Measuring Heterogeneity in Banks’ Interest Rate Setting in Russia

ABSTRACT

We use credit registry data on all corporate loans issued by Russian banks since 2017 to decompose bank interest spreads into a common factor, borrower- and lender-specific components. We find that the variation in loan rates associated with lender-specific factors (heterogeneity of banks) and borrower-specific factors (heterogeneity of borrowers) is substantial. We use the bank-specific components identified to measure the fragmentation of the corporate credit market in Russia. The results indicate that heterogeneity in banks’ interest rate setting is high and increased in the early stage of the pandemic. Finally, our results suggest that banks tightened non-interest loan conditions during the pandemic.

KEYWORDS:

• Bank interest spread
• credit registry
• financial stability
• credit market fragmentation
• pandemic

JEL CLASSIFICATION:

• E44
• E51
• E52
• E58
• G21
• G28

Acknowledgments

The authors thank José-Luis Peydró and the other participants of the NES-Bank of Russia research workshop for their helpful comments on and suggestions for an earlier version of this paper. The authors also thank Alexey Egorov (Bank of Russia) and all the participants of the internal research seminar for their comments on and suggestions for the later version of the paper.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Disclaimer

The views expressed in this paper are solely those of the authors and do not necessarily reflect the official position of the Bank of Russia. The Bank of Russia is not responsible for the content of this report.

Notes

1. Here and after, we consider newly issued loans, not the stock of loans (banks’ corporate loan portfolios).

2. Corporate credit interest rate spread is defined as the difference between the interest rate on a corporate loan and a particular benchmark. The credit spread in Figure 1 is calculated as the spread between the loan rates relative to the benchmark rate. As the benchmark rate, we use the average of the interest rates charged by the benchmark lender on loans issued to the benchmark borrower in each quarter from 2017 Q1 to 2021 Q2. This rate follows the dynamics of the money market rates common to all banks.

3. Anderson and Cesa-Bianchi (2020, p.11) note that ‘ … heterogeneity is multi-dimensional and that there are potentially other relevant empirical proxies for financial constraints – such as age, size, liquid assets, etc., which are frequently considered in the literature.’

4. Changes in the composition of borrowers, lenders, or loan terms for new loans may be an important factor in changes in the average interest rate spread. First, the level of the average corporate credit spread in the economy may change over time in dependence on macroeconomic factors common to all banks. The spread may also change due to changes in lending conditions (loan structure by maturity, loans to affiliated entities, or loans with collateral attached) or industry-specific and borrower-specific characteristics. Finally, some banks may decide to price loans to a certain borrower at lower spreads than those of other banks – this will then be a bank-specific factor of interest rate spread heterogeneity responsible for changes in the average interest rate spread. As a result, high heterogeneity reduces the information content of the mean/median interest rate.

5. See the work of Jiménez et al. (2014). There are several reasons why a bank may decide to charge a lower interest rate spread and take on relatively more credit risk as a result: a bank’s competitive strategy may induce it to price loans cheaper (Ross 2010), there may be an information asymmetry among banks (including relationship lending), or larger/lower income may stimulate a bank to take more risk (moral hazard).

6. When we discuss the main results we obtained with our quarterly sample, we consider 2020 Q2 as the quarter when the COVID-19 pandemic started and began to spread in Russia. For our monthly sample (in the robustness check), we consider March 2020 as the month the pandemic began.

7. As a robustness check, we also repeat the exercise on data with monthly frequency.

8. In the author’s notation, ‘countries’ correspond to the ‘banks’ in our study: ‘The coefficient associated with the country dummy represents our main measure of financial fragmentation.’

9. Horny, Manganelli, and Mojon (2018) define market fragmentation in a similar manner in their study of corporate bond pricing in the Europe. We follow their example here but apply the definition to the banking sector.

10. Note that in our analysis we employ a model with time-varying parameters (Eq.1)(Eq.1) $Sprea{d}_{i,t}={\beta }_{1,t}\cdot {\delta }_{quarter,t}+{\beta }_{2,t}\cdot {\delta }_{borrower,\hspace{0.17em}t}+{\beta }_{3,t}\cdot {\delta }_{lender,\hspace{0.17em}t}+{\beta }_{4,t}\cdot {\delta }_{loan,\hspace{0.17em}t}+\hspace{0.17em}{e}_{i,t}$(Eq.1) that may be regarded as a set of regressions estimated separately for each quarter. This means that, by design, the results for 2017–2021 will not be altered by adding more observations to the dataset (although, admittedly, it would provide more material for comparison). This also means that the presence of the relatively numerous observations that fall within the COVID period cannot affect the outcome for the tranquil periods.

11. Referred to hereinafter as the ‘credit registry’ (Form 0409303). The methodology and a detailed description of the form can be found at

http://www.cbr.ru/eng/statistics/pdko/sors/summary_methodology/#highlight=0409303.

12. For example, the European Central Bank’s AnaCredit:

https://www.ecb.europa.eu/explainers/tell-me-more/html/anacredit.en.html.

13. The data in the credit registry do not contain any indication of whether a given company is a subsidiary or affiliated with a larger business group. Group-based identification would demand much more data to control for intragroup variation in loan demand.

14. When we use a quarterly frequency for identifying firms with relationships with multiple banks, with maturities of less than 90 days, it might be the case that such relationships are a consequence of loans issued by one bank and repaid and a subsequent loan issued after that by another bank(or banks). For the quarterly frequency, considering a maturity of 90 days allows us to exclude such instances of incorrect identification of multiple-bank relationships. In the robustness check section, to maintain consistency with the monthly frequency of identifying relationships with multiple banks, we exclude loans with maturities of less than 30 days.

15. We also test seasonality in measures of market fragmentation – see Robustness.

16. See Jiménez et al. (2014).

17. Adverse selection in lending for lower values of interest spread is thought to result in the attraction of less risky borrowers applying for the loan. Thus, a lower spread may not imply larger risk taking. This argument does not apply to our identification scheme, as our method deals with the same borrower’s obtaining two loans with different interest rates. Another argument that challenges the idea that a lower spread corresponds to taking on more risk is that a lower spread makes it easier for the borrower to pay off the debt, while a larger spread increases the debt service burden on the borrower, which raises the probability of default.

18. This list does not include many other loan terms, such as conditionality or options attached. We use only those terms that 1) are included on the bank reporting form and 2) have a small number of missing values (as the reporting of some loan terms is not obligatory for reporting banks). We also do not control for relationship lending, as the credit registry data begins only in 2017, which is rather a short time period to unbiasedly identify the effect of relationship lending. For example, Banerjee, Gambacorta, and Sette (2021) track a bank-firm relationship beginning in 1998, 6 years before the credit registry time-series for Italy they use.

19. Cases when borrowers are not uniformly affected by external factors are left for future research, where we plan to introduce an industry-specific dummy.

20. These are the usual groupings for studies of the Russian banking sector (Simanovskiy et al. 2018). The ‘top 30’ banks constitute the 30 largest banks by assets as of March 31, 2021 (Bank of Russia 2021, 2).

21. See the work of Altig et al. (2020), Aramonte and Avalos (2020), and Banerjee et al. (2020).

22. Indeed, the Basel Committee on Banking Supervision recommends that longer-term loans be treated as riskier, as in the formula for maturity adjustment at the BIS (2019).