Our daily with-dividend stock returns data begin with all companies covered
by the Datastream information service as of January 1997. Datastream also
allowed us access to data for companies no longer traded, but whose stock
prices were formerly covered by their service. Our total cross section for 1995
thus contains 15,920 "rms spanning 40 countries. Newly listed or recently
delisted stocks are included in our sample only if more than 30 weeks of data is
available for the year in question. This requirement yields su$cient observations
to reliably assess the explanatory power of the market returns on each stock.
Thus, we omit newly traded stocks that have been traded for roughly less than
"ve months in a year, as well as stocks that are about to be delisted. When
trading of a stock is suspended, the returns data during the suspension period
are coded as missing and also excluded from our regressions. In addition, for
most countries, Datastream returns are either unavailable or seriously incomplete
until the mid-1990s. For this reason, we focus on 1993 through 1995, and
use only 1995 data in our international cross-sectional analysis. As a robustness
check, we reproduce our results using 1993 and 1994 data.
Datastream claims that its total returns are adjusted for splits and other
unusual events, but our data do contain some very large stock returns. If these
very large returns re#ect coding errors, they could add noise to our data or
create bias in our results. On the assumption that coding errors are overrepresented
in extreme observations, we trim our data by dropping biweekly
observations for which the stock's return exceeds 0.25 in absolute value.
The regression statistic for Eq. (6), R2
ij, measures the percent of the variation in
the biweekly returns of stock i in country j explained by variations in country j's
224 R. Morck et al. / Journal of Financial Economics 58 (2000) 215}260
2 Although the R2
j for the U.S. stock markets is lower than that reported by Roll (1988), note that
we use 1995 biweekly data while he uses monthly data from September 1982 to August 1987. Our
R2 estimate for the US market in the early 1980s ranges between 12% and 13% (see Fig. 3), and so is
much closer to the average R2 of 0.179 he reports.
market return and the U.S. market return. Given this statistic for each "rm i in
country j, we de"ne
R2
j
"+i
R2
ij]SS¹i,j
+i
SS¹i,j
(7)
as an alternative stock price synchronicity measure, where SS¹i,j is the sum of
squared total variations. We use this weighting rather than a simple average to
facilitate the decomposition of returns variation in Eqs. (16) and (17) (see in
Section 6). A higher R2
j indicates that stock prices frequently move together.
This measure of stock price synchronicity follows Roll (1988) and French and
Roll (1986).
Our daily with-dividend stock returns data begin with all companies coveredby the Datastream information service as of January 1997. Datastream alsoallowed us access to data for companies no longer traded, but whose stockprices were formerly covered by their service. Our total cross section for 1995thus contains 15,920 "rms spanning 40 countries. Newly listed or recentlydelisted stocks are included in our sample only if more than 30 weeks of data isavailable for the year in question. This requirement yields su$cient observationsto reliably assess the explanatory power of the market returns on each stock.Thus, we omit newly traded stocks that have been traded for roughly less than"ve months in a year, as well as stocks that are about to be delisted. Whentrading of a stock is suspended, the returns data during the suspension periodare coded as missing and also excluded from our regressions. In addition, formost countries, Datastream returns are either unavailable or seriously incompleteuntil the mid-1990s. For this reason, we focus on 1993 through 1995, anduse only 1995 data in our international cross-sectional analysis. As a robustnesscheck, we reproduce our results using 1993 and 1994 data.Datastream claims that its total returns are adjusted for splits and otherunusual events, but our data do contain some very large stock returns. If thesevery large returns re#ect coding errors, they could add noise to our data orcreate bias in our results. On the assumption that coding errors are overrepresentedin extreme observations, we trim our data by dropping biweeklyobservations for which the stock's return exceeds 0.25 in absolute value.The regression statistic for Eq. (6), R2ij, measures the percent of the variation inthe biweekly returns of stock i in country j explained by variations in country j's224 R. Morck et al. / Journal of Financial Economics 58 (2000) 215}2602 Although the R2j for the U.S. stock markets is lower than that reported by Roll (1988), note thatwe use 1995 biweekly data while he uses monthly data from September 1982 to August 1987. OurR2 estimate for the US market in the early 1980s ranges between 12% and 13% (see Fig. 3), and so ismuch closer to the average R2 of 0.179 he reports.market return and the U.S. market return. Given this statistic for each "rm i incountry j, we de"neR2j"+iR2ij]SS¹i,j+iSS¹i,j(7)as an alternative stock price synchronicity measure, where SS¹i,j is the sum ofsquared total variations. We use this weighting rather than a simple average tofacilitate the decomposition of returns variation in Eqs. (16) and (17) (see inSection 6). A higher R2j indicates that stock prices frequently move together.This measure of stock price synchronicity follows Roll (1988) and French andRoll (1986).
การแปล กรุณารอสักครู่..