Journal of Multinational Financial Management 11 2001 69 – 87
Pre- and post-1987 crash frequency domain analysis among Pacific Rim equity markets
Kenneth L. Smith
Department of Economics, Finance, and Banking, Kelce College of Business, Pittsburg State Uni6ersity, Pittsburg, KS
66762
, USA Received 12 February 1999; accepted 16 February 2000
Abstract
This paper uses the method of cross spectral analysis to examine pre- and post-1987 crash comovements among several Pacific Rim markets. The results support increased post-crash
comovements among the markets. Each of the pairwise median coherences was greater in the post-crash period. The same is true for the post-crash mean coherences, with one exception.
The nonparametric Wilcoxon Z statistics indicate that the pre- and post-crash coherences were drawn from different populations in seven of the eight pairwise market comparisons.
Phase lead results indicate that the post-crash phase leads in four of the eight pairwise comparisons were smaller than in the pre-crash period, while post-crash phase leads did not
fall in any of the market comparisons. © 2001 Elsevier Science B.V. All rights reserved.
JEL classification
:
G15 Keywords
:
Cross spectral analysis; Equity comovements; Diversification www.elsevier.comlocateeconbase
1. Introduction
Increased world trade has intensified the interest in the integration of world equity markets. Along with increased trade, increases in communications and
capital flows have contributed to the globalization of business activity. It has long been a tenet of investment that international diversification can reduce risk in
portfolios. Increasing equity market integration tends to reduce this diversification,
Tel.: + 1-316-2354592; fax: + 1-316-2354578. E-mail address
:
ksmithpittstate.edu K.L. Smith. 1042-444X01 - see front matter © 2001 Elsevier Science B.V. All rights reserved.
PII: S 1 0 4 2 - 4 4 4 X 0 0 0 0 0 4 3 - 8
reducing the benefit of reduced risk. The question is, ‘Are equity market movements around the world disparate enough to provide the diversification benefit?’ Even
though trade, communications, and capital flows have tended to make the world ‘smaller’, to what extent is international equity diversification a benefit to investors?
There are now many articles, using differing methods, which examine the comovement of equity markets around the world. One of the most used methods is
the ARCHGARCH family of models, which test for volatility spillovers. Darbor and Deb 1997 used bivariate GARCH models December 1989 through December
1992 to conclude that each bivariate pair JapanCanada, UKCanada, US Canada, UKJapan, USJapan, UKUS of markets showed evidence of ‘transitory
correlation’. Only the USJapan pair indicated a lack of ‘permanent correlation’. Employing the trivariate EGARCH model, Koutmos and Booth 1995 found price
spillovers from New York to Tokyo and London, and from Tokyo to London. Susmel and Engle 1994 used a bivariate ARCH model with hourly data and found
volatility spillovers last, ‘only an hour or so’. Theodossiou and Lee 1993 employed the multivariate GARCH-M model to find volatility spillovers among a subset of
five markets US, UK, Canada, Germany, and Japan from January 1980 to December 1991. Hamao et al. 1990 used a trivariate GARCH model with daily
data April 1985 to March 1988 to test for volatility spillovers from New York, London, and Tokyo. They presented evidence of volatility linkages. Booth et al.
1997 employed a trivariate VAR to investigate volatility spillovers in US, UK, and Japanese equity index futures markets. These results indicated that volatility in
the US and UK markets is a reaction to shocks from the other markets, supporting the argument that these markets are becoming more interdependent. The Japanese
stock index futures markets appear somewhat more isolated.
Other methods, which focus on price spillovers, have been used as well to study market linkages. Koch and Koch 1991 estimated an eight market Japan, Aus-
tralia, Hong Kong, Singapore, Switzerland, West Germany, UK, and the US simultaneous equations model for three separate years: 1972, 1980, and 1987. They
found evidence of, ‘several clusters of markets that display substantive interaction on the same day’. Brocato 1994 estimated six market US, UK, Canada, Ger-
many, Hong Kong, and Japan VARs: one for the early 1980s and another for the latter years of the 1980s. His variance decompositions yielded many linkages among
these markets, with the US role diminishing post-March 1984. Eun and Shim 1989 estimated a nine market US, UK Canada, France, Germany, Switzerland, Hong
Kong, Japan, and Australia VAR and concluded that a substantial amount of ‘interdependence’ existed among these markets: 1980 – 1985.
A further issue is the temporal nature of diversification. Do benefits from diversification accrue to international investors in the short term or is diversification
a long run concept? Taylor and Tonks 1989 employed cointegration analysis to conclude that low correlation among equity markets occurs in the ‘short term’.
Meric and Meric 1989 compared correlation matrices of international equity markets to test for intertemporal stability. Tests were performed which compared
the pairwise correlation matrices of nine adjacent 1.5-year subperiods for 17 countries. The tests were also conducted for four adjacent 3-year and two adjacent
5-year subperiods. The final test compared the correlation matrices for two 7.5-year subperiods. They rejected the null hypothesis that correlation matrices are the same
over shorter periods 1.5- and 3-year subperiods, but were not able to reject that the longer term 5- and 7.5-year subperiods correlation matrices are the same. In
both of these studies, longer term comovements were more correlated, making diversification more applicable in the near term. The results reported below, using
cross spectral analysis, lend support to this conclusion.
Institutional changes made following the crash period make 1987 a breakpoint from which pre- and post-crash observations can be compared. Lindsey and Pecora
1998 detailed the domestic regulatory changes to make the markets more efficient. Internationally, markets around the world have recognized the need for greater
coordination in regulation and have taken steps to ensure cooperation. Accord- ingly, there are several papers which use this breakpoint to compare pre- and
post-1987 crash comovements. Masih and Masih 1997 employed a battery of tests to determine if comovements increased in the post-1987 period. They reported,
using Granger causality tests, that since the crash the US market has led other markets Canada, UK, Germany, France, and Japan. Their vector error correction
model and variance decomposition tests indicated that the German and UK markets have become more dependent on shocks from other markets in the
post-crash period. Arsanapalli and Doukas 1993 used cointegration tests to show that comovements among UK, German, French, and Japanese markets increased
post-1987. Jeon and Von Furstenburg 1990 employed a four market US, UK, German, and Japan VAR analysis to conclude that, ‘The degree of international
comovements in stock price indexes has increased significantly since the crash of October 1987’.
Most of the above studies used time series methods to investigate market comovements. Granger and Morgenstern 1970 were the first researchers to apply
frequency domain methods to comovements among equity markets. Hilliard 1979 estimated mean coherences among ten markets US, Amsterdam, Frankfurt, Lon-
don, Milan, Paris, Sydney, Tokyo, Zurich, and Toronto. He concluded that ‘intra-continental’
prices moved
together, with
little ‘inter-continental’
comovements. Fischer and Palasvirta 1990 were the first to apply these methods to pre-post-
1987 crash data. They used frequency domain analysis to test for increasing interdependence. They concluded that, ‘the level of interdependence, as evidenced
by the co-movement of index prices in the world’s stock markets, has grown substantially from 1986 to 1988’. These results are based on a very short sample
period 1986 – 1988. Specifically, their Table 2 compared mean coherences between the 23 markets and a constructed world index. Twenty-two of the 23 mean
coherences were higher in 1988 when compared to mean coherences of 1986. Both years contained only about 250 daily observations. Furthermore, 22 of the 1987
mean coherences were greater than the 1988 mean coherences. Malliaris and Urrutia 1992 have shown that Granger causality increased during the month of
October 1987, with a lack of correlation for the four month periods immediately before and after the crash. These results beg an interesting question. If longer pre-
and post-crash samples are used, would the lack of causality continue to hold? Smith et al. 1993 have shown that, using a rolling Granger causality test,
somewhat greater causality existed after the crash their sample ends in June 1991. Even more high frequency data are now available to study the issue of both pre-
and post-crash comovements. Smith 1999 used cross spectral analysis along with a longer sample period. He found increased coherence among six major equity
markets, particularly among the three European markets studied UK, Germany, and France. Most pre-post-crash research described above has focused on the G-7
or a subset of the markets e.g. see Darbor and Deb 1997, Masih and Masih 1997 and Smith 1999.
1
It is the purpose of this paper to investigate these pre-post-correlations for Pacific Rim markets. This study includes two markets that are not part of the
Group of Seven. Cross spectral analysis is applied to eight pairwise combinations of coherence and phase leads from data for Australia and Hong Kong, as well as the
US, Canada, and Japan.
2. Data
