greater importance of attracting new customers, while greater out-migration should cause banks to offer lower deposit rates because locked-in depositors are less desirable. Finally, the positive effect of in-migration on deposit rates TheJournalofIndustrialEconomicsr 2011BlackwellPublishingLtd and theeditorialboardof TheJournalofIndustrialEconomics.

7 302 HANNAN AND ADAMS should be less pronounced (less positive), the greater is out-migration. Because these implications are quite intuitive, formal derivations are not presented here. See Hannan [2008] for a formal derivation of these implications in the case of deposit rate setting by banks. IV. THE TEST If the interaction between in-migration and out-migration can be captured by the product of these two variables in a linear deposit-rate regression, we can estimate: ð1þ r t d;i ¼ b 0 þ b 1 inmigrate t 1 m þ b 2outmigrate t 1 m þ b 3 outmigrate t 1 m ðinmigratet 1 m Þþb 4Mm t 1 þ b 5Bi t 1 þ n t þ m i þ e it ; t t 1 where r d.i denotes the retail deposit rate of bank i at time t, inmigrate m denotes the rate of in-migration observed for market m at time t 1, t outmigrate 1 m denotes the rate of out-migration observed for market m at time t t 1, M 1 m denotes a vector of market characteristics prevailing in market m t at time t 1, B 1 i denotes a vector of characteristics of bank i at time t 1, n t denotes a time-specific fixed effect, m i denotes a bank-specific fixed effect, and e it denotes an idiosyncratic error term. Explanatory variables are lagged because of the likelihood that it takes some time for banks to set deposit rates after a change in the characteristics that influence those rates (and because of the way deposit rates are calculated, as discussed below). In the absence of the interaction term, the hypothesized positive impact of in-migration and the hypothesized negative impact of out-migration on deposit rates imply that b and b 2 o 0, respectively. If, however, the product of in-migration and out-migration is included as an additional variable, then the hypothesized positive impact of in-migration on deposit rates implies ð2þ b 1 þ b 3 outmigrate t 1 m >0:; while the hypothesized negative effect of out-migration implies that ð3þ b 2 þ b 3 inmigrate t 1 m 0; The hypothesized interaction between in-migration and out-migration implies that b 3 o 0, because (focusing on (2)) new customers are less desirable if they on average do not stay as long, causing the positive impact of in-migration on deposit rates to decline with an increase in out-migration. With b 3 o 0, it follows from (2) that b 1, the coefficient of inmigrate in (1), is positive. If the influence of the interaction between in-migration and out-migration is adequately captured by the product of these two variables, The JournalofIndustrialEconomicsr2011BlackwellPublishingLtdandthe EditorialBoardof The JournalofIndustrialEconomics.

8 CONSUMER SWITCHING COSTS AND FIRM PRICING 303 then we predict (focusing on (3)) that b 2 5 0, since out-migration should have no effect on bank deposit rates if there is no in-migration. For purposes of statistical control, both firm and year fixed effects are used in all estimations. Only variation over time that differs across firms needs additional statistical control. 5 Market variables employed in the basic estimations reported below include the Herfindahl-Hirschman index of market concentration and the total real income in the market. Measures of market population growth, market income growth, and a measure of the importance in the market of large competitors that operate predominantly outside the market will also be included in some regressions. The coefficient of the Herfindahl-Hirschman index of concentration (defined as the sum of squared market shares and denoted hhi m t 1 )is predicted to be negative if, as implied by the traditional structure-conductperformance hypothesis, firms in more concentrated markets offer lower deposit rates because of the greater exercise of market power. Real market income, measured in natural logs and denoted ln(mktinc m t 1 ), plays an important role in these regressions, because it accounts for changes in the size of the market that might otherwise be incorrectly attributed to inmigration or out-migration. A shift in market income can influence deposit rates by increasing the supply of deposits, implying, in all likelihood, a negative relationship between ln(mktinc m t 1 ) and bank deposit rates. 6 An increase in market income, however, could also cause an increase in the demand for bank loans, and this could under some circumstances increase bank demand for deposit dollars, with a resulting increase in bank deposit rates. The natural log of this market variable is employed because it is highly positively skewed, and it is not reasonable to expect it to exhibit a linear relationship with deposit rates over the large range of values observed in the data. Some studies have found that competition with large banking organizations that operate in many different local geographic markets can influence the deposit rate of banks operating in a given market. One reason is that multimarket banks tend to offer the same deposit rates in many different markets, implying that their presence in a given market may influence the deposit rates of banks for reasons extraneous to the underlying conditions of any one market. 7 To control for this potential effect, we include in some 5 Of particular relevance in this regard are the general level of interest rates and other aspects of the macroeconomic environment. Assuming they apply to all areas of the country, their influence on bank deposit rates should be accounted for by the year fixed effects. 6 This result is more likely if the bank sets deposit rates such that the incremental outlay for a unit of deposits equates with the incremental gain obtained by holding an asset whose return to the bank is determined in a national or international market. See Klein [1971] for a classic discussion of this issue. 7 Yet single-market banks offer very different deposit rates in those same markets, implying that the uniform pricing observed for multimarket banks is not simply a matter of incorrect identification of market areas. See Heitfield [1999]. TheJournalofIndustrialEconomicsr 2011BlackwellPublishingLtd and theeditorialboardof TheJournalofIndustrialEconomics.

9 304 HANNAN AND ADAMS regressions reported below the share of market branches owned collectively by large, predominantly out-of-market banks (lpombshare m t 1 ). Following previous literature, this is measured as the collective branch share of banks that have less that 30 per cent of their deposits in the market and that have more than $1 billion in assets (measured in year 2000 dollars). 8 A bank-specific variable included in regressions reported below is the bank s market share. Market share has sometimes been used as an indicator of firm-specific market power. Its use in a price regression, however, raises serious problems of endogeneity, due in part to the fact that a firm s price is an obvious determinant of its market share. To ameliorate endogeneity bias when this variable is included, it will be measured as the bank s share of market branches (denoted branchshare i t 1 ) rather than the share of market deposits, and lagged values will be employed. Because this treatment may not eliminate all sources of endogeneity, results of regressions that exclude this variable are also reported. An additional bank-specific measure to be included in some regressions reported below will account for the size of the bank, measured by the natural log of the bank s total assets (denoted ln(bkassets i t 1 )) This variable too will be excluded from some regressions because of endogeneity concerns. To account for difficult-to-measure differences that may exist between urban and rural environments, a variable reflecting the extent to which a bank operates in urban markets (denoted urban i t 1 ) is also included. V. THE DATA The data set employed in the analysis consists of observations of individual banks observed annually from 1989 to 2006, yielding over 140,000 bankyear observations. For each bank and year, the interest rate measures were obtained for four different types of interest-bearing retail deposit accounts and a broader interest rate measure that incorporates service charges on deposit accounts. The four retail deposit rates examined are the rate for interest-bearing transactions accounts (denoted itrate), the rate for savings deposits (denoted svrate), the rate for time deposits less than $100 thousand (denoted smtrate), and the rate for time deposits greater than $100 thousand (denoted lgtrate). Interest bearing transaction accounts include NOW accounts, ATS accounts, and telephone and preauthorized transfer accounts, while savings accounts include money market deposit accounts and other savings accounts, as indicated on bank income and call report data. Interest rates for each account category were calculated by dividing the reported annual interest expense by the average of the beginning and endof- year dollar values of the accounts held. Due primarily to reporting errors, this procedure can produce some fairly unrealistic estimates of deposit rates. 8 See Hannan and Prager [2009]. The JournalofIndustrialEconomicsr2011BlackwellPublishingLtdandthe EditorialBoardof The JournalofIndustrialEconomics.

10 CONSUMER SWITCHING COSTS AND FIRM PRICING 305 To reduce the impact of such errors, observations containing the largest and smallest one per cent of values in each account category are eliminated from the analysis. A potentially useful alternative would be to calculate rates, net of the fees that depositors must pay to the bank. Unfortunately, information on deposit-related fees is not broken down by deposit category. Since a substantial portion of such fees should be associated with transaction accounts (both interest-bearing and non-interest bearing), we calculate the rate paid on all transaction accounts by subtracting fee income from interest expenses on transaction accounts (denoted transnetfeerate). 9 Since the fee data used in this calculation include fees that apply to accounts other than transaction accounts, some unavoidable measurement error may occur. The most original source of data employed in the analysis is that used to measure market-specific rates of in-migration and out-migration. These measures are calculated using the county-to-county migration data collected and reported annually by the Internal Revenue Service (IRS). These data are constructed from year-to-year changes in addresses shown on the population of returns from the IRS Master File system. For each county in the U. S., these data indicate the number of filers that immigrated into the county during the year (identified by an address in the county at the time of filing and an address outside the county at the time of the previous year s filing), the filers that emigrated out of the county during the previous year (identified by an address outside the county at the time of the filing and an address inside the county at the time of the previous year s filing, and the filers that did not migrate in or out of the county during the previous year (identified by an address in the county at the time of both filings). An issue associated with the use of these data concerns the choice between the number of returns and the number of exemptions associated with the returns, both of which are available from this data source. We employ migration data as reflected in the number of returns, since household migration would seem to be a better measure of bank account activity than a measure heavily influenced by the number of family members. Another issue concerns the treatment of multi-county markets. Following previous studies, 10 urban markets are defined as the county or the collection of counties that make up a metropolitan area. Rural markets are defined as labor market areas, as defined by the Bureau of Labor Statistics. These labor market areas are typically identical to counties, but sometimes they are larger areas formed by combining counties when 15 per cent or more of the 9 Of course, no interest expense would be registered for the transaction accounts in this category that do not pay interest. 10 See, for example, Hannan and Prager [2004], Berger and Hannan [1989], and Calem and Carlino [1991]. TheJournalofIndustrialEconomicsr 2011BlackwellPublishingLtd and theeditorialboardof TheJournalofIndustrialEconomics.

11 306 HANNAN AND ADAMS employed workers in one county commute to another. 11 For multi-county markets, an issue arises concerning the appropriate treatment of migration from one county to another within the same market. For the purpose of this paper, such migration is not counted as relevant to the pricing of banks in the defined market. Such moves are therefore netted out in calculating migration into and out of multi-county markets. For these defined markets, the rate of in-migration (inmigrate) is constructed as the number of new filers in the market, divided by the total number of filers, where both numerator and denominator are measured for the tax year previous to the year for which deposit rates are observed. 12 The rate of out-migration (outmigrate) is measured as the number of filers who left the market, divided by the total number of filers, where the numerator refers to the tax year for which deposit rates are observed, and the denominator refers to the previous tax year. This difference in timing used to measure the rates of in-migration and out-migration reflects the fact that the rate of out-migration logically requires information for two consecutive periods (the period in which filers were observed to have left and the period in which those same filers were observed to be in the market), while the rate of in-migration does not. Data to calculate market shares and the Herfindahl-Hirschman index, defined as the sum of squared market shares (measured in deposits) of all banks and thrift institutions operating in the market, are obtained annually from branch-specific information on institution deposits, as reported in the Federal Deposit Insurance Corporation s Summary of Deposits and the Office of Thrift Supervision s Branch Office Survey. These data are also used to calculate the proportion of branches in each market accounted for by large, predominantly out-of-market banks. Data on market income are obtained from the Department of Commerce s Regional Accounts Data, while data on market population are obtained from the U.S. Bureau of the Census. The assets of each banking institution are obtained from bank balance sheet data. In the case of banks that operate in more than one local market, all market-specific variables are calculated as weighted averages of market values, with the share of each bank s total deposits that are booked in each market serving as the weights. All of these variables are lagged one year because of a probable lag between the generation of revenue used in the calculation of the deposit rates and the setting of a deposit rate, as well as the possible lag between the setting of deposit rates and the observation of market and bank characteristics. 11 See for a detailed discussion. 12 For simplicity, subscripts and superscripts will be dropped from here on. This measure of inmigrate in essence indicates the importance of depositors that could be observed by banks at the beginning of the year for which deposit rates are measured, and this seems preferable to a measure that would apply to the end of the year. The JournalofIndustrialEconomicsr2011BlackwellPublishingLtdandthe EditorialBoardof The JournalofIndustrialEconomics.

12 CONSUMER SWITCHING COSTS AND FIRM PRICING 307 VI. THE RESULTS Table I defines all variables employed in the analysis, and Table II presents mean values, by year, of the different deposit rates and two relevant ratios. Note from Table II that the period from 1989 to 2006 was one of generally declining deposit rates. The mean values of the rates paid on interest-bearing transaction accounts (itrate) declined from.048 in 1989 to nearly zero by 2004, with a slight increase thereafter. The typically higher rates paid on large and small time deposits (and to a lesser extent savings accounts) remained substantially above zero for the entire period. The last two columns in Table II indicate that, while the average rate of interest paid on transaction accounts declined sharply by the end of the period, the average of the ratio of deposit fees to transaction accounts increased somewhat over the same period. itrate svrate smtrate lgtrate transnetfeerate hhi Table I Variable Definitions The interest rate offered on interest-bearing transaction accounts, calculated from bank income and balance sheet data (see text). The interest rate offered on savings accounts, calculated from bank income and balance sheet data (see text). The interest rate offered on small time deposit accounts (less than $100 thousand), calculated from bank income and balance sheet data (see text). The interest rate offered on large time deposit accounts (greater than or equal to $100 thousand), calculated from bank income and balance sheet data (see text). The interest rate paid on transactions accounts net of deposit fees charged (see text). The market Herfindahl-Hirschman index of concentration, calculated as the sum of squared deposit shares of all banks and thrift institutions in the market. 1 urban The share of total deposits booked at branches located in markets classified as urban. ln(mktincome) The natural log of total income in the market, adjusted for changes in the CPI. 1 branchshare The bank s share of all branches of banks and savings associations in the market. 1 inmigrate outmigrate outmig-decilei ln(bkassets) lpombshare The rate of migration into the market, calculated as the proportion of all IRS personal returns filed in the market that had addresses indicating a move into the market since the previous filing (see text). 1 The rate of migration out of the market, calculated as the proportion of all IRS personal returns filed in the market that had addresses indicating a move out of the market by the subsequent filing (see text). 1 A binary variable that receives the value of one if the value of outmigrate falls within the ith decile of its range and zero otherwise. Natural log of the assets of the bank. The collective share of branches in the market accounted for by banks that have more than $1 billion in assets (in year 2000 dollars) and that have less than 30 per cent of their deposits in the market. 1 popgrowth Annual rate of population growth of the market. 1 incgrowth Annual rate of income growth of the market. 1 1 Banks operating in multiple markets are assigned a weighted average, with the share of the bank s total deposits in each market serving as the weights. TheJournalofIndustrialEconomicsr 2011BlackwellPublishingLtd and theeditorialboardof TheJournalofIndustrialEconomics.

13 308 HANNAN AND ADAMS Table II Means ofdepositrates andvarious Ratios, byyear itrate svrate smtrate lgtrate Trans. int./ trans. accts Dep. fees./ trans. accts The mean values of the rate of in-migration and out-migration (inmigrate and outmigrate) varied from.057 to.066 over the period (not shown), and the maximum annual values for these variables ranged from.22 to.25. Differences in values of these variables between urban and rural markets are small. The correlations between the rate of in-migration and the rate of outmigration and correlations of these variables with other market characteristics are useful in providing a framework for the multivariate results reported below. Because reported estimations (with one exception) will include firm fixed effects, market-specific intertemporal variations are the most relevant for this purpose. The average market-specific correlation between the rate of in-migration and the rate of out-migration over the period is.127. The equivalently measured correlation between market population growth and the rate of in-migration is.558, while that between market population growth and the rate of out-migration is.325. Thus, it appears that on average, market-specific rates of in-migration and out-migration are not highly correlated. Not surprisingly, markets with greater population growth have greater rates of in-migration and lower rates of out-migration. 13 Correlations with another market characteristic, market concentration, are quite low. Table III and IV present the results of panel data estimations obtained using the entire sample of over 13,000 banks observed annually over the 13 The mildly positive correlation between in-migration and out-migration rates implies that population growth is not the only characteristic associated with migration flows. The JournalofIndustrialEconomicsr2011BlackwellPublishingLtdandthe EditorialBoardof The JournalofIndustrialEconomics.

14 CONSUMER SWITCHING COSTS AND FIRM PRICING 309 Table III Bank Deposit Rates and the Extent of In-Migration and Out-Migration in Local Banking Markets,1989^2006, with Bank andyear Fixed Effects itrate svrate smtrate lgtrate transnetfeerate constant (41.03) (39.96) (71.01) (45.58) (4.94) hhi.12e-6.17e-6 þ.11e-7.12e-6.96e-7 ( 1.16) ( 1.79) (.13) (.90) (.34) urban.88e e-3 þ ( 1.55) (4.77) (1.90) (2.50) (.43) ln(mktincome).18e-3.63e-4.83e-4.28e (2.23) (.62) (1.07) (.22) ( 3.13) branchshare.987e-3.50e e ( 1.40) (.66) ( 2.84) (.38) ( 2.63) inmigrate (3.78) (8.86) (9.94) (4.70) (.37) outmigrate ( 4.23) ( 8.65) ( 8.80) ( 3.79) (.97) year ( ) ( ) ( ) ( 67.60) ( 47.95)..... year ( ) ( ) ( ) ( ) ( 65.63)..... year ( ) ( ) ( ) ( ) ( 52.26) R Number of observations 142, , , , ,392 Number of banks 13,222 13,222 13,222 13,222 13,222 Note: t-statistics are presented in parentheses. The symbols þ,,and denote statistical significance at the 10, 5, and 1 per cent levels, respectively. period 1989 to All regressions include both bank and year fixed effects and allow for correlation of errors across banks in the same local market. 14 Table III presents results obtained for each measure of interest rates when the measures of both the in-migration rate (inmigrate) and the out-migration rate (outmigrate) are included as separate explanatory variables, with no interaction terms. As presented in Table III, the coefficients of inmigrate are positive and significant in all four of the retail deposit-rate regressions, consistent with the hypothesis that banks offer higher deposit rates to attract new customers, the greater is the share of customers in the market that can be more readily 14 For banks that operate in more than one market, the market in which the bank has the most deposits is used for this purpose. The vast majority of banks operate predominantly in one market. TheJournalofIndustrialEconomicsr 2011BlackwellPublishingLtd and theeditorialboardof TheJournalofIndustrialEconomics.

15 310 HANNAN AND ADAMS Table IV Bank DepositRates andthe Extent ofin-migration and Out-migration (includingtermstocapturetheirininteraction)inlocalbankingmarkets, 1989^2006, with Bank andyear Fixed Effects Dependent Variables: itrate svrate smtrate lgtrate transnetfeerate constant (38.05) (39.22) (67.21) (43.78) (4.60) hhi.11e-6.17e-6 þ.13e-7.12e-6.89e-7 ( 1.15) ( 1.78) (.16) (.88) (.32) urban.95e-3 þ e ( 1.68) (4.64) (1.58) (2.31) (.52) ln(mktincome).18e-3.68e-4.84e-43.26e (2.21) (.59) (1.08) (.20) (-3.19) branchshare.99e-3.52e e ( 1.42) (.68) ( 2.86) (.40) ( 2.65) Inmigrate (5.07) (11.30) (11.84) (6.53) (2.20) outmigrate ( 1.03) ( 3.12) ( 3.88) (.22) (.80) inmigrate outmig-decile ( 1.03) ( 3.12). ( 5.13) ( 3.33) ( 1.44) inmigrate outmig-decile ( 2.93) ( 5.55) ( 6.66) ( 4.71) ( 3.36) Inmigrate outmig-decile ( 3.25) ( 7.14) ( 7.92) ( 4.47) ( 3.26).. inmigrate outmig-decile ( 4.07) ( 7.18) ( 8.55) ( 4.49) ( 3.86) inmigrate outmig-decile ( 3.47) ( 6.74) ( 7.82) ( 4.71) ( 2.97) R Number of observations 142, , , ,392 14,392 Number of banks 13,222 13,222 13,222 13,222 13,222 Note: See Table III. attracted. This finding also is generally consistent with results reported by Sharpe [1997] and Hannan et al. [2003], who report similar findings using different measures of bank prices and much more restrictive data sets. Coefficient magnitudes suggest that a one standard deviation increase in the rate of in-migration would cause deposit rates to increase by 4.8 basis points in the case of interest-bearing transaction accounts, 7.4 basis points in the case of savings deposits, 8.7 basis points in the case of small time deposits, and 7.2 basis points in the case of large time deposits. One might suspect that differences in in-migration would have a larger effect on products for which switching costs should be more important. The JournalofIndustrialEconomicsr2011BlackwellPublishingLtdandthe EditorialBoardof The JournalofIndustrialEconomics.

16 CONSUMER SWITCHING COSTS AND FIRM PRICING 311 However, the coefficients of inmigrate are actually larger in magnitude in the case of large and small time deposit rates (smtrate and lgtrate) than in the case of interest-bearing transactions account rates (itrate), where one would expect switching costs to be the greatest This difference in magnitude is attributable to the fact that, as demonstrated in Table II, the levels of itrate (and to a lesser extent, svrate) approached zero during the later years of the period, thus reducing disproportionately the magnitudes of their coefficients in estimations that use data from the whole period. Reestimation using data for the earlier half of the sample (from 1989 to 1997) yields highly significant coefficients of inmigrate of.047 in the case of itrate and.043 and.049 in the case of smtrate and lgtrate, respectively (not reported). Since the levels of itrate averaged about half of the levels of smtrate and lgtrate during the period (see Table II), this suggests that the impact of inmigrate was about twice as large in percentage terms for interest bearing transaction accounts than for small and large time deposits, Nonetheless, the highly significant coefficients of inmigration (and outmigration)inthesmtrate and lgtrate regressions pose a potential challenge to the argument that they reflect switching costs. One would not expect such results if depositors in these accounts incurred no costs to switch from one bank to another and if their decisions regarding where to hold these accounts were not influenced by the location of their other accounts. In regard to the first of these conditions, the hassle cost of closing a time deposit at one bank and opening it at another may not be trivial. In regard to the second, if new customers tended to open several types of accounts, then economies associated with one-stop shopping might induce them to consider the combination of rates offered on several types of accounts in choosing their new bank. 15 A related explanation concerns the potential for existing depositors to arbitrage across their bank s deposit accounts. If the costs of switching from one account to another within the same bank are negligible, such within-bank arbitrage might cause the rates offered on some accounts to reflect the rates offered on accounts that are more directly influenced by switching costs. The coefficients of outmigrate are negative and significant in all four of the retail deposit-rate regressions, consistent with the hypothesis that banks tend to price less aggressively to attract new depositors, the less time they are expected to remain with the bank. Because of the potential importance of interaction terms involving this variable, results associated with it are discussed more fully below. The final column in Table III reports results obtained when transnetfeerate, an estimate of the rate paid on all transaction accounts net of deposit 15 Data from Survey of Consumer Finance show a strong tendency of bank customers to hold their various deposit accounts in one institution. See Amel, Kennickell, and Moore [2008], Table 7. This notion of a cluster of retail banking products has been used as a product market in banking antitrust for several decades. TheJournalofIndustrialEconomicsr 2011BlackwellPublishingLtd and theeditorialboardof TheJournalofIndustrialEconomics.

17 312 HANNAN AND ADAMS fees, is employed as the dependent variable. In this regression, neither the coefficient of inmigrate nor that of outmigrate is statistically significant. We have noted that this measure may be particularly flawed because the fee information used in its construction may apply to account types other than that for which the rate is calculated. Despite this, we will report results below that are more consistent with expectations when interaction terms are included in the analysis. The coefficients of the other explanatory variables are also of interest. Consistent with the greater exercise of market power in more concentrated markets, the coefficients of the Herfindahl-Hirschman index (hhi) are negative (with one exception). However, the coefficient is statistically significant (and then only marginally) only in the case of the rate offered for savings deposits, svrate. The coefficients of urban are statistically insignificant for transaction accounts but are positive and significant for the other three deposit rates examined. Thus, for non-transaction accounts, the deposit rate that a bank offers tends to rise, the larger its share of deposits that come from urban markets The per unit cost of transactions is probably greater in urban areas, and this may cause the relationship to be more negative in the case of transaction accounts. The natural log of real market income over time, ln(mktincome), is positive and significant in the case of itrate, but not in the case of the three other retail deposit rates. As noted above, this variable may capture both demand-side and supply-side differences, and thus its net effect is not suggested, a priori. The coefficients of branchshare are negative in all four deposit-rate regressions and statistically significant in only one of them. In the case of the transnetfee regression, the coefficients of both ln(mktincome) and branchshare are negative and significant. Because this rate includes deposit fees, these negative coefficients undoubtedly reflect the frequently observed phenomenon of higher fees associated with banks that are larger and that operate in larger markets. 16 All reported regressions include a full set of binary variables indicating the year, with the first year, 1989, serving as the excluded category. For reasons of space, only the coefficients for three relatively late years in the sample, 2000, 2003, and 2006, are reported. With 1989 serving as the excluded category, these coefficients may be interpreted as the changes in the deposit rates occurring over the relatively long periods of 1989 to 2000, 1989 to 2003, and 1989 to 2006 respectively, that cannot be explained by changes in the included explanatory variables. They document the substantial decline in interest rates that occurred over the period. Table IV reports results obtained when terms designed to account for the interaction between inmigrate and outmigrate are included in the analysis. 16 See Hannan [2002]. The JournalofIndustrialEconomicsr2011BlackwellPublishingLtdandthe EditorialBoardof The JournalofIndustrialEconomics.

18 CONSUMER SWITCHING COSTS AND FIRM PRICING 313 Interaction effects are typically accounted for by the inclusion of a single regressor defined as the product of the two variables in question. Inclusion of such a term in the regressions reported in Table III yields significant positive coefficients of inmigrate and significant negative coefficients of outmigrate, but the coefficient of the interaction term is significantly negative only in the case of the itrate regression (not reported). Further, the significant negative coefficients of outmigrate in this specification do not support the prediction that out-migration has no effect in the absence of in-migration, and failure to find significant negative coefficients of the interaction term in more than one case does not provide convincing support for the hypothesis that the predicted positive impact of in-migration on deposit rates is reduced by greater out-migration (results not reported). Because this specification imposes a linear relationship between this interaction term and the dependent variable, we estimate and report a more flexible functional form that allows for a nonlinear relationship. Instead of a variable calculated as the product of inmigrate and outmigrate, each regression includes 9 variables, each calculated as the product of inmigrate and a binary variable indicating for each decile of the range of outmigrate whether the observation falls into that decile (see Table IV). The results indicate that indeed the relationship is not linear. Consider first the results reported in column 1 for the itrate regression. We find that the coefficient of inmigrate is significantly positive, as predicted, and that the coefficient of outmigrate is not statistically different from zero when interaction terms are included, consistent with the hypothesis that outmigration has no effect in the absence of in-migration. 17 To understand the meaning of the interaction terms employed, note that outmig-decile2 is defined as a binary variable that receives the value of one if the value of outmigrate falls within the second decile of the values of outmigrate observed in the data, and zero otherwise. The variables outmig-decile3 through outmig-decile10 are defined similarly for the other deciles. The point estimate of.0086 obtained for the interaction of inmigrate with outmig-decile 2 implies that, for observations of outmigrate in this decile, the positive association between inmigrate and the rate paid on interest-bearing transactions accounts (itrate) is less by.0086 compared to observations in the first decile, which serves as the omitted category. The coefficients of the interactions involving higher deciles are highly significant and become more negative up to about the 5 th decile, where the value is.025, with no further reductions in coefficient values found for interactions involving higher deciles. The coefficient of the interaction involving the tenth decile is included to demonstrate this leveling off. 17 This follows because, with no in-migration, inmigrate 5 0. Thus, all interaction terms take the value of zero. If the coefficient of outmigrate equals zero, then out-migration has no effect on deposit rates under these circumstances. TheJournalofIndustrialEconomicsr 2011BlackwellPublishingLtd and theeditorialboardof TheJournalofIndustrialEconomics.

19 314 HANNAN AND ADAMS Interactions involving deciles 6 through 9 also have coefficients of about.025, but are not shown to save space. These results are consistent with the hypothesis that new customers are less attractive (as evidenced by less of a positive association between inmigration and deposit rates), the greater is out-migration. The results also suggest that this phenomenon exhausts itself at about the fifth decile of the values of outmigrate, since the observed decline in the coefficients of the interaction terms does note continue beyond this decile. The use of the product of inmigrate and outmigrate as a single interaction term would not capture this change in coefficients as out-migrate changes. Results obtained for the other three deposit rates examined (svrate, smtrate, and lgtrate) are similar in several regards. The coefficients of inmigrate are positive and highly significant in all cases, and all three regressions provide evidence of interactions between inmigrate and outmigrate (with the predicted signs) that are exhausted at about the fouth or fifth decile of the values of outmigrate. In the case of the svrate and smtrate regressions, we find negative and statistically significant coefficients of outmigrate a finding not consistent with the hypothesis that in the absence of in-migration, out-migration plays no role in influencing deposit rates. These coefficients, however, are considerably smaller in absolute value when the interaction terms are included, as may be seen by comparing the coefficients of outmigrate reported in this table with those reported in Table III. The final column of Table IV presents the results of an equivalent regression for transnetfeerate, the rate calculated for all transaction accounts (interest-bearing and not interest-bearing) net of deposit fees. In contrast to the results reported in Table III, and despite the potential measurement problems noted for this calculated rate, coefficient estimates are consistent with predictions. The coefficient of inmigrate is positive and significant, the coefficient of outmigrate is not statistically different from zero with the inclusion of the interaction terms, and the coefficient estimates of the interaction terms suggest again that the positive relationship between this calculated rate and the rate of in-migration declines with the rate of outmigration up to about the 5 th decile of outmigrate. 18 The coefficients of all other variables reported in these regressions are similar to their counterparts in the regressions presented in Table III. In what follows, we consider the robustness of these results to various changes in specification and sample. Most of these changes are presented explicitly in Table V. While our discussion of the robustness of results to various changes will focus on the results presented in Table IV, which 18 As noted, the numerator of this rate is calculated as interest expense minus deposit fee income, Use of the interest and fee components as dependent variables in this specification yields generally significant coefficients with predicted signs (not reported). The JournalofIndustrialEconomicsr2011BlackwellPublishingLtdandthe EditorialBoardof The JournalofIndustrialEconomics.

20 CONSUMER SWITCHING COSTS AND FIRM PRICING 315 Table V Bank DepositRates andthe Extent ofin-migration and Out-migration in Local Banking Markets, for Periods1989^2006, with Additional ExplanatoryVariables, Restricted Samples, and Bank andyear Fixed Effects Dependent Variables: Banks Predominantly in One Full Sample Market itrate svrate itrate svtrate itrate svrate itrate svrate constant (37.95) (39.21) (21.71) (29.47) (6.35) (12.71) (9.97) (13.34) hhi.15e-6 þ.19e-6.11e-6.17e-6 þ.89e-7.16e-6.17e-6 þ.12e-6 ( 1.64) ( 2.14) ( 1.14) ( 1.77) (.91) ( 1.56) ( 1.84) ( 1.11) urban.82e e-3 þ þ ( 1.46) (4.82) ( 1.72) (4.60) ( 1.70) (4.13) ( 1.29) (3.30) ln(mktincome).18e-3.60e-4.93e-4.20e-3.21e-3.15e-4.63e-3.43e-3 (2.19) (.60) ( 1.16) ( 1.99) (2.48) (.15) (2.05) (1.60) branchshare e-3.31e-6.61e-3.24e-3 ( 3.27) ( 1.49) ( 1.13) (.00) (.78) (.28) inmigrate (5.05) (11.30) (4.95) (11.54) (4.50) (9.90) (2.49) (4.99) outmigrate ( 1.02) ( 3.11) ( 1.37) ( 3.33) (.73) ( 2.96) (.40) ( 1.22) TheJournalofIndustrialEconomicsr 2011BlackwellPublishingLtd and theeditorialboardof TheJournalofIndustrialEconomics.