Economic Systems 38 (2014) 289–308

Contents lists available at ScienceDirect

Economic Systems

journal homepage: www.elsevier.com/locate/ecosys

How to foresee banking crises? A survey of the

empirical literature

Karlo Kauko *

Bank of Finland, P.O. Box 160, 00101 Helsinki, Finland

A R T I C L E I N F O

Article history:

Received 5 February 2013

Received in revised form 21 January 2014

Accepted 22 January 2014

Available online 5 July 2014

JEL classification:

E44

G01

G17

N10

Keywords:

Banking crisis

Financial crisis

Early warning

Financial instability

A B S T R A C T

A survey of the empirical literature on early warning indicators of

banking crises is presented. Descriptive analyses have been

published for decades, but cross-national panel data analyses have

only been performed since the late 1990s. More recently, the

severity of the subprime-Lehman crisis has been compared across

countries. Most findings corroborate the view that during a typical

build-up phase, banks borrow internationally to finance domestic

lending, boosting the current account deficit and causing a real

estate bubble. Increasing debt and imbalances lead to a crisis. Both

developing and developed countries have experienced these kinds

of boom-bust cycles.

� 2014 Elsevier B.V. All rights reserved.

1. Background

‘‘This time is different’’, claims the title of the popular book by Reinhart and Rogoff (2009). Theironic tone implicitly suggests that most banking and financial crises are caused by a group of fairlysimilar factors. Excessive loan growth and asset price bubbles may have preceded crises in the past,but in a climate of general euphoria no one wants to believe that history could repeat itself. This paperpresents an overview of the empirical literature on crisis-predicting early warning signs or ‘‘crystalballs’’, as named by Schwaab et al. (2010).

* Tel.: +358 10 831 2519; fax: +358 10 831 2294; mobile: +358 50 3870337.

E-mail address: karlo.kauko@bof.fi.

http://dx.doi.org/10.1016/j.ecosys.2014.01.001

0939-3625/� 2014 Elsevier B.V. All rights reserved.

K. Kauko / Economic Systems 38 (2014) 289–308290

The topic is highly relevant in terms of policy. Crises typically cause huge output losses and socialcosts. Hoggarth et al. (2002) estimated that the average crisis causes a production loss of 20% of annualGDP. A recent stimulus to this area of research is the countercyclical capital buffer (CCB) proposal bythe Basle Committee (2010). Regulators will impose additional capital requirements on all banks ifalarming signs are observed in the macrofinancial environment, but this is impossible unless theyknow what kinds of signs are alarming. Drehmann et al. (2011) explicitly motivate their analysis bythe need to identify good triggers for countercyclical capital buffers. The CCB was recentlyimplemented in the EU’s capital requirements directive (2013/36/EU) and similar regulatory reformsare coming on stream in the rest of the world as well.

Unfortunately, the terminology usage has been somewhat ambiguous and the frequently usedterm ‘‘financial crisis’’ may refer to many kinds of jitters, not only banking crises. Many interestingcontributions analyse the early warning signs of currency collapses and sovereign debt crises,1 butthese topics are beyond the scope of this survey. Here the focus is solely on large-scale banking crises,even though it is sometimes difficult to distinguish between banking and currency crises because theycoincide in so-called twin crises. As pointed out by van den Berg et al. (2008), crisis predictors maydiffer depending on the kind of crisis to be predicted. We also exclude studies on bank-level earlywarning indicators (see e.g. Patro et al., 2013, or Arena, 2008). There are dozens of studies reviewedhere despite our focus on narrowly defined large-scale banking crises.

During the Bretton Woods era, banking crises were exceptional or even non-existent (Reinhart andRogoff, 2008b). Consequently, academic economists did not find the topic important, at least not aftermemories of the Great Depression had faded. Crises again started to occur more often in the 1980s.Keeley (1990) argued that for the United States, the weakening of financial stability was due to adecline in the franchise value of banks, which was caused by intensified competition in the newlyderegulated environment. Banks had less to lose and so they consciously began to take more risks. Itmay not be coincidental that liberalisations in the 1980s were followed by crises in the 1990s in manycountries (Kiander and Vartia, 2011). Simultaneous banking and currency crises occurred in manyparts of the world in a liberalised environment, notably in emerging economies (Kaminsky andReinhart, 1998).

In August 2013, there were 6020 hits for the search term ‘‘financial crisis’’ in the Econlit data base,counting only articles in academic journals; 4480 of these papers had been published after January2009. For the search term ‘‘banking crisis’’ the respective figures were 497 and 270. Hence, it seemsthat a huge part of the crisis literature, probably most of it, has been published in the aftermath of therecent (and still ongoing) global crisis.

Manias, panics and crashes were discussed by Kindleberger (1978) before much econometricevidence on banking crises had been published. Minsky (1977) wrote about speculative euphoria andbubbles. In the late 1990s, a more empirical approach was taken and increasing numbers ofeconometric analyses of crisis predictors are now being published. As will be seen in the followingsections, most of the systematic evidence is consistent with the boom-bust cycle narrative. Bankingcrises tend to be preceded by periods of excessive loan growth and surging asset prices. A large yetstable loan stock does not normally pose major problems, but if the loan stock grows excessively largerelative to past levels, problems are likely to emerge within a couple of years. Accelerated house priceinflation is often observed before a crisis, but perhaps signs of stock market bubbles have lesspredictive power. Because not enough deposits can be collected domestically, banks finance theirincreasing loan stocks during boom phases by borrowing from abroad. A nation’s net foreign debttypically grows when banks borrow from abroad and grant loans domestically, thus enabling thepublic to spend more on imported goods. These boom-bust cycles have occurred in both developingand developed countries, even though this pattern is more typical for developed countries. Manydeveloping country crises seem to be closely related to price instability and currency collapses.

The literature on banking crises is less closely related to mainstream macroeconomics than it mighthave been a few decades ago. At least before the global financial crisis, mainstream macroeconomistsdownplayed the role of money and banking in their models. In fact, financial intermediation was seenas so insignificant that completely abstracting from it in DSGE models seemed harmless. These models

1 See e.g. Berg and Pattillo (1999a,b), van den Berg et al. (2008) and Frankel and Rose (1996).

K. Kauko / Economic Systems 38 (2014) 289–308 291

have no such variables as monetary aggregates or volume of bank lending. More recently, manymacroeconomists have included financial frictions (Hwang, 2012; Zhang, 2009) or bubbles (Wang andWen, 2012) in DSGE models. So far this has not become the standard approach, though this maychange and possibly upgrade our understanding of crisis dynamics.

Niemira and Saaty (2004) present some interesting insights on the history of economic thought andthe recent financial crisis literature. The growing number of crisis-related contributions can be seen asa return to the origins. In the 19th and early 20th century, in Pre-Mitchellian times, the predominantview was that normal, benign economic development was occasionally interrupted by anomalousrecessions. Recessions were seen as exceptional, extreme events, and the task was to explain why suchevents happen. Benign times, on the other hand, were not believed to require an explanation. Thismode of thinking disappeared for many decades, but has now re-emerged. Again, economists aretrying to explain why crises happen, but not why some countries are not likely to be hit by a bankingcrisis next year.

It is difficult to find comprehensive surveys of this literature. One of the very few available paperswith a comparable approach is by Quagliariello (2008), but this article has no particular focus onfactors that might be of use in spotting the risk of crises well in advance. Demirguc-Kunt andDetragiache (2005) and Staikouras (2004) have also published brief surveys, but most of the literaturein this field has been published after these were written.

The rest of this article is organised as follows. Section 2 presents a typology of the banking crisisliterature. Section 3 discusses the methods used in empirical research on early warning signals.Sections 4–9 present findings on the predictive power of individual variables; readers who areinterested in one particular variable can just read the relevant section. Section 10 compares resultsobtained with data from developed and developing countries, and Section 11 concludes and discussesthe findings.

2. The three waves of literature

To simplify things, the relevant literature can be classified into three waves. In this context, ‘‘wave’’may be more appropriate than ‘‘generation’’, because newer approaches have complemented ratherthan replaced older ones, and it may be difficult to defend the view that each wave has been animprovement over the previous one. All the waves still exist and new publications appearcontinuously.

2.1. The first wave

The first wave is descriptive, even though it tries to identify some regularities in run-ups to crises.Most of these studies concentrate on specific historic events, making it difficult to distinguish themfrom pure economic history. Unlike most contributions in the two following waves, many descriptivearticles also discuss crisis management policies and recoveries.

Detailed qualitative descriptions of US crises of past decades have been presented for example byFriedman and Schwartz (1963). Kindleberger (1978) discussed a large number of crises over a verylong period of time. In the early 1990s, the Nordic countries were among the first in the developedworld to experience serious post-World War II banking crises, and a small wave of largely descriptivecrisis literature was published by authors from these countries. For instance, Herrala (1999) describesFinnish crises from 1865 to 1998, while Jonung et al. (1996) and Kiander and Vartia (2011) describeand compare the 1990s crises in Finland and Sweden but do not focus solely on the banking sector, andKokko (1999) compares the crises in Sweden and Asia.

Recent descriptive analyses have been presented for example in Connor et al. (2012) for Ireland andthe US, while Reinhart and Rogoff (2008a) compare the US subprime crisis and other fairly recent ‘‘big’’crises in advanced economies. Razin and Rosefielde (2011) describe and compare the Japanese crisis ofthe 1990s, the ‘‘Asian flu’’ in 1997 and the subprime-Lehman crisis.

A few studies focusing on one country go beyond verbal descriptions and apply econometricmethods to the history of financial instability. Gorton (1988) presents econometric evidence ondeterminants of banking panics in the US prior to World War One. Singh (2011) reports on an

K. Kauko / Economic Systems 38 (2014) 289–308292

econometric analysis of the drivers of broadly defined financial fragility in India. Guo et al. (2011)analyse contagion effects among the stock market, real estate market, credit default market, andenergy market before the subprime crisis. These econometrically oriented contributions fall betweenthe first and the second waves of literature.

2.2. The second wave

The second wave emerged when there were enough relatively recent banking crisis observations foreconometric analysis with panel data. In a typical second wave paper, the incidence of wide-scalebanking crises is explained by macroeconomic and financial variables. What kinds of values do certainobservable variables obtain before a typical banking crisis? The samples usually consist of panel data on alarge number of countries, though in some analyses the focus has been on a very limited number of majordeveloped economies, often over a very long period of time. The second wave includes several analyses oftwin crises, i.e. simultaneous banking and currency crises (see e.g. Kaminsky et al., 1998). In most of thesestudies crises are seen as a being dichotomous in nature. Either there is a crisis or not; there are nointermediate cases. This view is consistent with the bank run model of Diamond and Dybvig (1983) withtwo Nash equilibria; each depositor finds it rational to withdraw if and only if others do.

The breakthrough with the second wave was also greatly facilitated by the instrumentalcontribution of Caprio and Klingebiel (CK, 1996a,b), who studied various publications and interviewedexperts in order to collect what is probably the first database of banking crises in different parts of theworld. This dataset is a list of country names, years, some rather basic economic indicators, and casualobservations on policy measures. The simple econometric analysis presented in CK (1996b) is the onlyanalysis the authors presented, but making the list of crises publicly available proved seminal. It isdifficult to see how the second wave would have taken off if no one had undertaken to systematicallycollect data for econometric analyses. The early pioneering studies of the second wave, including thewidely cited Kaminsky and Reinhart (1999) and Demirguc-Kunt and Detragiache (1998), explicitlyrefer to CK as a source of data, even though they mention other sources of information as well.

Most second wave contributions use very broad samples of different countries and focus on relativelyrecent crises that have taken place since the 1980s. A growing minority of second wave contributions,including Bordo and Meissner (2012), Schularick and Taylor (2012) and Jorda et al. (2011), combine dataon old and new crises, with sample periods ranging from the late 19th to the 21st century. These studieswith long-time perspectives cover a relatively narrow set of advanced countries.

2.3. The third wave

The recent global financial turmoil since 2007 has led to the emergence of a third wave of crisisliterature, namely cross-country econometric analyses of the subprime-Lehman crisis. This literaturecompares the crisis impacts of different countries and tries to explain the differences. The main focusmay be on the impact on either the financial sector (see Kauko, 2012; Kamin and DeMarco, 2012;Aizenman and Pasricha, 2012) or the real economy (Berkmen et al., 2012; Artha et al., 2011) or anextensive set of indicators (Rose and Spiegel, 2011, 2012; Acosta-Gonzalez et al., 2012; Frankel andSaravelos, 2010). As far as we know, this third wave approach has not been applied for instance to theprevious global crisis in the 1930s or the 1907 panic, possibly because of data availability problems.

3. Some comments on methodology

3.1. Defining the crisis

At least in second wave contributions, the first task is to define the term ‘‘crisis’’. Demirguc-Kuntand Detragiache (1998) defined a full-fledged crisis as an episode characterised by at least one of thefollowing:

(1) T

he share of non-performing assets exceeds 10%. (2) T he cost of rescue operations is at least 2% of GDP.

K. Kauko / Economic Systems 38 (2014) 289–308 293

(3) P

2

num

roblems lead to large-scale nationalisation of banks.

(4) E xtensive bank runs occur.

Various other criteria have also been applied. von Hagen and Ho (2007) define banking crises asepisodes of high demand for central bank reserves, i.e. periods of banks’ liquidity problems, measuredby changes in the ratio of reserves to bank deposits and changes in the short-term interest rate. Hahmet al. (2011) identify periods when banks’ funding costs are abnormally high relative to either theirlong-term average or the local Treasury bill rate. Herrala (1999) identifies episodes where banks’aggregate profit is negative.

Research on early warning indicators also implicitly reflects the ability and willingness of thegovernment to stabilise the system. No sample of crises contains cases where the governmentmanaged to prevent the crisis, possibly at the last moment, even though the banking sector was fragileand the environment had deteriorated. For instance, the German Landesbanken might have been hitby a crisis if some kind of government support had not been in place.

Not all econometric analyses adopt this dichotomous approach to crises. Klomp and de Haan(2009) use a continuous explained variable in a panel data analysis. They apply factor analysis to adataset consisting of potential instability indicators, employ a one-factor model, and derive factorscores for use as values of financial distress. Continuous explained variables are a key characteristic ofthe third wave of literature (see e.g. Rose and Spiegel, 2011, 2012; Kauko, 2012; Kamin and DeMarco,2012).

Lag-length selection may also be highly important. The standard view emphasises the role ofgradually evolving imbalances as a fundamental cause of crisis. In most studies, however, the lagstructure is simply assumed, and even variables that may affect financial fragility only after a lengthylag may be lagged by only one or two years, or even not at all. Many differenced variables reachalarming levels during the fastest build-ups of imbalances, which normally culminate years before acrisis. Drehmann et al. (2011) recommend flexibility in forecast horizons; early warning signals revealthe build-up of imbalances but do not tell us when problems begin, and if an alarmingly high value isobserved at least once during a lengthy pre-crisis period, the crisis must be regarded as predicted.

3.2. Analytical techniques

Since the early stages of the second wave there have been two dominant methods for predicting thedichotomous crisis variable, namely binary regressions and the signals method.

The signals method assumes that macroeconomic variables can safely fluctuate within certainboundaries, but beyond a threshold level the variation constitutes a menace to financial stability. Themethod was introduced in this literature by Kaminsky and Reinhart (1999) and has been used by Borioand Lowe (2002), Borio and Drehmann (2009) and Drehmann et al. (2011), among others. Thethreshold value is chosen to minimise the noise-to-signal ratio: the ratio of false alarms to all possiblefalse signals divided by the ratio of correct alarms to all possible correct signals.2 For instance, onemight conclude that the noise-to-signal ratio would have been minimised during the sample period ifa crisis had been predicted whenever house prices increased by more than 11% a year. The variablewith the lowest minimum noise-to-signal ratio has the strongest predictive power. One can imposethe additional restriction that at least a certain percentage of crises must be predicted (see e.g. Borioand Drehmann, 2009). The signals method allows strong non-linearities between the explanatoryvariable and crisis occurrence (Alessi and Detken, 2011). A notable limitation of the method is that thebasic version cannot be used to test the joint significance of several variables. Moreover, the methoditself does not include any tests of statistical significance, so that significance testing must beperformed separately.

However, most second wave contributions have used binary regressions such as logit and probit.Various types of regressions are widely used in economics, and one would expect that the readers ofacademic articles on economics are fully familiar with the basic concepts. Standard econometric

The noise-to-signal ratio equals [B/(B+D)]/[A/(A+C)], where A, number of correct alarms; B, number of false alarms; C,

ber of crises without alarm and D, number of cases without alarm and without crisis.

K. Kauko / Economic Systems 38 (2014) 289–308294

software packages can be used as such, and these packages routinely calculate results of statisticalsignificance tests. Using multiple explanatory variables simultaneously is easy. The early pioneersDemirguc-Kunt and Detragiache (1998, 2000, 2002) used the multivariate logit method. The samemethod, or comparable ones such as probit, have been used for instance by Ranciere et al. (2006), vonHagen and Ho (2007), Noy (2004), Angora and Tarazi (2011), Davis et al. (2011), Schularick and Taylor(2012), Angkinand and Willett (2011), Joyce (2011), Bunda and Ca’Zorzi (2010) and Lo Duca andPeltonen (2013).

Because the number of papers using the signals approach is relatively limited, it would not bemeaningful to test for the statistical significance of differences in results, e.g. whether house priceinflation appears more dangerous if one uses either of the two methods. However, in light ofpreliminary impressions, there are no major differences in the predictive power of the most widelyused indicators. It would be interesting to apply these two methods to the same data and compare theresults, but no one seems to have done this.

Evrensel (2008) used a survival analysis to measure the hazard rate of banking crises. van den Berget al. (2008) propose that when explaining currency crises countries should be clustered by statisticalcriteria; the causes of crises differ between such clusters, which might also be the case in bankingcrisis prediction. However, no one seems to have performed this type of analysis yet.

Surprisingly few second wave analyses have tried to analyse interaction effects of macrofinancialvariables. Among the few, Schularick and Taylor (2012) tested the interaction of the credit-to-GDPratio and credit growth and Noy (2004) the interaction of liberalisation and supervision. Interactioneffects of the income level and certain other variables have been implicitly tested where the samplehas been divided into developed and emerging markets, and the results differ to some extent (see e.g.Lo Duca and Peltonen, 2013). The binary classification tree or binary recursive tree method used byDuttagupta and Cashin (2011) and Davis and Karim (2008b) is well suited for analyses of interactioneffects. First, it is tested which variable has the strongest predictive power as a crisis predictor. Thesample is split into two child nodes according to the values of the best explanatory variable. Then, eachchild node is split according to the variable that best divides it into crisis and non-crisis cases; thevariable need not be the same one for both child nodes. In the following stage, the nodes are split again.Davis and Karim (2008b) use both multivariate logit and the binary recursive tree and obtainsomewhat different results.

3.3. Sample selection

Many second wave authors (e.g. Kaminsky and Reinhart, 1999; Lo Duca and Peltonen, 2013; Borioand Lowe, 2002; Joyce, 2011; Evrensel, 2008) do not include countries in their samples that have notexperienced any crises during the sample period. Non-crisis countries are an interesting referencegroup and have been included e.g. by Honohan (1997), Noy (2004), Demirguc-Kunt and Detragiache(1998), Hoggarth et al. (2005), Ranciere et al. (2006), Angkinand and Willett (2011), and Domac andMartinez Peria (2003). However, it is difficult to find examples of results that would clearly be affectedby pre-selecting samples consisting only of crisis ridden countries.

Recovery phases are different from stable, benign times, and classifying both kinds of observationsin the same non-crisis category may cause what Bussiere and Fratzscher (2006) call the ‘‘post-crisisbias’’ in their analysis of currency crises. Demirguc-Kunt and Detragiache (1998) chose to excludepost-crisis observations. Drehmann et al. (2011) performed two separate analyses, one for the build-up and another for the recovery.

Even though the number of post-WWII crises has been steadily increasing, the number of crisesincluded in any one analysis is still relatively limited. Hence, samples used by different researchersoverlap extensively. One reason for literature surveys of empirical research is to enable conclusionsbased on data from different parts of the world, different industries and different periods of time. Ifsimilar statistical interrelationships are observed over and over again, a universal regularity has beenfound. This may not be the case in the literature on early warning banking crisis indicators becausethese papers analyse almost identical sets of cases. Re-testing the same question with almost the samedata provides much less additional evidence than re-testing the same hypothesis with different data.On the positive side, the risk of publication bias is minimal. In other fields it may happen all too often

K. Kauko / Economic Systems 38 (2014) 289–308 295

that 20 researchers address the same question with different samples, each without knowledge ofwhat the others are doing. Then one of them gets statistically significant results only because of goodluck, and the journals’ uncoordinated reviewing processes accept nothing but significant results (seee.g. Stanley, 2005). This is not a problem if there is only one possible sample, which is almost the casein empirical analyses of banking crises.

4. The credit stock

Various indicators of excessive lending have probably been more extensively used as crisispredictors than any other variables. Common sense tells us that bank customers find it more difficultto service loans if they are overly indebted. Banks cannot substantially increase loan stocks withoutloosening credit standards. Two kinds of credit-based variables have been most often used in crisisprediction, namely the credit-to-GDP ratio and the growth rate of the credit stock in percent, i.e.relative to its own value. Evidence on the ability of these variables to predict crises is not highly robust,but the growth of credit seems a better choice than the absolute amount, even though it takes at least acouple of years before excessive credit growth causes problems. A stable but large credit stock is amuch weaker – but maybe not completely useless – crisis predictor.

Intuition says that the amount of debt must be somehow proportionate to the size of GDP. Thecredit-to-GDP ratio certainly should not be one thousand; even three would seem to be unsustainablyhigh. However, it would hardly be possible to set a precise maximum safe value for the ratio based onany theory. Empirically, the credit-to-GDP ratio as such is not a particularly robust crisis predictor.Using panel data, Davis et al. (2011) find that credit crises tend to be commonplace if the credit-to-GDPratio is high, but this result proved to be particularly sensitive to the inclusion of additional variables.Hahm et al. (2011) conclude that the level of the credit-to-GDP ratio as such is of no use as a crisispredictor in developing countries when controlled for banks’ foreign liabilities. Similarly, Joyce (2011)finds that domestic credit relative to GDP is of no use in predicting banking crises in emergingeconomies if one controls for foreign assets and liabilities. von Hagen and Ho (2007) do not findevidence that the ratio of private credit to GDP affects crisis probability.

According to Acosta-Gonzalez et al. (2012), the non-detrended credit-to-GDP ratio is a goodpredictor of cross-national differences in the severity of the 2008 crisis. Rose and Spiegel (2011) findonly weak evidence of the ability of the credit-to-GDP ratio in 2006 to explain cross-nationaldifferences in the severity of the 2008 crisis, even though the sample was basically the same as thatused by Acosta-Gonzalez et al. (2012). There is no obvious explanation for the disparity in results.

One can also define ‘‘excessive lending’’ as the growth rate of the credit stock. Lag-length selectionthen becomes highly relevant for the results, and few authors have obtained significant results whencredit growth is lagged by less than two years. Jorda et al. (2011) observe that over the period 1870–2008, in both national and international financial crises, the credit-to-GDP ratio grew quickly fouryears before the outbreak of a crisis. Demirguc-Kunt and Detragiache (2000) find a statisticalrelationship between credit expansion and crises with a two year lag. Bunda and Ca’Zorzi (2010)obtain strong evidence of a statistical connection between credit growth lagged two years and afinancial crisis, even though their definition of crisis refers to both currency and banking crises. Incontrast, Barrell et al. (2011) find that credit growth lagged one year is not a particularly goodpredictor of crises in developed countries. Bordo and Meissner (2012) show that credit growth laggedfive years has a strong positive impact on crisis probability, even though the one year lagged valueactually reduced crisis probability. Barrell et al. (2010) find that if the sample is limited to developedcountries, credit growth lagged one year does not appear to be a promising crisis predictor, evengetting the wrong sign.

Using a very long sample period, Schularick and Taylor (2012) find evidence of the ability of creditgrowth to predict financial crises in major developed countries. This variable worked better than anyof the other variables. Various lags were tested and the two year lag seemed to perform best. Perhapssurprisingly, granting loans to the corporate sector instead of other sectors did not affect theprobability of crisis. This contradicts the findings of Buyukkarabacak and Valev (2010), whodecomposed credit stock into household credit and enterprise credit; household credit appeared, atleast with no lag or a one year lag, to be much more dangerous than enterprise credit. At least one

K. Kauko / Economic Systems 38 (2014) 289–308296

reason for the difference may be the sample. Buyukkarabacak and Valev (2010) studied a very largenumber of different kinds of countries in 1990–2006, whereas Schularick and Taylor (2012) looked atonly a handful of developed countries with their sample covering a much longer period of time andemploying longer lags.

Davis and Karim (2008a) find that if a country has introduced a deposit guarantee scheme, creditgrowth predicts crises with a four to five year lag. For one to three year lags, the sign is reversed so thatcredit growth actually lowers the probability of an immediate crisis. The role of a deposit guaranteescheme in these findings probably indicates that the traditional boom-bust cycle does not normallyoccur without safety nets provided by the government. Demirguc-Kunt and Detragiache (2002) couldnot find a robust connection; credit growth lagged two years predicted crises only very weakly andonly if the existence of a deposit insurance scheme was included as an explanatory variable.

In recent times, the trend deviation of the credit-to-GDP ratio has been much discussed as it willbe the primary trigger for countercyclical capital requirements in the Basel III framework (see BasleCommittee, 2010, and EU directive 2013/36/EU). This ratio is comparable to the growth rate ofcredit; both indicators regard a large credit stock harmless if it stabilises at a roughly constantlevel. But the situation is not stable if the amount of debt has grown and reached a level that is highrelative to its past. Borio and Lowe (2002) may have been the first to use this variable and seem tohave introduced the so-called one-sided Hodrick-Prescott filter, a statistical technique to derivetrend values using only data that could have been available at each moment of time. They foundevidence on the general tendency of the credit-to-GDP ratio to reach its maximal trend deviationabout three years before the outbreak of a crisis. If stock prices are also above their trend level, therisk of a crisis in three years is very high. This finding applies to both developed and developingeconomies. Drehmann et al. (2011) used the same method and the same explanatory variable andconcluded that it seemed to perform the best among ten different potential variables based on thenoise-to-signal ratio.

The idea to base the countercyclical capital buffer of Basel III on the trend deviation of the loans-to-GDP ratio has been criticised by Repullo and Saurina (2011), as the credit-to-GDP gap is driven both byGDP and the loan stock. A slow GDP growth rate, let alone a negative one, is interpreted as excessiveloan growth in light of the credit-to-GDP ratio. Moreover, the credit cycle typically lags the businesscycle. In the sample of Repullo and Saurina the one-sided H-P residual is negatively correlated withGDP growth, which could induce policymakers to impose a capital requirement during recessions andto remove it during booms.

Domac and Martinez Peria (2003) conclude that credit growth is more dangerous in a liberalisedfinancial environment than in a less liberalised one. There may also be differences between continentsand crises; the growth of credit predicted crises in Asia, but not in Latin America (Davis et al., 2011).

Credit growth and the credit-to-GDP ratio could also be combined. Such a variable was tested byKamin and DeMarco (2012), who used the first differences of credit-to-GDP. This variable was not verysuccessful in explaining cross-national differences in banks’ CDS spreads or stock prices during thelatest financial crisis. Credit growth may weaken financial stability, but increasing income inequalitydoes not contribute to excessive credit growth (Bordo and Meissner, 2012), although Roy and Kemme(2012) found that inequality as such weakens banking stability in advanced countries.

5. The current account

Obstfeld (2012) proposes that the current account is relevant for policymaking for three reasons:imbalances may be a symptom of problems, the balance itself has macroeconomic implications andthe net figure may still be an indicator of cross-national investment flows. Corsetti et al. (1999) presenta theory on how moral hazard expectations contribute to excessive inflow of funding from abroad.Giannetti (2007) develops a formal model without assumptions of irrationality; after the liberalisationof cross-border capital flows, uninformed foreign investors grant funding to local banks in anemerging market, enabling even poorly managed banks to grant loans. Most econometric analyseshave found that banking crises are typically preceded by a current account deficit. In some papers aclosely related concept, such as the trade deficit or the quantity of foreign debt, has been used, but theresults seem to be broadly similar.

K. Kauko / Economic Systems 38 (2014) 289–308 297

Solow (2005) states in the foreword of the fifth edition of ‘Manias, Panics and Crashes’ byKindleberger and Aliber that, ‘‘The current U.S. international financial position in some ways parallels that

of Mexico, Brazil, and Argentina in the 1970s. These countries had unsustainably large current account

deficits and obtained the cash to pay the interest to their foreign creditors from the foreign creditors. The

implication is that the U.S. external payments position is not sustainable.’’ These words sound almostprophetic now. The recent crisis began in the US and culminated in the market for securitised housingloans, a market which probably would not have grown so rapidly without strong demand for securitiesby foreign investors. Caballero and Krishnamurthy (2009) present a model to explain how overseasdemand for safe assets pushed the US financial system towards advanced financial engineering forprocessing mortgages in order to provide the easily marketable low-risk assets demanded byforeigners and to keep a levered claim on domestic debt.

Reinhart and Reinhart (2008) introduce the concept of ‘‘capital bonanza’’: foreign investors becomeinterested in a developing country, capital flows into the market, the exchange rate appreciates andasset prices rally, which leads to a domestic credit expansion. The bust phase normally follows withina few years. Connor et al. (2012) conclude that both the US and Ireland were experiencing a typicalcapital bonanza before the subprime-Lehman crisis, although these countries are not part of thedeveloping world.

It would seem unreasonable to claim that financial crises can never occur without a currentaccount deficit. The world is a closed economy, but crises do happen. However, external deficitscan predict when and where crises occur, and large international imbalances are an alarming sign.Laeven and Valencia (2008) give details of 41 banking crises. Japan in the 1990s was the onlycrisis-ridden mature developed country running a current account surplus one year prior to theoutbreak of the crisis. Moreover, Estonia had a current account surplus in 1991 before theoutbreak of the crisis, but this may not be a representative case because of the extremely turbulentcircumstances amid the collapse of the Soviet Union. Reinhart and Rogoff (2008a) presentdescriptive statistics and conclude that the average current account deficit is about 3% of GDP oneyear before the outbreak of a serious crisis in an advanced economy. As to econometric studies, LoDuca and Peltonen (2013) find evidence on the role of the current account deficit in bothdeveloped and emerging economies. Kauko (2012) finds that a combination of rapid credit growthand a current account deficit made a national banking system more vulnerable in 2009. Rose andSpiegel (2012) obtain further evidence to support the view that the recent crisis was worse incountries with current account deficits than in surplus countries. The graphs presented by Sarlinand Peltonen (2011, p. 31) indicate that a large current account deficit is typical when a country ison the borderline of a crisis. Some other studies present weaker evidence. Roy and Kemme (2011)find that the current account is a fairly important explanatory factor when controlled for a numberof other variables including dwelling prices. In another study, the same authors (Roy and Kemme,2012) find that the current account is a good crisis predictor if and only if private debt and assetprices are omitted from the analysis.

A current account deficit normally occurs simultaneously with a trade balance deficit. Kaminskyand Reinhart (1999) conclude that exports tend to be weak before a financial crisis, contributing to aforeign trade deficit. Singh (2011) provides evidence on the role of both weak exports and strongimports as distress predictors for India. The impact of the current account seems to have varied overhistory. In the 1970s there was no detectable connection between the balance of payments and theoutbreak of a banking crisis, probably because of the tightly regulated financial market (Kaminsky andReinhart, 1999).

Hoggarth et al. (2005) find that the explanatory power of the current account is weak whencontrolled for different safety net relevant variables. Joyce (2011, p. 885) finds that controlling forcrises in neighbouring countries eliminates the predictive power of the current account variable foremerging economies.

There has been a connection between financial crises and the international mobility of capital.During times of high capital mobility, for instance before World War One or since the 1980s, financialcrises have been more commonplace (Reinhart and Rogoff, 2008b). Large capital inflows from the UKto the US and Latin America caused instability in the Americas in the late 19th and early 20th century(Reinhart and Rogoff, 2010). Jorda et al. (2011) observe that in major developed countries purely

K. Kauko / Economic Systems 38 (2014) 289–308298

domestic crises have typically been preceded by current account deficits, but national manifestationsof global crises cannot be predicted on this basis.

Mere gross flows may matter for the international transmission of crises; for instance, manyEuropean banks used to fund themselves in the US market and invest in securitised US assets. Thispractice was hardly reflected in net figures reported in balance of payment statistics, but it proved tobe a major channel of international contagion in 2007–2009 (see e.g. Obstfeld, 2012, and thereferences therein).

The bank loan stock cannot grow unless someone is willing to hold bank liabilities. Most of theevidence indicates that getting funding from abroad is more dangerous than relying on domesticsurplus sectors. Hahm et al. (2011) note that banks’ funding costs are more likely to reach a crisis levelif a large part of the funding is from foreign sources. In the panel logit analysis by Angkinand andWillett (2011), banks’ net foreign liabilities relative to GDP proved to be a robust predictor of crisis,and the impact seemed stronger if liabilities exceeded assets. Joyce (2011) concludes that at least indeveloping countries banks’ heavy foreign debt burden increases the probability of a crisis, whereasportfolio and direct investments from abroad are negatively related to crisis probability. Foreignliabilities are even more problematic for banking stability in countries with pegged exchange rates(Domac and Martinez Peria, 2003).

On the other hand, financial crises may abruptly eliminate the current account deficit. In theemerging markets data of Bussiere and Fratzscher (2006), the average current account-to-GDP ratio is�2.66% before the crisis, but two years after the crisis the average is +0.46%, which may be due to weakdomestic demand and export-boosting devaluations of the domestic currency during the crisis. Inboth Finland and Sweden, the current account switched from deficit to surplus during the crises of theearly 1990s (see e.g. Kiander and Vartia, 2011).

6. Asset prices

Kindleberger (1978) and Minsky (1977) focus on the role of asset price bubbles as drivers offinancial instability and crises. Indications of real estate bubbles seem highly promising as crisispredictors, although the evidence is sparse, especially regarding emerging economies. Banks’excessive lending may seem safe if it is collateralised. Real estate is often used as collateral, but if theprice level reaches an unsustainable level and debtors can no longer service the loans they have takento finance asset purchases, foreclosures may cause a market crash. A disproportionate part of thepapers addressing the stability consequences of real estate bubbles are purely descriptive, probablybecause of data availability problems, but the available evidence strongly indicates that abnormalhouse price increases systematically precede banking crises. The predictive power of the stock marketis somewhat less clear.

As to descriptive analyses of different cases, the available evidence indicates that acceleratedhouse price inflation is a warning sign. Reinhart and Rogoff (2008a) present some descriptivestatistics and conclude that house price increases are regularly observed before banking crises.Connor et al. (2012) emphasise the role of the housing market in the build-up of risks in the USand Ireland before the global financial crisis. Kokko (1999) emphasises the role of asset pricebubbles in both the Swedish crisis and the East Asian crises of the 1990s. On the other hand, theColombian banking crisis in 1998–1999 was not preceded by a housing price bubble (Gomez andRozo, 2008).

Econometric evidence on the predictive power of house prices has come almost exclusively fromdeveloped countries, possibly because it is even more difficult to obtain data from most emergingeconomies. Drehmann et al. (2011) conclude that the trend deviation of housing prices tends to peakabout two years before the outbreak of a crisis, but the limited availability of data weakens thereliability of the results. Barrell et al. (2010) find that an increase in housing prices predicts bankingcrises in developed countries with a lag of three years, although the effect seems to be weaker if onecontrols for international contagion (Barrell et al., 2011). Bunda and Ca’Zorzi (2010) find that houseprice growth lagged one year is an excellent crisis predictor. Roy and Kemme (2011) conclude thatboth real estate prices and stock indices (apparently in levels) predict banking crises within a period offour years, even when entered simultaneously in a logit estimation. Asset prices proved to be the most

K. Kauko / Economic Systems 38 (2014) 289–308 299

robust crisis predictor; they even Granger-caused the current account in 1970–2007 in Ireland, Spain,the UK and the US (Roy and Kemme, 2012).

Kemme and Roy (2012) discuss psychological biases prevailing prior to the 2008 crisis and presenteconometric evidence on developed countries that experienced a banking crisis after WWII. Theyconclude that housing bubbles preceded crises so regularly that the recent financial turbulence in theUS, Spain, Ireland and the UK could have been predicted.

Data on stock indices is easy to find, but surprisingly few authors have tried to systematicallymeasure their predictive power. The variable may have been tested, but it has been omitted from thefinal versions submitted to journals, possibly because of a lack of explanatory power. Some publishedpapers even fail to find a robust connection. Schularick and Taylor (2012) find very weak evidence ofthe ability of stock prices to predict financial crises, although the effect was somewhat stronger if thecredit-to-GDP ratio was high. Singh (2011) finds that rapid stock price increases boost the probabilityof banking distress in India with a remarkably short lag of less than two and a half years. Financialstock prices rose rapidly relative to other stocks in both the US and Ireland during the build-up phasein 2002–2007 (Connor et al., 2012). According to Drehmann et al. (2011), differenced equity prices area slightly weaker predictor of banking crises than differenced dwelling prices, and the trend deviationof equity prices is a clearly weaker predictor than the trend deviation of dwelling prices. Rose andSpiegel (2012) note that share price increases in 2003–2006 predicted cross-national differences inthe severity of the subprime-Lehman crisis; a rapid increase was typically followed by a worse-than-average outcome.

7. Level and growth of GDP

During a boom phase, one might expect rapid GDP growth. The crisis will probably reverse thesituation. As to empirical evidence, the connection between economic growth and banking crisesseems somewhat unstable. The impact may depend on whether the sample consists of developed oremerging economies. Most authors have used rather short lags, and these results probably tell us moreabout the immediate triggers than about the earlier build-up phase.

In the very short run, slow or negative growth is a worrying sign (see Beck et al., 2006; Davis et al.,2011; Davis and Karim, 2008b; or Klomp and de Haan, 2009). If one uses a somewhat longer but still shortlag, the results remain broadly unchanged. According to Demirguc-Kunt and Detragiache (1998, 2005),Angkinand and Willett (2011) and von Hagen and Ho (2007), economic growth has in most cases beenslow immediately before a crisis. Davis and Karim (2008a) extend the sample of Demirguc-Kunt andDetragiache by adding both years and countries; if lagged by two years, slow GDP growth is a warningsign and predicts crises. Joyce (2011) and Domac and Martinez Peria (2003) do not find a connection.

GDP growth may be rapid during the build-up phase, but the evidence is not unanimous. Kaminskyand Reinhart (1999) conclude that a national economy normally grows at higher-than-average speeduntil just eight months before the outbreak of a crisis. This finding contradicts some of the resultsreported above. Data charts presented by Drehmann et al. (2011) indicate that GDP normally slightlyexceeds its trend during the last few years before a crisis. If one focuses on the most advancedeconomies of each era, an international crisis is probably very severe in countries where economicgrowth was faster than average two to four years before the crisis, but in purely national crises thisphenomenon cannot be observed, which may imply that overheated economies are vulnerable tocontagion from abroad but do not trigger crises themselves (Jorda et al., 2011). On the other hand, Roseand Spiegel (2012) and Kauko (2012) find no evidence that economic growth during pre-crisis yearsexplains cross-national differences in the severity of the subprime-Lehman crisis.

Results concerning the level of GDP per capita are not particularly robust either. Rose and Spiegel(2011), Aizenman and Pasricha (2012) and Kauko (2012) find that low-income countries wererelatively mildly affected by the subprime-Lehman crisis. This contradicts some findings on earliercrises; problems become more likely if the income level is low (Davis and Karim, 2008a; Domac andMartinez Peria, 2003; Beck et al., 2006). Demirguc-Kunt and Detragiache (2000) do not find acorrelation between crisis occurrence and income level, but in a later publication the same authors(Demirguc-Kunt and Detragiache, 2005) find that poorer countries have been more likely to be hit bycrises. Income inequality has been a typical characteristic of pre-crisis eras (Roy and Kemme, 2012).

K. Kauko / Economic Systems 38 (2014) 289–308300

If we believe that some crises are caused by speculative euphoria and overheating but others bychronic economic weakness, lagged growth rates might have a non-linear impact on the probability ofa crisis: both rapid and slow growth may be worrying signs, whereas moderate growth could be a signof stability. Unfortunately, it seems impossible to find papers that test the non-linearity of thisrelationship.

8. Exchange rates and price stability

Twin crises, i.e. simultaneous banking crises and collapses of fixed exchange rate regimes, havetaken place mainly in emerging markets. Inspired by the Asian crises in 1997, Corsetti et al. (1999)present a theory on how moral hazard fuels excessive borrowing from abroad because evenunprofitable projects can be financed, and how expectations of inflationary finance in rescueoperations cause a collapse of the currency. Empirically, inflation is not strongly related to bankingcrises in developed countries but possibly somewhat more so in emerging economies, though theevidence is mixed.

By definition, exchange rates are key variables in studies on twin crises. According to Kaminsky andReinhart (1999), a banking crisis normally precedes a currency collapse, but the currency collapse mayworsen the banking crisis. Domac and Martinez Peria (2003) and Husain et al. (2005) conclude thatstable fixed exchange rate systems diminish the risk of crises in developing countries. Angkinand andWillett (2011) find support for their ‘‘unstable middle hypothesis’’: countries with a ‘‘soft’’ exchangerate peg are particularly vulnerable. Davis and Karim (2008b) find no impact in a logit analysis, butaccording to the binary recursive tree analysis, depreciation predicts crises if credit aggregates growrapidly (or credit contraction is slow) and real interest rates are not excessively high. In contrast,Duttagupta and Cashin (2011) find that nominal depreciation of the domestic currency is a regulardeterminant of banking crises.

Demirguc-Kunt and Detragiache (1998, 2005) do not find evidence of an instantaneous correlationbetween currency depreciation and the occurrence of crises in panel data consisting of many kinds ofcountries, and neither do von Hagen and Ho (2007) or Beck et al. (2006). Dollarization of the financialsystem may render a country more vulnerable to banking crises, but the effect seems to be weak(Hong, 2006). In contrast, Duttagupta and Cashin (2011) identify bank liability dollarization as one ofthe main crisis determinants. According to Lo Duca and Peltonen (2013), currency overvaluation doesnot lead to crisis, but von Hagen and Ho (2007) reach the opposite conclusion.

Some results indicate that rapid inflation increases the risk of banking crises. However, this findingis not at all robust. Demirguc-Kunt and Detragiache (1998, 2000) observe that rapid inflation isstatistically related to crises, but Davis and Karim (2008a), studying a sample consisting mainly ofdeveloping and emerging economies, find that the effect depends on the method and the precisedefinition of banking crisis. Davis et al. (2011), in contrast, reach the conclusion that inflation maypredict crises in Latin America but not in Asia. Joyce (2011) finds additional evidence on the impact ofinflation on the occurrence of crises in emerging markets. Angkinand and Willett (2011) obtain furtherweak evidence of this in a sample consisting of both developing and developed countries. Rose andSpiegel (2012) and Kauko (2012) do not find a correlation between inflation and the severity of thesubprime-Lehman crisis in cross-national comparisons. The findings of Buyukkarabacak and Valev(2010) indicate that rapid inflation predicts mild rather than severe banking crises. Lo Duca andPeltonen (2013) explicitly test the differences of early warning indicators between emerging anddeveloped countries and find inflation to be a useful early warning indicator in emerging economiesonly. Angora and Tarazi (2011) find, perhaps surprisingly, that inflation may impede crises in memberstates of the West African monetary union, a finding that is inconsistent with all other significantfindings and may somehow be related to the specifics of the CFA franc zone.

9. Banking sector-relevant structural factors

The conventional wisdom has been that many too-big-to-fail (TBTF) banks are willing to takeexcessive risks because of moral hazard. If the market is concentrated, TBTF institutions and theircounterparties expect the government to offer implicit guarantees. Alternatively, in an atomistic and

K. Kauko / Economic Systems 38 (2014) 289–308 301

highly competitive market, banks’ shareholders may have nothing to lose and therefore instructmanagers to accept excessive risks because it may be the only way to make profits (Keeley, 1990).Most of the evidence corroborates this latter hypothesis. Noy (2004) observes that banks’ monopolypower is stabilising. Based on panel data for 69 countries, Beck et al. (2006) find that concentratedbanking systems are less fragile than atomistic ones, and the relationship seems robust to theinclusion of a large number of control variables related to the institutional environment, such asopenness of the banking sector and economic freedom. However, the significance of the concentrationvariable weakens if a broader number of cases are classified as crises or if one controls for the tendencyof officials to deny banking licenses, which may be an indication of barriers to entry. Evrensel (2008)also concludes that concentrated banking systems are more robust and that the hazard rate of bankingcrises is lower. On the other hand, Aizenman and Pasricha (2012) fail to find much of a connectionbetween concentration and financial stability during the post-Lehman crisis.

Demirguc-Kunt and Detragiache (2002) test the impact of deposit insurance on the probability ofbanking crises. Explicit deposit insurance increases the likelihood of crises, especially if the scheme iseither funded or run by the government and the coverage is extensive. Beck et al. (2006) use the moralhazard variable of Demirguc-Kunt and Detragiache (2002) and obtain similar results. Davis and Karim(2008a), von Hagen and Ho (2007) and Evrensel (2008) find that deposit insurance increases the risk ofbanking crises. Angkinand and Willett (2011) and Davis and Karim (2008b), however, do not obtainevidence of this effect. Hoggarth et al. (2005) find that crises are more likely if the government eitheroffers an unlimited safety net for banks or alternatively claims, probably unconvincingly, not toprovide any safety net at all.

Banking sector liberalisation also seems to be a central variable. Statistically, the probability ofbanking crises increases in the aftermath of liberalisation (Kaminsky and Reinhart, 1999; Demirguc-Kunt and Detragiache, 1998). Evidence on permanent destabilising effects of liberalised capitalinflows is presented by Ranciere et al. (2006). Banking crises in Sub-Saharan Africa have been morecommonplace in liberalised financial systems (Misati and Nyamongo, 2012). The continuousinstability indicator of Klomp and de Haan (2009) is an increasing function of liberalisation. Noy(2004) concludes that the abolition of domestic interest rate regulations is a risk factor in emergingand developing markets, but that a relaxation of the regulations on international capital mobility hasno impact; this finding contradicts the idea that most banking crises are caused by excessive inflows offunding from abroad. Jonung et al. (1996) find that liberalisation and other changes in the monetaryregime were key drivers in the boom-bust cycles in Finland and Sweden in the 1980s and 1990s;Kiander and Vartia (2011) reach a similar conclusion.

It seems that no one found that deregulation reduces the occurrence of crises during the first yearsof liberalisation. On the other hand, Beck et al. (2006) obtain some evidence on the long-termstabilising effects: high values of the banking freedom index, which mainly indicate free entry, seem toreduce the risk of crisis; in contrast, restrictions on securities and real estate-related activitiesdestabilise rather than stabilise. Khan et al. (2013) find that central bank independence and stronglaw-and-order traditions reduce the probability of a banking crisis.

10. Other variables

The term ‘‘early warning’’ is often used when the focus is on the predictive ability of financialmarket data such as CDS spreads and stock volatilities. Market data have seldom been used forpredicting nationwide systemic distress, possibly because highly sophisticated financial instrumentshave been introduced in most parts of the world only recently, so that we do not have abundant pre-financial crisis observations on these variables.

The literature on currency crises has analysed cross-national contagion and speculative attacks(see e.g. Eichengreen et al., 2003a,b, or Haile and Pozo, 2008). Even though contagion might be possibleeven in the case of a pure banking crisis, i.e. a banking crisis without pressure on fixed exchange rates,it is difficult to find studies with a special focus on contagion. Its role has, however, been identified inthe analyses of Barrell et al. (2011) and Jorda et al. (2011), among others.

Lo Duca and Peltonen (2013) find that certain international or global variables can improve theaccuracy of crisis prediction models. There are differences between emerging and advanced

K. Kauko / Economic Systems 38 (2014) 289–308302

economies as regards the relevance of foreign variables, but for both country groups, foreignmacrofinancial data clearly has predictive power. Emerging economies seem more exposed to globalfactors. Aizenman and Pasricha (2012) report that more open countries experienced greater stressduring the crisis of 2008–2009.

Perhaps surprisingly, there are relatively few papers that focus on the crisis prediction power ofbanks’ profitability, solvency, liquidity or share quotations. One of the few analyses is that by Barrellet al. (2010). Their sample consists of developed economies in 1980–2007. A simple leverage indicatorproved to be a good predictor of future crises; a liquidity indicator lagged by one year was equallygood. Unfortunately, there were only 14 crises in the sample, which reduces the reliability of thefindings. Duttagupta and Cashin (2011) find further evidence of the role of liquidity. The deposits-to-loans ratio seems to be one of the best explanatory variables in analyses of the cross-national severityof the 2007–2008 crisis (Kamin and DeMarco, 2012) as well as the probability of crises in WesternAfrica (Angora and Tarazi, 2011); more deposits relative to assets make the system safer. Eichler et al.(2011) use a compound option model to estimate the risk of a banking crisis in data on major US banks.

The evidence is equally mixed for monetary aggregates. It is possible to find evidence of the impactof rapid growth of monetary aggregates on crisis occurrence among Asian countries in the 1990s(Davis et al., 2011). But Drehmann et al. (2011) find that M2 growth has not been a good crisispredictor in developed economies. Based on a sample covering a very long time period, Jorda et al.(2011) find that the amount of money relative to nominal GDP tended to be high four years prior to acrisis. Schularick and Taylor (2012), using similar data, reach the conclusion that the growth rate ofmonetary aggregates is a weaker predictor of crises than credit aggregates, although high correlationsbetween these variables make it difficult to differentiate between the two in data covering the erabefore World War II.

The M2 multiplier, i.e. the ratio of M2 to the monetary base, often grows substantially and reacheshigh levels before the outbreak of a banking crisis (Kaminsky and Reinhart, 1999; Demirguc-Kunt andDetragiache, 1998, 2000; Buyukkarabacak and Valev, 2010). The M3 multiplier seems to be a goodpredictor of banking distress periods in India (Singh, 2011). These ratios also measure central bankreserves, with a shortage of reserves probably increasing the risk of twin crises. Again, it is possible tofind contradicting evidence. Joyce (2011) does not obtain evidence of the relevance of the moneymultiplier and Beck et al. (2006) find only weak evidence.

High interest rates affect debtors’ solvency in a straightforward manner by weakening theirfinancial viability and capacity to service debt. Demirguc-Kunt and Detragiache (1998, 2000, 2002)and Evrensel (2008) conclude that high real rates of interest systematically precede banking crises.Jorda et al. (2011) find that short-term real interest rates have no explanatory power as such, but thatthe difference between economic growth and real rates of interest does. If the crisis is limited to onecountry, real interest rates are high four years prior to the crisis, but in the case of global crises theytend to be low, at least in the sample of Jorda et al. (2011). Bordo and Meissner (2012) conclude thatlow interest rates promote credit cycles, leading to a heightened risk of a banking crisis with some lag.

Large fiscal deficits seem to predict banking crises – defined as periods of banks’ liquidity problems(von Hagen and Ho, 2007). Central bank independence seems to prevent broadly defined financialinstability in panel data (Klomp and de Haan, 2009).

It is also interesting to examine what kinds of explanatory variables have not been used. Crises mightoften be due to expectations that later turn out to be far too optimistic. This was largely the point of theFinancial Instability Hypothesis presented by Minsky (1977); when good times continue for some time,it seems safe to increase leverage by using both old and new financial instruments. As to existing sourcesof data, the optimism of expectations is sometimes directly observable in different surveys such as theconsumer confidence indicator in various countries or business climate indices. If these expectations areexcessively positive relative to any objective measure of the state of the economy, observable in real timeor not, the risk of a crisis within a few years may be above its long-term average.

11. Are there differences between developing and developed countries?

Table 1 presents a summary of the reported significance of different potential early warning signsin a large number of papers on banking crises. The table oversimplifies previous findings for various

Table 1aSignificance of potential explanatory variables as early warning indicators.

(Mainly) developing and

emerging countries

(Mainly) advanced countries Heterogenous samples

Amount of (private) debt/leverage

Significant von Hagen and Ho (2007) Bordo and Meissner (2012), Roy

and Kemme (2012), Lo Duca

and Peltonen (2013)

Lo Duca and Peltonen (2013)

Non-significant Joyce (2011) von Hagen and Ho (2007) Domac and Martinez Peria

(2003), Demirguc-Kunt and

Detragiache (1998), von Hagen

and Ho (2007), Davis and Karim

(2008a), Hoggarth et al. (2005)

Growth of (private) debt or its trend-deviation

Significant Domac and Martinez Peria

(2003), von Hagen and Ho

(2007), Davis et al. (2011)

(Asia)

Bordo and Meissner (2012),

Drehmann et al. (2011), Barrell

et al. (2011), Schularick and

Taylor (2012), von Hagen and

Ho (2007), Jorda et al. (2011)

Demirguc-Kunt and Detragiache

(1998, 2000, 2005), Domac and

Martinez Peria (2003),

Buyukkarabacak and Valev

(2010), Kauko (2012), Borio and

Lowe (2002), Davis and Karim

(2008a), Angkinand and

Willett (2011)

Non-significant Davis et al. (2011) (Latin Am);

Angora and Tarazi (2011), von

Hagen and Ho (2007)

Barrell et al. (2010), Lo Duca

and Peltonen (2013)

von Hagen and Ho (2007), Beck

et al. (2006), Lo Duca and

Peltonen (2013), Demirguc-Kunt

and Detragiache (2002)

House prices

Significant Roy and Kemme (2011), Barrell

et al. (2011), Roy and Kemme

(2012), Drehmann et al. (2011),

Kemme and Roy (2012)

Bunda and Ca’Zorzi (2010)

Non-significant

Current account or trade balance deficit, foreign net debt

Significant Lo Duca and Peltonen (2013) Bordo and Meissner (2012), Roy

and Kemme (2011, 2012), Jorda

et al. (2011), Lo Duca and

Peltonen (2013)

Kaminsky and Reinhart (1999),

Kauko (2012), Lo Duca and

Peltonen (2013)

Non-significant Joyce (2011), Domac and

Martinez Peria (2003)

Domac and Martinez Peria

(2003), Angkinand and Willett

(2011), Hoggarth et al. (2005)

K. Kauko / Economic Systems 38 (2014) 289–308 303

reasons. Many authors present multiple equations and the significance of a variable may vary acrossthem. The significance may also vary depending on the lag structure. In such cases the classification isbased on either the most comprehensive equation or the one presented as the main model by theauthors themselves. If any lag of the potential explanatory variable is statistically significant, thevariable is regarded as significant. In some cases no explicit significance test results are reported; inthese cases, the most promising early warning indicators are regarded as ‘‘significant’’ based onsomewhat arbitrary criteria.

Nevertheless, as a rule, differences between developed countries and emerging markets aremoderate. To a large extent, the same macrofinancial phenomena precede banking crises. However,one can tentatively conclude that a current account deficit is more dangerous for advanced countriesthan for emerging economies whereas inflation is more problematic for developing countries, possiblybecause of higher inflation rates. Differences in results may be due to differences either in the realitybehind the figures or in the quality of statistics. Statistics on many developing economies may beunreliable. In the worst case, the so-called statistics might not differ much from random numbers,which make poor early warning indicators. In fact, there is a slight tendency of results for emerging

Table 1bSignificance of potential explanatory variables as early warning indicators.

(Mainly) developing and

emerging countries

(Mainly) advanced countries Heterogenous samples

Real interest rate

Significant Roy and Kemme (2011,

2012), Barrell et al. (2011),

von Hagen and Ho (2007),

Jorda et al. (2011)

Angkinand and Willett (2011),

Demirguc-Kunt and Detragiache

(1998, 2000, 2002), Beck et al.

(2006), Kaminsky and Reinhart

(1999), Evrensel (2008)

Non-significant Domac and Martinez Peria

(2003), Davis et al. (2011),

von Hagen and Ho (2007)

Domac and Martinez Peria (2003),

Davis and Karim (2008b)

Monetary aggregate relative to e.g. GDP or forex reserves

Significant Domac and Martinez Peria

(2003), Davis et al. (2011)

(Asia)

Schularick and Taylor (2012),

Jorda et al. (2011)

Domac and Martinez Peria

(2003), Demirguc-Kunt and

Detragiache (1998, 2000),

Davis and Karim (2008b),

Buyukkarabacak and Valev

(2010), Kaminsky and

Reinhart (1999)

Non-significant Joyce (2011), Davis et al.

(2011) (Latin Am)

Roy and Kemme (2012),

Drehmann et al. (2011),

Barrell et al. (2011)

Demirguc-Kunt and Detragiache

(2002), Beck et al. (2006),

Kauko (2012)

Inflation

Significant; +, rapid

inflation precedes

crises; �, slow

inflation precedes

crises

Noy (2004) (+), Joyce

(2011) (+), Domac and

Martinez Peria (2003) (+),

Lo Duca and Peltonen

(2013) (+), Angora and

Tarazi (2011) (�); Davis

et al. (2011) (+, Latin Am,

recursive tree method)

von Hagen and Ho (2007) (+) Domac and Martinez Peria

(2003) (+); Demirguc-Kunt and

Detragiache (2000) (+);

Demirguc-Kunt and Detragiache

(1998) (+); Buyukkarabacak

and Valev (2010) (+)

Non-significant Davis et al. (2011) (Asia and

Latin America with logit)

Barrell et al. (2011), Lo Duca

and Peltonen (2013)

Angkinand and Willett (2011),

Demirguc-Kunt and Detragiache

(2002), Beck et al. (2006), Lo Duca

and Peltonen (2013), Kauko

(2012), Davis and Karim (2008a,b),

Rose and Spiegel (2012)

Econometric studies with data from several countries only.

K. Kauko / Economic Systems 38 (2014) 289–308304

market data to be less frequently significant than results for data from developed countries. Morevariables might prove to be significant for emerging economies if the data were more accurate.

12. Final comments

The early warning signs of banking crises have been studied econometrically since the late 1990s.Even though there are many inconsistencies in the findings of different authors, the typical mainfindings of this literature can be briefly summarised as follows. Most banking crises are preceded by acredit-driven boom period. This boom period manifests itself as abnormally rapid growth of the loanstock. Whatever the metric of loan growth, dangerous imbalances probably exist if the amount ofcredit is excessively large relative to its own past. A large yet stable credit stock is probably lessdangerous. Moreover, an asset price bubble often emerges before the crisis and bursts when the worstphase is approaching. House prices seem a better indicator than equity quotations, although thenumber of papers with results on house prices is fairly limited. A current account deficit typicallyprecedes a banking crisis, at least in developed countries. Historically, banking crises have been more

K. Kauko / Economic Systems 38 (2014) 289–308 305

commonplace during eras of free capital mobility. Increasing amounts of banks’ foreign liabilities are aparticularly worrying signal. GDP growth tends to exceed its long-term average during the pre-crisisbuild-up period, but immediately before the crisis economic growth slows down sharply. In twincrises, banking instability and pressures on a fixed exchange rate are intertwined. In light ofdescriptive contributions, for example by Kindleberger (1978), these findings are not particularlysurprising.

Although the differences in crisis dynamics are quite similar in both developed and emergingeconomies, factors related to inflation and exchange rates may be more important in emergingeconomies. As to non-cyclical variables, many studies have found that deposit insurance weakensrather than improves financial stability, which is in line with the moral hazard hypothesis. Moreover,banking systems consisting of a large number of relatively small banks are inherently unstable; thereis no evidence to support the view that banking systems dominated by a handful of too-big-to-failinstitutions regularly end up in moral hazard-induced problems. Tight regulations make crises lesslikely, and the first years after liberalisation are particularly crisis-prone.

Anomalies on the stock exchange seem to offer a free lunch to investors, but they have a veryunpleasant tendency to disappear after their existence becomes generally known, probably becauseinvestors try to benefit from them (Marquering et al., 2006). Analogically we may, at least in the bestcase, see something similar happening in financial crises. Excessive loan growth, house price bubblesand current account deficits may lead to banking crises, but these kinds of imbalances may becomeless commonplace if their dangers are understood. If and when research results on early warningsignals for banking crises become more widely known, policymakers, potential providers of fundingand banks themselves will pay increasing attention to signs of financial fragility. This is more likely toreduce the frequency of crises than to increase it. There are encouraging signs of increasing awareness.As seen in the introduction, research in the field is expanding. Scattered crises in the 1980s, 1990s andearly 2000s did not lead to major changes in the international regulatory system. In contrast, theinternational regulatory community has reacted to the recent global financial crisis by launching BaselIII, a more demanding and presumably safer capital requirement system. An interesting part of theBasel III initiative is the countercyclical capital buffer (see Basle Committee, 2010), which should beoperated according to lessons learned from the crisis prediction literature. It may be unrealistic toargue that the history of financial crises has ended, but lessons learned, growing general concern, andsystematic research on the topic will help policymakers to avoid repeating the old mistakes. So, thistime may be different after all – unless we forget the lessons we have learned.

Acknowledgement

Thanks are due to an anonymous referee and seminar participants at the Bank of Finland for severalvaluable comments.

References

Acosta-Gonzalez, E., Fernandez-Rodrıguez, F., Sosvilla-Rivero, S., 2012. On factors explaining the 2008 financial crisis. Econ. Lett.115, 215–217.

Aizenman, J., Pasricha, G.K., 2012. Dereminants of financial stress and recovery during the great recession. Int. J. Finance Econ.17, 347–372.

Alessi, L., Detken, C., 2011. Real time early warning indicators for costly asset price boom/bust cycles: a role for global liquidity.Eur. J. Polit. Econ. 27, 520–533.

Angkinand, A.P., Willett, T.D., 2011. Exchange rate regimes and banking crises: the channels of influence investigated. Int. J.Finance Econ. 16, 256–274.

Angora, A., Tarazi, A.E., 2011. Crises bancaires dans les pays de l’UEMOA: un systeme d’alerte avancee fonde sur une approchelogit multinomiale.(Banking crises in UEMOA countries; an early warning system based on multinomial logit). Bruss. Econ.Rev. 54, 21–50.

Arena, M., 2008. Bank failures and bank fundamentals: a comparative analysis of Latin America and East Asia during the ninetiesusing bank-level data. J. Bank. Finance 32, 299–310.

Artha, I.K.D.S., de Haan, J., 2011. Labor market flexibility and the impact of the financial crisis. Kyklos 64, 213–230.Barrell, R., Davis, E.P., Karim, D., Liadze, I., 2010. Bank regulation, property prices and early warning systems for banking crises in

OECD countries. J. Bank. Finance 34, 2255–2264.Barrell, R., Davis, E.P., Karim, D., Liadze, I., 2011. How idiosyncratic are banking crises in OECD countries? Natl. Inst. Econ. Rev.

216 (May), R53–R58.

K. Kauko / Economic Systems 38 (2014) 289–308306

Basle Committee on Banking Supervision, 2010. Guidance for national authorities operating the countercyclical capital buffer.Bank Int. Settl., 27 pages.

Beck, T., Demirguc-Kunt, A., Levine, R., 2006. Bank concentration, competition and crises: first results. J. Bank. Finance 30, 1581–1603.

Berg, A., Pattillo, C., 1999a. Predicting currency crisis: the indicators approach and an alternative. J. Int. Money Finance 18, 561–586.

Berg, A., Pattillo, C., 1999b. What caused the Asian crisis: an early warning system approach. Econ. Notes 28, 285–334.Berkmen, S.P., Gelos, G., Rennhack, R., Walsh, J.P., 2012. The global financial crisis: explaining cross-country differences in the

output impact. J. Int. Money Finance 31, 42–59.Bordo, M.D., Meissner, C.M., 2012. Does inequality lead to a financial crisis? J. Int. Money Finance 31, 2147–2161.Borio, C., Drehmann, M., 2009. Assessing the risk of banking crises – revisited. BIS Q. Rev. (March), 29–46.Borio, C., Lowe, P., 2002. Assessing the risk of banking crises. BIS Q. Rev. (December), 43–54.Bunda, I., Ca’Zorzi, M., 2010. Signals from housing and lending booms. Emerg. Mark. Rev. 11, 1–20.Bussiere, M., Fratzscher, M., 2006. Towards a new early warning system of financial crises. J. Int. Money Finance 25, 953–973.Buyukkarabacak, B., Valev, N.T., 2010. The role of household and business credit in banking crises. J. Bank. Finance 34, 1247–

1256.Caballero, R.J., Krishnamurthy, A., 2009. Global imbalances and financial fragility. Am. Econ. Rev. Pap. Proc. 99, 584–588.Caprio Jr., G., Klingebiel, D., 1996a. Bank Insolvencies: Cross-country Experience. World Bank Policy Research Working Paper

1620.Caprio Jr., G., Klingebiel, D., 1996b. Bank insolvency: bad luck, bad policy or bad banking? In: Paper presented at the Annual

World Bank Conference on Development Economics 1996.Connor, G., Flavin, T., O’Kelly, B., 2012. The U.S. and Irish credit crises: their distinctive differences and common features. J. Int.

Money Finance 31, 60–79.Corsetti, G., Pesenti, P., Roubini, N., 1999. Paper tigers? A model of the Asian crisis. Eur. Econ. Rev. 43, 1211–1236.Davis, E.P., Karim, D., 2008a. Comparing early warning systems for banking crises. J. Financial Stab. 4, 89–120.Davis, E.P., Karim, D., 2008b. Could early warning systems have helped to predict the sub-prime crisis? Natl. Inst. Econ. Rev. 206,

35–47.Davis, P.E., Karim, D., Liadze, I., 2011. Should multivariate early warning systems for banking crises pool across regions?

Weltwirtschaftliches Archiv (Rev. World Econ.).147, 693–716.Demirguc-Kunt, A., Detragiache, E., 1998. The determinants of banking crises in developing and developed countries. IMF Staff

Pap. 45, 81–109.Demirguc-Kunt, A., Detragiache, E., 2000. Monitoring banking sector fragility: a multivariate logit approach. World Bank Econ.

Rev. 14, 287–307.Demirguc-Kunt, A., Detragiache, E., 2002. Does deposit insurance increase banking system stability? An empirical investigation.

J. Monet. Econ. 49, 1373–1406.Demirguc-Kunt, A., Detragiache, E., 2005. Cross-country empirical studies of systemic bank distress: a survey. Natl. Inst. Econ.

Rev. 192, 68–83.Diamond, D.W., Dybvig, P.H., 1983. Bank runs, deposit insurance and liquidity. J. Polit. Econ. 91, 401–419.Domac, I., Martinez Peria, M.S., 2003. Banking crises and exchange rate regimes: is there a link? J. Int. Econ. 61, 41–72.Drehmann, M., Borio, C., Tsatsaronis, K., 2011. Anchoring countercyclical capital buffers: the role of credit aggregates. Int. J. Cent.

Bank. 7, 189–240.Duttagupta, R., Cashin, P., 2011. Anatomy of banking crises in developing and emerging market countries. J. Int. Money Finance

30, 354–376.Eichengreen, B., Rose, A.K., Wypolsz, C., 2003a. Contagious currency crisis. In: Eichengreen (Eds.), Capital Flows and crises. MIT

Press, Cambridge, Massachusetts, pp. 155–185.Eichengreen, B., Rose, A.K., Wypolsz, C., 2003b. Exchange market mayhem: the antecedents and aftermath of speculative

attacks. In: Eichengreen (Eds.), Capital Flows and Crises. MIT Press, Cambridge, Massachusetts, pp. 99–154.Eichler, S., Karmann, A., Maltritz, D., 2011. The term structure of banking crisis risk in the United States: a market data based

compound option approach. J. Bank. Finance 35, 876–885.Evrensel, A.Y., 2008. Banking crisis and financial structure: a survival time analysis. Int. Rev. Econ. Finance 17, 589–602.Frankel, J.A., Rose, A.K., 1996. Currency crashes in emerging markets: an empirical treatment. J. Int. Econ. 41, 351–366.Frankel, J.A., Saravelos, G., 2010. Are Leading Indicators of Financial Crises Useful for Assessing Country Vulnerability? Evidence

from the 2008–2009 Global Crisis. NBER Working Paper Series 16047. .Friedman, M., Schwartz, A. Jacobson, 1963. A Monetary History of the United States 1867–1960. NBER, Princeton University

Press, New Jersey.Giannetti, M., 2007. Financial liberalization and banking crises: the role of capital inflows and lack of transparency. J. Financial

Intermed. 16, 32–63.Gomez, E., Rozo, S., 2008. Beyond bubbles: the role of asset prices in early-warning indicators. Ensayos sobre polıtica economica

56, 114–148.Gorton, G., 1988. Banking panics and business cycles. Oxf. Econ. Pap. 40, 751–781.Guo, F., Chen, C.R., Huang, Y.S., 2011. Markets contagion during financial crisis: a regime switching approach. Int. Rev. Econ.

Finance 20, 95–109.Hahm, J.-H., Shin, H.S., Shin, K., 2011. Non-core bank liabilities and financial vulnerability. In: Paper Presented at the FED and

JMCB:n Conference in September 2011.Haile, F., Pozo, S., 2008. Currency crisis contagion and indentification of transmission channels. Int. Rev. Econ. Finance 17, 572–

588.Herrala, R., 1999. Banking crises vs depositor crises: the era of the Finnish markka 1865–1998. Scand. Econ. Hist. Rev. 47, 5–22.Hoggarth, G., Reis, R., Saporta, V., 2002. Costs of banking system instability: some empirical evidence. J. Bank. Finance 26, 825–

855.Hoggarth, G., Jackson, P., Nier, E., 2005. Banking crises and the design of safety nets. J. Bank. Finance 29, 143–159.

K. Kauko / Economic Systems 38 (2014) 289–308 307

Hong, A., 2006. Is there a link between dollarization and banking crises? J. Int. Dev. 18, 1123–1135.Honohan, P., 1997. Banking System Failures in Developing and Transition Countries: Diagnosis and Prediction. BIS Working

Paper 39.Husain, A.M., Mody, A., Rogoff, K.S., 2005. Exchange rate regime durability and performance in developing versus advanced

countries. J. Monet. Econ. 52, 35–64.Hwang, Y.-N., 2012. Financial friction in an emerging economy. J. Int. Money Finance 31, 212–227.Jonung, L., Soderstrom, H.T., Stymne, J., 1996. Depression in the North—Boom and bust in Sweden and Finland 1985–93. Finn.

Econ. Pap. 9, 55–71.Jorda, O., Schularick, M., Taylor, A.M., 2011. Financial crises, credit booms and external imbalances: 140 years of lessons. IMF

Econ. Rev. 59, 340–378.Joyce, J.P., 2011. Financial globalization and banking crises in emerging markets. Open Econ. Rev. 22, 875–895.Kamin, S.B., DeMarco, L.P., 2012. How did a domestic housing slump turn into a global financial crisis? J. Int. Money Finance 31,

10–41.Kaminsky, G., Reinhart, C.M., 1998. Financial crisis in Asia and Latin America: then and now. In: The American Economic Review

88: Papers and Proceedings of the Hundred and Tenth Annual Meeting of the American Economic Association. pp. 444–448.Kaminsky, G.L., Reinhart, C.M., 1999. The twin crises: the causes of banking and balance-of-payments problems. Am. Econ. Rev.

89, 473–500.Kaminsky, G., Lizondo, S., Reinhart, C.M., 1998. Leading indicators of currency crises. IMF Staff Pap. 45, 1–48.Kauko, K., 2012. External deficits and non-performing loans in the recent financial crisis. Econ. Lett. 115, 196–199.Keeley, M.C., 1990. Deposit insurance, risk and market power in banking. Am. Econ. Rev. 80, 1183–1200.Kemme, D., Roy, S., 2012. Did the recent housing boom signal the global financial crisis? South. Econ. J. 78, 999–1018.Khan, A.H., Khan, H.A., Dewan, H., 2013. Central bank autonomy, legal institutions and banking crisis incidence. Int. J. Finance

Econ. 18, 51–73.Kiander, J., Vartia, P., 2011. Lessons from the crisis in Finland and Sweden. Empirica 38, 53–69.Kindleberger, C., 1978. Manias, Panics and Crashes. Basic Books, New York.Klomp, J., de Haan, J., 2009. Central bank independence and financial instability. J. Financial Stab. 5, 321–338.Kokko, A., 1999. Asienkrisen: manga likheter med den svenska krisen.(The Asian crisis, many similarities with the Swedish

crisis). Ekonomisk Debatt 27, 81–92.Laeven, L., Valencia, F., 2008. Systemic Banking Crises: A New Database. International Monetary Fund, WP/08/224.Lo Duca, M., Peltonen, T.A., 2013. Assessing systemic risks and predicting systemic events. J. Bank. Finance 37, 2183–2195.Marquering, W., Nisser, J., Valla, T., 2006. Disappearing anomalies: a dynamic analysis of the persistence of anomalies. App.

Financial Econ. 16, 291–302.Minsky, H.P., 1977. The financial instability hypothesis: an interpretation of Keynes and an alternative to ‘‘standard’’ theory.

Neb. J. Econ. Bus. 16, 5–16.Misati, R.N., Nyamongo, E.M., 2012. Financial liberalization, financial fragility and economic growth in Sub-Saharan Africa. J.

Financial Stab. 8, 150–160.Niemira, M.P., Saaty, T.L., 2004. An analytic network process model for financial-crisis forecasting. Int. J. Forecast. 20, 573–587.Noy, I., 2004. Financial liberalization, prudential supervision and the onset of banking crises. Emerg. Mark. Rev. 5, 341–359.Obstfeld, M., 2012. Does the current account still matter? Am. Econ. Rev. 102, 1–23.Patro, D.K., Qi, M., Sun, X., 2013. A simple indicator of systemic risk. J. Financial Stab. 9, 105–116.Quagliariello, M., 2008. Does macroeconomy affect bank stability? A review of the empirical evidence. J. Bank. Regul. 9, 102–

115.Ranciere, R., Tornell, A., Westermann, F., 2006. Decomposing the effects of financial liberalization: crises vs growth. J. Bank.

Finance 30, 3331–3348.Razin, A., Rosefielde, S., 2011. Currency and Financial Crises of the 1990s and 2000s. Centre for Economic Policy Research

Discussion Paper 8264.Reinhart, C.M., Reinhart, V.R., 2008. Capital Flow Bonanzas: An Encompassing View of the Past and Present. NBER Working Paper

Series 14321.Reinhart, C.M., Rogoff, K.S., 2008a. Is the 2007 US sub-prime crisis so different? An international historical comparison. In:

American Economic Review 98: Papers and Proceedings of the One Hundred Twentieth Annual Meeting of the AmericanEconomic Association. pp. 339–344.

Reinhart, C.M., Rogoff, K.S., 2008b. Banking Crises: An Equal Opportunities Menace. NBER Working Paper 14587.Reinhart, C.M., Rogoff, K.S., 2009. This Time is Different: Eight Centuries of Financial Folly. Princeton University Press, New

Jersey.Reinhart, C.M., Rogoff, K.S., 2010. From Financial Crash to Debt Crisis. NBER Working Paper Series 15795.Repullo, R., Saurina, J., 2011. The Countercyclical Capital Buffer of Basel. III: A Critical Assessment. CEPR Discussion Paper Series

8304.Rose, A.K., Spiegel, M.M., 2011. Cross-country causes and consequences of the crisis: an update. Eur. Econ. Rev. 55, 309–324.Rose, A.K., Spiegel, M.M., 2012. Cross-country causes and consequences of the 2008 crisis: early warning. Jpn. World Econ. 24, 1–

16.Roy, S., Kemme, D.M., 2011. What is really common in the run-up to banking crises? Econ. Lett. 113, 211–214.Roy, S., Kemme, D.M., 2012. Causes of banking crises: deregulation, credit booms and asset bubbles, then and now. Int. Rev.

Econ. Finance 24, 270–294.Sarlin, P., Peltonen, T.A., 2011. Mapping the State of Financial Stability. European Central Bank Working Paper 1382.Schularick, M., Taylor, A.M., 2012. Credit booms gone bust: monetary policy, leverage cycles and financial crises 1870–2008. Am.

Econ. Rev. 102, 1029–1061.Schwaab, B., Andre Lucas, A., Koopman, S.J., 2010. Systemic Risk Diagnostics. Duisenberg school of finance, Tinbergen Institute

Discussion Paper TI 10-104/DSF2.Singh, T.R., 2011. An ordered probit model of an early warning system for predicting financial crisis in India. IFC Bull. 34, 185–

201.

K. Kauko / Economic Systems 38 (2014) 289–308308

Solow, R.M., 2005. Foreword in: Charles P Kindleberger – Robert Z Aliber. In: Manias, Panics and Crashes. 5th ed. John Wiley &Sons Inc., New Jersey.

Staikouras, S.K., 2004. A chronicle of the banking and currency crises. Appl. Econ. Lett. 11, 873–878.Stanley, T.D., 2005. Beyond publication bias. J. Econ. Surv. 19, 309–345.van den Berg, J., Candelon, B., Urbain, J.-P., 2008. A cautious note on the use of pane models to predict financial crises. Econ. Lett.

101, 80–83.von Hagen, J., Ho, T.-K., 2007. Money market pressure and the determinants of banking crises. J. Money Credit Bank. 39, 1037–

1066.Wang, P., Wen, Y., 2012. Speculative bubbles and financial crises. Am. Econ. J.: Macroecon. 4, 184–221.Zhang, L., 2009. Bank Capital Regulation, the Lending Channel and Business Cycles. Deutsche Bundesbank Discussion Paper

Series 1; Economic Studies 33.