Vertical Differentiation, Risk-Taking and Retail Funding

Vertical differentiation occurs when customers rank products from the best to the worst using an objective measurement, such as quality. This stands in contrast to horizontal differentiation when customers choose between products based on personal preference. In this paper, we apply these concepts to depositors and differentiation among banks.

Matutes and Vives (1996) develop a theory model where depositors’ perception of a bank’s quality depends on its probability of default. Because the premise of Kim et al. (2005) is that clients have a relative preference for more solvent banks, their study computes vertical differentiation on the basis of measures that reflect a bank’s relative probability of failure.

In this paper, a banking correspondent is a third party that establishes business relationships with a bank to offer, on behalf of and for it, financial services to the bank’s clients. Legal responsibility for the financial operations carried out by banking correspondents remains with the bank. The correspondents, in turn, form a network of access points through which clients can carry out different financial transactions such as withdrawals, deposits, credit payments with a charge to a debit card, payments of basic services with a charge to a debit or credit card, balance and transactions inquiries, etc. In Mexico, the network of banking correspondents is formed by a number of commercial firms including, among others, Oxxo, Wal-Mart, and Telecomm.

Berger et al. (2009) use a sample of twenty-three developed economies and different proxies for risk-taking and competition. The proxies for banks’ risk-taking are the ratio of non-performing loans to total loans, the z-score, and the equity-to-assets ratio. The proxies for competition are the Lerner index, a Herfindahl–Hirschman index (HHI) for deposits, and a HHI for loans.

Many studies have used concentration as a competition measure even if they don’t relate concentration to competition.

Beck et al. (2006) use indicators of entry barriers, banking and economic freedom, and bank activity restrictions, among others, as proxies for competition. They show that an increase in the concentration ratio of the three largest banks reduces the probability of a banking crisis. They conclude that such a reduction has to do with “something other than” than market power.

Related to this point, Carlson and Mitchener (2006) find that profits were lower in states of the US that allowed branch banking, suggesting that branching led to increased competition rather than monopoly power. In contrast, it is worth noting that Allen and Gale (2004) consider a model of spatial competition in which competition can be consistent with diversification and stability. If they allow a bank to occupy several locations, the bank can pool several independent risks, and in some cases, there is no trade-off between competition and financial stability.

This database can be accessed at: http://​portafoliodeinfo​rmacion.​cnbv.​gob.​mx/​Paginas/​defaultGOBMX.​aspx.

This database can be accessed at: http://​www.​banxico.​org.​mx/​estadisticas/​statistics.​html.

This yields a number of components and a number of uncorrelated linear combinations of the variables in which weights are chosen so that the variance of the corresponding combination is maximized. We retain the first component.

Precisely, we use: ,\(ATM_{i,t}={\textstyle\sum_k}\frac{p_{k,2010}}{p_{2010}}ATM_{i,k,t}\), \(POS_{i,t}={\textstyle\sum_k}\frac{p_{k,2010}}{p_{2010}}POS_{i,k,t}\), \(AP_{i,t}={\textstyle\sum_k}\frac{p_{k,2010}}{p_{2010}}AP_{i,k,t}\), where \(ATM_{i,k,t}\),  \(POS_{i,k,t}\) and \(AP_{i,k,t}\) are the numbers of ATMs, point-of-sale terminals, and access points of bank i in municipality k at time t, respectively; \(p_{k,2010}\) and \(p_{2010}\) are municipality k’s and Mexico’s population in 2010. Thus, \(ATM_{i,t}\), \(POS_{i,t}\) and \(AP_{i,t}\) are the population weighted averages of ATMs, point-of-sale terminals, and access points of bank i in time t.

Collateral could be an additional important dimension. Nonetheless, payroll loans are credits for which the principal and interest payments are directly deducted from the bank account where borrowers receive their salary (in Mexico, payroll checks are directly deposited in these accounts by the employer). Thus, in this sense, future borrower’s income serves as collateral for this type of loans and, therefore, banks generally do not demand additional collateral to grant them. Indeed, in Mexico more than 99 percent of the total amount of payroll loans are secured neither by cash nor by any other collateral.

The relative position of each bank as shown in Table 1A of Section A of the Appendix is based on publicly available quantitative information. These relative positions do not reflect an opinion on the banks’ services.

Besides demand deposits, there is another category of deposits, i.e., called time deposits. These deposits represent an important source of funding for banks in the US, but they are less important in Mexico: they represented, on average, 60 percent and 20 percent of total liabilities over 2008-2017 in the former and latter countries, respectively. We do not include time deposits in the study and choose only demand deposits as the type of funds whose costs more strongly depend on a retail approach to vertical differentiation. This is because the information on time deposits provided by the CNVB is somehow linked to money markets and debt instruments, whose costs cannot be interpreted as being dependent on that approach, given that many of these deposits belong to financial institutions instead of individuals.

Based on the evidence that the IVD captures the business strategies of commercial banks, the empirical exercise in Sect. 6 is re-performed as a robustness check, but this time considering solely commercial banks (i.e., investment banks are excluded). Tables 12A-14A of the Appendix show that the results of Sect. 6, which focuses on both commercial and investment banks, are robust to excluding investment banks.

The index not only captures the business strategies relevant to commercial banks and, thus, the type of vertical differentiation relevant in financial systems that rely heavily on retail funding, but it also shows substantial variation over time (see Table 5A in Section A of the Appendix). Those temporal variations make it possible to estimate more precisely the link between vertical differentiation, market power, and risk-taking—a task we undertake in Sect. 6.

The linearity of   P(yj), makes it possible to derive a closed-form solution.

They derive this demand function from a standard consumer maximizing-utility framework, like the one used by Dixit and Stiglitz (1977); Singh and Vives (1984); and Hackner (2000).

On the basis of the same best response function, Fig. 2B in Section B of the Appendix studies the impact of different target returns by considering different values, \(y_1\) and \(y'_1\), where \(y_1<y'_1\). Intuitively, an increase in the target return raises the profits associated with the transformation of deposits into investments, shifting the best response function of Bank 1.

The relative indices are calculated in two different manners. First, they are calculated for each bank and each period as the difference between the IVD and the maximum IVD; that is, we measure the preference for each bank relative to the preference for the most preferred bank. Second, relative indices are computed for each bank and each period as the difference between the IVD and the weighted average of the IVDs of the other banks; here, the weights are given by a bank’s share of the total amount of demand deposits in the previous period; here, we measure the preference for a bank relative to the mean preference for the other banks. Tables 6A-11A of Section A of the Appendix show that the results of Sect. 6 are robust to using either one of these two relative indices of vertical differentiation rather than absolute indices.

This score is constructed as Z_it = (ROA_it + EA_it) / σ(ROA)_it, where ROA and σ(ROA) are the rate of return on assets and an estimate of the standard deviation of this rate, respectively; and EA is the equity-to-assets ratio.

The table reports standard deviations for the entire sample in parenthesis and mean standard deviations across banks in brackets. The last ones are calculated as follows: for each bank, we compute the standard deviation of the corresponding variable and, then, we take the simple average across all banks.

To quantify the effect, we use the mean standard deviation of the IVD across banks rather than its standard deviation over the whole sample (i.e., the number in brackets in Table 2 rather than the number in parenthesis). The rationale for that decision is that because the standard deviation over the whole sample is very large it reflects changes in the IVD that are unlikely to take place in the short term for a particular bank. In contrast, the mean standard deviation across banks is based on observed changes within banks and, hence, represents more realistic adjustments to the IVD. These adjustments are more likely to occur in a particular bank in the short term (for the construction of the mean standard deviation of the IVD across banks, see the notes to Table 2).