24
25
tempt is made to repair this neglect by exploring the degree to which markets in the Czech
Republic, Hungary and Poland are integrated with markets in Germany and the UK. The
economies of Germany and the UK are generally considered economically important in a
European context and, in particular, account for a fair share of East European trade, with
each also providing an important source of foreign direct investment. The motivation for
considering the three East European markets arises from the following reasons. First, the
liberalisation of these economies and the subsequent opening of their capital markets has,
since the late 1990s, resulted in favourably equity returns that can be explained by the success of public policy measures designed to effect structural changes which improved the
macroeconomic performance of their economies. This, in turn, has led to changes in the
economic position of the East European economies in relation to trade, resulting in growing
interest from institutional investors seeking to diversify in East European stocks, and thus
finding it difficult to disregard the potential from investing in these markets. Notably also,
the East European economies have benefited from the flow of inward investment from the
European Union, and especially from countries that have always had close political and
trading links. Second, the economies of Eastern Europe are interesting if only because they
provide an opportunity for economists to mark the presence of strong economic linkages
between their markets and the markets of the EU. Third, our interest is justified as the growing tide of foreign investment in Eastern Europe has helped to tighten their price linkages to
global financial centres within a short period of reform years. Therefore, it is only natural to
ask whether the importance of international investors and, especially, the activities of multinational corporations can induce the long run relationship between stock prices of these
countries and member countries of the EU. Evidence of increased market integration of
East European countries with the markets of the EU may be found in the joint movements of
returns realised by investors in both Eastern Europe and EU markets alike. It is with regard
to this that the present paper tests for the existence of a longterm relationship among the
capital markets of Eastern Europe and the markets of Germany and the UK by way of cointegration analysis. Most notably, the presence of cointegration has implications for the extent to which risk can be reduced by exploiting the nonzero longrun correlation among
indices when investing in these markets.
The following section presents the data and some preliminary statistics. Section 3 contains a summary of the methodology employed, while section 4 outlines the model and reports results pertaining to the analysis of market linkages. Concluding remarks are set forth
in the final section of the paper.
2. Data and descriptive statistics
This study employs weekly data sets for the stock indices of the markets in the Czech Republic (PX50), Hungary (BUX), Poland (WIG), Germany (FAZ General), and the UK (FT
All Share).1 The period under investigation extends from 28 April 1991 through to 30 April
1995. For the Czech republic the sample period begins at 28 April 1994 because data for the
Prague stock exchange index are discontinuous prior to that time period. The indices are expressed in natural logarithms and are calculated from prices expressed in local currency.
Accordingly, continuously compounded (percentage) returns, denoted by Rt, are calculated
as the difference of the (log) closing index value (Pt), i.e., Rt = 100(lnPt  lnPt1). The indices
are value weighted and not adjusted for dividends. Thus, on the basis of the evidence furnished by French et al. (1987) and Poon and Taylor (1992), it is expected that adjustment for
dividends would not affect the results reported in this paper. Preliminary statistics on the
percentage returns are reported in Table 1.
Managerial Finance
26
Table 1.
Descriptive Sample Statistics on Weekly Stock Returns (1)
Staistic
Czech Rep.
Hungary
Poland
Germany
UK
Mean
0.334
0.148
0.942
0.19
0.151
Std. Dev.
3.007
3.446
8.861
2.992
2.357
0.11
0.04
0.11
0.06
0.06
.637***
2.3***
.448***
1.12***
2.65***
2.8***
19.3***
4.0***
4.98***
28.7***
7.73
11.63
8.74
9.86
14.98
83.7***
3678.9***
157.34***
724.2***
20712.1***
16.92
20.38*
14.04
13.44
16.53
3.52***
1.33
2.28***
7.17***
0.327
1994/04/10
1995/08/06
 1991/01/06
 1987/04/30
 1991/04/14
 1995/08/06
 1987/07/12
 1999/02/07
 1987/07/12
 1999/02/07
Skewness
Excess Kurtosis
Studentised Range
Normality (3)
c2 (2)
Autocorrelation (4)
Q(12)
ARCH(5)
Sample Period
Notes: (1) Defined as Rt = 100 ln(Pt/Pt1) , (2) Ratio of Mean to Standard Deviation, (3)
JarqueBera statistic; (4) LjungBox statistic. (5) Ftest of no ARCH effects with 12 lags. ***, **
and * denote significance at 1, 5 and 10%, respectively.
Over the period considered, mean returns are positive except for the Czech Republic
and larger than returns in UK and Germany. A look at the variances explains why. Ranging
from 3.007 (Czech Republic) to 9.404 (Poland), from which it may be concluded that the
emerging Eastern European markets are riskier and, consequently, command, in the least, a
higher risk premium. Even so, the Sharpe ratio suggests that both Poland and Hungary fare
well when compared to the more established EU markets of Germany and the UK. With the
exception of Hungary, whose index is positively skewed, all series exhibit significant negative skewness. The positive skewness may, however, be due to some extreme positive return that the Budapest market realised over the sample period examined. Further evidence
of this can be gleaned from Fig. 1.
Coupled with the evidence from the excess kurtosis measure and the studentised
range, there is clear indication that the distributions of returns are not normal. When formally tested, the normality hypothesis is firmly rejected by the JarqueBera test at all levels
of statistical significance.
27
The presence of intertemporal dependencies in the returns is tested by means of the Ljung
and Box portmanteau test (LB) for (up to) 12th order correlation. Overall, there is little evidence of correlation in stock returns over the period. Thereby confirming that there is widespread evidence of dependence in conditional second moments. This also validates the
indication of the leptokurtosis by the excess kurtosis measure of which the ARCH tests  for
up to 12 lags  rejects the null of no ARCH effects in the case of Czech Republic, Poland and
Germany.
3. Methodology
The degree of market integration or linkage between two or more markets may be examined
by investigating the existence of a longrun relationship between the stock market indices
by drawing on Granger (1986) and Engle and Grangers (1987) concept of cointegration.
The methodology of cointegration requires that longrun equilibrium constraints be imposed on shortrun asset return equations allowing estimation of market dynamics. Here,
Engle and Granger (1987) instruct us that multivariate time series models which exclude
longrun constraints, such as vector autoregressive models (VAR) of stock returns, will be
misspecified if stock prices are cointegrated. They note, in particular, that imposing longrun equilibrium constraints results in efficiency gains and improved multistep forecasting.
And since the cointegration methodology does not distinguish, at least in principle, between exogenous and endogenous variables, there is therefore no simultaneity bias.
If a variable is stationary after first differencing that it is integrated of order one, I(1).
On this basis and, in the multivariate context of this paper, we consider the vector zt p of
stock price indices generated at time t, on which observations t =1, ..., T are available and let
Z t = (z1,z2,...,zt ) be the (tp) matrix of price index histories. Here, we assume a linear pro
Managerial Finance
28
29
cess2 for zt and let its conditional density D(ztZt1, l) be multivariate normal. In other words,
zt Zt1,Dt ~ NID(A1zt1+...+Akztk+m+FDt,W), where zk+1,... z0 are fixed, Dt is a vector of
nonstochastic variables  including seasonal and, if needed, intervention dummies  and
possibly weakly exogenous variables, zt is at most I(1),3 and the parameters (A1,...,Ak, F
,W.) are unrestricted. The implied vector autoregressive (VAR) model can then be rewritten in vector errorcorrection form (VECM) as follows:
Dz = G Dz
t
t 1
where P =
+...+G
k 1
Dz
+ Pz
t k +1
A  I , G =
i=1
t 1
A and G = I 
j =i+1
(1)
+ m + FD + e ,
i=1
tem incorporates a structure whereby shortrun dynamics in stock returns as given by (Dzt)
respond to deviations from the longrun equilibrium relationship among stock index levels.
The hypothesis of cointegration is formulated as a reduced rank condition on the matrix P. Specifically, for r = 0, 1, ... p, the hypothesis of at most r cointegrating vectors is thus
defined:
(2)
H :P = ab
r
where a and b are (pr) matrices of full rank. The hypothesis Hr implies that the process Dzt
is stationary, meaning that zt is nonstationary, and the vector of linear cointegrating relations bzt is stationary (see Johansen, 1991). Using the reduced rank condition, the VECM
becomes:
Dz = G Dz  +...+G
t
k 1
Dz t
k +1
+ abz  + m + FD + e
t
(3)
This parameterisation offers a clearer interpretation of the coefficients Gi which describe the shortterm dynamics of the process, while the effect of the levels is isolated in the
matrix ab. The r cointegrating relations contained in bzt can be considered as the longrun
relationships among stock price indices, whilst the loadings included in the feedback matrix a summarise the behaviour of stock returns when deviations from longrun equilibrium
occur.
In our subsequent analysis, we are interested not only in the features of the longrun relationship(s) bzt, but also in the nature of the market response to departures from it (a). So
assuming the cointegrating rank r is known, we can achieve our objective by partitioning
the VECM, which will then allow us to deliberate the exogeneity properties of the stock
markets considered in this study. In particular, the partitioning of the a , G and F matrices to
correspond to the partitioning of zt into two subsets of stock price indices, say xt and yt, results in the following:
a
a =
a
G
F
W W
, G = G , F = F and W =W W .
11
12
21
22
Dx = a bz
t
t 1
+ G Dz
1i
i=1
t 1
+ m +F D +e ,
1
1t
(4)
Managerial Finance
30
k 1
Dy = a bz
t
t 1
+ G Dz
2i
i=1
t 1
+ m +F D +e
2
2t
(5)
Conditional on past history, Dxt and Dyt can be reformulated from equations (4) and (5) as
k 1
Dy = wD + ( a  wa ) bz
t
~
+ G Dz
t 1
2i
i=1
t i
~
+~
m + F D + ~e ,
2
(6)
2t
~
~
~
w = W W1
11 , G i = G  wG , F = F = wF , m = m  wm and
where ~e = e  we ,
21
2t
2i
1i
1t
with variance W
22.1
= W  W W1
11 W . If a = 0,( 4) and (6) become
22
21
12
k 1
Dx = G Dz
t
1i
t 1
(7)
+ m + F D + e and
1
1t
i=1
k 1
Dy = wDx + a bz
t
t 1
~
+ G Dz
2i
t 1
~
+~
m + F D + ~e
2
2t
(8)
i=1
so that equation (7) can now be treated as a marginal model, whilst equation (8) remains
conditional. Under a1 = 0, this parameterisation operates a sequential cut (Engle, Hendry
and Richard, 1983) in that it allows us to examine the conditional model by taking the marginal as given.3 In this case, if a1 = 0, the stock price processes in xt are said to be weakly exogenous for the longrun parameters of interest ( a2 , b). Hall and Wickens (1993) and Hall
and Milne (1994) for example interpret weak exogeneity in the context of cointegrated systems as longrun causality. In practical terms, weak exogeneity for a subset of stock price
indices rather implies that the relevant stock markets are only affected by shortrun fluctuations in the system. Notwithstanding, national markets that are not weakly exogenous may
also be affected by other markets, both as a result of shortrun fluctuations and by deviations from the longrun equilibrium.
4. The Model and Empirical Results
This section examines the linkages between the three East European capital markets Czech Republic, Poland and Hungary  and the two major European Union equity markets,
namely Germany and the UK. The first step in our cointegration analysis is to assess the stationarity of each series. In table 2, we perform a series of tests in order to assess the extent to
which shocks to each process are (not) reversed at any finite horizon. Drawing on the Augmented DickeyFuller (ADF) (1979) and the PhillipsPerron (PP) (1998) tests of the null
hypothesis of a (univariate) unit root, table 2 reveals widespread acceptance of the null hypothesis, except for the UK.
To uncover more about the dynamic properties of the processes under consideration,
we adopt a number of alternative methodologies in order to shed further light on the behaviour of our stock markets at different horizons. Specifically, the Variance Ratio (VR) test as
advanced by Lo and MacKinlay, (1988, 1989) are carried out on returns obtained over a pe
31
riod (k) of 2, 4 and 12 weeks. Using heteroskedasticityconsistent variances for testing, the
results fail to denounce the null of a random walk process for twoweek returns for all markets. Over the much longer horizon, however, nonrandom walk behaviour can be detected
over 12 periods for the Czech, Polish and Hungarian markets. In the case of the Hungarian
market, it is notable that the market also fails the test on fourweek returns. This aspect is
further analysed under a different null hypothesis: that of no longrange dependence, otherwise known as fractional integration. Here, the Normalised Rescaled Range (R/S) (Lo,
1991) is obtained for windows (q) of 0, 4 and 12 weeks. The first case is the Classical Range
measure, whilst the other two (q = 4,12) correspond to Los Modified Range. The test is formulated so as to be more sensitive to longrun correlation, and so any rejection of the null
hypothesis should be ascribed to slow meanreverting or meanaverting behaviour, rather
than to shortterm activity. It is well known that the Classical measure is relatively more
sensitive to correlation at low lags than Los measures (see Lo, 1991). But interestingly, Table 2 shows that any significant rejection occurs under the Classical Range measure. Taking
into consideration the VR results, the R/S results would seem to bear some implications for
the (weakform) efficiency of the stock markets under consideration. From this, it may be
argued that over longer horizons, in particular, the emerging East European markets exhibit
dynamic properties that may be inconsistent with the efficiency hypothesis5 and, furthermore, the perceived mean dependence is likely to occur at shorter horizons. In addition, the
general conclusion about inefficiency is also buttressed  for Hungary and the Czech Republic  by a nonparametric procedure such as the Runs test.
Having uncovered some univariate features, we now turn our attention to the multivariate dynamics of the markets under study. It is in this connection that we perform two
sets of analyses. In the first case, the vector zt includes all five markets and t spans a period
of just over a year (April 1994April 1995). While in the second analysis, we exclude the
Table 2.
Unit Roots and Efficiency
Market
ADF (1)
PP (2)
Runs
Czech
Rep.
2.75
3.338* 2.985***
Poland
1.16
1.16
Hungary
1.91
1.524
Germany
2.23
2.23
.144
.945
UK
4.1***
4.1*** 1.176
1.028
k=4
q=0
q=4
q = 12
1.029
1.155
1.4**
1.753*
1.404
1.294
1.35
1.08
1.143
1.468***
1.993**
1.827*
1.361
2.646***
1.175
1.509*** 2.054***
1.998**
1.41
1.256
1.01
1.118
1.129
1.095
1.079
1.111*
1.026
.942
.917
1.095
Notes: (1) Augmented DickeyFuller ttest of the unitroot null in log stock price indices.
Number of lags chosen by the Schwartz Information Criterion. (2) PhillipsPerron ttest of
the unitroot null in log stock price indices. (3) Uses heroskedasticityadjusted asymptotic
variances. ***, ** and * denote significance at 1, 5 and 10%, respectively.
Managerial Finance
32
Prague stock market in an attempt to obtain more robust results by virtue of allowing the series to extend for approximately four years, from April 1991 through to April 1995. In all
cases, Dt contains (centred) seasonal dummies to allow for monthly fluctuations in conditional means. The results are summarised in Table 3, panels A and B.
Our approach follows five separate steps, reflected in five separate sections in Table 3.
First, we choose the optimal lag length for the VAR model outlined in the previous section.
Contrary to common practice, we do not select the VAR lag length by conventional methods such as the Akaike or Schwartz Information Criteria. Instead, we follow Johansen
(1995a) and select the number of lags, which then allows us to obtain reasonably wellbehaved (multivariate) residuals. The second step involves the determination of the rank (r)
of the P matrix. And following the reformulation of the VAR into a VECM, the rank is then
determined by first using Johansens (1998, 1991) Full Information Maximum Likelihood
(FIML) procedure to estimate P. From which we then proceed to test the hypotheses about
its rank by means of the Trace and Maximum Eigenvalue statistics (see Johansen and Juselius, 1990). In both cases,6 however, the evidence points towards the existence of one longrun relationship among stock indices.
As a third step, we use the VECM framework to carry out multivariate c2 tests of the
null hypothesis of stationarity for each of the series. This is accomplished by imposing, in
turn, zero restrictions on all but one of the elements of bvector, i.e. by evaluating whether
the cointegrating vector consists of solely one market index, rather than a combination of
some; see Johansen and Juselius (1990) and Johansen (1995a). This procedure differs from
the traditional univariate tests in two important respects. First, it takes into account the dynamics of other processes as well as those of the series being tested. Second, and unlike univariate procedures, its null hypothesis is one of stationarity, and not of unit root. The results
forcefully reject the null and confirm (except for the UK market) the conclusion obtained
from univariate tests that the five indices are indeed nonstationary during the periods considered.
Our next step in this process is to uncover the nature of the dynamic interactions  if
any  among the five stock markets, and to test each component of z for weak exogeneity to
the longrun parameter vector b.7 This is achieved by permitting x in equation (7) to contain
the stock price index being tested, and allowing y in equation (8) to contain the remaining
series and, in the process, testing the zero restriction on the relevant element of a. The likelihood ratio (LR) tests for the 5equation system (panel A) provide partly unexpected results.
As Table 3A shows, both the London and Frankfurt markets are exogenous to the system; in
addition, the Budapest and Prague markets also appear exogenous  thus leaving the Warsaw market as the only market whose shortrun behaviour is affected by departures from the
longterm equilibrium relationship among the stock markets considered. This limited evidence in favour of crossborder feedback may well be due to the limited horizon used in the
analysis. When the Prague market is excluded and the analysis repeated, the results are
more in line with expectations (panel B). Over a much longer horizon, both the Budapest
and Warsaw markets appear endogenous to the system. In fact their shortrun behaviour appears to be affected by any price fluctuations that propels the markets out of equilibrium.
Once identified the markets that reveal themselves as being weakly exogenous to the system, their exogeneity is (jointly) imposed 8on the VECM and the cointegrating relationship
is finally made explicit. The identified vectors are presented in the Restricted Cointegration
Analysis section of Table 3.
33
As can be seen from Table 3, panel A and B, none of the markets in our sample appear
redundant in the cointegration space, since all the longrun coefficients are significant at
conventional significance levels.9 This evidence is taken as a clear indication of financial
market integration, at least with reference to the markets we consider. Furthermore, the
longrun coefficient vector b also bears some clues as to the nature of such links among
price index levels. Here, conditional on movements in other stock markets, a positive
movement in Frankfurt, for example, may be deemed to translate in a negative shock to the
Warsaw exchange if equilibrium is to be maintained in the long run. On a more general
note, the evidence which stems from the 4equation system points toward positive partial
correlation between the Warsaw, London and Frankfurt markets, whilst the Budapest market appears negatively (partially) correlated to all the other markets, in the long run.
Our VECM formulation also allows us to draw some firm conclusions as to the impact
that departures from equilibrium have in the short run. As the Frankfurt and London equity
markets reveal themselves as weakly exogenous, violations of the longrun equilibrium
will not produce any direct impact. But in the case of the Budapest market, there is, ostensibly, (i) strong evidence of mean reversion (error correction) when the shock originates in
that market and (ii) indication of positive feedback to positive shocks to stock prices in the
Warsaw, London, and Frankfurt markets. Interestingly, the evidence on the shortrun behaviour of the Warsaw market points towards mean aversion (lack of error correction), as
well as positive responses to surges in index levels in both the Frankfurt and London equity
markets.
The last part of Table 3 sheds further light on the role played by crossborder shortrun
dynamics. For instance, the LR test results assess the possibility that shocks to any one market only affects the market from which they originate. Specifically, the tests impose zero re~
strictions on w, and the relevant subset of G in equation (8) above. If the LR test fail to
reject, then the market may be said not to cause any other in the system, in the short run.
The evidence summarised in Table 3 is quite telling. Here, the general impression gleaned
from the results is that once the longrun effects are accounted for, there is little role for (unconditional) shocks originating across the border. Interestingly, while the 5equation system reserves a leading role to the London market, even in the short run, this role is assigned
to the Warsaw market when the Prague market is excluded and a longer time horizon is considered.
2i
Overall, the above findings furnish a clearer picture of the extent to which the three
emerging East European stock markets are linked with the German and UK equity markets
and of the nature of such linkages both in the short and in the long run. The evidence points
strongly towards market integration and also highlights the role played by these markets beyond their national borders. The implications of this evidence can be appreciated from a
portfolio diversification perspective. Indeed, it is interesting to mention that for some time
now, institutional investors worldwide have closely observed the investment opportunities arising in East European markets as a result of macroeconomic policies and widespread
liberalisation and privatisation programmes. Yet, the dynamic behaviour of such markets
remains largely unknown. The presence of integration documented in this paper suggests
the existence of longrun gains in risk reduction from diversifying in East European stocks
as well as stocks in the two major European Union markets considered. In addition, the
documented shortrun behaviour will also prove helpful, especially in evaluating shortterm investment strategies.
Managerial Finance
34
(.296)
(.181)
Feedback:
(.807)
Coefficient
.685

DYCZE
DYPOL
DYHUN
DYGER
DYUK
tvalue
5.633

Shortrun Dynamics
Noncausality Tests c2(4)
LR test
pvalue
DYCZE
3.89
.42
DYPOL
5.85
.21
DYHUN
7.72
.10
DYGER
4.33
.36
DYUK
9.41
.05
Notes: ( 1 ) Abbreviations: YPOL = Log stock price Series for Poland, YCZE = Log stock price Series for the Czech Republic, YHUN = Log stock price Series
for Hungary. YGER = Log stock price Series for Germany. YUK = Log stock price Series for the UK. The test for (upto4 order) autocorrelation is described
in Hansen and Juselius (1995). ( ) ***, ** and * indicate significance at 1%, 5% and 10%, respectively, ( ) A tick denotes rejection of the stationarity null at the
5% level. ( ) The test of the stationarity null hypothesis is carried out in the unrestricted cointegrating space. The statistic is distributed as c (5r ) . See Johansen
(1995) for details. ( ) Standard errors in parenthesis. ( ) The maximum eigenvalue statistic rejects the null of r = 0. Accordingly, we choose r = 1.
th
35
Trace statistic
54.69***
30.41**
[
[
[
[
LR test
pvalue
8.42
7.89
.37
.02
.01
.02
.83
.99
(11.734)
(6.883)
Feedback:
DYPOL
DYHUN
DYGER
DYUK
Coefficient
tvalue
.004
.001

3.467
3.523

LR test
pvalue
Shortrun Dynamics
Noncausality Tests c2(3)
DYPOL
DYHUN
DYGER
DYUK
9.15
.03
1.11
.77
4.62
.20
3.98
.26
Notes: ( ) Abbreviations: YPOL = Log stock price Series for Poland, YHUN = Log stock price Series for Hungary. YGER = Log stock price Series for
Germany. YUK = Log stock price Series for the UK. The test for (upto4 order) autocorrelation is described in Hansen and Juselius (1995). ( ) ***, ** and *
indicate significance at 1%, 5% and 10%, respectively, ( ) A tick denotes rejection of the stationarity null at the 5% level., ( ) The test of the stationarity null
hypothesis is carried out in the unrestricted cointegrating space. The statistic is distributed as a c (4r).
See Johansen (1995) for details. ( ) Standard errors in parenthesis.
1
th
Managerial Finance
36
37
Endnotes
*Corresponding author. Email address: E.Dockery@staffs.ac.uk
1. Data on Germany and the UK have been extracted from Datastream. The series on Eastern European Stock Market have been obtained from the relevant Exchanges.
2. Nonlinear cointegration is a distinct alternative but will not be taken up in this paper.
3. The treatment of I(2) variables requires different likelihood analysis and ensuing distributions. See Johansen (1992a, 1995a,b).
4. Since the parameters of the marginal model are lm =(G11 ,..., G1k, m1,W1, W11 )and the ones
~
~
~
for the conditional model are lc = (a2, b, G ,..., G  , ~
m , F 2 , w,W ), the whole parameter space is l=(a2, b, G1,..., Gk1, , F, W).
21
2k
22.1
5. Of course, this claim is subject to all the limitations typical of studies in the market efficiency literature such as, for example, timevarying expected returns and market microstructure factors (although to a lesser extent, given that our returns are computed over a
relatively long period for microstructure effects to survive).
6. In the 4equation system, the null that r 1 could not be rejected at the 1% significance
level. In the 5equation system, the Trace statistic fails to provide evidence of cointegration
whilst the Maximum Eigenvalue statistic suggests the presence of one cointegrating vector.
Accordingly, we select r = 1.
7. Note that, as there only exists one cointegrating vector, identification is trivial and is
achieved by imposing a normalisation restriction on one of the elements of b.
8. The joint restrictions are not rejected by the LR test.
9. Moreover, tests for longrun exclusion were carried out on each of the indices in the unrestricted VECM and the exclusion null was always rejected.
Managerial Finance
38
References
Allen, D. and MacDonald, G. (1995) The longrun gains from international equity diversification: Australian evidence from cointegration tests, Applied Financial Economics, 5,
3342.
Arshanapalli, B. and Doukas, J. (1993) International stock market linkages: evidence from
the preand postOctober 1987 period, Journal of Banking and Finance, 17, 193208.
Bekaert, G. and Harvey, C. R. (1997) Emerging equity market volatility, Journal of Financial Economics, 43, 2977.
Byers, J. D. and Peel, D. A. (1993) Some evidence of interdependence of national stock
markets and the gains from international portfolio diversification, Applied Financial Economics, 3, 239242.
Chan, C. K., Gup, B. E. and Pan, M. S. (1992) An empirical analysis of stock prices in major
Asian markets and the United States, Financial Review, 27, 289307.
DeFusco, R. A., Geppert, J. M. and Tsetsekos, G. (1996) Longrun diversification potential
in emerging stock markets, Financial Review, 31, 343363.
Dickey, D. A. and W. A. Fuller (1979) Distribution of the estimators for autoregressive
time series with a unit root, Journal of the American Statistical Association, 74, 427431.
Engle, R. F. and Granger, C. W. J. (1987) Cointegration and error correction: representation, estimation, and testing, Econometrica, 55, 251276.
Engle, R. F., Hendry, D. F. and J. F. Richard (1983) Exogeneity, Econometrica, 51, 2,
277304.
Hall, S. G. and A. Milne (1994) The Relevance of PStar Analysis to UK Monetary Policy,
Economic Journal, 104, 597604.
Hall, S. G. and M. Wickens (1993) Causality in Integrated Systems, London Business
School Discussion Paper 2793, November.
Hansen, H. and K. Juselius (1995), CATS in RATS: Cointegration Analysis of Time Series,
Estima, Evanston, Illinois.
Johansen, S. (1988) Statistical analysis of cointegrating vectors, Journal of Economic Dynamics and Control, 12, 231254.
Johansen, S. (1991) Estimation and hypothesis testing of cointegration vectors in Gaussian
vector autoregressive models, Econometrica, 59, 15511580.
Johansen, S. (1992a) Testing weak exogeneity and the order of cointegration in U. K.
Money demand data, Journal of Policy Modeling, 14, 3, 313334.
Johansen, S. (1992b) Determination of the cointegration rank in the presence of a linear
trend, Oxford Bulletin of Economics and Statistics, 54, 383397
39
Molto più che documenti.
Scopri tutto ciò che Scribd ha da offrire, inclusi libri e audiolibri dei maggiori editori.
Annulla in qualsiasi momento.