STOCKS AFFECTED BY LOCATION OF TRADE? 1135.2. EstimatesThe results in tables 3.2–3.4 strongly reject the perfect-integration hypothe-
sis. The signs of virtually all coefficients line up with our alternative hy-
pothesis, and most are significantly different from zero at the 1 percent
level.^15 The estimates are also economically large. In table 3.2, for example,
the one-day Royal Dutch/Shell return differential yields coefficients of
about 0.15 on the S&P, −0.50 on the FTSE, and 0.30 on the Dutch index.
The coefficients on the exchange rate changes are also large, at −0.10 and
−0.50 for the guilder/dollar and guilder/pound exchange rates. A 1 percent
appreciation of the guilder against the dollar and pound, respectively, in-
creases the relative price of Royal Dutch over Shell by about 10 and 50
basis points. These coefficient values also imply that a 1 percent apprecia-
tion of the dollar relative to the pound increases the relative price of Royal
Dutch over Shell by about 40 basis points.
It is also interesting to note that much of the variation in return differen-
tials (which have an average annualized standard deviation of about 17
percent) is explained by Eq. (1). The R^2 s in table 3.2 are surprisingly high,
around 20 percent for one-day returns and up to 50 percent for longer-
horizon returns.
The coefficient estimates appear reasonably stable over time. Interest-
ingly, a large change in Shell ownership occurred in 1985, when U.S. hold-
ings rose to 8 percent from under 1 percent. Table 3.2 suggests that this
change in ownership was associated with a decline in the S&P coefficient,
consistent with our alternative hypothesis. Specifications 3 and 4 yield esti-
mates of the lagged-dependent variable coefficient, θAB, of about −0.2,
which is strongly statistically significant. This implies that the short-horizon
beta coefficients are about 20 percent greater than their long-horizon coun-
terparts. While this estimate is not small economically, it suggests that the
comovements we measure persist over longer return horizons.
Tables 3.3 and 3.4 reveal a similar story for Unilever N.V./PLC and
SmithKline Beecham. We reject the null hypothesis in most cases at the 1 per-
cent level.
These results provide evidence of comovement between relative twin
prices and market indexes for both short and long horizons. The data actu-
ally reveal an even stronger finding: in our sample, we find no statistical ev-
idence that the comovement is at all transient. Specifically, we cannot reject
the hypotheses that: (1) the price differentials contain unit roots, and (2)
the price differentials and stock indexes are cointegrated.
In table 3.5 we investigate whether the price differentials contain unit
roots using the augmented Dickey–Fuller test. The data cannot reject the
(^15) The significance tests are F-tests on the sum of the lead, current, and lag coefficients for
each index.