Do Markets Cointegrate after Financial Crises ? Evidence from G-20 Stock Markets

The results of the single-equation cointegration tests indicate that patterns of cointegration in the two main and four sub-periods are not homogeneous. Two key findings emerge from the study. First, fewer stock markets cointegrated with S&P 500 during the crisis period than they did during the pre-crisis. In other words, as the 2008 financial crisis deepened, S&P 500 and G-20 stock indices moved towards less cointegration. The decreasing number of cointegrating relationships implies that the U.S. stock markets and other G-20 markets have experienced different driving forces since the start of the U.S. crisis. Second, among those markets that are cointegrated with S&P 500, they happened to be deeply affected by S&P and the shocks emerging from it. The 2007–2009 financial crises can be considered a structural break in the long-run relationship and may have resulted from effective joint intervention/responses taken by members of G-20 nations.


Introduction
The rising number of financial crises that happened in recent times and studies looking at these events from various perspectives has enriched the literature on financial crises.The world witnessed the dreadful events of 11 September 2001, the attack on the Twin Towers in New York, USA, which caused the stock markets to plunge in the USA.The aftermath of the tragedy was visible worldwide as its impact was felt in major equity markets, which suffered sharp declines, signifying that market participants perceived the event as a global shock.The 2001 event was followed by another crisis of greater magnitude in the United States, namely the housing bubble.The subprime mortgage crisis of 2007-2009, in which the housing market collapsed, causing the values of securities connected to housing prices to tumble thereafter, damaged major financial institutions.In recent years, due to the increase in the degree of integration of world capital markets, financial crises originating from one country have had a worldwide impact.The tragedy of 11 September 2001 and the financial crisis that followed affected more economies than the world has ever seen.Several other crises followed, such as the 2008-2009 Russian financial crises, the 2008-2012 Icelandic financial crises and the 2008-2010 Ireland banking crisis, and the news of the European sovereign debt (Euro) crisis followed, shattering investors' confidence and causing the global stock markets to plummet.
Since the seminal work of King and Wadhwani (1990) [1], international finance literature has examined how shocks are spread across the borders 1 .Despite the fact that much of the literature studies the cointegration between the U.S. stock markets and other countries, very little has explored the co-movement of the U.S. markets and the rest of the G-20 markets.Our paper joins this crisis transmission literature and investigates the transmission of shocks from the U.S. market (S&P) to those of the G-20 nations.The U.S. financial crisis had global implications and brought about a fundamental change in the global economic governance, with the G-20 taking over the leadership of the world economy from the G-7.The G-20 was formed as a group in 1999 after the Asian crisis of 1997, and is an international forum of finance ministers and central bank governors from the twenty most economically developed countries that meet annually to discuss the critical issues affecting the global economy.The G-20 countries, which constitute over three-quarters of the global GDP (on a market exchange rate basis) and over two-thirds of the world's population, became the de facto major global grouping of countries that is pushing responses to the crisis.The G-20's work only gained importance in recent years, especially after the Pittsburgh summit in September 2009, though the diplomatic unanimity was formed at the London summit in April 2009.To ease the 2007-2009 financial crises the leaders of G-20 agreed on an action plan, which included reinforcing international cooperation, reforming the international financial institutions and ensuring that the IMF, World Bank and other multilateral development banks have sufficient resources to continue playing their role in overcoming the crisis 2 .Building the resilience of the financial sector has been at the heart of the G-20's work since 1 Taylor and Tonks (1989) [2], Kasa (1992) [3] and, subsequently, Masih and Masih (1997) [4], Chowdhry (1994) [5] and Chowdhry et al. (2007) [6], among several others, have used the cointegration hypothesis to assess the international integration of financial markets.Rao and Naik (1990) [7], Chan et al. (1997) [8], Kasa [3] and Kwan et al. (1995) [9] have examined the integration of financial markets before the Asian economic crisis.The second group of studies examined the effects of the economic crisis on the financial integration after the Asian crisis.
the global financial crisis.To a large degree, the actions of the G20 economies helped to reverse the direction of the crisis and our findings lends credence to that fact.
In this study, we investigated if any cointegration exists between the G-20 markets with the U.S. after the stabilizing measures put into action by the G-20 countries during the global financial crisis of 2007-2009 3 .We also attempted to identify whether the G-20 markets moved toward more or toward less integration after the financial crisis of 2007 4 .Our findings provide evidence of the patterns of cointegration and of the effectiveness of G-20 intervention/responses to the crisis.We applied the following methodologies: (1) Cointegration (CI); (2) Vector Auto regression (VAR); (3) Granger Causality (GC) and (4) Variance decomposition (VC) to perform two levels of analysis: bivariate analyses, using the U.S. (S&P) and each individual country, and multivariate analyses, using regional cointegration.
The paper is organized as follows.Section 2 presents the main contributions of the literature.Section 3 discusses the data and the sample, while Section 4 deals with the methodology.Section 5 reports and discusses the empirical results, while Section 6 concludes the study.

Literature Review
Market contagion and co-integration/co-movement related to financial crises and their responses to the market are issues of enormous interest in the literature.Bekaert, Harvey and Ng (2005) [13] have identified contagion in equity markets.Papers have been written proposing quantitative measures of contagion (Karolyi (2003) [14], Dungey et al. (2004) [15]) and developing theories to explain it (Allen and Gale (2000) [16]).Concerning the U.S. financial crisis, Wei and Hui (2011) [17] found that the average decline in stock prices during the crisis in a sample of 4000 firms in 24 emerging countries was more severe for those firms intrinsically more dependent on external finance (in particular on bank lending and portfolio flows).Hau and Lai (2011) [18] state that stocks with a high share of equity funds ownership performed relatively well during the crisis, whereas stocks with ownership links to funds that were heavily affected by portfolio losses in financial stocks severely underperformed.Yang et al. (2003) [19] examined whether long-run integration between the United States and many international stock markets has strengthened over time.Their results show that there is no long-run relationship between most of these markets and the United States5 .
3 Angeloni, I. and J. Pisany-Ferry (2015) [11].The G20 acted as a crisis manager when global financial markets were under threat in 2008 and 2009, and contributed to a positive outcome.4 Duca and Stracca (2014) [12] ran an event study to test whether G20 meetings at ministerial and leaders level have had an impact on global financial markets.By focusing on the period from 2007 to 2013, looking at equity returns, bond yields and measures of market risk such as implied volatility, skewness and kurtosis.They found that G20 summits have not had a strong, consistent and durable effect on any of the markets that we consider, suggesting that the information and decision content of G20 summits is of limited relevance for market participants.
Many researchers have also pointed out the increased vulnerability to crises that comes with financial and economic integration 6 .Bekaert et al. (2011) [13], using the 2007-2009 financial crisis as a laboratory case, analyzed the transmission of crises to country-industry equity portfolios in 55 countries.They find statistically significant evidence of contagion from U.S. markets and from the global financial sector, but the effects are economically small.By contrast, there has been substantial contagion from domestic equity markets to individual domestic equity portfolios, with its severity inversely related to the quality of countries' economic fundamentals and policies.Their findings confirm the old "wake-up call" hypothesis, with markets and investors focusing substantially more on country-specific characteristics during the crisis.Slimane et al. (2013) [28] found that the spread of the global financial crisis of 2008/2009 was rapid and affected the functioning and the performance of financial markets.Their paper investigates the patterns of linkage dynamics among three European stock markets, France, Germany and the U.K., during the global financial crisis by analyzing the intra-day dynamics of linkages among these markets during both calm and turmoil phases and applying a VAR-EGARCH framework to high frequency five-minute intra-day returns on selected representative stock indices.It found evidence that the interrelationship among European markets increased substantially during the period of crisis, pointing to an amplification of spillovers.Furthermore, during this period, French and U.K. markets herded around the German market, possibly due to the behavior factors influencing the stock markets on or near dates of extreme events 7 .Wasim et al. (2014) [37] examined the contagion effects of the stock markets of Greece, Ireland, Portugal, Spain and Italy (GIPSI), as well as the U.S. stock markets, on seven Eurozone and six non-Eurozone stock markets.Empirical results suggest that among GIPSI stock markets, Spain, Italy, Portugal and Ireland appear to be most contagious for Eurozone and non-Eurozone markets.Their study found that the Eurozone countries of France, Belgium, Austria and Germany, as well as the non-Eurozone countries of UK, Sweden and Denmark, were strongly hit by the contagion shock.Prorokowski (2013) [38] combined quantitative and qualitative research methods and painted the picture of the contemporary European financial markets with particular attention paid to the existing cross-market linkages, vulnerabilities, systemic risks and flawed regulations that altogether constituted a group of factors propagating the financial crisis contagion.Thakor (2015) [39] reviewed the literature results point towards a decreasing number of common stochastic trends influencing the stock markets, i.e., the degree of convergence among European stock markets has been increased during the recent two decades.6 Mendoza and Quadrini (2010) [23] for a theoretical analysis, and Fratzscher (2012) [24] for empirical evidence during the 2007-2009 crises.Khan Taimur A. (2011) [25], paper examines the long-run convergence of the United States and 22 other developed and developing countries.Using daily data to run the Johansen (1988) [26] and the Gregory and Hansen (1996) [27] test, and find stock markets of most countries have become cointegrated by 2010.Also using the relative risk of each country (the CAPM model) to measure performance of each country over the recession of the 2000s and finds that the relative risk of a country is a good predictor of country performance in a recession.
on the 2007-2009 crises and discusses the pre-crisis conditions, the crisis triggers, the crisis events, the real effects and the policy responses to the crisis.The author states that the pre-crisis conditions contributed to the housing price bubble and the subsequent price decline that led to a counterparty-risk crisis in which liquidity shrank due to insolvency concerns.The policy responses were influenced both by the initial belief that it was a market-wide liquidity crunch and the subsequent learning that insolvency risk was a major driver.Gennaioli et al. (2015) [40] modeled financial markets in which investor beliefs are shaped by representativeness.The authors express that the investors overreact to a series of good news because such a series is representative of a good state.A little bad news does not change the minds of investors because the good state is still representative, but enough bad news leads to a radical change in beliefs and a financial crisis.The model generates debt over-issuance, "this-time-is-different" beliefs, neglect-of-tail risks, and under-and over-reaction to information, boom-bust cycles and excessive volatility of prices in a unified psychological model of expectations.Reinhart et al. (2014) [41] examined the evolution of real per capita GDP around 100 systemic banking crises.Part of the costs of these crises is due to the protracted nature of recovery.On average, it takes about 8 years to reach the pre-crisis level of income; the median is about 6.5 years.Five to six years after the onset of the crisis, only Germany and the United States (out of 12 systemic cases) have reached their 2007-2008 peaks in real income.Forty-five percent of the episodes recorded double dips.Post-war business cycles are not the relevant comparator for the recent crises in advanced economies.

Data and Sample
The indices (equity daily price indices (PI)) in U.S. dollars (USD) or conversions to USD are used in the study and include Australia, Brazil, India, France, Germany, UK, Italy, Indonesia, South Korea, Argentina, Mexico, Japan, Russia, Canada, China, and South Africa8 .After matching the sample periods for each time series, a common sample period from 1 January 2000 to 30 April 2013, with the number of daily observations for each panel, is selected mainly from Yahoo Finance, the Federal Reserve St. Louis database, and Quandl.This period encompasses the three major events that have occurred since the advent of the 21st century-11 September 2001 (hereafter referred as 9/11), the 2007-2009 subprime mortgage crises in United States, and the burst of Europe's sovereign debt (Euro) crisis in 2010.The purpose was to conduct extensive empirical research on the three events and to compare the impact these events had on the major economies.The sample period has been divided as follows: Two (2) main periods, four (4) subsample periods, and one (1) overall period.Main Period 1 ranges from 1 January 2000 to 31 December 2008.This period coincides with U.S. President George W. Bush's two terms in office, which also coincided with 9/11 and the start of the subprime mortgage crises in the United States.Main Period 2 spans from 1 January 2009 to the end of our sample period, i.e., 30

Methodology
Engle & Granger's (1987) [42] residual-based single-equation of cointegration was employed to analyze the data and estimated the following long-run equilibrium equation: where yt represents S&P 500 and Xt are individual stock market indices of the G-20 nations.
The augmented Dickey-Fuller (ADF) was used to check whether our time series data are I (1).For a variable to be I (1), the variable must be non-stationary at its level and become stationary after the first difference.We estimated ADF in Equation ( 2), shown below, in which Dt is a vector of deterministic terms.The single-equation technique was preferred over Johansen cointegration because of its intuitive interpretability.While the Johansen methodology is suitable for a system that involves more than two variables, Engle-Granger cointegration has an advantage when performing bivariate testing (Alexander, 1999 [43]).In this study, we performed bivariate testing between S&P 500 and the stock market of each G-20 country.In addition to the cointegration test, VAR and innovation accounting was also applied to analyze the series that are cointegrated with S&P 500.Regarding the choice of U.S. stock markets, we used S&P 500 instead of the Dow Jones Industrial Average (DJIA) because S&P 500 is a broader measure of market movements than DJIA.

Empirical Results and Discussion
The unit root test was conducted on all the periods.The results in Table 1 show that the time series process of all stock indices in all periods are non-stationary at their levels, except British and German Indices in Sub-Period 1, Chinese Index in Sub-Period 2 and South African Index in Sub-Period 4. Their first differences are stationary in all periods.Table 2 presents the Pearson's Correlation.The average correlations with S&P 500 of Periods 1 and 2 are very similar, with the correlation coefficients of 0.92 and 0.87 respectively.Among the four sub-periods, Sub-Period 4 shows lowest average correlation of 0.32.The average correlation of Sub-Period 1 is 0.65, while those of Sub-Period 2 and 3 are 0.92 and 0.87 respectively.During the Sub-Period 1, the stock markets of France, Japan and Canada show highest correlations with S&P 500.The markets that show highest correlation in Sub-Period 2 are Germany, the United Kingdom and France.The French and Italian markets exhibit the highest correlations with S&P 500 in Sub-Period 3. We find the markets that are highly correlated with S&P 500 during Sub-Periods 1 to 4 are in Europe and Asia.In the next section, we perform the bivariate cointegration test between the stock market in the G-20 country with S&P 500.Table 3 shows the results of single-equation cointegration tests.Nikkei is the only index that cointegrates with S&P during the Main Period 1.In Main Period 2, only Mexico's Mexabol cointegrates.Testing cointegration during sub-periods, we find that three stock markets namely Canada, Japan and France are cointegrated with S&P 500 in Sub-Period 1.Like Sub-Period 1, three markets cointegrate with S&P 500 in Sub-Period 2; including Germany, UK and Italy.The number of cointregating relationships reduces in Sub-Periods 3 and 4.Only two markets, France and Italy, are cointegrated with S&P 500 during Sub-Period 3. Sub-Period 4 shows no evidence of cointegration.In addition to performing single-equation cointegration tests, we also used the Johansen methodology 9to estimate the same data set to obtain the results presented in Tables A6-A11 of the Appendix section.
The results of the Johansen estimation indicate that no markets are cointegrated with S&P during the two main periods and four sub-periods, except French and UK Indices, which cointegrate with S&P in Sub-Period 3. Regarding the choice of using either the Engle-Granger or the Johansen methodology to estimate the bivariate cointegration relationships discussed in the methodology section, we decided to adhere to the cointegration tests performed using the Engle-Granger methodology.The overall result shows that there are more stock markets of the G-20 nations that are cointegrated with S&P 500 during the pre-crisis era than during the post-crisis periods.The results of the cointegration analysis in Table 3 reveal two observations.First, it shows that the markets cointegrated with S&P 500 were deeply affected by S&P 500 and shocks emanating from it.This finding is robust, as the summary of the impulse-response analyses shown in Table 4 and the vector auto regression analyses shown in Table 5 and Tables A1, A2 in the Appendix confirm, and is consistent with the study of Slimane et al. (2013) [28].Second, in general, an increasing number of cointegration relationships indicate that stock markets become more integrated over time because they are being driven by the same common stochastic trends (Rangvid 2001 [21]).However, our results show the opposite.Fewer stock markets cointegrate with S&P 500 during the crisis than during the pre-crisis.In other words, S&P 500 and G-20 stock indices moved toward less cointegration after the 2008 global financial crisis.The findings are similar to those of Bekaert et al. [13], which found weak evidence of contagion from U.S. markets to equity markets globally.The decreasing number of cointegration relationships in our findings may imply that the U.S. stock markets and other G-20 markets have experienced different driving forces since the start of the U.S. crisis.It may also imply that the 2008 financial crisis can be considered a structural break in the long-run relationship.For the sake of brevity, the sub-periods' tables (Tables A1-A5) are not reported here.
Next, we tested to see if the regional market cointegration exhibits the same pattern as that of each individual market and S&P 500.We classify G-20 markets into Asia, Europe, and Latin America.The results in Table 5 below illustrate that only Sub-Period 1 European markets exhibit regional cointegration.Sub-periods 2, 3, and 4 show no cointegration 10 .This implies that, as the crisis deepened, fewer G-20 markets were cointegrated with S&P, and no regional markets were integrated.

Conclusions
The results of the single-equation cointegration tests indicate that patterns of cointegration in all main and sub-periods are not homogeneous.Two major findings emerge from the study.First, fewer stock markets cointegrated with S&P 500 during the crisis period than they did during the pre-crisis period.As the 2007 financial crisis deepened, S&P 500 and G-20 stock indices moved toward less cointegration.The decreasing number of cointegrating relationships may indicate that the U.S. stock markets and other G-20 markets have experienced different driving forces since the start of the U.S. crisis.Second, among those markets that were cointegrated with S&P 500, they happened to have been deeply affected by S&P and the shocks that emerged from it.The 2007-2009 financial crises can be considered a structural break in the long-run relationship and may have resulted from effective joint intervention/responses taken by members of G-20 nations.For international investors, findings suggest that, in the long run, there were probable rewards, which may have been acquired by smart investors through portfolio diversification.While the global financial markets were being assimilated, with the economies turning out to be more interdependent, the instantaneous outcomes of the markets may not have been associated to the rising ability of information-processing by the financial markets.Our results for the sample periods, including the sub-periods, support the findings of Bekaert et al. (2011) [13], which pointed out that, during most of the global crisis, the market's external exposure played a very small role in determining its equity market performance.Though Prorokowski (2011) [38] states that the role of the USA in propagating the financial crisis was far more important, his study considers the financial crisis contagion in Europe only and recommends a future study that would investigate the role of the USA in propagating the global financial crisis.We believe that the findings from our study fill the gap and contribute to the literature.It is apparent from the study that, as the crisis deepened, the G-20 markets moved toward less cointegration with the U.S. market.G-20 markets perhaps should be investigated more intensely in a future study to determine whether the degree of contagion was lessened by a single country's domestic intervention or by the G-20's joint international responses to the crisis.Note: 1 = cointegrated in Sub-period 1; 2 = cointegrated in Sub-period 2; 3 = cointegrated in Sub-period 3.

Appendix
April 2013, and includes the post U.S. financial crises and Europe's sovereign debt (Euro) crisis.Subsample Period 1 covers 1 January 2000 to 10 September 2001 and is labeled as "Pre-9/11."Subsample Period 2 extends from 15 September 2001 to 31 December 2006.U.S. stock markets were closed for a few days immediately following 9/11.This subsample period is labeled as "Post 9/11 and Pre-Financial Crisis."Subsample Period 3, which extends from 1 January 2007 to 31 December 2009, is labeled as "Peak Financial Crises," while the last subsample, Period 4, covers 1 January 2010 to 30 April 2013 and is labeled as "Post U.S. Crisis and Euro Crisis."The overall period of the data sample is from 1 January 2000 to 30 April 2013.

Table 1 .
Unit Root Test Results.
Note: Since the results of Tau and Z-statistics do not contradict, we report only Z statistics.*, ** and *** represent levels of significance at 10%, 5% and 1%.NA = Data not available.Time Periods: Main Period 1
Note: * Average 10-period response to a one standard shock.The table shows that the response of the markets to the shock that emanated from S&P is substantially greater than that of the stock markets to S&P.

Table 5 .
Test for Possible Regional Cointegration.

Table A6 .
Test for Johansen Cointegrtion in Main Period 1.

Table A7 .
Test for Johansen Cointegrtion in Main Period 2.

Table A8 .
Test for Johansen Cointegration in Sub-period 1.

Table A9 .
Test for Johansen Cointegration in Sub-period 2.

Table A10 .
Test for Johansen Cointegration in Sub-period 3.

Table A11 .
Test for Johansen Cointegration in Sub-period 4.

Table A12 .
Test for Possible Regional Cointegration in Asia.

Table A13 .
Test for Possible Regional Cointegration in Latin America.

Table A14 .
Test for Possible Regional Cointegration in Europe.

Table A15 .
Test for Possible Regional Cointegration in BRIC.