Stock Returns Dynamics around U.S. Stock Market Crises and Inverted Smiles

AM AL-Rjoub Volume 8, Issue (1) Journal of New Business Ideas & Trends 1, 8(), pp. 7-. http://www.jnbit.org Stock Returns Dynamics around U.S. Stock Market and Inverted Smiles Samer AM AL-Rjoub Department of Banking and Finance Hashemite University B.O.Box 33195 Zarqa,13133 Jordan E-mail: salrjoub@hu.edu.jo Abstract Purpose- This study examines the historical episodes of financial crises in the United States over the last one hundred years of major stock market crashes in the Dow Jones industrial average, the S&P 5 and the NASDAQ indices. Design/methodology/approach- Patterns of U-shaped first, third, and forth moments of stock retunes are recorded around sixteen financial crises in the U.S history. Findings- The probability of extreme outcomes increases before and after the crises signaling the beginning of the crisis but not the ending of a one. Patterns of an inverse U- shaped volatility around crises are also reported. The major result from VaR calculations shows that Value at Risk increase dramatically during crises and takes the shape of an inverted smile. Results are robust across the DJIA, S&P 5 and NASDAQ. Keywords: Dynamics,, Inverted Smile. JEL Classification Numbers: G1 JNBIT Vol.8, Iss. (1) 7

AM AL-Rjoub Volume 8, Issue (1) Introduction Global economies have experienced crises regardless whether they are strongly or poorly developed financial systems, whether they own weak or strong banking systems, and whether they are highly dependent on external capital flows or lowly dependent. Actually in the last three decades there have been roughly more than 1 crises around the globe some of which are catastrophic and some fade away. For this reason a renewed calls for increased concern about asset price bubbles, their bursting, and aftermath have emerged and encourages the search for developing alternative early warning models that send timely and correct signals about the start of a financial crises and hence to reduce the risks of future ones. Joseph Stiglitz, the Nobel laureate in economics, once remarked, It is becoming rarer for a country not to have a crisis than to have one (3a, p. 31). Although we cannot be sure that all these crashes were bubbles, a bursting of a bubble surely results in a stock market crash, and so analyzing the price behavior before and after the crashes can provide some clues as to the timing of a bursting bubble and the ending of a one. This knowledge might serves as an early warning system to stock markets and help policymakers to take their precautions and to arrange for proper actions. In addition studying the stock return behavior after the crises can give us some clue whether a crisis ends and whether the threat has gone away. To understand the behavior of stock market bubbles around market crashes, we pursue a historical approach. This paper examines the historical episodes in the United States over the last one hundred years of major stock market crashes in the Dow Jones industrial average, the S&P 5 and the NASDAQ indices. This paper contributes to the existing literature in three ways: (1) by providing information about stock market behavior before and after the occurrence of a financial crises and by providing some clues as to the timing of a bursting bubble and the ending of a one, () by adding new methodology to the existing methods used in the literature of early warning systems based on fundamentals or signal (Such as, Probit /logit, Markov Switching Regimes with time varying probabilities and leading and lagging indicators, and the signals approach )1, and (3) by calculating Value at Risk (VaR) for the stock markets before, during and after the crises. By doing so the study provides some light on how bad things can get when financial crises occur. This article provides an alternative framework by measuring the probability of extreme outcomes before, during and after the crises, for use in early warning systems constructions, and for use in announcement mechanism of an ending crash to allow policymakers to take appropriate steps. This should provide much more information about the dynamics of crises before and after a bubble burst. This article use the crisis specification adopted by Mishkin, Eugene and White () where they defined stock market crash as percent decline in the stock market. The study focuses on 16 crises dates : 193,197,1917, 19,199, 193-33, 1937, 19, 196,196,1969-7, 1973-7,1987,199,, and 8. The rest of the paper is organized as follows. Literature review and brief history of crises are in section, section 3 describes the data and the procedures used to identify the stock market crashes in the United States over the last one hundred years. Section pursue a sequence of events approach to discuss the patterns of price movements and volatility months before and after the defined collapses, and extends further by calculating VaR in 1 Berg and Pattillo (1998) assess the performance of these early warning systems and find that false alarms almost always out-counter appropriate warnings JNBIT Vol.8, Iss. (1) 8

AM AL-Rjoub Volume 8, Issue (1) the methodology and VaR approximation. We report empirical results in section 5 and section 6 concludes. Literature Review The subject at hand deals with two interrelated issues discusses separately in the literature of financial crises. The first issue related to volatility and return patterns around crises and the second issue related to early warning systems developed to predict crises. In what follows the paper examines the two issues separately. Stock return behavior around crises The literature on the stock returns behavior around the crises are rare, most existing empirical work is on examining the relationship between stock returns and conditional volatility and the behavior of stock prices after the crises, but fewer has examined the stock return behavior and changes in volatility around (before and after ) the financial crisis. Li and Kwok (9) study the growth volatility of GDP and its components and stock market prices, using quarterly data of five crisis affected East Asia economies. To study GDP and stock market volatility in the pre- and post- Asian financial crisis (AFC) period, and to study volatility clustering and volatility asymmetry, they use Hodrick- Prescott (HP) filter to get smoothed trend series, then they use ARCH family of models. Their results show a decrease in GDP volatility at Japan, Hong Kong SAR and Chinese Taipei, while Singapore and South Korea showed a great increase in volatility in the post- AFC period. Moreover, stock markets of Japan and Singapore show a slightly increase in the post-crisis while South Korea had the most stock prices volatility. There was also evidence for volatility clustering, and it shows weak or no evidence of volatility asymmetry except in South Korea. Leon (7) examines the relation between stock returns and volatility in the BRVM (the regional stock market of the West African Economic and Monetary Union). He used EGARCH in mean framework assuming two distributions for error terms; the normal and the student, and weekly closing prices for ten of the most actively traded BRVM stocks. Leon finds a positive but insignificant relationship between stock market return and volatility, and finds that volatility change over the business cycles where it becomes higher during booms. Caporale and Spagnolo (3) examine the effects of financial crisis on the relationship between stock prices (stock markets) and output growth (real economy), using bivariate GARCH (1, 1) model and monthly data for three Asian countries and three industrialized countries (for comparison purposes). In their results, Caporale and Spagnolo show that the stock market volatility positively and significantly influence the output growth volatility. For the crisis which affected East Asia countries there was a stronger and more significant spillover of the volatility from stock markets to output growth in the post-crisis period comparing with the pre-crisis period. Choudhry (1996) study the stock return volatility persistence in emerging markets before and after 1987 crash, using GARCH in mean approach and monthly data from six emerging stock markets. He finds changing in volatility before and after the crisis of October 1987, but these changes were not uniform and related to factors other than this crisis depending on individual markets. JNBIT Vol.8, Iss. (1) 9

AM AL-Rjoub Volume 8, Issue (1) Schwert (199) study daily stock returns and volatility behavior during and around crisis, focusing on the crash of October 1987 whether it differ from the average for the previous crashes, using daily data from 1885 to 1987 and lagged return shocks simultaneously with lagged volatility measures plus lagged high-low spreads. The study shows that stock return volatility increases when stock prices collapse, also during business cycle recessions and bank crisis, which verify Schwert (1989) results. Furthermore Schwert shows that the volatility before 1987 crash was lower than the average of other crisis, above the average for the five days followed October19, but on the whole it was lower than the average, it rose faster than the normal levels at the time of the crash and it returned to lower afterward. Schwert (1989) investigates the behavior of stock prices and volatility in the United States documenting its relation with business cycles and financial crashes, in the period between 183 and 1987. For this purpose he characterized the normal stock volatility behavior, and then he analyzed the sudden changes in volatility that related to the financial crisis and recessions. He uses two different methods of modeling volatility; linear autoregressive model for the conditional means and conditional standard deviations of stock returns, and nonlinear regime switching model. Schwert shows that stock return volatility increases after stock prices falling, during recessions, around financial crisis. He also shows that stock markets respond deeply to the banking crisis, where stock prices drop before the major crisis, and stocks volatility increase after the major crisis. Early warning systems There have an ongoing effort by the International Monetary Fund (IMF) to develop early warning systems since the 1999. Researcher's interest where directed more toward emphasizing leading fundamentals or signals that preceded the crisis and developing early warning systems based on the recurrences of some incidents, than searching for recurring stock returns patterns before and after the crisis ( which this paper actually do). Most early warning systems models developed so far concentrated on currency, debt and banking crises ( see for example, Ciarlone and Trebeschi (5), Berg et al.(), Kaminsky et al. (1998), and Frankel and Rose (1996)). Until now, however, little work has been done on studying the dynamics of returns around the crises and the recurring of a pattern. Recently, Rose and Spiegel (9), model the causes of the 8 financial crisis together with its manifestations, using a Multiple Indicator Multiple Cause model. Their analysis is conducted on a cross-section of 17 countries; and focuses on national causes and consequences of the crisis, ignoring cross-country contagion effects. They combine 8 changes in real GDP, the stock market, country credit ratings, and the exchange rate and explore the linkages between these manifestations of the crisis and a number of its possible causes from 6 and earlier. Despite the fact that Rose and Spiegel use a wide number of possible causes in a flexible statistical framework, they were unable to link most of the commonly-cited causes of the crisis to its incidence across countries. Rose and Spiegel (9) states "This negative finding in the cross-section makes us skeptical of the accuracy of early warning systems of potential crises, which must also predict their timing." Berg, Borensztein and Pattillo () assess different early warning systems developed by the International Monetary Fund (IMF) since 1999. Their main results were mixed. They found that one of the long-horizon models has performed well relative to pure guesswork and to available non-model-based forecast such as agency ratings and private analysts' currency crisis risk scores. They also reports that the two short-horizin private sectors models performed poorly. JNBIT Vol.8, Iss. (1) 3

AM AL-Rjoub Volume 8, Issue (1) Kaminsky, Lizondo and Reinhart (1998) propose a nonparametric approach, called the signals approach, to predict banking and currency crisis. The signals approach involves the monitoring of several indicators that tend to behave differently in periods preceding a crisis compared to their pre-crises behavior. The model attempts to predict the probability of a crisis within the next months. This model had some success in predicting crises out of sample (the Asia crisis of 1997/1998). Berg and Pattillo (1999a,b) retest the predictability of Kaminsky, Lizondo and Reinhart (1998) approach in out-of-sample to predict the Asian crisis and extend further to test the usefulness of interpreting predictive variables in terms of discrete threshold, the crossing of which signals a crisis. They called their approach the Developing Country Studies Division (DCSD) model. Berg and Pattillo found that a better simple assumption is that the probability of crises goes up linearly with changes in the predictive variables. Mulder, Perrelli and Rocha (1) have developed a new early warning system that adds balance sheet variables and proxies for standard DCSD model. Major variables such as leveraged financing, high ratio of short-term debt to working capital, balance sheet indicators of a bank and corporate debt to foreign banks as a share of exports and regime switching proxies for shareholders rights was found to be important in predicting the probability of a crisis. A Brief History of the Major Financial in the World Economic and financial crises that hit the world economies have long history. In what follow we list the most sever crises in the twentieth and twenty first centuries, starting from the crash of 197, and continue to the current global financial crises of 8. Detailed information about the crisis, its origin, date of peaks and troughs in cycles, reasons and solutions are all reported in Table 1. By reading through Table 1 we can notice that the effects of crises are reflected mainly on stock volatility and returns. It is clear that during the crises stock volatility register high levels and stock prices falls strongly in both developed and emerging markets. In emerging markets the effects are rapid, steep, and prolonged. This is the aftermath of a crisis but what about the period preceding it. Data In order to account for different index specifications and different categorizations of companies that comprises these indices. We look at historical episodes in the United States over the last one hundred years of major stock market crashes in the Dow Jones industrial average (DJIA), the S&P 5 and the NASDAQ indices. Data are monthly and spans from 19 to 9 for the DJIA, from 1971 to 9 for the NASDAQ and from 195 to 9 for the S&P 5. Based on this identification, DJIA index will account for all crisis date specification, NASDAQ will cover five crises after 195 and S&P 5 will cover seven. Summarized information about the crisis; date of peaks and troughs in cycles, and percentage of declines in indices are all reported in Table 1. We examine our assumption under three different scenarios, namely the index for the blue-chip companies (DJIA) and the index for the largest 5 companies (S&P 5) and a comprehensive index across different company sizes (NASDAQ). JNBIT Vol.8, Iss. (1) 31

AM AL-Rjoub Volume 8, Issue (1) Table 1: History of the United Sates of America Financial Methodology and VaR approximation The first steps is to monitor signals around crises by monitoring return data properties reflected in the four moments of distributions. The basic idea of the signals approach, suggested by this paper, is to monitor several indicators that tend to exhibit an unusual behavior in the periods surrounding a crisis. We implement this signaling approach for the U.S. and watch the distributional prosperities of the stock market returns data. The search is for a signal in the first four moments (Mean, Standard Deviation, Skewness and Kurtosis) of the return distributions of DJIA, S&P 5 and NASDAQ. Then we continue by calculating Value at Risk (VaR) approximation to calculate the loss level that will not be exceeded in N days with a specified probability. For this research the 99% confidence level is used when calculating VaR. The simplest assumption is that daily gains/losses are normally distributed and independent. It is then easy to calculate VaR from the standard deviation (1-day VaR=.33σ ), where.33 is the Z-score at the 99% confidence interval and 1% confidence level, and σ is the standard deviation of returns for the three time specifications, namely; before, during and after the crises (for more details see Hull, 9). If data distribution is far from normal the VaR measurement will be adjusted for non-normality by calculating a new measure of volatility extracted from the Generalized Autoregressive Conditional Heteroscedasticity (GARCH) model proposed by Bollerslev in 1986. By this we get an approximation of what the VaR estimate will be around and during JNBIT Vol.8, Iss. (1) 3

AM AL-Rjoub Volume 8, Issue (1) crises. This estimate constitute an approximation for the market risk VaR for the three indices. Empirical results Mean A summery statistic for an interactive dummy that accounts for the months during the crisis and the six months surrounding the crises, namely six months before and six months after. Results are reported in Table, 3 and respectively. Interesting results emerge. Table shows,on average,the mean average return of interactive dummy that accounts for returns during the crises is consistently negative and equal to, -.5, -.39, and -.67 for DJIA, S&P 5, and NASDAQ respectively under different sample specifications. Table : Summery statistics for the return and the interactive dummy for the return on different indices during the crises RTDJIA RTDJIAD RTSP RTSPD RTNASDAQ RTNASDAQD Std. Dev. Skewness Kurtosis Jarque-Bera Probability.38381 5.89 -.75879 8.77676 196.76 -.5136 3.3515-3.7891 33.99 589.8.58386.1531 -.68931 5.555679 51.91 -.388333.18597-3.11699 9.1636 17.37.6685 6.18669 -.86113 5.76367 6.33 Observations 13 13 7 7 67 RTDJIA stand for the return for the whole sample, RDJIAD, is an interactive dummy during the crisis. RTSP stand for the return for the whole sample, RTSPD, is an interactive dummy during the crisis. RNASDAQ stand for the return for the whole sample, RNASDAQD, is an interactive dummy during the crisis -.679.1857 -.73339 1.7135 693.391 67 Mean Table 3 shows,on average, the mean average return of interactive dummy that account for returns before the crises is consistently positive and equal to,.8,.9, and. for DJIA, S&P 5, and NASDAQ respectively and for different sample specifications. Table 3 : Summery statistics for the return and the interactive dummy for the return on different indices before the crises RTDJIA RTDJIAB RTSP RTSPB RTNASDAQ RTNASDAQB Std. Dev. Skewness Kurtosis Jarque-Bera Probability.38381 5.89 -.75879 8.77676 196.76.836 1.36356-1.976158 91.8619 381..58386.1531 -.68931 5.555679 51.91.93.78571.573.963 91.7.6685 6.18669 -.86113 5.76367 6.33 Observations 13 13 7 7 67 RTDJIA stand for the return for the whole sample, RDJIAB, is an interactive before the crisis. RTSP stand for the return for the whole sample, RTSPB, is an interactive dummy before the crisis. RNASDAQ stand for the return for the whole sample, RNASDAQB, is an interactive dummy before the crisis.19968 1.58781 9.3611 99.8889 188989.1 67 Table, which summarize the statistics for the return and the interactive dummy for the return on different indices after the crises, shows, on average, that mean returns register positive numbers across all the indices. JNBIT Vol.8, Iss. (1) 33

AM AL-Rjoub Volume 8, Issue (1) Mean Table : Summery statistics for the return and the interactive dummy for the return on different indices after the crises RTDJIA RTDJIAAFT RTSP RTSPAFT RTNASDAQ RTNASDAQAFT Std. Dev. Skewness Kurtosis Jarque-Bera Probability Observations 13 13 7 7 67 RTDJIA stand for the return for the whole sample, RDJIAAFT, is an interactive after the crisis. RTSP stand for the return for the whole sample, RTSPAFT, is an interactive dummy after the crisis. RNASDAQ stand for the return for the whole sample, RNASDAQAFT, is an interactive dummy after the crisis Plot 1.a shows that mean returns take a U-shape for the three indices and across all time specifications that was tested in Tables, 3, and respectively. Plot 1.a: Mean return around and during crises for the DJIA, S&P 5 AND NASDAQ indices respectively.. -. -. -.6.38381 5.89 -.75879 8.77676 196.76.183 1.6313 7.8556 13.333 933555.3.58386.1531 -.68931 5.555679 51.91.1767 1.17355.69 38.17598 3976.7.6685 6.18669 -.86113 5.76367 6.33.9165 1.6678.59376 37.6 37. 67 -.8 Return DJIA After Return DJIA During Return DJIA Before.. -. -. -.6 -.8 Return S&P After Return S&P During Return S&P Before.. -. -. -.6 -.8 Return NASDAQ After Return NASDAQ During Return NASDAQ Before When comparing the dummies before, during and after the crises Table 5 emerges. Table 5 shows that returns possess a U-Shaped pattern around financial crisis. It moves from positive and moderate before the crises to negative and low during the crises to positive and high after the crises. JNBIT Vol.8, Iss. (1) 3

AM AL-Rjoub Volume 8, Issue (1) Moreover, Table 5 shows that standard deviation takes an inverse U-Shaped curve. It is much higher during the crises than before and after the crises, actually the standard deviation of DJIA during the crises is.5 folds DJIA before, the standard deviation of S&P during the crises is 3 folds of S&P before, and finally the standard deviation of NASDAQ during the crises is.6 folds than that of NASDAQ before. Plot 1.b shows that standard deviation of returns take the shape of an inverse U- Shaped curve for the three indices and across all time specifications that was tested in Tables, 3, and respectively and summarized in 5. Plot 1.b: Standard deviation of return around and during crises for the DJIA, S&P 5 AND NASDAQ indices respectively 5 3 1 Return DJIA After Return DJIA During Return DJIA Before 5 3 1 Return S&P After Return S&P During Return S&P Before 5 3 1 Return NASDAQ After Return NASDAQ During Return NASDAQ Before Skewness statistics is positive and very high before the crises for the S&P 5 and NASDAQ except for the DJIA index where the skewness parameter was negative for the six month before the crises dummy. This can be justified by the fact that the DJIA sample was the longest and account for the major crises of 193,197,1917, 19,199, 193-33, 1937, 19, 196. Plot 1.c shows that Skewness statistics of returns take the shape of a U-Shaped curve for the three indices and across all time specifications that was tested in Tables, 3, and respectively and summarized in 5. JNBIT Vol.8, Iss. (1) 35

AM AL-Rjoub Volume 8, Issue (1) Plot 1.c: Skewness of return around and during crises for the DJIA, S&P 5 AND NASDAQ indices respectively 1 8 6 - - -6 Return DJIA After Return DJIA During Return DJIA Before 6 5 3 1-1 - -3 - Return S&P After Return S&P During Return S&P Before 1 8 6 - - Return NASDAQ After Return NASDAQ During Return NASDAQ Before More interestingly when comparing the Kurtosis parameter results shows that the Kurtosis is much higher before the crises than during it. This result holds for the three indices. The distribution is peaked (leptokurtic) relative to the normal. Plot 1.d shows that Kurtosis parameter of returns also takes the shape of a U-Shaped curve for the three indices. Plot 1.d: Kurtosis of return around and during crises for the DJIA, S&P 5 AND NASDAQ indices respectively 1 1 1 8 6 Return DJIA After Return DJIA During Return DJIA Before 5 3 1 Return S&P After Return S&P During Return S&P Before JNBIT Vol.8, Iss. (1) 36

AM AL-Rjoub Volume 8, Issue (1) 1 1 8 6 Return NASDAQ After Return NASDAQ During Return NASDAQ Before This result provides evidence that the possibility of extreme outcomes is much higher around the crises than before the crises. The possibility of extreme outcomes persist also after the crises, the signs that might signals a crises might also signals the ending of a one. Even though the two distributions are far from normal, the distribution tails for returns before and after crises is much thicker than that of the distribution for returns during the crises. Extreme outcomes can happen. This finding is extremely important and constitutes an important block in early warning systems that can be developed to predict crashes. The fact is that kurtosis will register high numbers before and after the crises. A signal of an upcoming and ending of a crash. Since the assumption of normally distributed returns is naïve we extend further by calculating the VaR under the assumption of non-normality. Actually this assumption is strongly supported by the Jarque-Bera statistics for testing normality in Table. For this purpose the time varying structure of variances is applied using GARCH methodology. GARCH (1, 1) can be used to extract new estimate of volatility and account for non-normality in returns issue. The new volatility estimate is based on the most recent observation of the change in the stock index and the most recent estimate of the variance rate. Applying that, the new estimate of VaR will be: σ 1-day VaR=.33 GARCH Where.33 is the Z-score which has a normal distribution with zero mean and unit variance at the 99% confidence interval and 1% confidence level, σ GARCH is the standard deviation of returns extracted form GARCH calculations for the three time specifications, namely; before, during and after the crises When applying VaR estimation approximation for market risk and the maximum loss that the market will exert at any given period at the 99% standard confidence level we get Table 6. Interesting results emerge; Table 6 shows that VaR take the shape of an inverted smile around and during crises. VaR for the DJIA index starts from approximately 9.56 to 39.7 1 then to 9.9, before, during and after crises respectively. For the S&P5, VaR starts from.73, and then peaked to 17.6 then decline to.56. For the NASDAQ index VaR follow similar pattern and starts form 8.855 to 9.5367 then to 7.8877. The major results from VaR calculations are that the value at risk is the highest during crises and take the shape of an inverted smile. JNBIT Vol.8, Iss. (1) 37

AM AL-Rjoub Volume 8, Issue (1) Table 5: Comparison between the periods before, during and after the crises Variables are defined as in tables, 3 and. Table 6: Value at Risk (VaR) around the crises using volatility estimate from GARCH calculation Variables are defined as in tables, 3 and. Plot shows that VaR estimate of loss takes the shape of inverse U-Shaped curve for the three indices and across all time specifications that was tested in Tables, 3, and respectively and summarized in 5. Plot : Value at Risk (VaR) of return around and during crises for the DJIA, S&P 5 AND NASDAQ indices respectively 1 8 6 Return DJIA After Return DJIA During Return DJIA Before 1 8 6 Return S&P After Return S&P During Return S&P Before 1 8 6 Return NASDAQ After Return NASDAQ During Return NASDAQ Before JNBIT Vol.8, Iss. (1) 38

AM AL-Rjoub Volume 8, Issue (1) Conclusions The paper examined historical episodes in the United States over the last one hundred years of major stock market crashes in the Dow Jones industrial average, the S&P 5 and the NASDAQ indices. Analysis focused on the price behavior round and during the crashes in a try to provide some clues as to the timing of a bursting bubble and the ending of a one and serves as an early warning system to stock markets and help policymakers about what they should do about it. Part of this article tell us much more about the dynamics of stock returns before and after the bubble burst, the second part continue by calculating Value at Risk (VaR) to account for the loss level that will not be exceeded with a specified probability. After accounting for six-months-before and six-months-after the crises dummies, interesting results emerge. On average, the mean average return of interactive dummy that account for return during the crises is negative; the mean average return of interactive dummy that account for return before and after the crises is consistently positive for different sample specifications. When comparing the dummies before, during and after the crises results show that returns take the shape of an inverse U-shaped curve, where it change its pattern from positive and high before the crises to negative and low during the crises. The standard deviation is much higher during the crises than before and after the crises, Skewness statistics is very high and positive before and after the crises for the S&P 5 and NASDAQ except for the DJIA index where the skewness parameter was negative for the six-month-before the crises dummy. More interestingly when comparing the Kurtosis parameter results shows that the Kurtosis is much higher before the crises than during it. This result holds for the three indices. The distribution is peaked (leptokurtic) relative to the normal. Furthermore, when applying VaR estimation to calculate the maximum loss that the market will exert at any given period at the 99% standard confidence level, VaR takes the shape of an inverted smile. JNBIT Vol.8, Iss. (1) 39

AM AL-Rjoub Volume 8, Issue (1) Reference List Berg, A, Borensztein, E and C. Pattillo, (). Assessing Early Warning Systems: how Have they Worked in Practice? IMF Working Paper. /5. Berg, A, and C. Pattillo, (1998). Are Currency Predictable? A Test, IMF Working Papers, 98/15. Berg, A, and C. Pattillo, (1999a). Are Currency Predictable? A Test, IMF staff papers, 6, 17-38. Berg, A, and C. Pattillo, (1999b). Predicting Currency Crisis: The Indicators Approach and an Alternative, Journal of International Money and Finance, 18,561-86 Bollerslev, T., (1986). Generalized Autoregressive Conditional Heteroscedasticity, Journal of Econometrics, 31,37-37. Ciarlone, A. and Trebeschi,G.,(5). Designing an Early Warning System for Debt, Emerging Markets Review 6, 376-395. Caporale, G. and Spagnolo N., (3). Asset Prices and Output Growth Volatility: the Effects of Financial. Economics Letters, 79, 69-7. Choudhry, T., (1996). Stock Markets Volatility and the crash of 1987: Evidence from Six Emerging markets. Journal of International Money and Finance, 15 (6), 969-981. Frankel, J., and Rose. A.,(1996). Currency Crashes in Emerging Markets: an Empirical Treatment, Journal of Economics 1,351-366. Hull, J.C., (9), Risk Management and Financial Institutions, nd Edition, Pearson education. Kaminsky, G., Lizondo, S and Reinhart, C.,(1998), Leading Indicators of Currency, IMF Staff Papers 88 (),1-8. Kui-Wai Li and Ming-Lok Kwok (9), Output Volatility of Five Crisis-Affected East Asia Economies, Japan and the World Economy, 1 (), 17-18 Leon, N., (7). Stock Market Returns and Volatility in the BRVM. African Journal of Business Management, 1(5), 17-11. Mishkin, F., and Eugene N. White (). U.S. Stock Market Crashes and Their Aftermath: Implications For Monetary Policy. National Bureau of Economic Research 899, June. Mulder,C., and R. Perrelli, and M. Rocha ().The Role of Corporate, Legal and Macroeconomic Balance Sheet Indicators in Crisis Detection and Prevention, IMF working Paper WP//59. Rose, Andrew K., and Mark M. Spiegel. (9). Cross-Country Causes and Consequences of the 8 Crisis: Early Warning. FRBSF Working Paper 9-17 Schwert, W., (1989). Business Cycles, Financial, and Stock Volatility. Carnegie-Rochester Conference Series on Public Policy 31, 83-16. Schwert, W., (199). Stock Volatility and the Crash of 87. National Bureau of Economic Research, Working Paper No. 95. Stiglitz, J. (3a). Dealing with Debt: How to Reform the Global Financial System. Harvard International Review, 5(1), 5 JNBIT Vol.8, Iss. (1)