A recent VoxEU column by Boot et al. (2020) contributed to the debate on changes to financial intermediation brought about by new technology or FinTech, discussing how online platforms can, in general, lower entry barriers in banking. While Boot et al. (2020) focused mostly on non-banks, in this column, based on our recent CEPR discussion paper (Basten and Ongena 2020), we show more specifically how online platforms can also allow smaller banks to expand to areas beyond their branch network, with the magnitude of expansion similar to that of non-banks.
Our analysis is based on an empirical investigation of bank responses to household mortgage applications through a Swiss web platform. It yields three important findings:
First, we find that through the online platform, banks make more attractive offers to local markets that are more concentrated, i.e. less competitive. We surmise that this surprising finding is caused by banks’ desire to enter particular markets with higher follow-up earnings potential, as product bundling and switching costs allow banks to benefit repeatedly from each new client they win through an online mortgage. Clients in more concentrated, i.e. typically more rural, local markets benefit the most.
Second, lenders also make more attractive offers to areas where unemployment as a driver of probabilities of default, or house price changes as a driver of loss given default, are less positively correlated with those at home. This adds to the findings portrayed in the 2016 VoxEU column by Götz et al. (2016) whereby geographic expansion, in their case across national borders, can materially reduce the riskiness of bank holding companies. The advantage of improving geographical diversification by lending through online platforms over seeking it through bank holding companies, or for that matter asset securitisation, is the avoidance of now well-known agency problems. Geographical diversification through this channel can benefit not only the risk management of the lenders studied but arguably also the financial stability more widely.
Third, lending through online platforms can allow banks to increasingly automate decision-making and thereby lower their operational costs.
Data from online platform
Our research analyses empirically the responses each household received from different mortgage lenders through the Swiss online platform Comparis between 2010 and 2013. To apply for a mortgage, households had to supply information on their finances including their income, wealth, and age; and on the property they sought to finance including its age, location, size, and purchasing price. In addition, they were asked what amount they sought to borrow, and for what period between 3 months and 10 years they sought to fix the interest rate. After this hard information had been forwarded to all possible lenders, lenders responded with an offer or a rejection, also providing an offered rate for the former.
Beyond an indicator of whether or not a provider made an offer, we analyse the provider’s eagerness to lend as measured by the spread between the rate offered and the CHF interest swap rate applicable on that day for that rate fixation period. This spread is already cleaned for any interest rate risk and hence isolates the sum of assessed credit risk and the margin desired by the lender. We then relate offer propensity and pricing to borrower characteristics including the loan-to-value (LTV) and loan-to-income (LTI) ratio as well as lender characteristics including total assets, mortgage specialization, and capitalisation. More importantly, we focus on the effects of prior market concentration in the applicant’s canton (state) and on how well a loan to the applicant would complement the lender’s pre-existing portfolio.
Exploiting exogenous shifts in prior local market concentration
We measure the pre-existing concentration of each of 26 cantonal (state) mortgage markets with the Herfindahl-Hirschmann Index (HHI). But the eagerness to lend to a local market may be also be spuriously correlated with local market concentration if e.g. unobservable factors which influence the attractiveness of local lending have shaped both prior market entry and current offer attractiveness. To obtain the causal effects of current market concentration, we therefore focus on recent changes therein triggered by the need of the Swiss ‘Big Two’ universal banks UBS and Credit Suisse to restrict domestic mortgage lending following losses in the US subprime crisis. The higher a canton’s prior Big Two market share, the greater was the negative impact of growth reductions on market concentration. Exploiting prior variation in exposure to exogenous supply shifts has been previously used by Mian and Sufi (2012) and others, but is particularly clean in our setup where the losses occurred overseas and are very plausibly exogenous to later online bids of small Swiss banks with no noteworthy US exposure.
Observing responses from different lenders for each potential borrower
When estimating lenders’ responses to how well mortgage borrowers from different regions would complement their existing portfolios, we can benefit from the fact that we observe not only how each lender responds to applications from different regions, but also how each applicant receives responses from lenders in different regions. This allows us to use both borrower and lender fixed effects, and thereby control for both observable and unobservable borrower and lender characteristics. This type of identification has been used repeatedly to analyse lending to large firms, as in Khwaja and Mian (2008) and here with applications as in Jimenez et al. (2012). But it has been used for analysing lending to households only by Basten (2020) given that it is not common to observe interactions with multiple possible lenders for households.
Measuring the automation of banking decisions
After having estimated rules of how offer propensity and pricing depend on specific characteristics of borrower, lender, and their interaction, we can also investigate how closely lenders follow these rules and in which cases they diverge from them. To do so, we follow the framework on multiplicative heteroscedasticity first suggested by Harvey (1976) and applied to bank lending by Cerqueiro et al. (2011): We first estimate the mean equation or rule and obtain the squared residuals. Then we estimate the variance equation by regressing the log of the squared residual of each observation on characteristics of interest. Consistent with prior literature, we find more discretion or diversion from the estimated rules for riskier applications, which lenders may want to escalate to manual or even senior manual inspection. We also find it among smaller or less mortgage-specialized lenders with less experience on which to base automated decision-making. More interestingly, in the context of the new web platform, we find lenders reduce the amount of discretion and hence increase automation with an increase in responses that they have already submitted through the platform. This suggests that they can increasingly reduce their operational costs.
Our work shows the enormous potential for web platforms to shake up local lending competition, open up new ways for geographical diversification, and facilitate automation of lending decisions.
The net effects on the quality of bank risk management from increasing geographical diversification and automating decision-making are likely positive in the studied context of fairly standardised lending with reliable hard information, including reliable hedonic evaluations of the collateral value. But it is important to emphasise that they may well turn negative in contexts where soft information is more important.
Basten, C and S Ongena (2020), “The Geography of Mortgage Lending in Times of FinTech”, CEPR Discussion Paper 14918 and SFI Research Paper 19-39.
Basten, C (2020), “Higher Bank Capital Requirements and Mortgage Pricing: Evidence from the Counter-Cyclical Capital Buffer”, Review of Finance, 24(2), 453-495.
Boot, A, P Hoffmann, L Laeven and L Ratnovski (2020), “Financial Intermediation and Technology: What's Old, What's New?”, CEPR Discussion Paper 15004 and ECB Discussion Paper 2438.
Cerqueiro, G, H Degryse and S Ongena (2011), “Rules versus discretion in loan rate setting”, Journal of Financial Intermediation 20: 503-529.
Goetz, M R, L Laeven and L Levine (2016), “Does the geographic expansion of banks reduce risk?”, Journal of Financial Economics 120(2): 346-362.
Harvey, C (1976), “Estimating Regression Models with Multiplicative Heteroscedasticity”, Econometrica 44(3): 461-465.
Khwaja, A I and A Mian (2008), “Tracing the impact of bank liquidity shocks: Evidence from an emerging market”, American Economic Review 98: 1413-1442.
Mian, A and A Sufi (2012), “The Effects of Fiscal Stimulus: Evidence from the 2009 Cash for Clunkers Program”, The Quarterly Journal of Economics.
Jiménez G, S Ongena, J L Peydró and J Saurina (2012), “Credit Supply and Monetary Policy: Identifying the Bank Balance-sheet Channel with Loan Applications”, American Economic Review 102(5): 2301-26.