A New Measure of the Financial Cycle: Application to the Czech Republic

Abstract

The recent financial crisis has demonstrated the importance of the linkages between the financial sector and the real economy. This article proposes a suitable and easy-to-apply method for assessing the position of the economy in the financial cycle in order to identify emerging imbalances in a timely manner. The method utilizes a composite indicator, constructed by the authors, that combines variables representing risk perceptions in the financial sector and their reinforcing interactions over the financial cycle. The indicator is calibrated to capture the future credit losses of the Czech banking sector. This method can be used by policymakers for a wide range of policy decisions, including the setting of a countercyclical capital buffer.

Keywords:

• Bayesian model averaging
• countercyclical capital buffer
• credit risk
• financial cycle

JEL Classification:

• C11
• E32
• E37
• E58

ACKNOWLEDGMENTS

The authors would like to thank Mathias Drehmann, Jakub Matějů, Štefan Rychtárik, and Roman Horváth for their helpful comments. All remaining errors are our own.

FUNDING

This work was supported by Czech National Bank Research Project No. C4/13 and by Czech Science Foundation Grant No. P402/12/G097.

Notes

1. Credit-to-GDP measures can deliver highly misleading signals in some countries, as shown in Geršl and Seidler (2011). This is particularly true when the credit aggregate is polluted by structural breaks, as is the case in the Czech economy.

2. Forward-looking indicators also differ across studies, depending on the set of countries examined. For emerging economies, foreign exchange reserves and the equilibrium real exchange rate are often appropriate indicators (e.g., Frankel and Rose 1996).

3. In contrast to common factor models, idiosyncratic developments should not automatically be seen as noise, because they may still contain valuable information from the macroprudential perspective.

4. We would argue that bundling together risk accumulation and risk materialization indicators lacks theoretical interpretation and cannot result in any meaningful measure.

5. The choice of variables would be rather different for other countries where the data quality is higher and longer time series for relevant indicators are available. Some very useful variables (e.g., the debt service ratio) could not be used in the present case, since the time series necessary for calculating them start only in 2004.

6. This ranking is used later in simulating the weights of the input variables in the resulting composite indicator.

7. On the other hand, figures for new credit can be tainted by the phenomenon of loan refinancing—the situation where one bank takes over a client’s debt from another bank to enlarge its credit portfolio (refinancing becomes part of the “new credit aggregate” despite the fact that no additional credit was extended to the sector in economic terms). Since the available data do not allow the series to be adjusted for this phenomenon, one should take it into account when interpreting the final outcome.

8. The original CISS applies a rather simpler transformation using the empirical cumulative distribution function. Some of the variables (spreads, current account deficit/GDP) had to be multiplied by a coefficient of −1 before the transformation, so that low financial risk aversion corresponded to higher values for all the variables.

9. We are not the first to propose a measure of the financial cycle based on the comparison of indicators with their own history and subsequent aggregation; see Rychtárik (2014). However, the mapping and the aggregation algorithm (as well as our conceptual approach to the variable selection) are different in the present case.

10. Current experience seems to support this claim, as ex post revisions of the past values of the FCI were negligible when data for 2013Q4–2014Q4 were added to the sample (not reported).

11. Definitions of the time and cross-sectional dimension of risk can be found in Frait and Komárková (2012). In this article, however, the cross-sectional dimension of risk is defined rather differently. The original concept defines the cross-sectional dimension as the degree of financial interconnection between economic agents, which can generate financial risks, whereas here the cross-sectional dimension is taken to mean the degree of interconnection between the various aspects of financial risk, which can amplify the overall level of financial risk.

12. This is not necessarily a very precise approximation of the initial values, but we are mainly interested in the relative value of the correlations vis-à-vis their value in other periods, rather than their absolute value.

13. As an alternative, the weights were determined with regard to the predictive power of the FCI for the 12-month default rate in the nonfinancial corporations sector and for the first difference of the ratio of nonperforming loans to total loans in the private sector. The results were similar.

14. The weights are rounded and their order corresponds to that in Table 1. Rounded values were used to calculate the FCI.

15. Hlaváček and Komárek (2011) point to some overvaluation on the property market in 2007 and 2008. This reflected, among other things, a pre-announced increase in VAT on residential property construction.

16. In recent years, moreover, the FCI values have been further overestimated due to the phenomenon of mortgage refinancing, which is inflating the total amount of new loans to households. It is not yet possible to fully filter out this effect based on the available statistics.

17. As a rule of thumb, Drehmann and Tsatsaronis (2014) propose the use of credit gaps only when at least ten years of data are already available for the credit-to-GDP ratio. Another option is to drop the initial data points.

18. Note that the weights w in the FCI were already chosen optimally with respect to the predictions for loan losses six quarters ahead. As loan losses and NPL growth are clearly correlated, there is a sort of circularity in this exercise. However, optimal values (leading to the minimum RMSE) do not necessarily imply good predictions. Thus, they are still considered to show how good the obtained optimum is.

19. Splitting the sample into calibration and verification subsamples, however, would hardly have been possible in the given conditions due to the short time series available. To obtain reasonable estimates of the cycle, a span covering expansion and contraction phases is needed. Moreover, this article does not use the vintage data which were available to policymakers at time t.

20. Unity minus the posterior variance over the dependent variable.

Additional information

Funding

This work was supported by Czech National Bank Research Project No. C4/13 and by Czech Science Foundation Grant No. P402/12/G097.

Notes on contributors

Miroslav Plašil

Miroslav Plašil is senior economist at the Financial Stability Department, Czech National Bank, Prague, Czech Republic.

Jakub Seidler

Jakub Seidler is chief economist at the ING Bank NV, Prague, Czech Republic and was at the Financial Stability Department, Czech National Bank, Prague, Czech Republic at the time this paper was written.

Petr Hlaváč

Petr Hlaváč is not currently affiliated with an institution, but was senior economist at the Financial Stability Department, Czech National Bank, Prague, Czech Republic at the time this paper was written.The views expressed in this article are those of the authors and not necessarily those of the Czech National Bank.