New Credit Drivers: Results from a Small Open Economy

ABSTRACT

In developed economies, macroeconomic indicators such as unemployment and price indices tend to predict new credit expansion. We explore whether business and consumer surveys complement traditional macroeconomic variables in predicting new household and corporate loans in the following 3, 6, 9 and 12 months. Using monthly data for Slovakia, starting in 2009 and ending in 2019, we use Bayesian model averaging to examine 102 potential credit drivers. Our results show that, with the exception of interest rates and unemployment, traditional macroeconomic variables do not seem to drive credit market development. Instead, survey-based perceptions, calendar effects and policy uncertainty show relevant predictive power.

KEYWORDS:

• Credit growth
• sentiment
• early warning systems
• Bayesian model averaging
• new loans
• macroprudential policy

JEL Classification:

• C40
• G01
• E44
• E61
• G21

Funding

Košťálová, Horvátová and Lyócsa acknowledge that this work was supported by the Agentúra na podporu výskumu a vývoja (Slovak Research and Development Agency) under the Contract no. APVV-18-0335. Gernát and Horvátová acknowledge the support from Vedecká Grantová Agentúra MŠVVaŠSR a SAV with a project no. 1/0607/21.

Disclosure Statement

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

Notes

1. Nevertheless, such a study does not exclude foreign variables from entering predictive models.

2. For example, in Hungary, loans were backed by a foreign currency, whereas this practice is nearly nonexistent in Slovakia. Both Hungary and the Czech Republic have their own currencies, while Slovakia has already adopted the Euro. Therefore, even for such geographically and historically related countries, conditions differ significantly, suggesting that a single-country study approach is preferable to a panel framework.

3. Moreover, even though business and consumer surveys are designed to measure the same constructs, given language differences, they are country specific.

4. For a recent use at the regulatory level, see Kupkovic and Suster et al. (2020), who use credit standards as one of the building blocks in estimating the financial cycle of Slovakia.

5. See Figure 1 in Section 4.1 for the dynamics of the time series of new loans or Figure A1 for the joint dynamics with selected sentiment indicators.

6. This fact is related to the problem of nowcasting, which acknowledges that in reality, many historical data for a given time period are actually observed with a certain lag. We do not address this issue in this paper.

7. To facilitate the temporal disaggregation via the Silva and Cardoso (2001) method, for i) GDP growth rate, we use monthly data from the industrial production index, ii) house price index, we use interest rates, housing Harmonized Index of Consumer Prices (HICP) and Eurozone shadow rate, iii) net acquisition of financial assets, we use the Slovakia Stock Market (SAX) index, debt securities held – General Government and average overall interest rate of existing loans, iv) debt instruments, we use loans (monetary financial institutions (MFI)), equity capital and reserves and a dummy for the electoral cycle, v) credit standards on housing loans, we use a Eurozone crisis dummy, debt securities held (MFI), consumer confidence index, vi) credit standards on consumer and other loans, we also use a Eurozone crisis dummy, debt securities held (MFI), and consumer confidence index, and finally for vii) credit standards on corporate loans, we use the public sector purchase program and consumer confidence index.

8. 1 was set for the period from January 2010 until December 2015 and 0 otherwise. We chose this period as a compromise between several sources, Lo Duca et al. (2017), Laeven and Valencia (2013), Stracca (2015).

9. Captures the electoral cycle. It returns 1 for 12 months prior to the scheduled election or 1 if early elections are announced; otherwise, 0.

10. It returns 1 for a period when Lex Beblavy was passed in the parliament until it came to have legal force; otherwise 0. The second dummy variable returns 1 for 6 months after Lex Beblavy came into effect; otherwise, 0.

11. There were two cases where the model has not converged well – for 3- and 6-month-ahead new corporate loans. For these two cases, the hyperparameter was set to the number of observations.

12. Detailed BMA model results for all variables are available upon request.

13. Note that we use differences in the Euro area shadow rate, while % changes are used for other interest rate variables that have only positive values. This explains the differences in the size of the estimated PMs between interest rates on new household loans and on the Euro area shadow rate.

14. Note that in the case of the order book variable, we use a trend filtered series, as suggested by the KPSS test.

15. Data retrieved on 26 July 2021 from <http://www.bsse.sk/>

16. Note that when 6-, 9- and 12-month-ahead new household loans are of interest, the two variables can have an overlapping effect, which is likely the reason behind the fact that only the passing of the legislation is relevant for the model predicting new household loans in the 12-month-ahead period.

17. We included proxies for house prices in two forms, house price index directly and harmonized consumer price index.

18. Closer investigation of such exogenous factors is reserved for future research.

19. See https://www.nbs.sk/en/statistics/news (retrieved June 2021) for detailed statistics.

20. These research paths will be the subject of our future research.

Additional information

Notes on contributors

Zuzana Košťálová

Zuzana Košťálová is a junior researcher at the Institute of Economic Research of the Slovak Academy of Sciences. Her research interests are centered around big-data analysis, machine learning and financial stability. In her current research projects, she has two research paths. The first one uses machine learning methods to estimate financial cycles and financial distress. The second one uses machine learning methods to model various aspects of labor market.

Eva Horvátová

Eva Horvatova works as a professor at the University of Economics in Bratislava (Slovakia) and at the Masaryk University in Brno (Czech Republic). Her research focus is centered around banking, finance, financial regulation and behavioral finance. The actual research interests are mainly in the field of banking in conditions of the European Union and signaling of possible imbalances, the counter-cyclicality of regulation, the search for suitable indicators, financial development, and the motivation of investors in capital markets. She has published monographs oriented on banking, real estate market and papers in peer-reviewed journals.

Štefan Lyócsa

Štefan Lyócsa works in academia as a professor at the Masaryk University (Czech Republic) and University of Prešov (Slovakia). His research interests are centered around: modeling financial market risks, machine learning in credit-risk modeling and network analysis. He has published in over 50 peer-reviewed journals, such as: International Journal of Forecasting, Journal of Economic Dynamics and Control, Quantitative Finance, The European Journal of Finance, etc. He is a fellow of the Slovak Economic Association.

Peter Gernát

Peter Gernát is a PhD student at the University of Economics in Bratislava (Slovakia). His research interest covers financial cycles, consumer data analysis for academic and industrial purposes.