Summary
In our paper of interest, Reinhart and Kaminsky analyse the interdependency between banking and currency crises, in a relationship which has been dubbed the “Twin Crises”. The Mexican and Asian financial crises of the 90’s was a key motivation in their attempts to model the effects of banking-sector complications onto currency problems. Empirical literature on the matter is silent, and theoretical literature is weak, thus the authors sought to solve the ambiguous answers of former literature; in which direction does the causality of these two crises move? Studying 76 currency and 26 banking crises over 25 years, they aim to gauge whether the crises share a common macroeconomic background. Of further interest is; which crisis, if any, precedes the other? Will this worsen the effects of the crisis, causing a more severe recession than a crisis which occurs in isolation? Finally, does financial liberalisation fuel the boom phase of the business-cycle leading to a subsequent bust?
The preliminary analysis involves a comparison of conditional and unconditional probabilities to assess whether a crisis is more likely, knowing that the other has occurred. An analysis into the severity of crises highlights that twin crises are costlier than individual crises, accentuating the need for this understudied area to be explored. The core methodology is the signalling approach, which is the workhorse of many early warning systems adopted in the private and public sector to pre-empt crises. This approach involves monitoring the evolution of indicators which typically exhibit unusual behaviour prior to a crisis. When an indicator exceeds a certain threshold value, it is considered a "signal" of an impending crises.
The authors do not find a linkage between currency and banking crisis during the 1970’s, however this relationship comes to fruition in the 80’s following the liberalisation of financial markets. Banking-sector problems are found to predominantly predate currency crises, but the causal relationship is not unidirectional. Frequently the two crises share common causes, and which crisis surfaces first is a matter of circumstance. Furthermore, the collapse of the currency is found to deepen the banking crisis, activating a vicious spiral. In 18 of the 26 banking crises studied, the financial sector was liberalised within the previous five years. Thus, the probability of a banking crisis conditional on financial liberalisation is higher than its unconditional probability. The common origins of Twin Crises in the deregulation of the financial system and boom-bust cycles point to the need for tougher banking regulations.
Evaluation
Following a period of financial liberalisation in the 80’s, banking catastrophes quadrupled and subsequently became one of the main hindrances to a fixed rate regime whilst also amplifying the severity of balance-of-payment crises. Consequently, analysing the relationship between the two crises became increasingly topical in a phenomenon which became known as ‘Twin Crises’. Traditional literature on currency crises is ascribed to Krugman (1979) who focuses on the inconsistency of fiscal and monetary policies with the commitment to a fixed rate regime. The tranquillity of currency markets led literature in this field to stagnate until the Asian and Mexican Peso crisis brought around renewed attention. Subsequently, a new form of self-fulfilling models emerged, these second-generation models presumed crises are difficult to predict as they may occur amongst immaculate economic fundamentals, driven by the speculators expectations alone. Empirical work on the potential links between what they dub to be twin crises has been largely non-existent. Thus, the authors fill this void by examining currency and banking crises episodes for ‘small open economies’. It recognises what many others have overlooked – that countries experiencing currency crises often simultaneously encounter banking crises. Given that financial markets were headed for an era of great financial integration and capital mobility; twin crises were on the rise and this became an increasingly insightful piece of work.
This pioneering notion of twin crises was first explored by Kaminsky and Reinhart (1996) in the working version of this paper. They estimated a probit model where the binary measure of currency crisis was regressed against an index of banking crisis under alternative lag specifications. Similar to the revised paper, they found the occurrence of banking crises helps to predict currency crises. Beyond this, there has been a clear absence of empirical work, however some theoretical models address the connection. Some believe the causation runs from a currency crisis to a banking crisis. Stoker (1996) argues that an external shock which is incompatible with a fixed exchange rate will lead to loss of reserves that can escalate to a financial crisis if not sterilized, however Mishkin (1996) attributes the causation to ‘original sin’. In contrasting studies, Velasco (1987) and Diaz-Alejandro (1985) find the causality is reverse; a bank run can trigger a currency attack if the credit creation from the bailout of the banking system is inconsistent with a fixed exchange rate. Miller (1998a,b) argued that although a domestic banking crisis might cause a domestic currency crisis, or vice versa, this link can operate across international borders, pointing to the issue of contagion. The last family of contention is that both crises have common cause. Despite not providing a clear-cut answer as to the direction of the causality, the literature provides a multitude of economic indicators which offer insights into the causes of twin crises.
To resolve the ambiguous theoretical explanations, the links between the crises are examined empirically by constructing indexes to identify the beginning of each crisis. The currency market turbulence index is constructed as a weighted average of exchange-rate changes and reserve changes. However, given that currency crises are frequently resolved using contractionary monetary policy, the failure to incorporate interest rates is a clear deficiency of the index. The start of banking crises is identified using market events such as forced mergers, bank closures or government intervention. The disparity in the methods defining and identifying the two crises is a clear shortcoming. The subjective nature of an events-based definition is quite arbitrary and can lead to inconsistency; it only recognises crises which are severe enough to trigger market events such as bank closures but neglects crises which are pre-emptively prevented. To avoid dating crises too late, the authors specify the peak – when intervention or closures are at their highest. However, we felt this was an inadequate correction as even the peak can be mis-specified. Banking problems manifest themselves well before a run occurs, consider Lehman Brothers, it would be irrational to mark the start of the crisis on the day the bank shut up shop. Ideally an index should reflect a deterioration in asset quality, increased leverage and/or an increase in defaults which typically precede a bank crisis. To address this, Von Hagen and Ho (2013) develop an index of money market pressure which is the weighted average of changes in the ratio of reserves to deposits and changes in the short-term real interest rate. This is more appropriate as it recognises that a banking crisis is characterised by an increase in the banking sector's demand for central bank reserves.
The data sample covers “small open economies”, a classification which is quite misleading given that developing countries such as Bolivia and Peru are included alongside more developed countries like Spain and Norway. A key finding within the paper is that twin crises are more severe than crises which occur in isolation. However, examining developed countries alone may not reproduce these results given that crises are typically less severe and prolonged in developed nations. In a similar study Glick and Hutchinson (1999) perform signal-to-noise ratios and several probit regressions to analyse the existence of Twin Crises. However, they divide their sample into industrial, developing and emerging countries. Consequently, they found that Twin Crises occur predominantly in financially liberalised emerging economies, their results highlight the need for such a distinction. A further question mark is raised over the exclusion of the Exchange Rate Mechanism crisis from the sample, despite its reference in the introduction. We assume the crisis is excluded because it doesn’t classify as a “small” economy, but surely, it’s just as valid as Norway and including as much information as possible will lead to more accurate inferences.
The conditional and unconditional probabilities of individual and twin crises are compared to determine the direction of causality. The unconditional probabilities for currency and banking crises are 29% and 10% respectively. While these differences are mainly due to the higher frequency of currency crises during the sample perhaps an evenly split divide would garner different results. The probability of a currency crisis conditioned on the beginning of banking problems is 46%, much higher than the unconditional estimate, verifying the Krugman (1979) notion that bailouts contribute to currency crises or weak banking systems constrain the defence of the currency. While a currency crisis does not help predict a banking crisis, it helps predict the probability that the crisis will worsen, confirming that twin crises are more severe. These results emphasise the importance of this study, there is a clear link between the two crises, and given its severity it clearly needs addressed by policymakers.
Using a composite measure of average reserve losses and exchange rate depreciation, the authors gauge the severity of currency crises. Banking bailout costs are more than double in twin crises and reserve losses during a twin crisis are much higher than a standalone currency crisis. The use of the real exchange rate is an advantageous measure as it accounts for devaluation fuelled by inflation, however measuring reserve changes in a six-month window may not reflect all losses; countries can spend a long time protecting the currency in the FX market before the crisis ‘begins’. For example, according to Forbes, Venezuela foreign reserves have depleted from $30bn in 2013, to less than $10m today. Perhaps the biggest inadequacy is the timeframe in which Twin Crises are defined; it is marked as a currency crisis which follows the beginning of the banking crisis within the next 48 months. While this seems to be a benchmark among similar studies such as Glick and Hutchinson (1999) we felt this was too long of a timeframe and may generate spurious relationships.
The heart of the methodology is the signals approach which involves analysing the evolution of 16 macroeconomic and financial variables around crises. This assesses to what extent crises share common roots and which indicators best signal impending trouble. The key concept behind the approach is that the economy behaves differently on the eve of financial crises and this behaviour has a repetitive pattern. Early literature on this approach dates to Bilson (1979), however it took off again during the 90’s with a renewed focus. Perhaps most comparable to our paper is that of Kaminsky, Lizondo and Reinhart (1998) who develop an early warning system for currency crises. When variables deviate from tranquillity and exceed a threshold, they are considered as signalling a coming crisis. Similarly, our paper conducts ex-post analysis on key economic variables pre- and post- crisis, however it applies it to the area of Twin Crises.
Selecting financial liberalisation indicators is difficult, given its multifaceted nature, and while the study captures most manifestations, a better measure is Quinn’s (1997) international financial liberalization index. This measures the degree of a country’s restrictions on the flow of international finance using the IMF’s Annual Report of Exchange Arrangement and Exchange Restrictions. By incorporating capital account openness, current account openness, and international agreements we felt it was an all-encompassing measure. The list of 15 variables addressed isn’t exhaustive, perhaps most crucially, political variables are ignored. Political instability can exacerbate crises, consider the assassination of Mexican PRI presidential candidate in 1994. Incorporating the sovereign credit rating by Moody’s would be an ample proxy for political risk, an indicator incorporated in Reinhart, Goldstein and Kaminsky (2000). Currency crises often boil down to weak institutions and the sovereign credit rating will also account for that. The quadrupling of banking crises during 80’s is indicative of contagion, a concept noted by Eichengreen, Rose, and Wyplosz (1995). If contagion is unaccounted for domestic currency crises may be triggered by international banking crises and falsely attributed to domestic banking crises. Other omitted variables include capital flight and the budget deficit relative to GDP.
The methodology begins with a graphical analysis of the indicators, comparing the pre- and post- crises behaviour to average behaviour during tranquil periods. While this informs our intuition for the threshold test, one clear issue is that it does not distinguish between developing and developed nations. The developing nation is likely to be more severely impacted and take longer to respond than the developed nation, but this won’t be reflected.
The signalling approach involves monitoring the evolution of indicators that behave systematically different prior to a crisis. Each time an indicator exceeds a predetermined threshold, it is interpreted as a signal that a crisis may take occur. The conventional indicator methodology is employed by Diebold and Rudebusch (1989); however, this study is unique in that it has never been applied to analyse twin crises. Any signal given within the 24-month period before the beginning of the currency crisis is deemed a good signal. While any signal given within the 12-month period before or after the beginning of a banking crisis is labelled a good signal. This asymmetry of horizons is necessary to deal with the different timing of the peaks of both crises, however we felt it might unduly punish signals which tend to provide a premature warning before the window starts such as M2/reserves.
Selecting the optimal threshold involves a trade-off between Type I and Type II errors, i.e balancing false alarms and missing valid signals. Thus, they select the threshold value on an indicator- by-indicator basis whereby the noise-to-signal ratio is minimized. While we felt this was the most accurate means of selecting the threshold, a major shortcoming is that an indicator may give a valid signal, but if policy makers heed the signal and pre-empt a crisis, it is deemed a false alarm. Ideally, we could quantify an interventions-to-signal ratio, to signify if monetary authorities acted to prevent a crisis after noting the signal. A further criticism is that any signal within the window is treated the same irrespective of when the crisis occurs. Naturally, from the vantage point of policymakers an earlier signal is the more valuable one as they have more time at their disposal to put out the fire. Incorporating the average number of months prior to the crisis is one such solution, which is carried out in Kaminksy, Reinhart and Lizondo (1997) who analyse the average lead time of each indicator.
The capital account indicator comes out as most accurate in predicting crises, this is intuitive if one considers the use of reserves to defend the currency and the rise in interest rates following financial liberalisation. Second best is the real sector, which comprises of output and stock prices, however its predictive power is stronger in forecasting banking crises. _ Deciphering the extent to which crises are predictable offers policy makers potential early warning systems (EWS) to forecast crises ahead of time. We may not always have the privilege of hindsight, but we can use ex-post data to forebode future crises and prevent their manifestation, or even diminish their severity. Goldman Sachs have their own EWS known as GS-Watch which predicts the likelihood of a crises in three-months. Their measure incorporates some of Kaminksy and Reinhart’s most successful indicators mentioned above, namely: reserves, stock market prices and exchange rate depreciation.
Prior to crises, economies are distinctly weak, with a high proportion of the indicators signalling. This suggests that most crises have weak and deteriorating economic fundamentals at their core. The authors argue these results imply that crises cannot be defined as self-fulfilling. However, we felt this ignores the fact that crises can be self-fulfilling in the sense that investors’ expectations worsen an already deteriorating set of fundamentals. Capital flight may be the final nail in the coffin of an ailing exchange rate, a crisis that may not have come to fruition if speculation was restrained.
One main criticism of the paper is the generic treatment of every currency crises, in a “one size fits all” approach. First-generation models, inspired by Krugman, focus on the incompatibility of fiscal and monetary policies with a fixed exchange rate, while second-generation models emphasis self-fulfilment. The emergence of the “Tequila” and South East Asian crisis represented a new breed of crisis, those fuelled by contagion. This highlights the fact that currency crises can manifest themselves in multiple forms and this should be accounted for example, those powered by fiscal deficits, self-fulfilment or contagion. Accounting for different kinds of currency crises would provide more granularity on their relationship with banking crises. Some currency crises might be more linked to banking crises than others, whereas this paper would lead you to believe all forms of currency crises are equally linked to banking crises. Although Kaminsky (2006) doesn’t examine the link between banking and currency crises, she acknowledges the different manifestations of currency crises and her investigations found there to be six varieties of currency crises.
Considering the previous belief that there are different breeds of currency crises, we feel that there needs to be a separation of developing and developed. The need for this distinction is because developed and developing countries react differently to crises and they are typically affected by different forms of currency crises. For example, Kaminsky (2006), notes that currency crises in the form of financial excesses, are more frequent for developing economies whereas self-fulfilling crises only affect mature economies.
A neglected issue within this study, and existing literature is the interaction of currency crises with a third form of financial crises; sovereign debt crises. Debt and currency crises may be correlated due to common causes or due to direct links between the two. Consider a fall in GDP, this may lead to a budget deficit and increase the probability of default. Likewise, a fall in GDP may lead the government to devalue the currency to jump start the economy. This concept of this triangular relationship, adding debt crises to affair appears has rarely have been explored. In exception, Bauer et al. (2007) contend that twin debt and currency crises should be regarded as a specific type of crisis and analysed separately from pure currency and pure debt crises Such a distinction may be important to gauge the likelihood of the crises and its severity
Conclusively, the work of Kaminsky and Reinhart (1999) represents a monumental insight into the much-neglected field of Twin Crises. While the signalling approach could be bettered methodically through a more apt identification of the beginning of crises and the shortening of the Twin Crises time window, the concept is intuitive, and its success has been demonstrated by its usage in private and public institutions such as Goldman Sachs and the IMF. Further research into the signalling approach may lead us to achieve what many deem to be an elusive goal, the ability to forecast crises.