The Icelandic Crisis:
A study towards leading indicators of financial crises
Key words: Financial crisis, leading indicators, Iceland JEL Classification: E44, E58, F30, F34, G21
Master thesis by Jorn D. Douwstra1 Supervised by Prof. Dr. K.W.H. Knot
MSc Economics
University of Groningen, The Netherlands Faculty of Economics and Business
August 2009
Abstract
Content
I. Introduction ... 5
II. Theoretical Background ... 8
2.1 Financial crises... 8
2.1.1 Currency crises... 9
2.1.2 Banking crises... 11
2.1.3 Twin crises ... 11
2.1.4 Foreign debt crises ... 13
2.1.5 Connection between foreign debt and currency crises ... 14
2.1.6 Systemic crises... 14
2.1.7 Common origins and consequences... 16
2.2 Leading Indicators ... 17
2.2.1 Currency crisis indicators ... 17
2.2.2 Banking crisis indicators... 21
2.2.3 Foreign debt crises ... 24
2.2.4 Leading indicators considered ... 25
III. Icelandic crisis ... 29
3.1 Overview... 29
3.1.1 Credit Rating Agencies ... 30
3.1.2 Credit default swaps... 32
3.1.3 Overextended economy ... 33
3.1.5 Effects of the international financial crisis on Iceland... 35
3.2 Graphical data analysis ... 35
3.2.1 Multiple crisis indicators... 36
3.2.2 Banking crisis indicators... 39
3.2.3 Currency crisis indicators ... 41
3.2.4 Foreign debt crisis indicators ... 42
3.3 Empirical data analysis ... 44
3.3.1 Stationarity tests... 44
IV. Policy before and during the crisis ... 53
VI. Conclusions/limitations ... 55
References... 57
Appendix... 63
Appendix 1: Various types of crises defined ... 63
Appendix 2: Leading indicators of currency crises ... 64
Appendix 3: Data ... 65
Appendix 4: Results of the stationarity tests... 66
Appendix 5: Adjusted variables... 67
Appendix 6: Time intervals available per indicator... 68
Appendix 7: Sensitivity analysis regarding only 2008 as a crisis in Iceland... 69
Appendix 8: Summary of logit estimations for all unadjusted variables... 70
I. Introduction
The topic of financial crises (severe disruptions of financial markets) has become increasingly interesting again in the light of recent developments on world's financial markets. From July 2007 onwards, the financial world saw a spiking TED-spread2 (Bloomberg), the first bank run in 140 years in Great Britain (The Economist, 2007) and a takeover of the three large Icelandic banks (FME, 2008a, b, c) by the Icelandic Financial Supervisory Authority3. The problems in Iceland were noticed by several Dutch newspapers which reported about the possible bankruptcy of Iceland (Groot, de, 2008) on October the 7th, 2008, as well as “panic in Iceland” which was put on the front page of one of the leading newspapers (Kalse, 2008).
The effects of the financial crisis in Iceland on other countries can be illustrated by the Icesave case4 in The Netherlands. Icesave entered the Dutch market for saving deposits in May 2008 (Hofs, 2008). Landesbanki was in need of more liquidity since the credit crunch which became apparent in August 2007 (Elliot, 2008). Savers were attracted to Icesave by the high interest rates it offered and Icesave grew to 120.000 customers in the mere four months they were active on the Dutch market. Icesave raised an estimated total of 1.6 billion Euros of savings from Dutch customers from May to October 2008 (NRC, 2008a). The effects on the Dutch public became more and more clear when money transfers from the Icesave savings account failed, the office and call center in Amsterdam were closed (Witt Wijnen, de, 2008) and the newspaper questioned the statement by the Dutch Minister of Finance that all Dutch saving deposits in Icesave could be guaranteed (NRC, 2008b). Matters turned for the worse when the Icelandic Prime Minister Haarde stated that the severe banking problems could well lead the country into a national bankruptcy (NRC, 2008c).
2
The TED-spread is an indicator of lack of trust in the financial market, and marks the difference between the interest rate on three month US treasury bills and LIBOR (Krugman, 2008). See Bloomberg for TED-spread charts.
3
This is abbreviated to FME. 4
One of the possible causes of the near bankruptcy of Iceland was the large amount of external debt the country accumulated. The three main Icelandic banks (Glitnir, Kaupting and Landesbanki) made outstanding loans which accumulated up to nine times the size of their economy (measured by GDP). Four-fifth of this external debt was denoted in foreign currency (Economist, The, 2008b). Debt problems can arise if income of the debtor is insufficient or the debtor cannot meet its debt obligations due to a lack of sufficient assets (IMF, 2003). These problems are deepened when a considerable amount of this debt is denoted in foreign currency and the exchange rate falls, because the debt becomes more expensive when transferred in the domestic currency. The Economist (2008a, b) noted that the liquidity of Icelandic banks dried up because they could not refinance their debts. Matters turned for the worse when the Icelandic central bank failed to act as a lender of last resort, partly since their small economy was uniquely overextended by having too big a banking industry relative to the size of their economy. Thereby leading to the major financial crisis in which the Financial Supervisory Authority of Iceland (FME, 2008a, b and c) was compelled to take over Glitnir, Kaupthing and Landesbanki.
Although each financial crisis has their unique features, they also share some striking similarities; run-up of asset prices; debt accumulation; growth patterns; and current account deficits (Reinhart and Rogoff, 2008c). Reinhart and Rogoff place the current U.S. sub-prime crisis in a historical context and a quick comparison shows that, at first sight, the Icelandic crisis has similarities with historical crises. However, Iceland was the first industrialized country to request support from the IMF in over thirty years (Danielson, 2008).
In order to answer these questions, studies of previous financial crises are reviewed (for instance the research performed by Kaminsky (1999), Obstfeld (1996) and Von Hagen and Ho (2007)) to determine possible leading indicators that could predict an upcoming crisis. These possible indicators are investigated for the specific Icelandic situation in order to conduct an analysis of the predictability of the crisis. An overview will be given of the events that occurred in the run-up to the crisis and various linkages will be described. Such as the impact from the crisis in other countries on Iceland and which role the Icelandic supervision played in the crisis.
II. Theoretical Background
The theoretical background to financial crises is discussed in this chapter and from this research follows a study of various indicators that could predict an upcoming crisis. The first paragraph deals with financial crises in general and the second paragraph discusses the various leading indicators.
2.1 Financial crises
A comprehensive explanation of financial crises is provided by Mishkin (1992); he defines a financial crisis as:
“a disruption to financial markets in which adverse selection and moral hazard problems become much worse, so that financial markets are unable to efficiently channel funds to those who have the most productive investment opportunities.”
Mishkin indicates that a financial crisis can dispel the economy from high output equilibrium with high performing financial markets to an equilibrium of sharply declining outputs, as a result of the inability of financial markets to channel the funding efficiently.
which he has got the loan. The cost of failure is shifted to the lender while the borrower will receive the higher pay-off in the case of a successful project due to limited liability.
The IMF (1998) distinguishes several forms of financial crises. These are; banking crises; currency crises; foreign debt crises; and systemic financial crises. These various forms are covered below and the definitions are summarized in appendix 1.
2.1.1 Currency crises
When a speculative attack on the exchange rate of the currency results in a devaluation of this currency, is it called a currency crisis (IMF, 1998). Note that, a devaluation of the currency is not the only possible reaction to a speculative attack. The national authorities can also intervene in order to defend the currency.
Obstfeld (1994) showed that speculative attacks on a currency peg could lead to a crisis originating as a self-fulfilling prophecy. He relates the EMS-crisis5 in the early 1990s to the theory described by Krugman (1979). The model developed by Krugman describes how a currency peg must be abandoned once the international reserves of the pegging country are depleted. Krugman finds that speculators with foresight attack the currency inevitable before the reserves are depleted, which leads to a buy up of the remaining reserves by these speculators. This Krugman type of model is sometimes referred to as a first generation model of currency crises (Dreher et al., 2006).
Because the EMS countries had easy access to world capital markets where they could borrow, a depletion of their reserves was not of major concern. Obstfeld therefore argues that effects like high interest rates and unemployment were perhaps more important to these governments, and that these effects played a role in the decision making process to
5
abandon the peg. He states that: “In a setting of purposeful action by the authorities, however, the possibility of self-fulfilling crises cannot be easily dismissed.” He comes to this statement by observing a “circular dynamic” in the market in which speculative anticipations depend on assumptions of government responses which again depend on how price changes (which are also influenced by expectations) affect the economic and political situation of the government. This “circular dynamic” is exactly why crises could occur because agents expect them to occur and would not have happened otherwise, hence the self-fulfilling prophecy.
2.1.2 Banking crises
According to the IMF (1998) a banking crisis is when bank runs (either real or potential), or bank failures, cause banks to suspend the convertibility of their liabilities or makes it necessary for the government to intervene in an attempt to prevent this by extending large scale assistance. Demirgüç – Kunt and Detragiache (1997) discuss reasons why banks are prone to crises. Banks are financial intermediaries that transform maturities since they hold long-term loans to businesses and consumers as assets and have short-term deposits as liabilities. Shareholder pressure could drive banks to widen the maturity gap in an upsurge in search for more profit (Gerlach et al., 2009), which leads to increased financing difficulties in economic downturns. Banks will become insolvent when the value of the bank’s assets drops below the value of their liabilities, because borrowers are unable or unwilling to repay their debt. This could create a bank-run when all depositors want to withdraw their money at the same time.
2.1.3 Twin crises
Crises do not stand alone but often show linkages between them, as pointed out by Kaminsky and Reinhart (1999) as they call banking crises and currency crises to be “twin crises”. They point out that many of the countries that have had a currency crisis also had a banking crisis around the same time. They name for instance the Thai, Indonesian, Korean, Finnish, Norwegian, and Swedish crisis6. Kaminsky and Reinhart find that the linkages between currency and banking crises were more often to be found from the 1980s onwards, going hand in hand with the liberalization of the financial markets. They have observed as well that a banking crisis often begins before the currency crisis, but that the collapse of a currency often deepens the banking problems as the banking crisis will reach it peak after the currency crisis.
6
However, Kaminsky and Reinhart point out that the banking problems are not the immediate cause of the currency problems, but that both types of crises face common causes. Both types tend to be preceded by below normal economic growth, an overvalued exchange rate and a rising cost of credit. Also a shock to the financial system, for instance one caused by financial liberalization, provides more access to leverage. The possibility of the occurrence of a currency crisis is increased when a country is faced with a banking crisis. The vulnerability also increases as the liabilities of the banking system rise substantially, especially when they are not supported by a rise in international reserves of the country. It is shown that these crises tend to be preceded by deteriorating economic fundamentals. These fundamentals were worse in the case of a twin crisis.
Different varieties of models have been developed to explain the linkages between the “twin crises”. One of the models Kaminsky and Reinhart (1999) refer to is developed by Velasco (1987), who describes how a currency crisis can be a reaction to problems in the banking sector. This can happen when a central bank needs to bail out financial institutions that have run into trouble and finances this by seigniorage, from which a currency crash via excessive money creation is induced. Kaminsky and Reinhart (1999) also point to another model which is developed by Stoker (1994) who indicates how problems develop from a currency problem to a banking sector problem. He argues that the problems begin with a loss of reserves as a direct effect from an external shock, such as an increase in foreign interest rates. Whilst at the same time the country has a fixed exchange rate. Since a higher foreign interest rate will make the home country less attractive for investors (because they will receive a relative lower yield on capital). As well as putting a downward pressure on the exchange rate (as investors will sell the home currency). This will usually lead to a buy up of the domestic currency by the authorities, and thereby depleting their international reserves.7 If this effect cannot be offset by the authorities could it lead to a credit crisis (due to the deterioration of the central banks position), more bankruptcies (less possibilities for borrowing), and financial crisis. Another problem that could lead to these crises is the currency mismatch that arises when most of the external liabilities of the country are denominated in foreign currency and its
7
assets are denominated in domestic currency. This makes the country unable to hedge the risk of exchange rate fluctuations according to Eichengreen and Hausmann (1999), which is one part of their “original sin” hypothesis. An eroding real value of external debt arises when devaluation occurs of the domestic currency when the country would be able to generate the external debt in their own currency. Therefore Eichengreen and Hausmann explain that foreigners could be unwilling to lend in a denomination that the borrower can manipulate unless compensation is given to such an extent that only those borrowers that plan to devaluate are prepared to pay.
2.1.4 Foreign debt crises
confidence in the country can also be considered as a start of a debt crisis. This broader indicator also notices turbulence in bond markets; therefore countries do not have to go into full default before their problems are registered as a debt crisis.
2.1.5 Connection between foreign debt and currency crises
Some argue that interrelation exists between foreign debt crises and currency crises. Both debt crises and currency crises can originate from common origins (Dreher et al., 2006). A negative demand shock can for instance leads to a breakdown in real economic activity, and creates market pressure on the exchange rate to devaluate, as there will be less demand for the domestic currency. When governments have the incentive to abandon the currency peg and fight recession by monetary expansion, rational economic agents will withdraw capital in the anticipation of the devaluation, which would make their investments less profitable. This withdrawal leads to further pressure to devaluate, as more domestic currency will be offered since the agents will want to convert it back in their own currency. As availability of capital will be reduced, the government will need to borrow at a higher interest rate which increases the risk for debt default, at a moment in which the economy is already weakened by the economic downturn. A rising world interest rate could also trigger both a debt and a currency crisis, when capital is more expensive and will be drawn away from the country. However, Dreher et al. (2006) only find weak evidence that currency and debt crises arise from common causes. The only variable that proved a significant effect simultaneously on both types of crises was public debt.
2.1.6 Systemic crises
“where the release of “bad news” about a financial institution or crash of a financial market leads in a sequential fashion to considerable adverse effects on one or several other financial institutions or markets, e.g. their failure or crash.”
De Bandt and Hartmann point out that the essential concept is that there will be a so-called “domino effect” in contrast to an idiosyncratic shock, which is an isolated shock. De Bandt and Hartmann expand the definition of a systemic event which leads to the definition of a systemic crisis:
“a systemic event that affects a considerable number of financial institutions or markets in a strong sense, thereby severely impairing the general well-functioning of the financial system.”
A narrow shock is thus different from a broad shock in the initial impact level, which affects more institutions in a broad event. A credit crunch is an example of a systemic financial crisis as is affects a lot of institutions at the same time.
2.1.7 Common origins and consequences
These various forms of crises discussed above can originate from common underlying problems as described by the IMF (1998). These problems could be, for instance, a buildup of unsustainable economic imbalances and misalignments in exchange rates or asset prices. These problems often occur in a context of structural rigidities and financial sector distortions. Economic imbalances do not always have to lead to a crisis, but when it does happen, it is often caused by a sudden loss of confidence in the banking system or the currency. This loss of confidence arises from the exposure of economic and financial weaknesses, uncovered by a sudden correction in asset prices or by disruption to credit or external flows. The vulnerability to a crisis of an economy depends on the magnitude of the crisis, the credibility of policies to correct imbalances and the robustness of the financial system.
The IMF further showed that consequences of financial crises of all sorts are usually characterized by sharp declines in asset prices, failures of financial institutions and failures of non-financial organizations. Financial crises can be very costly, not only in the negative effect on economy activity or because the financial markets cannot function effectively, but also in fiscal costs of restructuring the financial sector. The costs of resolving the financial crisis can exceed ten percent of GDP in emerging market countries. These costs are less high in industrialized countries but are still considerable, e.g. five percent for the United States.8 The IMF compiled a study towards the effects on GDP of currency and banking crises. For currency crises, average output growth returned to the trend in approximately one and a half years and the loss in output growth relative to the trend was four and a half percentage points for emerging and developed countries together, these numbers increase for severe currency crises to a recovery time of two and a quarter years and a loss in output growth of seven percentage points. For banking crises, the average recovery time was three years and the cumulative loss in output growth was eleven and a half percentage points.9
8
See table 14 in IMF, 1998. 9
Reinhart and Rogoff (2009) showed that the consequences of financial crisis have three main characteristics in common. First, the collapse of the asset market is often deep and lengthened; they show that real housing prices decline an average of 35 percent over six years and a collapse of equity prices of 55 percent of three and a half years. Second, output and employment will decline severely. Output falls with an average of nine percent for two years and the unemployment rate rises on average of seven percentage points. The third shared distinctive feature is a huge rise in government debt, averaging at 86 percent. With this study, Reinhart and Rogoff show that financial crises leave large marks on society. The next section deals with leading indicators to predict crises, so that measures could be taken in order to prevent them.
2.2 Leading Indicators
2.2.1 Currency crisis indicators
Kaminsky (1999) argues that the vulnerability of an economy makes a very useful indicator for a possible currency crisis as most currency crises happen in fragile economies, with different sectors of the economy already in distress. Because of the existence of the twin crises it is also possible to assess the vulnerability of the banking sector in order to predict a currency crash. With this in mind Kaminsky describes techniques to indicate the degree of vulnerability of the economy using leading indicators and she constructs different baskets of indicators. One of the main results of her study is that:
“crises have developed in the midst of multiple economic problems with basically no crises occurring following a unique bad shock.”
In this light and in the spirit of Eichengreen et al (1995), Kaminsky and Reinhart (1999) find that turbulence in the currency market is reflected in multiple ways. When a central bank tries to fight off a speculative attack on the currency, it will usually raise interest rates and buy up domestic currency. A loss of foreign reserves and a rise in interest rates can point at currency market turbulence. Another option, as pointed out before, is to devalue the currency or abandon the peg, in either way the exchange rate will change. Thus a basket of indicators that serves as index of currency market turbulence needs to reflect these different variables.
A large group of empirical studies on the different indicators used to identify currency crises is reviewed by Kaminsky et al. (1997). They have assessed a grand total of 105 indicators, subdivided in six broad categories. Their discussion focuses on papers in which “indicators were used to estimate the probability of a crisis” or “indicator’s behavior was systematically compared with its behavior in a control group” or “the indicator’s ability for signaling future crises was systematically assessed in quantitative terms.” They derive some main conclusions from the analysis of the different studies. Since currency crises are often lead by a wide variety of economic disturbances, is it rational to also use a wide variety of variables as an indicator, will the warning system be effective.
preceded by a deterioration of the trade balance, because expansionary fiscal and credit policies lead to higher demand for goods and thereby imports. The appreciation of the currency before the crisis leads to a decrease in exports.
The main variables that proof to be good indicators as concluded by Kaminsky et al. (1997) are: international reserves, real exchange rate, domestic credit, credit to the public sector, and domestic inflation. Other useful indicators are trade balance, export performance, money growth, M2-to-international reserves, real GDP growth, and the fiscal deficit. Other variables that give a significant prediction only arise in a very narrow selection of studies and are therefore not reliable enough. The first signal from most of the previously mentioned indicators occurs between 24 and 12 months prior to the crisis, which indicates the forecasting abilities of these variables. Remarkable is that it turned out that the current account balance is not useful as a forecasting indicator; neither did the external debt profile prove to be useful. From this analysis Kaminsky et al. (1997) construct a “signals” approach in which the probability of a crisis is conditional on the signal that the indicator gives. A warning is registered when a variable deviates from its normal level beyond a certain threshold. This can either be a true warning (a crisis is at hand) or noise (false signals). It is less advisable to use the variable when the number of noise signals increases.
Kaminsky (1999) develops a method to assess how vulnerable the economy is by computing the number of signs that indicate an economy in distress. She studies the distribution of the various indicators to define the critical cut-off point, the threshold beyond which level a vulnerable economy is indicated by the variable. She defines the signaling state as equation (1):
}
,
{
}
1
{
tj j t j t j tS
X
X
S
=
=
>
(1)indicator could have a positive or negative sign depending on the variable. Two major problems can arise while selecting the threshold, namely one that does not indicate mild crises and the other that gives too many noise signals (indicating non-existing crises). Kaminsky also defines a “noise-to-signal” ratio which indicates how accurate a certain threshold of an indicator functions. When the ratio is close to zero, the threshold is optimized. The indicators that provide the best noise-to-signal ratio with their corresponding thresholds are summarized in appendix 2. Kaminsky explains that indicators with a ratio above unity are not useful and can be dropped from the sample10. The thresholds in appendix 2 are extracted from the study by Kaminsky and Reinhart (1999).
The comprehensive overview of early warning signals by the IMF (1998) gives another view on these indicators. The study indicates that a large number of economic indicators are suggestive for crises but could not be used as an effective early warning system. A couple of reasons are mentioned: “statistical significance of the differences between tranquil and crisis times has not been established”, “an early warning system should signal well in advance and some variables signal vulnerability when the crisis has already started”, and “information about the behavior of the economic variables are available only with a delay which is too long to serve as an effective indicator.” The IMF concludes that only a few variables are effective as a leading indicator. The real exchange rate tends to be relatively appreciated to its norm before a crisis and devaluated during and after a crisis. The domestic credit and the M2-to-reserves ratio rose substantially in the months prior to a crisis and plummeted during the crisis. These three variables are the regarded by the IMF as effectively assessing the vulnerability of the economy when they are consequently above average levels.
10
2.2.2 Banking crisis indicators
The IMF (1998) notices that the construction of leading indicators of banking crises are hampered by the lack of high frequency data that could mark the beginning of a banking crisis. However, a study towards the empirical determinants of banking crises has been adopted by Von Hagen and Ho (2007). They point out that recent studies have shown that there are three main groups of determinants; domestic macroeconomic disturbances; shocks from the external sector; and institutional factors. The authors reach the conclusion that poor domestic macroeconomic policies could lead to a banking crises induced through large monetary or fiscal expansions which lead to large credit expansions and result in subsequent contractions to keep the inflation controllable. Another option, as stressed by Goldstein et al. (2000) is that an overvaluation of the exchange rate is the best leading indicator for banking crises. Loss of international competitiveness of domestic industries can lead to problems in the banking sector, when the industries become insolvent when sales collapse.
will arise, because of a decline in the quality of bank loans or an increase in non-performing loans. The second reason is that a large removal of deposits by the general public compels the banks to refinance themselves. The third explanation is that the inter-bank lending dries up because inter-banks prefer to hold safer assets. The central inter-bank can react to this increased demand, according to Von Hagen and Ho, by raising the short-term interest rate, if bank reserves are at the operating target. However it can also provide more liquidity in the market through open market operations or discount window lending.11
Following from the reasoning above Von Hagen and Ho develop an “index of money market pressure” (called IMP), since a banking crisis is either characterized by a large increase in the short-term interest rate or by an increase in the volume of central bank reserves, or both. IMP is defined below in equation (2) as:
γ γ
σ
σ
γ
∆ ∆∆
+
∆
=
t t tr
IMP
(2)Von Hagen and Ho explain that this index of money market pressure is made up from the weighted average of changes in the ratio of reserves to bank deposits and changes in the short-term interest rate. The reserves to bank deposits ratio is denoted by γ, which is the ratio of total reserves of the banking system to total non-bank deposits in the bank sector. This ratio will increase in turbulent times, since either the depositors will withdraw their funds or central banks will make new reserves available to the banks. Moreover ∆ indicates the difference in the variables between periods and σ denotes the standard deviation of the different components. Von Hagen and Ho further define a banking crisis, using this indicator, when the IMP exceeds the 98.5 percentile of the sample distribution of the IMP for the country under consideration and the increase in the IMP needs to be at least 5% compared to the previous period.
11
For examining banking crises, while using the IMP, a few problems may arise, as the authors have noticed. The IMP only identifies banking crises of a reasonable size, since they need to be large enough to increase the aggregate demand from central bank reserves.
Studies performed by the IMF (1998) and Drees and Pazarbasioglu (1995) have shown that banking crises are often preceded by financial liberalization. According to the IMF is this liberalization is reflected by an increasing ratio of broad to narrow money12, growing deposits and high real interest rates. The IMF also showed that there is some evidence that banking crises have often occurred in countries where the system was subjected to excessive government influence, or where liberalization took place before regulations and supervisory arrangements were adequately implemented. Drees and Pazarbasioglu suggest that a crisis following a financial deregulation does not necessarily imply that the crisis was caused by the liberalization process. But they do point out that financial deregulation expands the possibilities of borrowers and lenders to engage in risk taking. A severe shock to the system, which was the financial liberalization in the previously heavy regulated Nordic countries, can trigger major changes, such as a surge in lending by banks that were available to more sources of funding. The liberalization also weakened the bank’s ability to assess credit risks and monitor borrowers. Together with insufficient measures taken by the authorities to minimize the adjustment costs after the deregulation and failing authorities in creating adequate supervisory and tightening prudential bank regulation, this lead to a magnifying impact of the negative shock to the financial system. The stability of the system was thereby put at risk and contributed to the banking crisis that emerged.
12
2.2.3 Foreign debt crises
Roubini (2001) showed that it can be difficult to distinguish between insolvency and illiquidity. He indicates that it is possible to use a range of criteria to assess the solvency of a country and thereby the risk to a foreign debt crisis. Traditional indicators of debt sustainability are widely discussed and Roubini summarizes them for both the external debt (external debt-to-GDP, external debt-to-export, debt service-to-GDP, and debt service-to-export) as well as the public debt (public GDP, public debt-to-government revenues, debt service-to-GDP, and debt service-to-debt-to-government revenues). He also points out that an analysis of historical data of macroeconomic variables (such as GDP growth, trade balance and interest rates) can indicate if the country is likely to service its debt obligations. A significant shortfall of available resources in relation to payments due can indicate a solvency problem. A method that indicates if the debt burden is sustainable is to look at the discounted value of debt, by which Roubini means the discounted value of the expected cash flows regarding the debt service in both the amortization of the principal and the coupon rate. However, mispricing problems can arise due to panic or that the market is unable to accurately assess the country's risk.
Amadou (2007) also show, using extreme value theory13 as well as kernel density estimations,14 that when the secondary-market bond spreads are 1,000 basis points (bps)15 higher than normal, it can be described as a foreign debt crisis.
A comprehensive study into various debt indicators has been put up by Carlione and Trebeschi (2006) in which they develop an early warning system for debt crises. They use an extended range of indicators and show that the variables that measure the external debt, the service of this debt and short-term debt ratios increase considerably in the periods preceding the crisis. On the other hand, variables that indicate revenues, such as exports and private capital flows, deteriorate, as well as the international reserves held by the country. They used a multinomial logit approach16 for the analysis of defining significant indicators, which is based on the work by Bussiere and Fratzscher (2002) who conduct this analysis for currency crises. Three different states are defined in the analysis; tranquil (no crisis will occur in the coming two years); pre-crisis (a crisis will erupt in the next two years); and adjustment periods (the crisis has already started). When the multinomial logit approach has been conducted, only eight variables (out of a 28 at the starting point) remain as indicators which are debt-to-exports ratio, federal funds rate, interest payments on external debt-to-international reserves, real GDP growth, short term debt-to-total external debt ratio, total private capital flows on GDP, annual inflation, and the international reserves-to-total external debt ratio.
2.2.4 Leading indicators considered
A lot of different indicators have been covered in the previous sections. To keep this information manageable is it convenient to distinguish the variables here which will be used in the next chapter. Multiple studies have shown (such as Kaminsky, 1999) that a basket of indicators often proofs to give the best results; therefore a wide range of
13
The extreme value theory is an statistical analysis that deals with extreme deviations from the median of a distribution (see De Haan and Ferreira, 2006).
14
The Kernel density estimation is a method used for estimating the probability density function of a variable (see Scott, 1992).
15
This corresponds to 10 percentage points. 16
variables will be used in the analysis. The most variables that will be used in the next section are the results of multiple studies mentioned before. However, some variables are omitted from the analysis due to problems with data availability.
The first set of variables that is discussed could serve as leading indicators for multiple crises. A decrease in export performance (exports is divided by GDP here) is indicated by Kaminsky et al. (1997) and Carlione and Trebeschi (2006) to serve as both a currency as well as a foreign debt crises indicator because this indicates a decrease in revenue for the country. These authors also point out that a rise in inflation could indicate these crises. An increase in the interest rate could indicate a banking, currency or foreign debt crisis according to the IMF (1998), Kaminsky and Reinhart (1999), and Roubini (2001). A decrease in international reserves could indicate a currency crisis according to Kaminsky and Reinhart (1999), Von Hagen and Ho (2007) highlight that this could also indicate a banking crisis. An increase in the exchange rate could also indicate a currency or a banking crisis (IMF, 1998 and Kaminsky and Reinhart, 1999), this indicates a depreciation since the real exchange rate in this study is expressed as amount of domestic currency per unit of foreign currency. A decrease in real GDP growth could indicate both a currency and a foreign debt crisis (Carlione and Trebeschi, 2006 and Kaminsky et al., 1997). Kaminsky et al. (1997) have shown that a decrease in the trade balance is also a possible indicator of a currency crisis since this also leads to reduced income for the country. Roubini highlights that the trade balance deterioration could also indicate a foreign debt crisis.
Kaminsky et al. (1997) have researched various variables that could indicate an upcoming currency crisis, most of them are already covered in the section above that deals with multiple crises indicators. A few variables that are specific currency crisis indicators remain. However, not all these variables are considered in the analysis in the next section data problems. The two variables that are available for the period that covers the 2008 Icelandic crisis are the M2-to-reserves ratio and the money growth. Both variables should be increasing since Kaminsky et al. (1997) explain that money is created to support the economy and fight of currency attacks. The variables that could not be retrieved are the domestic credit, credit to public sector and fiscal deficit17.
In the case of foreign debt crises Roubini (2001) suggests to use the debt-to-GDP and debt to-GDP ratios for external and public debt; debt-to-export and debt service-to-export ratios solely for external debt; and debt-to-government revenues and debt service-to-government revenues ratios for public debt. The data for debt service (both external and public) was not available and thus are these variables also omitted from the analysis. Carlione and Trebeschi (2006) also recommend using the external debt ratios but also point at exports, private capital flows, and international reserves. Private capital flows was also not available and is therefore not analyzed. They further recommend looking at the short term debt-to-external debt ratio. All these variables are summarized in table 1 below where the columns define the variable, the direction it is expected to reveal (in extreme values) to indicate market turbulence, and at which crises they indicate.
17
Table 1: Summary of leading indicators that will be used in the analysis
Indicator Direction Crisis and Source
Multiple crises indicators
Export/GDP Decrease Currency: Kaminsky et al. (1997)
Foreign debt: Carlione & Trebeschi (2006) Inflation Increase Currency: Kaminsky et al. (1997)
Foreign debt: Carlione & Trebeschi (2006) Interest rate Increase Banking: IMF (1998), Von Hagen and Ho (2007)
Currency: Kaminsky and Reinhart (1999)
Foreign debt: Carlione & Trebeschi (2006), Roubini (2001) International reserves Decrease Banking: Von Hagen and Ho (2007)
Currency: Kaminsky et al. (1997), Kaminsky and Reinhart (1999) Real exchange rate Increase
(depreciation)
Banking: IMF (1998), Von Hagen and Ho (2007)
Currency: Kaminsky et al. (1997), Kaminsky and Reinhart (1999) Real GDP growth Decrease Currency: Kaminsky et al. (1997
Foreign debt: Carlione & Trebeschi (2006), Roubini (2001) Trade balance Decrease Currency: Kaminsky et al. (1997
Foreign debt: Roubini (2001)
Banking crisis indicators
Deposits Increase IMF (1998)
M2 multiplier Increase IMF (1998)
Currency crisis indicators
M2-to-reserves ratio Increase Kaminsky et al. (1997), IMF (1998) Money growth (M3) Increase Kaminsky et al. (1997)
Foreign debt crisis indicators
External debt-to-export ratio Increase Carlione & Trebeschi (2006), Roubini (2001) External debt-to-GDP ratio Increase Roubini (2001)
Public debt-to-GDP ratio Increase Roubini (2001) Public debt-to- government
revenues ratio
Increase Roubini (2001)
Short term debt-to-external debt ratio
Increase Carlione & Trebeschi (2006)
Source: Carlione and Trebeschi (2006), IMF (1998), Kaminsky et al. (1997), Kaminsky and Reinhart (1999), Roubini (2001), and Von Hagen and Ho (2007).
III. Icelandic crisis 3.1 Overview
The Icelandic financial supervisory authority (FME) took a 75 percent in Glitnir on September 29, 2008 in which it provided the bank with new equity worth of 600 million euro to help the bank with its liquidity difficulties (MFA, 2009). The FME further announced on October 7, 2008 that it would proceed to take over Landesbanki and Glitnir to ensure “continued commercial bank operations in Iceland” (FME, 2008a, b). A like statement about the takeover of Kaupthing was made on October 9 (FME, 2008c). This takeover was a result of lack of available credit to finance its debts (MFA, 2009). Beside this, the authorities of the United Kingdom froze the UK assets of Kaupthing and Landesbanki on October, 8. Using the anti-terrorism law instated after 9/1118 (OPSI 2008) as the Icelandic authorities supposedly informed the UK authorities not to honor claims for deposit guarantees for clients of UK subsidiaries of the Icelandic banks. Buiter and Sibert (2008) question this statement and call it false. They base this on the transcripts of the key conversation on the issue between the British and Icelandic authorities. Instating this act accelerated the collapse of Iceland as the payment system effectively came to a standstill, and lead to very large difficulties in transferring money between Iceland and abroad. Which is very detrimental for a trade dependent economy like the Icelandic one (Danielsson, 2008). Moreover did it aggravate the downfall of Kaupthing, the last surviving Icelandic bank at that time. The Icelandic central bank tried to peg the Icelandic Kroner on October 7th to stabilize the exchange rate (Buiter and Sibert, 2008) but it did not have the reserves to support it. Speculative attacks on the Icelandic Kroner were the result and the peg lasted only a very short period. The downfall of the three banks resulted in a contraction of the Icelandic economy. External debt, unemployment and inflation all rose substantially, which has negative effects for the Icelandic population.
18
3.1.1 Credit Rating Agencies
Table 2: Explanation of various credit ratings given by the rating agencies.
Standard & Poor's; Fitch Moody's Explanation
AAA Aaa Highest grade, lowest risk
AA Aa High grade, low risk
A A Above average grade, relative low risk
BBB Baa Average grade, medium risk
BB Ba Payment likely, but uncertain
B B Currently able to pay, risk of future default
CCC Caa Poor liquidity, clear risk of default
CC Ca Very doubtful liquidity, frequent default
C C Lowest grade, extremely poor outlook for repayment
D In default
Source: Sedlabanki, 2009a
Table 3: Overview of government bonds rating for Iceland by Moody's
Government bonds
Date Foreign currency Domestic currency Outlook
2008 - December Baa1 Baa1 Negative
2008 - October A1 A1
2008 - September Aa1 Aa1
2008 – May Aa1 Aa1 Stable
2008 - March Aaa Aaa Negative
2002 - October Aaa Stable
1997 - July Aa3 Aaa Stable
1996 - June A1
Source: Sedlabanki, 2009b
Table 4: Overview of sovereign credit ratings of Iceland by Standard & Poor's and Fitch
Standard and Poor's Fitch
Date Long-term foreign currency Outlook Long-term foreign currency Outlook
2008 – October, 24 BBB- Negative 2008 – October, 9 BBB- 2008 – October, 6 BBB 2008 – September, 30 A- 2008 – September, 29 A- 2008 - April, 1 Negative 2008 A 2007 - November Negative 2007 - March A+ Stable 2006 – December Stable 2006 - June AA Negative
Source: Sedlabanki 2009a
3.1.2 Credit default swaps
denominated in Icelandic Kronur (ISK) is reflected. Pires et al. (2008) define a CDS as follows:
“A credit default swap (CDS) is a bilateral over-the-counter financial contract in which one counterparty (the protection buyer) pays a periodic fee (the CDS spread), typically expressed in basis points per annum, paid on the notional amount, in return for a contingent payment by the other counterparty (the protection seller) following a credit event with respect to a reference entity.”
The CDS spread is often used as the measure to determine the riskiness of the assets. The CDS spread for the Icelandic government bonds started to rise as early as late 2007 (Sedlabanki, 2009c), this was only a minor increase but it soon climbed to 400 basis points in the first half of 2008. It remained between 200 and 400 basis points until it spiked at 1400 basis points in October 2008 after which it gradually declined towards 800 basis points. The market was already aware of the increased risk of the Icelandic government. Other advanced countries which had a large rise in the CDS spread remain at 200 to 400 basis points (such as Ireland, Greece and Portugal), indicating that Icelandic government bonds were an investment with considerable more risk attached to it.
3.1.3 Overextended economy
Icelandic banks have, as most banks in financial advanced economies, illiquid assets of relative long maturity compared to its liabilities. Therefore, Buiter and Sibert argue that the Icelandic banks face a possible run on liabilities, when available assets prove to be insufficient to cover the short-run liabilities. Banks are in essence vulnerable to possible illiquidity due to the maturity transformation process. When a possible run would focus on the foreign currency denominated liabilities, problems will arise because the Icelandic central bank cannot act as an effective lender of last resort. The foreign exchange resources of the Icelandic authorities are relatively small compared to the short-term foreign currency exposure of its domestic banks and a therefore more vulnerable. In graph 1 below, the foreign assets of the Icelandic central bank are compared to the foreign liabilities of the Icelandic deposit money banks (DMBs). It shows that the foreign assets of the central bank fell significantly short of the liabilities of the DMBs. The liabilities of the DMBs showed a tremendous growth in the last year, where they almost doubled.
Graph 1: Foreign assets of the Icelandic central bank compared with the foreign liabilities of the Icelandic deposit money banks
Source: Sedlabanki jul-07 aug-07 sep-07 okt-07 nov-07 dec-07 jan-08 feb-08 mrt-08 apr-08 mei-08 jun-08 jul-08 aug-08 sep-08 0 1.000.000 2.000.000 3.000.000 4.000.000 5.000.000 6.000.000 7.000.000 8.000.000 9.000.000 10.000.000
Foreign Assets Central Bank Foreign Liabilities DMBs
M
ln
I
3.1.5 Effects of the international financial crisis on Iceland
The difficulties in the international financial markets started with the burst of the United States housing market bubble in 2006 (Lahart, 2007). House prices began to fall and difficulties with subprime mortgages increased in early 2007. Mortgage defaults rose and housing prices continued to fall in the following months. Investors pulled back their money from the credit markets and financial institutions cut back on the lending because the investments prove to be less save than expected and confidence eroded (Bernanke, 2009). The interbank lending market dried up because of the loss of confidence and difficulties with valuations of complex financial products, such as repackaged collateral debt obligations. The exposures of the banks lead to increased counterparty risk and a rise in the global price of liquidity (Buiter and Sibert, 2008). Icelandic banks were unable to borrow in the international financial wholesale markets, despite having more than adequate capital ratios, liquidity provisions and profitability of their operations because of an extreme international wholesale liquidity shortage. These problems translate to the Icelandic banks in three ways (Buiter and Sibert, 2008):
1. The likelihood of banks becoming insolvent increased because the funding cost increased.
2. By coordinating market beliefs about banks, bank runs that are based on self-fulfilling expectations are more likely.
3. Greater difficulties for banks to insure themselves against bank runs.
These problems did emerge for the Icelandic banks and government takeovers took place to protect the banks from falling.
3.2 Graphical data analysis
3.2.1 Multiple crisis indicators
The theoretical analysis in section two has shown that seven variables are said to indicate multiple crises. Decreasing exports relative to GDP could indicate an upcoming foreign debt crisis since lower exports makes it more difficult for a government to get international reserves (Dreher et al. 2006); it could also point at a currency crisis. Rising inflation, a decrease in real GDP growth and a deterioration of the trade balance can point at a currency or foreign debt crisis. A rising real interest rate could indicate all types of crisis. A decrease in international reserves and a depreciation of the real exchange rate could indicate a currency or banking crisis. Further information is available in section 2.2.4 and table 1. The graphs of these variables are depicted below in graphs 2 and 3.
Graph 2: Graphical distribution of multiple crises indicators
Graph 3: Graphical distribution of multiple crises indicators
Source: IFS, Sedlabanki
A couple of remarks can be made about these graphs at first sight. It is clear that inflation rose and real GDP fell in 2008, this is just as the theory predicts. Both variables are said to indicate a general deterioration of the economy. The volume of international reserves shows an increase contrary to an expected decrease. The volume of international reserves grew only very small and steadily until 2004. It started to rise rapidly in and after 2004, with large spikes in 2007 and 2008. Buiter and Sibert (2008) argue that the Icelandic authorities were already trying to accumulate more reserves as a buffer for the tremendous growth of their banking system; this is a possible explanation for the increase.
crisis and that interest rates could rise when authorities try to defend the currency from a speculative attack, in the case of a currency crisis. Von Hagen and Ho (2007) also expect the nominal interest rate to rise as a reaction on the increased demand for bank deposits. The real interest rate can be calculated via a deduction of the inflation of the nominal interest rate according to the Fisher equation (Walsh, 2003). If the inflation would rise more than the nominal interest rate, a negative effect is still noticeable in the real interest rate. Graphical descriptions can be made of the nominal short-term interest rate and the 3 month interest differential between the REIBOR and EURIBOR which are the interbank nominal interest rates for the Iceland and the Euro-areas; these could have a maturity from 1 week to 12 months (ECB, 2004). These are shown below in graph 4, from which it is clear that the expected rise on the onset of the crisis can be noticed. It is noticeable from graph 2 that inflation rises; this could indeed have caused the real interest rate to fall as it grew faster than the nominal interest rate.
Kaminsky et al. (1997) predict that the overvaluation of the currency in the run-up to a crisis lead to a deterioration of the export performance. However, the export performance increases in the wake of the crisis at hand in 2008 as can be noticed from graph 2 above. This is remarkable since the real effective exchange rate (measured as amount of Icelandic Kroner per SDR19) appreciates. A different story arises when the Icelandic Kroner per US dollar exchange rate is examined, which can be seen from graph 5 below. The devaluation can be seen here at the moment that the crisis starts, which is according to the theory described above. The graph for the trade balance is in line with the Kroner/Dollar exchange rate graph since a worsening of the trade balance happens in the years preceding the crisis due to an appreciation of the exchange rate. This leads to higher imports and lower exports. The trade balance showed a huge decline in the years before 2008 (where the KR/US $ exchange rate rose) and shifted back after the exchange rate had fallen.
19
Graph 4: Nominal interest rate and 3 month interest differential between REIBOR and EURIBOR
Source: Euribor, Sedlabanki
Graph 5: Graphical distribution of Kroner/US Dollar exchange rate
Source: IFS
3.2.2 Banking crisis indicators
changes in the ratio of reserves to bank deposits and changes in the short-term interest rate. The IMP tries to indicate upcoming banking crises.20 The nominal value of bank deposits shows a positive trend except for the decrease in 2008. The IMF (1998) indicated that growing deposits and a rise in the M2 multiplier indicate financial liberalization. Both bank deposits and the M2 multiplier confirm the financial liberalization starting in the early 2000s. This can be noticed from the spikes from 2004 until 2007. The last two years are remarkably characterized by declines in the M2 multiplier. The reserves-to-bank deposits ratio shows a negative trend from the 1980s onwards. A peak can be noticed in early 2007 and the start of a peak is visible for 2008. These could point at an upcoming crisis. The data for deposits is unfortunately only available until the third quarter of 2008, thus is the full crisis period not entirely incorporated in the data. The IMP shows an increased turbulence in the market in 2005 and in 2008. This could also indicate an upcoming crisis.
Graph 6: Banking crisis indicators
Source: IFS, Sedlabanki
20
3.2.3 Currency crisis indicators
Graph 7: Currency crisis indicators
Source: IFS, Sedlabanki
Graph 8: M3 used instead of M2 in currency crisis indicator variables
Source: IFS, Sedlabanki
3.2.4 Foreign debt crisis indicators
indicators lead in the direction they should be pointing to indicate an upcoming crisis. The short term debt-to-total external debt ratio shows an interesting history. Lots of fluctuations are noticeable in the years until 2004 when it drops and stays very low for a few years until it rises again at the onset of the crisis. This variable has been magnified a little bit because short term debt is a relatively very small part of total debt and fluctuations would not be visible otherwise. Five of the variables that are covered in graphs 2 and 3 are also variables that could indicate a foreign debt crisis, but they are already covered in the previous section.
Graph 9: Foreign debt crisis indicators
Graph 10: Foreign debt crisis indicators
Source: IFS, Sedlabanki
3.3 Empirical data analysis
3.3.1 Stationarity tests
The second step in analyzing the data after the graphical analysis is an empirical analysis using a logit model to estimate the predictive behavior of the various indicators. Test for stationarity of the data needs to be conducted before the analysis can be undertaken. Time series are stationary when its mean and variance are constant over time. Also the covariance between two values needs to depend only on the length of time separating those values, and not at the actual time the variables are observed. It is important to use stationary time series in a regression analysis. Because danger of spurious results arises when nonstationary series are used. A spurious result is an apparently significant result from unrelated data and therefore meaningless (Hill et al., 2001).
t t
t
y
v
y
=
+
+
∆
α
0γ
−1 (3)Where yt is a regular time series, α0 is a constant and vtis a random error term. The time
series will be stationary if γ < 1. This Dicky-Fuller test is adapted to the augmented Dicky-Fuller test to control for the possibility that the error term in one of the equations is autocorrelated21 with additional terms. The null hypothesis is γ = 0, or more formally (4):
H0: γ = 0 (data is nonstationary and needs to be adjusted to make it stationary)
H1: γ < 0 (data is stationary and applicable in the analysis) (4)
When the augmented Dicky-Fuller test yields more negative outcomes than the corresponding critical values, the hypothesis that the series contain a unit root is rejected. A rejection of the null hypothesis is evidence in favor of a stationary series which can be used in regression analysis. Appendix 4 gives the outcomes of the augmented Dicky-Fuller tests for the various variables and which adjustments are made to ensure stationarity of the variables. A large number of variables in this analysis are nonstationary and are therefore adapted using the log difference of these variables. Only seven variables are stationary before the adaption and one variable, money growth, needs to be omitted from the dataset because it stays nonstationary after adaptations. Results of the stationarity tests for the adjusted variables can be found in appendix 5.
3.3.2 Logit model
The logit model is a frequently used model for binary choice situations (Hill et al., 2001). Another binary choice model is the probit model, where its estimation is based on the normal distribution. The logit model uses a particular S-shaped curve to constrain the probabilities to the [0, 1] interval. Jacobs et al. (2008) point out that there is only a slight
21
difference in the predicted probabilities of these models. A logit model is less numerically complicated and therefore easier to implement. This paper therefore uses the logit model. The dataset runs from the first quarter of 1983 to the last quarter of 2008 since later data barely existed at the time of writing. An overview of the data periods available per variable is given in appendix 6. Reinhart and Rogoff (2008a) did not find any major banking crises in Iceland in their comprehensive historical overview, but in another study (2008c), they do point at a financial crisis in Iceland in 1985. Two incidents have taken place in the 1983 – 2007 periods (before the current crisis) which were: a government capital injection into a state- owned commercial bank in 1993, and a bank insolvency of one of the three state-owned banks in 1985-1986. Both incidents are regarded to be a borderline or smaller banking crisis by Caprio and Klingenbiel (2003). A dummy is constructed to indicate the crises periods in Iceland (1985, first two quarters 1986, 1993, and the last two quarters of 2008). This leads to a slightly odd dummy variable with only a very small number (12) of observations indicating a crisis out of the 104 observations in the dataset.
Hill et al. (2001) show that the logit model can be specified as equation (5):
) ( 1 2
1
1
)
(
]
1
[
xe
Y
F
Y
P
p
− α +α+
=
=
=
=
(5)In this model P is the probability that Y takes the value 1, F is the cumulative distribution function for a logistic random variable, x is a set of regressors and α1 and α2 are
parameters. The regression equation (6) is then equal to (Jacobs et al., 2008):
x
Y
P
P
2 11
ln
=
=
α
+
α
−
(6)Table 5: Summary of logit estimations for various variables of entire dataset and without 1993 as a dummy
Entire dataset 1993 omitted
Variable Coefficient Z-statistic Probability Coefficient Z-statistic Probability
Exports-to-GDP 1.115 0.560 0.575 0.870 0.365 0.715
International reserves 0.674 0.382 0.703 1.979 1.044 0.297
Inflation 0.072 2.757 0.006 0.122 3.266 0.001
Real exchange rate -0.088 -1.888 0.059 -0.107 -1.951 0.051
Nominal interest rate 0.034 0.378 0.706 0.081 0.779 0.436
Real interest rate -0.073 -2.025 0.043 -0.096 -2.203 0.028
KR/US $ exchange rate 5.758 1.521 0.128 6.560 1.509 0.131
Trade balance 8.39E-08 1.936 0.053 6.71E-08 1.615 0.106
Bank deposits -0.330 -0.076 0.939 3.913 0.769 0.442
Data starting after 1985
M2 multiplier -3.083 -1.490 0.136 -9.101 -1.924 0.054
M3 multiplier -0.971 -0.454 0.650 -4.626 -1.370 0.171
M2-to-reserves ratio -1.463 -2.632 0.009 -6.313 -1.251 0.211
M3-to-reserves ratio -1.744 -2.603 0.009 Na Na Na
Short term debt-to-external debt ratio 0.254 0.753 0.452 0.013 0.706 0.482 Three month interest rate differential 9.202 1.451 0.147 9.202 1.451 0.147 External debt to export ratio -1.227 -0.422 0.673 1.038 0.201 0.841 External debt to GDP ratio 0.117 0.557 0.579 0.870 0.365 0.715 Public debt to GDP ratio 0.800 0.215 0.830 -2.336 -0.299 0.765 Public debt to government revenue 2.617 0.816 0.415 2.251 0.424 0.672
Real GDP growth -0.237 -2.213 0.027 -0.150 -1.146 0.252
IMP 0.026 0.099 0.921 0.116 0.382 0.702
Source: Euribor, Eviews, IFS, Sedlabanki, Statice
omitted (at a ten percent confidence level).22 Some variables show an improvement when the government intervention of 1993 is not regarded as a crisis, but other variables deteriorate, leading to less significant responses. The IMP is never significantly indicating a crisis. This is very unfortunate and proves already that predicting a banking crisis is very hard. The three month interest differential stays the same since the Euribor time series are only available from 1998 onwards. An analysis has also been made to determine if 1985 should be considered as a crisis. The results of this sensitivity analysis are given in appendix 7. Conclusions are drawn that the 1985 crisis is needed in the data set as a crisis.
Another analysis is made using the unaltered data (with the risk of nonstationary analyses). These results can be found in appendix 8. Only four variables improve so much that they proof to be significant. These are the exports-to-GDP ratio, external debt-to-GDP ratio, the public debt-debt-to-GDP ratio and the public debt-to-government revenue ratio. The other variables that significantly explain the crises were stationary in their original form and did not need to be altered. The research conducted above indicates that is very difficult to predict a financial crisis using macroeconomic variables. But still seven variables proved some significance explaining the various crises.
3.3.3 Principal components analysis
The data set used in this paper consists of a lot of economic indicators. An analysis with multiple variables could lead to multicollinearity. Therefore this dataset is reduced to a limited number of components using a principal components analysis. The components are used in the logit model as explanatory variables, following Jacobs et al. (2008). A principal components analysis is designed to transform a set of random variables into new random variables (Venables and Ripley, 2002). The different factors are constructed in such a manner that they are uncorrelated. Jacobs et al. (2008) explain the principal components analysis as (where a factor is a component):
22
“The first factor is the normalized linear combination of the original set of random variables with maximum variance; the second factor is the normalized linear combination with maximum variance of all linear combinations uncorrelated with the first factor; and so on.”
Different criteria exist for dropping less important components. All components with an eigenvalue below one will be dropped in the Kaiser criterion. A graphical analysis is used in the Cattel spree test; eigenvalues are plotted on vertical axis and factors on horizontal axis. All factors below the point on the curve where the smooth decrease levels off will be dropped. Jacobs et al. suggest that the Kaiser criterion is widely used in the existing literature, therefore they adopted this criterion. It will also be used in the principal components analysis in this paper.
Four groups of data are examined in the factor analysis. These groups consist of the different explanatory variables per crisis and consist of the same variables explained above. The groups are given below and indicated to how many factors they are reduced:
• Multiple crises indicators (7 variables, 3 principal components) • Foreign debt crises indicators (10 variables, 5 principal components) • Currency crises indicators (8 variables, 3 principal components) • Banking crises indicators (5 variables, 2 principal components)
A table can be found explaining how these different factors are made up of the variables in appendix 9.
