No Contagion, only Globalization and Flight to Quality

In this article, tests for globalization and contagion are separated using an ex ante definition of crises, and contagion tests are neutralized with respect to globalization effects. A large database is constructed to study the stability of correlation matrices for four asset classes: equities, government bonds, investment grade corporate bonds, and high-yield corporate bonds, in four geographical zones. Overall, the results confirm the instability of correlations and point to a combination of globalization and flight to quality, while emphasizing that contagion on the equity markets appears as an artifact due to globalization.


A. Introduction
The interdependence of financial markets is a serious concern for investors looking to diversify their portfolios internationally. Yet, recent articles have observed an uptrend in correlations among financial centers -a major warning signal. Two analytical frameworks exist side by side on this issue. Some see economic globalization, coupled with the growing integration of financial markets, as the main reason for these developments. Others attribute the empirical findings to market contagion during crises.
On the one hand, the globalization phenomenon, i.e. the general increase of correlations within asset classes and across geographical areas over the past decades, is well documented, both for equities (Berben and Jansen (2005)) and for government bonds (Hunter and Simon (2004)) On the other hand, crises can be transmitted to markets other than those in which they originate, leading to a contagion effect. Empirical studies (Lin et al. (1994); De Santis and Gérard (1997); Corsetti et al. (2005); Wälti (2003)) found that correlations increased in equity markets during hectic periods, pointing to the presence of contagion. However, according to Hartmann et al. (2004), equity markets are twice as likely as bond markets to crash simultaneously.
Besides, correlations across different asset classes are shown to decrease in times of crises, creating potential for diversification through asset allocation (Smith (2002); Hunter and Simon (2004)). This is particularly the case for correlations between bonds and equities (Connolly et al. (2005)). The contrast between the global increase within each asset class and the correlation decrease across asset classes seems to be explained by the effect known as "flight to quality" (Hartmann et al. (2001)) where investors shift funds towards safer assets, leading to "decoupling": higher correlations within the equity markets but negative correlations between government bonds and equities (Gulko (2002)). The decrease in equity and bond correlations during crises, attributable to flight to quality effects, may be present whether associated or not with contagion.
Contagion can be confused with globalization since both have a tendency to increase correlations among assets, especially during periods of high volatility coupled with bear markets (Longin andSolnik (1995), (2001); Silvapulle and Granger (2001) ;Chesnay and Jondeau (2001)). In a theoretical paper, Calvo and Mendoza (2000) show that globalization may promote contagion by weakening incentives for gathering costly information. On empirical grounds, Forbes and Rigobon (2002) deny the existence of contagion as such. They exhibit a high level of market co-movement in all periods, not only crises-a phenomenon they refer to as interdependence. Similar results are found by Flavin and Panopoulou (2008). Our paper attempts to go further in dissociating globalization and contagion phenomena by testing them separately while including all market and banking crises from 1978 to 2007. Contagion and globalization are not necessarily mutually exclusive, but they are difficult to separate econometrically (Bekaert et al. (2005)). One major problem consists in identifying precisely what constitutes a crisis period. For investors, though, the practical consequences will be different depending on whether these developments are attributable to increasing market globalization or to crisis contagion. In the first case, a gradual but unstoppable movement can be expected. In the second, investors will have to be especially careful when international volatility is high, because increased risk will be compounded by a decline in diversification protection. Optimal portfolio management depends on proper identification of the effects at work. This article makes use of the tests for correlation stability that were laid down by Jennrich (1970) and refined by Goetzmann et al. (2005) through new advances in asymptotic theory. We propose an original empirical study that is broadly scoped in terms both of geographical coverage and of asset classes. We abide by established crisis definitions to avoid a personal classification that might be tainted by endogeneity.
Although most research has concentrated on equity markets, we broaden our scope to include government and corporate bonds, the latter being almost completely uncharted in the literature on globalization and contagion i . We also distinguish between investment grade (IG) and high yield (HY) bonds, so as to segment bond products according to whether they are primarily dependent on interest rate risk or on default risk. Furthermore, we simultaneously analyze the impact of 15 crises on asset markets between 1978 and 2007. Securities are divided into 15 categories depending on their financial characteristics and geographical zone.
Our results confirm the presence of globalization, with several nuances. In particular, the bond market segments do not appear to be greatly affected. By contrast, contagion effects are not corroborated by the data when corrected for globalization. In addition, our findings suggest that the tendency towards flight to quality dominates during crisis periods.
The remainder of the article is organized as follows. Section 2 presents the tests for correlation stability that will be used in the empirical section. Section 3 describes the database used. Sections 4 and 5 form the heart of the article, proposing globalization tests followed by contagion tests. In the latter case, the definition of crises necessitates some documentary research, which we describe in Appendix 1. Section 7 concludes.

B. Testing the stability of correlations between financial series
Correlations among financial data series are a key tool in portfolio management and risk control. Markowitz's classic model is based on knowledge of the entire variancecovariance matrix of returns, and hence of all correlations within the set of securities analyzed. The assumption that these parameters remain stable over time guarantees the consistency of forecasts based on past data. But this stability has recently been challenged by a large body of econometric research.
In recent years, analyses of the stability of second-order parameters, i.e. variances, covariances and correlations, have developed considerably from a theoretical perspective and have been tested empirically many times. Regime-switching tests could be a simple option in this regard. Unfortunately, testing the existence of two correlation regimes may prove difficult. The main problem lies in identifying the observation dates corresponding to each of the two possible states. Crises are generally identified by high volatility in one or more asset classes that are being tested for correlations. But splitting the sample ex post creates potential selection bias distortions (Boyer et al. (1999)).
It is nevertheless possible to test the stability of correlations versus the onset of contagion during crises provided that crises are delineated beforehand. Therefore, we identify crises based on their fundamental determinants and not on equity or bond volatility (see Appendix 1). Once crises periods have been precisely delineated, we test the null hypothesis of equality between all correlations across assets during both crisis and normal periods.
The test, proposed by Jennrich (1970), is based on the chi-square distance between two correlation matrices. Its validity is established under the normality assumption. The null hypothesis is the equality of the correlation matrices and of the asymptotic variancecovariance matrices, which represents a significant restriction. In particular, because crises occur relatively infrequently, there is a greater possibility of measurement error for correlations during turbulent periods. The Jennrich (1970) test has been applied by Kaplanis (1988) and Annaert et al. (2006), among others.
Goetzmann, Li and Rouwenhorst (2005) (GLR) employ the same test statistic as Jennrich (1970), but draw on knowledge of the asymptotic variance-covariance matrix, denoted Ω , developed in Browne and Shapiro (1986) and Neudecker and Wesselman (1990). Consider two sub-periods of lengths 1 n and 2 n , along with correlation matrices p p × measured in the two corresponding sub-samples, denoted 1 R and 2 R . In a case where the asymptotic variance-covariance matrices are identical ( 1 2 Ω = Ω = Ω ), GLR obtain the following result ii : ( ) The test statistic is asymptotically distributed as where the asymptotic matrix Ω may be determined analytically when observations are assumed to come from i.i.d. p-variate distributions with finite fourth-order moments.
Although it simultaneously tests the equality of correlation matrices and asymptotic variance-covariance matrices iii , the GLR method remains the most effective way of dealing with the case of p-variate distributions where 2 p > .

C. Data
The database includes weekly returns to indices for equities, government bonds and corporate bonds, based on geography and, in the case of bond indices, on ratings. The series are the longest we could find for each asset class since the purpose is to study the impact of globalization which is, by definition, a long-term phenomenon.
Our analysis focuses on four geographical areas: the U.S.A., the Eurozone, Japan and the U.K. For equities, we use the indices constructed and supplied by Datastream (DS indices) for the period from August 1978 to May 2007. These indices are denominated in local currencies and include dividends. They are weighted and cover at least 75% of the total capitalization of the markets they represent.
For government bonds, we take the 10-year benchmark indices supplied by Datastream iv . These indices, which include coupon returns, are usually based on a single bellwether, generally the last bond issued by the country's Treasury in a given maturity.
Factors such as liquidity, issue size and coupons are also taken into account when choosing the index components. Weekly data are available from January 1980 onwards, except for Japan, where the series begins in January 1984. Accordingly, the period under review goes from January 1984 to  For corporate bonds, we use two categories: investment grade, with ratings between AAA and BBB -, and high yield, rated from BB + to CCC. The indices are denominated in local currencies and include coupon returns. Convertible bonds are excluded. The weekly data cover the period between July 1998 and May 2007. They are sourced from Merrill Lynch (i.e. bids quoted by traders at the ML desk) at the market close v . All indices have been hedged in dollars.
As some data series (equities) are longer than others (HY bonds), the descriptive statistics in Table I have been established on the common observation period stretching from July 1998 to May 2007 (except for Japanese HY corporates) to allow for comparisons. Unsurprisingly, the government bonds are the assets with the lowest annualized return, while IG corporate bonds display returns that are higher than those on HY bonds and similar to equity returns. This is typical of a period characterized by decreasing long-term interest rates, like the one markets have experienced until recently. More interesting is the low level of standard deviations of HY bond returns over the period. The reason lies probably in the decorrelation between the interest rate component and the credit risk component, which evolve in opposite directions when the economic situation changes. This creates a compensating effect in HY bond portfolios, decreasing the overall volatility at index level. This effect is found partly, although to a much lesser extent, in corporate bonds.
Skewness takes a negative value for all the assets under review. Kurtosis exceeds the reference value of the normal distribution (equal to 3) for all countries and asset classes.
This leptokurticity is typical of financial data series. The non-normality of returns is confirmed by the Jarque-Bera test. Phillips-Perron tests (not reported here) confirm that all the series are stationary.
[Insert Table I here] Table II shows all the correlations for the same period, marked by high equity market volatility, the "tech bubble" and a string of crises in bond markets and emerging economies. Broadly, correlations are negative between equities and government and IG bonds in all countries. By contrast, the correlations between high yielders and equities are positive. This last result is consistent with the findings of several authors (Fama and French, 1993;Alexander et al., 2000). Co-movements between low-rated bonds and equities are commonly attributed to the importance of the credit risk component in HY bonds -a factor shared with equity returns. Likewise, correlations between HY and IG bonds are generally close to zero or even negative. Within the same asset class, the strongest geographical correlations are found between the Eurozone and the U.K., with a maximum of 86% for equity markets; and the weakest are those for Japan, as other research has shown (Hunter and Simon (2004); Berben and Jansen (2005)).
[Insert Table II

D. Globalization tests
The recent literature tends to suggest that geographical correlations within asset classes have increased over the last 20 years. This is true for equities and government bonds.
This situation is linked to the rise of globalization.
Relying on the methodological analysis presented in Section 2, we test the equality of correlation matrices using the GLR (2005) test. The sample is broken into two subperiods of equal length. The break date thus varies according to the dataset under consideration. Since the aim of the test is to detect an evolving phenomenon, the precise break date is not vital. Moreover, the results are not affected if the date is shifted slightly. We have therefore opted for a symmetrical choice, which is more accurate.
[Insert Table III (1) and (2). The result of the test carried out on all asset classes (16 indices, minus Japanese HY bonds, for which data are unavailable) is given in the first row of Table III. It shows that the differences in correlation between the two sub-periods are significant for all asset classes under consideration, thus confirming the impact of globalization on market interdependence during the previous decade.
But this finding, which confirms those established previously for international equity markets (Berben and Jansen (2005); Chesnay and Jondeau (2001)), should be treated with caution. This is because the GLR test is bilateral, and the statistic measures the correlation differences, both positive and negative, between sub-periods. To give a clearer picture of the impact for each asset category, we have shown the correlation differences in Table IV: If all the correlations had increased, the table would show positive items only. But this is certainly not the case. Taking a closer look, however, we can see that the negative items in Table IV mainly concern the correlations between different types of asset. For example, the correlation between US Treasuries and European equities fell 2.3%.
Interpreting this type of observation is obviously problematic and the link with the intuitive idea of globalized financial markets vi remains vague.
We therefore ran a second set of intra-asset class tests using the three 4 X 4 matrices and the 3 X 3 matrix from the lower rows of Table III. The results point clearly to a globalization effect in the equity and government bond market but none whatsoever in the corporate bond markets (HY and IG, separately). Accordingly, there appears to be no globalization in these two bond market segments.
In terms of methodology, there is a major difference between the first test and the last four. Whereas the statistics from the former set mix geographical and inter-class globalization, the latter take account of purely geographical correlations only. In sum, our results point to globalization in equity markets combined with decorrelation between equities and bonds. The data for same-type geographical corporate bonds lead us not to dismiss the stable correlation hypothesis.
It has been observed that Japan plays little part in financial globalization. Even today, the Japanese market is only loosely correlated with other world markets. The last column in Table IV, which shows Japanese equities, stands out singularly from the other columns showing the equity markets of the other three regions.
[Insert Table IV here]

E. Contagion tests
Our definition of "crisis" is broad. It encompasses five types of movement: currencies, sovereign debt, events arising from a bond or equity crash, corporate bankruptcies or loss of confidence (Enron, WorldCom), and other crises of confidence, such as terrorist attacks. We have deliberately omitted crises of a purely banking nature unless they are related either to currency crises, where the impact on financial assets is more diffuse, or to economic crises such as recessions or oil shocks. The real difficulty lies in establishing precise timeframes for the crises we have selected.
The start and end dates used in this article (Table V) have been chosen solely on the basis of previous papers (Appendix 1), thereby avoiding, at least partially vii , the problem of endogeneity raised in Section 2. Admittedly, while the onset of a crisis is usually easy to identify, the end date is much harder to pinpoint. This awkward problem is highlighted by the Asian crisis (Appendix 1), which several authors have studied.
[Insert Table V here] Figure 1 plots the dates of the crises, regardless of type, and shows that the majority occurred from the 1990s onwards. This may be due to pure randomness or to a shortsighted choice of turbulent periods, i.e. a tendency to choose only the most recent crises.
[Insert Figure1 here] The GLR (2005) contagion test consists in comparing correlations among all markets, segregating crisis periods from other periods. Table VI gives the results of those tests.
The results of the first four rows show that contagion is observed neither globally nor in the bond segments of the world markets.
[Insert VI here] Contagion in the equity market is significantly detected at the 5% level, but not at the 1% level. In light of this result, we wanted to rule out the possibility that globalization could spill over to contagion. Financial globalization at world level viii , which basically corresponds to the closer synchronization of economic cycles, can manifest itself in different ways. If, in addition, there is a contagion effect, this compounds the globalization effect. Since the crises identified earlier, shown in Table V and Figure 1, are over-represented in the second half of the sample period, there is indeed a risk that globalization will be confused with contagion.
To overcome the awkward problem of identification, we adjust the time periods to ensure that, for the entire period tested, crises no longer appear systematically at the beginning or the end of the sample. If the crises are spread evenly over the time interval under consideration, then the globalization effect will be "neutralized". As reported in Table VI (last two columns), adjusting the intervals in the cases does indeed affect the equity contagion result. Indeed, taking into account the adjusted sample period, contagion in the equity market is no longer significant, even at the 10% level. We therefore conclude that the contagion primarily detected in the unadjusted (full) sample period actually appears to be an artifact caused by globalization. This observation probably explains the confused interpretation of some of the results presented in the literature.
The mixed case of the equity-bond link is harder to deal with because, by nature, it cannot be segregated in a specific correlation matrix, since the matrix always includes geographical correlations between equities and bonds as well. Therefore, we adapt the GLR test to partial correlation matrices by isolating the cross-correlations only, i.e.
correlations between assets of different categories. For instance, in the first reported test of this category (see Table VI, second part, first row) the correlations between the U.S.
sovereign and EU IG bonds are taken into account because the assets belong to different classes, while the correlations between the U.S. and EU sovereign bonds (same class assets) are excluded. In other words, these additional tests pick only on the pairs of securities that could generate flight to quality effects and rule out the ones that are more likely to be associated with contagion.
Among the six possibilities, only two lead to significant differences in correlation: GVT bonds/IG bonds and GVT bonds/equities. Moreover, these findings are not affected by the correction for globalization. Thus, crises do indeed affect the bond markets, but through cross-correlations, not through intra-class correlations. Moreover, the presence of a flight to quality in times of crisis is observed with no doubt. Scared by turbulence, investors pull out of the markets they consider too risky and seek safety in reliable bond issuers, especially governments. This flight to quality effect drives risk premia higher and reduces the correlations -some being already deeply negative -between asset categories. The movements can be very large. Table VII shows the correlation differences between crisis and quiet periods for the two pairs of assets which tested positively for this effect.
In conclusion, to prepare for crisis periods, diversifying between equities and bonds while employing an appropriate fixed income management strategy is just as important, if not more so, as managing the portion of the portfolio reserved for equities, even global equities. In this respect, there is good news for investors: even though equity volatility rises during periods of turmoil, it is offset -at least partially -by a steep fall in correlations with high quality bonds. The flight-to-quality phenomenon acts as an antidote to the perverse effects of crises on the global financial markets. Detecting it should therefore help to prevent the harmful effects of stock market crises.
[Insert Table VII here]

F. Conclusions
The recent empirical literature on market financing contains a strong message with major practical implications for risk management, namely that correlations on markets are broadly unstable. Two main factors are usually cited to explain regime breaks in correlations: economic globalization and crisis contagion. Structurally, the two factors are very different. Confusing them would then have a harmful impact on portfolio management.
For analysts, therefore, distinguishing between globalization and contagion is a real challenge. However, econometric research often tries to detect one or other of the effects, without considering the possibility that the results could be misinterpreted. To avoid that pitfall, we have used a sequential process that considers, firstly, the possibility of globalization and, secondly, overlying contagion.
Empirically, the data examined in this study are original in two regards: the asset classes covered and the number of market crisis analyzed. There is a vast literature on the behavior of international correlations in equity markets and, to a lesser extent, in the government bond market, but very little has been written about corporate bonds. We have split corporate bonds into IG and HY in order to measure more accurately the flight-to-quality that occurs in periods of high volatility -an occurrence that market practitioners are thoroughly familiar with. Although the literature on this subject is evolving rapidly, we are not aware of any other articles that address this topic in such a general framework.
Our second contribution is the exhaustive nature of our crisis study. We have not limited ourselves -as is often the case in the literature -to one or two crises, such as Russia, Asia, LTCM or Enron. Instead, we have dealt simultaneously with all identifiable crises in an effort to test as exhaustively as possible the assumption that asset correlations change during periods of turmoil. We selected the start and end dates of these periods with the utmost care, drawing on previous research but without using our database. In this way, we have been able to avoid the distorting effects of endogeneity, which would have arisen had we used realized volatilities to establish the dates.
In sum, our results confirm that globalization is present in all markets, with the exception of the corporate HY bond segments (IG bonds being borderline), where correlations are stable.
We therefore look for contagion, first disregarding the results of the globalization tests and then factoring them in. Contagion is immediately rejected for the fixed income assets. Concerning equities, contagion is detected at the 5% level in the first test irrespective of globalization bias, but disappears when the appropriate correction is incorporated. Therefore, we confirm the results obtained by Forbes and Rigobon (2002) and conclude that the initially detected contagion is an artifact caused by globalization.
Our conclusion rests heavily on the assumption that globalization is checked for before contagion, and not the reverse. Among the motivations for this assumption are the different natures of the two phenomena. While globalization is a permanent technologically -and economically-sensible financial driver, contagion is often thought of as an easy way to represent the excess financial movements, i.e. those for which no fundamental explanatory variables have yet been found, as testified by the literature on speculative bubbles (Adam and Szafarz (1993); Sornette and Malevergne (2001); Salge (1997)). So, by taking into account globalization first, we reduce as much as possible the residual volatility to be attributed to contagion.
Methodwise, the GLR (2005) test consists in opposing the null hypothesis of equal correlation matrices and the alternative of separate matrices, whatever the sign of the differences between entries. Conversely, the highly restrictive view states that globalization/contagion on a market must be characterized by an increase in correlations for any pair of securities in that market. A middle approach would be to introduce an asymmetric GLR-type test that makes it possible to consider only increases in correlations. Thus, a "signed" matrix generalization of the test used in this article would open up new horizons for investigating both globalization and contagion. Moreover, non-normality distortions could be taken into account (see, e.g., Campbell et al. (2008)).
Finally, the flight to quality effect has been shown to remain after globalization has been taken into account. This observation is good news for investors, who can partially hedge against the crises by benefiting from a decorrelation between risky assets and safer bonds. While the amplitude of this hedge deserves further investigation, the effect might decrease as traders will realize that fleeing all risky assets when a crisis is feared is not the best option. In this respect, the flight to quality, like other market anomalies, is bound to disappear precisely because it has been identified. However, as pointed out by the behavioral finance stream of literature, some anomalies can prove self-fulfilling and persist much longer than expected under the rationality assumption. If indeed the flight to quality appears to be a consequence of irrational fears rather than of smart hedging attitudes during crises, then it will presumably last a long time.

Appendix: Crises selected for study
In this study, we examine five types of crisis: (1) currency crises , (2) sovereign debt crises, (3) crises triggered by an equity or bond crash, (4) corporate bankruptcies or loss of confidence (e.g. the collapse of Enron), and (5) crises of confidence arising from severe external events (e.g. 9/11).

Mexico 1976
The onset of the Mexican crises is usually dated to August 31, 1976, when the authorities decided to allow the peso to float (Bordo and Schwartz, 1996). That decision sparked a dramatic rise in inflation. According to the authors, the crisis ended on October 26, 1976, when the authorities devalued the peso by 27% against the dollar.

Chile 1982
The Chilean crisis began on June 15, 1982, when the government devalued the peso by 18% (Bordo and Schwartz, 1996). The end of the crisis is generally dated to August 5, 1982, when the currency was left to float freely (De Gregorio, 1999;Cowitt, 1984).

Mexico 1982
The  Bordo and Schwartz (1996), the crisis ended on September 1, 1982, when Mexico nationalized the banking system and imposed currency controls.

European Monetary System 1992
The EMS crisis began on September 16, 1992 when the Bank of England raised the base lending rate from 10% to 12% and announced the intention of raising it to 15% the next day (which it did not do). As a result, sterling dropped below its EMS floor rate.
On September 19, the pound was ejected permanently from the exchange rate mechanism (ERM), followed by the Italian lira. In the aftermath, the currencies of Sweden, France, Spain and Portugal came under attack. The crisis ended with the adoption of an exchange rate mechanism very similar to a system of floating exchange rates, with the authorized fluctuation bands broadened to 15% (Bordo and Schwartz, 1996).

Mexico 1994
The crisis began on December 20, 1994 when Mexico decided to widen the peso's fluctuation band against the dollar. The end is generally dated to March 10, 1995 and the announcement of an austerity plan (Bordo and Schwartz, 1996;Whitt 1996).

Asia 1997
According to the IMF, Chakrabarti and Roll (2002), and Dungey et al. (2004, the crisis began on July 2, 1997 when Thailand decided to allow the baht to float after it had come under attack on May 14 and 15. The Philippines, Hong Kong, South Korea, Malaysia, Indonesia and Singapore were caught in the downdraft. According to Kaminsky and Schmukler (1999), the end of the crisis can be dated to January 13, 1998, when investors were reassured by the announcement of government reforms in Indonesia and a merger between two Singapore banks, as well as by upbeat comments from Morgan Stanley strategists about the "end of the Asian bear market". Candelon et al. (2005) examined the Hong Kong crisis, which they situate in the period from October 17 to 31, 1997, while Caporale et al. (2005) deal with the entire Asian crisis.
Lastly, Ball and Torous (2006) consider three possible durations for the crisis period: 1 year, 2 years and 3 years.

Brazil 1999
Dungey et al. (2006) say that the crisis began on January 13, 1999 with the devaluation of the real. It is hard to establish an end date because no landmark events occurred.
However, the crisis is generally referred to as the "January 1999 Brazilian crisis". We have therefore taken the final date to be the end of January 1999.

Russia 1998
The Russian crisis began on August 17, 1998, when the country defaulted on its debt, and continued until September of that year, when another crisis was triggered by the collapse of the hedge fund LTCM. We have therefore considered these two crises jointly, setting the end date for both at the end of the LTCM crisis.

Argentina 2001
The crisis began on November 1, 2001 when Argentina announced a debt restructuring plan. On December 5, the IMF refused to release funds to help the country, and the Argentine president was forced to resign on December 20. On December 23, 2001 the country announced that it was in default. For investors, the announcement marked the end of the crisis, and emerging spreads began to narrow (BIS, 2002).

Crashes 1987 equity crash
The steep drop in prices that occurred on October 19, 1987 lasted just one day, but it took several months to return to pre-crash levels. It is therefore difficult to set a precise end date. We have assumed that the crisis lasted until December 7, 1987, the day that prices troughed but before the market began to rally.

bond crisis
On February 4, 1994 the US Federal Reserve announced it was increasing its policy rate, taking the bond market by surprise (BIS, 1995). The announcement triggered a wave of panic and resulted in a massive bond sell-off in all industrial countries. We have dated the end of the crisis to November 3, 1994 when the steep rise in long-term interest rates came to an end (by which time, 10-yr yields in the USA had reached 8%).

E-crash
Triggered by the crash in tech stocks, the equity meltdown began on March 28, 2000.
We have dated the end of the crisis to April 14, 2000 when prices stopped falling.
Thereafter, the market entered a period of stagnation.

Corporate bankruptcies and crises of confidence LTCM 1998
The hedge fund Long Term Capital Management (LCTM) collapsed on September 23, 1998. Dungey et al. (2004 consider that the crisis ended when the US Federal Reserve decided to cut interest rates in order to contain the fallout. The Fed's decision was taken unexpectedly between two FOMC meetings on October 15, 1998.

Enron 2001
The onset of the crisis can be dated to November 28, 2001 when Moody's Investor Services decided to downgrade Enron, taking it from investment grade to high yield.
Although it was Moody's decision that sparked the mood of wariness which spread to all financial markets, signs that Enron was in trouble had emerged much earlier. On October 16, 2001 the company lowered its earnings guidance (BIS, 2002), and on November 8 it announced a retroactive adjustment to all its results since 1997. Enron filed for bankruptcy on December 2. It is extremely difficult to set a precise end date, and we consider that the crisis lasted throughout December.

WorldCom 2002
The crisis related to the bankruptcy of WorldCom began on June 25, 2002 when the company revealed accounting inaccuracies concealing losses of $3.8 billion in 2001 and 2002; it also announced 17,000 job cuts, equivalent to 20% of the workforce.
WorldCom filed for bankruptcy on July 11, and its shares fell 80% over the next four months. Once again it is very hard to establish an end date because the loss of confidence was exacerbated by fears relating to terrorist attacks in May and June 2002 and to political tensions between India and Pakistan. According to the BIS (2002a), the most significant crisis-related market movements occurred between July 10 and 23. We therefore consider that the crisis lasted until end-July 2002.

Subprime 2007
The Other crisis of confidence

9/11
The terrorist attacks on the USA on September 11, 2001 sparked a crisis of confidence across markets worldwide. It is hard to say precisely when the crisis ended, but we have considered that it lasted for the whole of September.
iv For Eurozone, we use the German bond index.
v The indices have minor differences. For IG indices, we selected a maturity of 7 to 10 years. However, for HY indices, maturity was not proposed as a selection parameter, so there are small differences in durations.
vi In fact, the literature focuses mainly on the increasing correlations between equity markets. To our knowledge, the expected impact of globalization on inter-class correlations has not been addressed.
vii The reference to earlier paper does not fully protect our results from endogeneity biases, as the way other authors have dealt with this issue might well have consequences on our results. Nevertheless, as far as volatility tests are concerned, no full protection against endogeneity does exist currently. Moreover, endogeneity pushes toward the acceptance of contagion (if crises are determined on the basis of high correlations, then the test statistic which measures the difference between quite-time and crisis correlation matrices will tend to be biased upwards). Therefore, the fact that this paper ends up rejecting contagion testifies against the presence of any significant endogeneity bias.
viii Or at least in so-called developed countries (Shackman, 2006).