How Well Do Aggregate Bank Ratios Identify Banking Problems?

1
How Well Do Aggregate Bank Ratios Identify
Banking Problems?

• by
• Martin Cihák and
• Klaus Schaeck
• Comments
• Kenneth D. Jones
• Federal Deposit Insurance Corporation
• Identifying and Resolving Financial
  CrisesFederal Reserve Bank of Cleveland andFDIC
  Center for Financial Research
• Cleveland, OH
• April 17, 2008

2
These deep thoughts are my own and not those of
the FDIC or the U.S. Government
3
The paper in perspective

• The authors use a proprietary data set containing
  financial soundness indicators (FSIs) to
  investigate whether aggregated banking data are
  useful in predicting the occurrence and timing of
  banking crises.

4
Literary Perspective

• The development of predictors to financial crises
  has been a particularly active area for research
  in the 1990s and 2000s.
• Bussierre and Fratzscher (2002) characterize two
  broad empirical approaches
• Leading Indicator Approach
• The Discrete-Dependent-Variable Approach

5
Literary Perspective

• Leading Indicator Approach
• Considers vulnerability indicators (from single
  variables or combinations of variables) and
  transforms them into binary signals if a given
  indicator crosses a critical threshold, or lies
  within a set interval, or is an outlier, then a
  red flag is raised or a signal is sent. The
  number and strength of the signals are predictors
  of an impending crisis.
• Discrete-Dependent-Variable Approach
• Emphasis is on identification of variables that
  are statistically correlated with financial and
  banking crises. A logit or probit regression is
  used to estimate the probability of a crisis
  occurring in the near term.
• The empirical performance of crisis predictors is
  rather poor.
• Rather high Type I and Type II errors
• Generally fail to perform when tested
  out-of-sample

6
Main Findings of Cihák and Schaeck

• Find evidence using a binomial logit regression
  model that suggests that at least some aggregate
  bank ratios are indicative of oncoming systemic
  banking crises.
• Indicators
• Return on equity
• M2 / international reserves
• Degree of corporate leverage
• Offer results of a duration model to show that
  some aggregate banking system ratios provide
  signals about the timing of systemic banking
  crises.
• Indicators
• Return on equity
• GDP per capita
• GDP Growth

7
General Comments

• Methodology
• The approach used falls somewhere between the
  discrete-dependent-variable and the leading
  indicator approach.
• Fairly good job of exploratory data analysis
• Doesnt go quite so far toward developing
  variables for an early warning system based on
  leading indicators
• Need to investigate threshold values, or trip
  wires, test combinations of indicators, and
  report noise-to-signal ratios

8
General Comments

• Contrary to the papers conclusion, I thought
  that the findings suggested the opposite that
  aggregate banking ratios were not useful
  indicators of future systemic banking crises (at
  least when used in isolation).
• The findings actually highlighted the
  difficulties in finding predictors that could
  presage banking crises across a wide
  cross-section of countries with an acceptable
  proportion of missed crises and false alarms.
• Good support for this conclusion exists in the
  literature
• Even the caveats discussed by the authors seem to
  support this conclusion
• Limitations of the data set
• Backward-looking nature of bank accounting data
• Aggregation problems

9
General Comments

• Duration Analysis
• The short time period studied and the arbitrary
  nature of the start date cast some doubt on the
  robustness of these results.
• add additional years (2005-06)
• Limit sample to those countries with similarly
  strong/weak economies at the beginning of the
  period
• With first differences, all bank variables become
  insignificant but GDP growth emerges as an
  indicator why?
• Other
• The insignificance of some variables that other
  studies have found to be important needs to be
  explained
• Credit and loan growth to GDP
• Short-term capital flows to GDP
• Measures of nonperforming loans

10
Some Suggestions

• Test the bank ratios with macroeconomic controls
  and controls for the regulatory and institutional
  environments of each country
• The authors suggest this as an area for further
  research (I concur)
• Kaminsky (2000) I have only considered the
  macroeconomic data in the list of univariate
  indicators, but data from the balance sheets of
  financial institutions would be an important
  complement to the macro data.
• The idiosyncratic nature of banking crises makes
  it difficult to find predictors
• One size does not fit all
• Possible solution Conduct analysis on split
  samples
• Different types of banking crises
• Depositor runs / Liquidity crises / Macroeconomic
  shocks
• Macroeconomic cycles
• Different types of economies and banking systems
• Developed economies vs. emerging economies

11
Some Suggestions

• Expand prediction horizon
• Allow two or three year lags to possibly increase
  predictive ability this also allows sufficient
  time for corrective actions by markets or
  regulators
• Control for the effect of currency or financial
  crises that precede a banking crisis
• Further explore the trade-offs between Type I
  (missed crisis) and Type II errors (false
  signals) by determining the optimal cut-off or
  critical level for important variables.
• For example, in figures 10 and 11, there are
  potential trade-offs that would reduce one type
  of error while increasing the other type only
  slightly. This could be explored over a range of
  policy weighting schemes (Bussiere and
  Fratzscher, 2002).

12
Where to from here?Perhaps someday the authors
will have something akin to the DHS Threat
Advisory System
13
Or perhaps

• the holy grail of crisis prediction is
  intrinsically unattainable (Sharma, 1999)
• In which case
• the intellectual capital of the economics
  profession could be more productively expended
  devising appropriate changes in the overall
  regime in which investors operate (such as
  measures that compel changes in financial
  strategies) rather than searching for the correct
  set of crisis predictors (Grabel, 2003).