In both research and policymaking, there is a pressing need for a comprehensive international database that identifies the characteristics of financial systems that are vulnerable to crises. The analysis of such a database would enhance our understanding of the principal mechanisms through which crises are initiated and propagated. Even when financial fragility is not severe enough to precipitate a crisis, it is part of the explanation for disruptions to the finance-growth relationship (Loayza and Ranciere 2006).1 Fragile banks are reluctant to make new loans (Andrianova et al. 2015b). Instead, they focus their efforts on deleveraging their balance sheets and strengthening their liquidity buffers in order to cope with deteriorating depositor confidence.
In order to address this need, a new database on financial fragility for 124 countries over 1998 to 2012 has been constructed as part of an ongoing DFID-ESRC project.2 This database – the International Database on Financial Fragility (IDFF) – can be downloaded from the project website at the University of Leicester, together with a discussion paper that describes its construction, advantages, and limitations in some detail (Andrianova et al. 2015a).3 Here, we introduce and briefly summarise the construction of the database and highlight some of its main characteristics.
Novelties of the new database on financial fragility
The IDFF utilises bank-level data from Bureau van Dijk’s Bankscope to construct eight different measures of financial fragility, each focusing on a different aspect of vulnerability in the financial system. Seven of these relate to bank-level measures of vulnerability, including capitalisation, asset quality, managerial efficiency, earnings, liquidity, and risk exposure. The eighth measure is the Z-score measuring the distance of the entire banking system from insolvency.
Various types of deposit-taking institutions are included in the national aggregates, and commercial banks account for only two-thirds of the total asset value of banks in the IDFF – a feature which distinguishes it from previous databases that have focused exclusively on commercial banks (e.g. Beck et al. 2000, Cihak et al. 2013). Other institutions incorporated into the IDFF aggregates include cooperative banks (11% of total assets), savings banks (9%), investment banks (9%), real estate and mortgage banks (5%), and Islamic banks (0.1%). In many countries, commercial banks do not completely dominate the banking system, and ignoring other types of banks can lead to under-measurement or over-measurement of financial fragility at the national level. One example is the Netherlands, where commercial banks account for 80% of total banking assets, while cooperative banks account for 18%. The aggregate Z-score for Dutch commercial banks is 9.2, but this figure jumps to 13.4 when other types of bank are included, which probably reflects greater risk aversion among cooperative banks. Similarly, in Germany, where savings banks and cooperative banks together account for over 40% of assets, the inclusion of these other types of bank raises the Z-score from 14.1 to 29.8. On the other hand, in the US where commercial banks account for 58% of total assets, the addition of other types of banks results in a significant worsening of financial fragility indicators. US financial fragility will be under-estimated if one focuses on commercial banks alone.
- One novel characteristic of the IDFF is that it includes alternative estimates of individual fragility variables based on different rules for the inclusion of individual banks in the national aggregate.
Banks are not required to report any financial data to Bankscope, and the amount of information reported by some banks does vary over time. Including data from the most capricious banks may bias the resulting estimates of year-on-year changes of national aggregates, because the propensity to report data may be correlated with bank performance. On the other hand, including only those banks that always report a wide range of data may bias the resulting estimates of long-run national averages, because certain types of banks may report data more consistently than others. For this reason, the IDFF reports alternative aggregate measures based on a variety of selection rules that are more or less restrictive in terms of the consistency of bank reporting.
- Under the most restrictive rules, a bank must report each one of a range of financial indicators for at least two-thirds of the time it is in operation (or alternatively, for at least eight years) in order for its figures to contribute towards any aggregate.
- Under somewhat less restrictive rules, a bank must report a particular indicator for at least two-thirds of the time it is in operation (or alternatively, for at least eight years) in order for its figures for that indicator to contribute towards the aggregate.
- Under the least restrictive rule, all available figures are used in constructing the aggregates.
Table 1 illustrates how the different selection rules affect one of the aggregates – bank capitalisation. It is recommended that researchers using the database choose the measure that is best suited to the particular question they are addressing, or compare findings using different measures in order to judge the robustness of their results.
Table 1. Total annual national bank equity as a percentage of bank assets using five different selection rules
Figures 1 to 3 illustrate some of the variation in asset quality, liquidity, and stability (Z-score) across country-income groups.4 The graphs suggest that low-income countries (LICs) have the most fragile banking systems, reflected in low Z-scores and poor asset quality (impaired loans as a fraction of total loans). By contrast, they have the most liquid banking systems. As most low-income countries are located in sub-Saharan Africa, geographical variation patters are similar. While the high liquidity ratios may appear somewhat puzzling, they could reflect banks’ unwillingness to lend to local borrowers, as suggested by Andrianova et al. (2015b) in their analysis of why African banks lend so little. It is hoped the IDFF will facilitate research that deepens our understanding of the factors driving these characteristics and help policymakers to mitigate the causes and consequences of financial fragility.
Figure 1. Asset quality
Figure 2. Liquidity
Figure 3. Stability (Z-Score)
More generally, it is intended that this dataset will be used both for academic research and to inform policymaking. Analysis of the dataset might shed new light on questions such as how financial fragility influences economic growth, whether countries that liberalise their financial systems too quickly become more vulnerable to financial fragility, and whether there are indicators of fragility that can be used for predicting financial crises. Moreover, it can be used to examine questions relating to the weak, if not entirely absent, finance-growth link in low-income countries.
Andrianova, S, B Baltagi, T Beck, P Demetriades, D Fielding, S G Hall, S Koch, J Rewilak and P L Rousseau (2015a) ‘A new international database on financial fragility.’ Working Paper 15/18, University of Leicester.
Andrianova, S, B H Baltagi, P O Demetriades and D Fielding (2015b) ‘Why do African banks lend so little?’ Oxford Bulletin of Economics and Statistics 77(3), 339-359.
Beck, T, A Demirgüç-Kunt and R Levine (2000) ‘A new database on financial development and structure’, World Bank Economic Review 14, 597-605.
Čihák, M, A Demirgüç-Kunt, E Feyen and R Levine (2013) ‘Financial development in 205 economies, 1960 to 2010,’ Journal of Financial Perspectives 1, 1-19.
Demetriades, P O & G A James (2011) "Finance and growth in Africa: The broken link," Economics Letters 113(3), 263-265.
Loayza, N V, and R Ranciere (2006), "Financial development, financial fragility, and growth,” Journal of Money, Credit and Banking: 1051-1076.
1 This feature is particularly evident in Africa (Demetriades and James 2011).
2 “Politics, Finance and Growth”, DFID-ESRC Award Reference ES/J009067/1.
4 The selection rules document the ‘base’ rule where the country level data is generated using all available observations.