Longitude matters: time zones and the location
of foreign direct investment∗
Ernesto Stein†,‡
Inter-American Development Bank
Christian Daude§
University of Maryland at College Park
This Version: September, 2005
Abstract
Using bilateral Foreign Direct Investment (FDI) data, we find that differences in time zones have a negative and significant effect on the location
of FDI. We show that this finding is robust across different specifications,
estimation methods and proxies for time zone differences. Time zones
also have a negative effect on trade, but this effect is smaller than that on
FDI. Finally, the impact of the time zone effect has increased over time,
suggesting that it is not likely to vanish with the introduction of new information technologies.
Keywords: Foreign direct investment; transaction costs; time zones
JEL classification: F21; F23
∗ We would like to thank Josefina Posadas for superb research assistance, Eduardo LevyYeyati, Ernesto Shargrodsky, Christian Volpe, two anonymous referees, as well as conference
participants at LACEA 2001 in Montevideo and seminar participants at the Worldbank, IDB
and George Washington University for helpful comments and suggestions. The views expressed
in this document are those of the authors’ and do not necessarily reflect those of the InterAmerican Development Bank.
† Corresponding Author.
‡ Inter-American Development Bank, 1300 New York Avenue NW, Washington DC 20577,
USA. Tel. +1(202)623-2823, Fax +1(202)623-1772, Email: [email protected].
§ Department of Economics, 3105 Tydings Hall, College Park 20742, MD, Email:
[email protected]
1
1
Introduction
A lot has been written in the literature on geography and economic development
about the importance of latitude. Hall and Jones (1999), for example, find a
positive correlation between the distance to the equator and output per worker,
mediated by the social infrastructure. Gallup, Sachs and Mellinger (1999) argue
that the tropical climate in locations near the equator has an adverse effect on
human health, agricultural productivity and consequently on economic growth.
Acemoglu, Johnson and Robinson (2001) claim that tropical diseases, associated with latitude, affected the kind of settlements and institutions colonizers
established in the colonies, and thus had an important impact on their later development path. However, little attention has been paid to the economic effects
of longitude.
In this paper, we center our attention on a variable that is closely related
to longitude: the difference in time zones between locations. In particular, we
estimate the effects of time zone differences on bilateral stocks of foreign direct
investment (FDI), using OECD data for 17 OECD source countries and 58 host
countries from 1997 - 1999. We present evidence that longitude - in the form of
time zones - imposes important costs between the parties in a transaction.
In July of 2004, an article in the Miami Herald by Daniel Grech analyzed
the consequences of a sudden change to daylight saving time - due of important
energy shortages - in nine of the twenty-three provinces of Argentina, which
historically has had only one time zone. While the potential reduction in energy consumption of daylight savings in Argentina was estimated at 2 percent,
within two weeks seven provinces abandoned the time zone change because of
the confusion and coordination problems created for trade, transportation, and
traveling across provinces. Clearly, the costs imposed by time zones seem to
be important. Although in everyday life there are many anecdotes related to
the jet lag and problems related to time zones, this paper is the first, to our
knowledge, to present systematic evidence of this effect and quantify the costs
that time zone differences impose on doing business.
The empirical literature, in particular, the one related to the gravity model
of bilateral trade, has used two types of variables to control for transaction costs.
On the one hand, geographical characteristics of countries or country pairs, such
as: distance, adjacency, remoteness, and whether one or the two countries in the
pair are landlocked or islands, are used to capture mainly transportation costs.
On the other hand, variables related to cultural and historical ties between the
countries, such as common language, cultural similarities or past colonial links,
are frequently used to account for other transaction costs that may affect the
cost of doing business.1 However, none of these variables captures the trans1 Rauch (1999) analyzes the effects of distance, common language and colonial links - proxies
for search barriers in the matching process between sellers and buyers of goods - on bilateral
trade for different types of goods. Using gravity equation estimates, he finds that cultural
”distance” and physical distance have a significantly negative effect on trade. This effect is
greater for differentiated goods than goods that are traded in organized exchanges - like many
commodities - and homogenous goods. While he does not address formally the effect of search
2
action costs related to the need for frequent interaction in real time between
the parties. Distance, in particular, does not capture this effect. If telephone,
e-mail and teleconference communication are close substitutes for face-to-face
interaction, North-South distance should not be such a large problem. In contrast, differences in time zones can matter even given today’s easy and low-cost
communications, for the obvious reason that people at night usually prefer to
sleep.
An alternative way to communicate in real time is to travel. In this case,
again, East-West transaction costs may be more important than North-South
ones, since jet lag can affect the effectiveness of business travelers, or require
longer trips to adjust to the time difference. Jet lag, in effect, may lead to an
exception to the rule that people tend to sleep at night: those severely affected
by jet lag in fact tend to sleep during the day!
The transaction costs associated to the difference in time zones should be
important in activities that are intensive in information and require a great deal
of interaction in real time. For this reason, we think that FDI offers a perfect
setting in which to study the effects of time zone differences. Frequent real
time communications should be particularly important between headquarters
and their foreign affiliates, as well as between a firm and its prospective foreign
partners.2 In this sense, while time zones may also affect bilateral trade, the
need for real time communication is probably smaller among trading partners,
so we expect their effects to be relatively more important for FDI.
In a related paper, Kamstra et al (2000) investigate the effect of daylight
saving time changes on equity returns. They find that returns are significantly
lower immediately after the weekend the change occurred. If sleeping disorders
caused by minor time changes can affect the patterns of judgment, reaction time
and problem solving of stock market participants, the jet lag associated to intercontinental travel should also lead to important adverse effects.3 Portes and
Rey (2005) have addressed the empirical determinants of bilateral equity flows,
finding a negative and significant impact of the bilateral distance. Given that
transportation costs should have no effect on asset trading, the authors interpret
this finding as distance being a proxy for informational costs and asymmetries
between local and foreign investors. Once they include proxies for informational
aspects like the volume of bilateral telephone call traffic, the estimated elasticity
of distance is reduced, but remains highly significant. Interestingly, when they
barriers on bilateral FDI, he claims that it seems natural to assume that proximity plays an
important role in explaining bilateral FDI.
2 It is possible that the effect may go in the opposite direction in some specific sectors (such
as perhaps software development), in which differences in time zones may allow multinational
firms to gain some advantage by working around the clock. Unfortunately, our data does
not report FDI by sectors. Interestingly, studies in the business and information technology
literature identify time zones as a major obstacle of geographically dispersed teams in the
software sector(see e.g. Espinosa and Carmel,2005). Thus, this issue remains an open issue
for future research.
3 The problems associated with the jet lag are obviously well known, as are the difficulties
posed by time zone differences for live communications, even in the era of internet and e-mail.
In fact, the inspiration for this paper came from the difficulties experienced while writing a
recent paper with a co-author in Lebanon.
3
include the number of hours that trading in the host’s and source’s stock markets
overlap - a variable closely related to the bilateral time zone difference used in
this paper - it has a significant and positive effect. Loungani et al. (2002)
extend Portes and Rey (2005) analysis to the case of bilateral FDI flows. Using
the same data source that we use in the present paper, they estimate gravity
models for a panel of FDI flows. They find that the coefficient of distance
is significantly reduced and turns actually insignificant in some specifications,
once they account for the“transactional distance” approximated by the bilateral
telephone traffic volume. However, they do not include in their analysis variables
that may capture the time zone difference effect.4
The rest of the paper is organized as follows. Section 2 describes the data
and discusses our empirical strategy. In Section 3, we present our main results
on the estimated effect of time zones differences on the location of FDI. We
find that time zones have a negative and economically significant impact on the
location of FDI. Furthermore, once we introduce time zones into the analysis,
the negative impact of distance on FDI is significantly reduced, by around 50
percent in our baseline regression. We also presents several robustness checks
based on alternative measures of the time zone difference and empirical specifications, including a model that follows more closely recent FDI theory, the so
called knowledge-capital model first used as an empirical framework by Carr,
Markusen and Maskus (2001).5 In addition, we check whether our results are
robust to the use of different sub-samples and estimation methods. In Section
4, we present some extensions by analyzing the importance of time zones as a
determinant of bilateral trade. As expected, time zones matter also for trade,
but their impact is much smaller than that on FDI. In addition, we look at
the evolution of the impact of time zones over time, using panel data from the
same OECD database, from 1988 through 1999. We find that the impact of
time zones increases over time, a result that we attribute to the development of
new communications technologies, which have reduced dramatically the cost of
North-South distance, but have not reduced the cost of East-West distance to
the same extent. Finally, in Section 5, we conclude.
2
Data and empirical methodology
In order to analyze the effects of time zones on foreign direct investment, we
use bilateral data on FDI stocks from the OECD Direct Investment Statistics.
This variable is available for 17 source countries - all of them from the OECD
- and 58 host countries, which results in a total of 986 observations. We use
stocks rather than flows, because we are interested in the level of activity of
multinational enterprises, and capital stocks are a closer proxy to multilateral
4 It is interesting to point out that the partial correlation - after controlling for the GDP of
the host and source country - between the difference in time zones and the volume of telephone
traffic in 1997 is - 0.37 and significant at conventional levels. This suggests that the volume
of telephone traffic is potentially endogenous to the time-zone difference.
5 Therefore, in what follows we will refer to this specification as the CMM model.
4
activity than investment flows.6 In our cross-section regressions, we consider the
average values for 1997 to 1999 to avoid the potential influence of sudden changes
in valuation of FDI. This database has been used previously by Wei (1997, 2000)
to study the effect of corruption on FDI; by Stein and Daude (2002) to address
the impact of the quality of institutions on the location of FDI; by Daude et al
(2003) to analyze the relationship between FDI and regional integration; Egger
and Pfaffermayr (2004) to analyze the impact of bilateral investment treaties
(BITs) on FDI, and by Blonigen et al (2003) in order to test empirically different
theories of FDI.7
Although there have been recent developments in laying the theoretical foundations of FDI activity, as of today there is no clearly dominant or universal
theoretical framework to guide the empirical work. Therefore, our empirical
specification is designed to capture all potentially relevant determinants of FDI.
In the literature, different versions of the gravity model - which is the standard specification in empirical models of bilateral trade - have also been used
recently to estimate the determinants of bilateral FDI stocks and flows. In its
simplest formulation, the gravity model states that bilateral trade flows (in our
case bilateral FDI stocks) depend positively on the product of the GDPs of
both economies and negatively on the distance between them. While in the
trade literature the gravity model has good theoretical foundations, the use of
this model for the case of FDI is somewhat ad-hoc. An alternative specification,
which follows more closely the recent theoretical work on multinational activity,
is given by the CMM model and slightly different versions by Blonigen et al
(2003) and Carr et al (2003).
Given that our priority is to identify the effect of a variable corresponding
to the country-pair that is invariant over time, we control for source and host
characteristics by using source-country and host-country dummies. By doing so,
we also take care of potentially omitted variables and unobserved heterogeneity,
6 Clearly the ideal case would be to use production or affiliate sales, but given this data is
not available for a large sample of country pairs, foreign investment positions are - although
imperfect - the best available proxy. It is reassuring that Blonigen et al (2003) find that the
main results regarding their analysis are robust to using U.S. affiliate sales or the same FDI
stock data we use here.
7 According to the OECD Benchmark Definition of Foreign Direct Investment, FDI is as
an investment with “. . . the objective of obtaining a lasting interest by a resident entity in one
economy (“direct investor”) in an entity resident in an economy other than that of the investor
(“direct investment enterprise”). . . OECD recommends that a direct investment enterprise be
defined as an incorporated or unincorporated enterprise in which a foreign investor owns 10
per cent or more of the ordinary shares or voting power of an incorporated enterprise or the
equivalent of an unincorporated enterprise.” However, some countries use different criteria.
In terms of valuation, the OECD recommends market valuation, but also in this case many
countries follow: “. . . book values from the balance sheets . . . to determine the value of the
stock of direct investment.” Thus, there are potentially important differences across countries
in the definition and valuation of FDI. Also, the database presents some problems regarding
the reporting of zeros and missing values. In particular, source countries are not required to
report values for hosts with which FDI is zero. For this reason, the data on bilateral FDI
stocks required some adjustments. The criteria we used to distinguish the true missing values
from the zeros are described in detail in the appendix. More detailed information is available
at http://www.oecd.org/dataoecd/10/16/2090148.pdf.
5
as well as the measurement errors and differences in reporting methodologies
across source countries previously discussed. In addition, we include other bilateral variables that might affect the cost of doing business and whose effect if omitted from the analysis - could be picked up by the time zone difference,
in particular several of the standard gravity variables (like distance, adjacency,
common language, common membership in a free trade agreement, past colonial
links), as well as a common bilateral investment treaty (BITs), a common capital taxation treaty, and common legal origin. We also include some variables
from the CMM model, like the sum of GDPs, and the absolute difference in
GDP per capita between the source and the host country. Thus, the preferred
specification is given by:
ln(F DIij ) = βxij + αi + γi + δtimezoneij + ij ,
(1)
where F DIij is the average outward stock of FDI from source country i in host
country j between 1997 and 1999 in millions of dollars, αi is a source country
fixed effects, γi is a host country fixed effect, xij is a vector that includes all
bilateral control variables mentioned above, timezonesij is the variable that
captures the time zone difference between both countries, and ij is a random
error term.8
The bilateral distance used is the great circle distance between the countries’
capitals.9 The adjacency dummy was constructed using information from the
CIA’s World Factbook on whether the host and the source countries share a
common border. The past colonial links dummy was also constructed using
information from this source, as well as the common language dummy. The past
colonial links dummy has a value of one if the source and the host were colonized
by the same country or if either of the two countries was the colonizer of the
other during the twentieth century. The common language dummy takes a value
of 1 if the language spoken by the majority of the population in each country is
the same.10 We include a dummy for common membership in a regional trade
bloc, since Daude et al (2003) find that membership in a common regional
integration agreement increases FDI significantly. The information regarding
entrance dates and agreements was taken from Frankel (1997).11 In addition,
Egger and Pfaffermayr (2004) find that BITs have a positive and significant
impact on FDI. Therefore, we control for BITs including a dummy for a common
BIT. The information about existing BITs was taken from UNCTAD’s online
database.12 We also include a dummy that takes the value one if the country8 We also used a common religion dummy, but it is insignificant when the common legal
origin dummy is included. Therefore, we do not report results using this variable.
9 Following previous literature, in some cases other more centrally located cities were used
in place of the capitals, e.g. Chicago in the United States in place of Washington DC, or
Shanghai in China instead of Beijing. See Table A.1 for details.
10 This information is available at http://www.cia.gov/cia/publications/factbook/index.
11 We only considered agreements where a significant portion of total trade (more than 50
percent) is liberalized, rather than specific preferential trade agreements on only some specific
goods. These decisions were based also on information from Frankel (1997).
12 http://www.unctadxi.org/templates/DocSearch779.aspx.
As Egger and Pfaffermayr
6
pair has a bilateral capital tax treaty in place that year, which di Giovanni
(2005) has found to have a positive and significant impact on mergers and
acquisitions of FDI.13 La Porta et al (1998), and subsequent research in the law
and finance literature, stress the importance of legal origin on financial market’s
development and institutions. Therefore, as Lane and Milesi-Ferretti (2004) for
the case of bilateral portfolio positions, we include a dummy for common legal
origin, constructed using data from La Porta el al (2005).
Two additional control variables are included in our baseline regressions.
First, we include the sum of the GDPs (in logs) of the source and the host
countries. The rationale for the inclusion is to control for the joint ”mass”
that under almost any specification - whether the CMM or any gravity-type
equation - is a relevant determinant of bilateral FDI stocks.14 Second, we include
the difference in GDP per capita (in logs) between the source and the host.
This variable can be seen as a proxy for differences in factor endowments or
alternatively capture relevant determinants related to the differences in the level
of development of both economies.
Finally, in order to measure the importance of time zones on FDI, in our
benchmark regressions we use the time difference in hours between the countries’
capitals. This variable varies from 0 to 12. Table A.1 in the appendix presents
the time zone corresponding to each of the countries in the sample.15 In the section on robustness, we use two alternative measures of time zone differences, as
well as a decomposition of distance into a North-South and an East-West component. Table 1 presents summary statistics of the main variables, while Figure
A.1 presents the histogram of the time zone differences used in the baseline
regressions.
[Insert Table 1]
The log specification is used because it has typically shown the best adjustment to the data in the empirical trade literature. A problem that arises
when using the log of FDI as a dependent variable, however, is how to deal
with the observations with zero values. Our dataset includes a large number
of observations where FDI stocks are zero (about one quarter), which would be
dropped by taking logs. The problem of zero values of the dependent variable
is typical in gravity equations, and it has been dealt with in different ways.
Some authors (e.g. Rose, 2000) simply exclude the observations in which the
dependent variable takes a value of zero. A problem with this approach is that
those observations may convey important information for the problem under
(2004), we also recorded the time by which the agreement was signed rather than ratified
by both countries. However, results are quite similar such that we report only those using the
effective date when the treaty was ratified by source and host.
13 These data were kindly provided by di Giovanni. The primary data is from Taxanalysts’
website http://treaties.tax.org.
14 The correlation with the more traditional ”gravity” approach of considering the log of the
product of GDPs is around 0.92.
15 We construct the variable based on standard time zones, abstracting from the issue of
daylight savings.
7
consideration. Zero observations could for example be more prevalent among
countries that are far apart in terms of their longitude. Given the importance of
zero observations in our sample, this strategy could lead to a serious estimation
bias. Eichengreen and Irwin (1995, 1997) have proposed a simple transformation to deal with the zeros problem: work with log (1 + trade), instead of the
log of trade. This has the advantage of simplicity, and the coefficients can be
interpreted as elasticities when the values of trade tend to be large, since in this
case log (1 + trade) - or in our case log (1 + FDI) - is approximately equal
to log (trade). In turn, following Greene (1981), they scale up the coefficients
obtained from the OLS by a factor equal to the ratio between the total number
of observations and the number of non-zero observations. A disadvantage of
this approach is that it is somewhat ad hoc. Another approach has been to use
Tobit instead of OLS. Since we prefer this latter approach, we will derive our
main results using the Tobit specification. For robustness, however, we show
that our results are robust to the use of alternative approaches.
3
Empirical results
In the first column of Table 2, we present OLS estimations of the baseline
equation using the log of 1 + FDI as dependent variable, without including
the time zone variable and without source and host fixed effects. As discussed
above, we prefer the TOBIT estimate, but it is interesting to point out the
baseline OLS regression explains successfully more than two thirds of the total
cross-country variation in bilateral FDI stocks. In addition, most estimates
show the expected sign and are statistically and economically significant. In
particular, the distance between both countries has a negative and significant
impact on FDI. Its coefficient suggests that when distance increases by one
percent, bilateral stock of FDI falls by about 0.96 percent, an effect that is
slightly higher than that usually obtained in gravity models of bilateral trade,
and in line with the estimates of Portes and Rey (2005) on equity flows. In the
next column, we present the TOBIT estimates for the same specification. The
statistical significance of all variables remains the same, except for the dummy
for adjacency which becomes significant. In addition, the absolute size of all
point estimates increases in line with the argument of Greene (1981) discussed
above.16 In columns 3 and 4 we include the source and host country fixed effects
into the regressions. The OLS regression shows an increase in the goodness of
fit, increasing the R-squared to 0.82. In addition, for both estimation methods
the exclusion tests of these fixed effects are statistically significant. Therefore,
in addition to the considerations discussed above, from an empirical point of
view it is relevant to include these effects.
[Insert Table 2]
16 For example, the OLS coefficient on GDP Sum is 1.023 and the proportion of non-censored
observations is 0.767 = 665/867. Therefore, adjusting the OLS according to Greene (1981)
would yield an estimate of 1.334 = 1.023/0.767, which is very similar to the TOBIT estimate,
1.318.
8
In the last two columns of Table 2, we include our time zone variable in the
regression. The time zone variable has a negative and significant effect on FDI.
The TOBIT estimate shows that a time zone difference of one hour reduces the
bilateral FDI stock by around 26 percent.17 In addition, while the magnitude
and significance of all other regressors remains unchanged, the magnitude of the
distance is reduced by more than 50 percent, from -1.899 to -0.881. In the case
of the OLS regression the point estimate of the time zone effect is significant but
somewhat lower, around 20 percent, after adjusting the estimate by the fraction
of non-censored observations. Again, as in the case of the TOBIT estimation,
the inclusion of the time zone difference reduces the coefficient of distance by
around 50 percent.
Next, we test the robustness of our results in several ways. First, we consider
different measures of time zone differences. Second, we analyze if our results
are robust if we restrict our sample only to industrial-developing country pairs.
Third, we analyze the possibility of non-linearities in the effect of time zones.
We also check whether using a different specification, in particular the CMM
model, affects our results.
Tables 3 present the results of the TOBIT regressions using alternative specifications to capture the time difference effects. In column 1, we consider the
minimum time difference between both countries, in order to address the problem posed by countries with multiple time zones. In column 2, we use the
number of hours of overlap in office hours, assuming a standard workday from
9AM-5PM in each of the locations.18 Both variables yield similar results to our
basic time zone variable.
[Insert Table 3]
We also decompose the distance between the source and the host countries
into a latitudinal and a longitudinal component, and include both together in
our regressions. Let the capital of the source country be located at (Las, Los)
and that of the host country at (Lah , Loh ), denoting longitude and latitude in
gradients, respectively. We define latitudinal (or North-South) distance as the
logarithm of the great circle distance in miles from (Las, Los) to (Lah , Los ), i.e.
holding constant the longitude at the source country’s coordinate. Notice that
this distance would be the same if we held the longitude constant at the host
country’s coordinates instead. The measurement of longitudinal (or East-West)
distance is not as straightforward; a given difference in longitude can represent
a long distance, if countries are close to the equator, or a very short distance, if
they were close to the pole. Thus, the measure of longitudinal distance would
differ according to whether we hold latitude constant at the source coordinate,
or at the host coordinate. Our measure of longitudinal distance is just the
17 exp(-0.303)-1=-0.261.
18 This variable varies between 0 and 8 and, in contrast with the time zone difference variable,
is expected to have a positive coefficient. As mentioned above, Portes and Rey (2005) use a
similar variable - the overlap in trading hours in the respective stock exchanges - and find a
positive and significant effect on bilateral equity flows.
9
average of the two.19
The results using the decomposition of distance in its North-South and EastWest components are presented in column 3. While both measures of distance
are significant, the effect of longitude distance is significantly greater than latitude, with a point estimate that is roughly three times larger. Furthermore,
column 4 shows that the difference between the East-West and the North-South
components is mainly driven by the time zone difference. When including the
time zone variable in the regression with the decomposition of distance, the
longitudinal distance turns insignificant, while the estimated time zone effect is
significant and in line with the baseline estimate from Table 2.20 Taking all the
previous results together, our findings are clearly robust to different measures
of time zone differences.
In our data more than 80 percent of the total FDI stock is between industrial
countries, which tend to be located in the northern hemisphere. In addition,
much of that FDI occurs between countries in Europe, which are bundled together within a couple of time zones. To check whether this is driving our
results, we restrict our sample to industrial-developing country pairs.21 The
results presented in the last two columns of Table 3 show that the results with
respect to the effects of time zones differences on FDI remain similar to that
for the whole sample. The longitude coefficient is significantly greater than the
latitude dimension, and once we include the time zone difference variable, this
variable is negative and highly significant, while the longitudinal distance loses
its significance.
In order to explore the possible existence of non-linear effects of distance,
we add a quadratic distance term to the basic regression. In column 1 of Table
4, when both terms - distance and the quadratic term - are included, they
turn insignificant while the estimated time zone coefficient remain significant,
suggesting that non-linearities in distance do not play an important role.22 Next,
we estimated the model including dummy variables for each possible value of the
19 It is convenient here to clarify this with an example: The coordinates for Oslo, the
capital of Norway, are (59◦ 540 N, 10◦ 450 E), while those of Bogotá, the capital of Colombia, are
(4◦ 370 N, 47◦ 50 W). The longitudinal distance between Oslo and Bogotá is calculated as follows:
we first compute the great circle distance between (59◦ 540 N, 10◦ 450 E) and (59◦ 540 N,47◦ 50 W).
This turns out to be 2729 miles. Next, we repeat the exercise, now computing the great circle
distance between (4◦ 370 N,10◦ 450 E) and (4◦ 370 N,47◦ 50 W), i.e., holding the latitude constant
at the coordinate of Bogotá. This distance turns out to be 5834 miles. Not surprisingly,
the second value is much greater than the first, since Oslo is (relatively) close to the North
Pole, while Bogotá is very close to the equator. Our measure of longitudinal distance between
Norway and Colombia is simply the average of these two values: 4281.5.
20 Although the correlation between the time zones variable and the longitude distance in
0.82, the significance and relative stability in the estimated coefficient of the time zone effect
compared to the baseline model indicate that multicollinearity is not a serious issue.
21 All OECD countries except Korea, Poland, Mexico and Turkey were considered industrial
countries and the remaining countries in our sample were classified as developing. With this
classification out of the final number of 982 observations, 364 are North-North and 618 are
North-South.
22 We also estimated the model including the distance in levels rather than in logs and the
results do not change. In particular, the estimate of time zone differences is -0.334.
10
bilateral difference in time zones.23 Figure 1 presents the estimated coefficients
and the respective confidence intervals at a 95 percent of confidence.
[Insert Table 4]
[Insert Figure 1]
Two interesting conclusions can be drawn from this graph. First, FDI declines as the time zone difference increases. Second, the estimated effect becomes significant for differences in time zones greater or equal to 6 hours. The
figure suggests that there are some non-linearities associated with time zone
differences, with the impact being larger once the difference in time zones goes
beyond a certain threshold. However, the fit of the hour-by-hour effects with
respect to a linear trend is quite good - with an R2 of 0.81 - indicating that
the linear structure of the effect of time zone differences used in the previous
estimations seems not to impose too rigid a structure.24 In the second column
of Table 4, we included a dummy for country-pairs with a time zone difference
greater or equal to six hours. While the point estimate of the coefficient for this
variable is negative, it is not significant. Furthermore, the estimate for the time
zone differences remains stable.
In Table 5, we analyze the sensitiveness of our results to the use of an alternative specification, the CMM model. The CMM specification is based on the
so-called knowledge-capital model developed by Markusen (1997, 2001). This
theory tries to bring the vertical and horizontal approaches of FDI into a unified
framework.25 Within this model, the type (horizontal versus vertical) and the
volume of FDI between the two countries depends on the size of each economy,
differences in the size between the host and the source country, relative factor
endowments, trade costs and investment costs. When countries differ in size,
but not in factor endowments, there is no vertical FDI, and horizontal FDI follows an inverted U-shape, reaching its highest point when the size of the two
countries is similar. 26 In this sense, the empirical specification should include
the difference in size and its squared value to account for this relationship. Vertical FDI takes place if the difference in the size of the economies is significant
and the small country is skilled labor intensive, so that the production facility
tends to be installed abroad. Since headquarters’ decisions on location are based
23 Given that only 13 country pairs have a time zone difference of 12 hours, we aggregated
them with those that have 11 hours because of the imprecision in the estimate. The constant
term in the regression captures the 0 hours time zone difference case, such that all dummies
are incremental with respect to this case.
24 In addition, the slope coefficient of this linear trend is -0.33, which is in line with previous
estimates.
25 See Helpman (1984) and Helpman and Krugman (1985) for early model of vertical FDI,
and Markusen (1984), Horstmann and Markusen (1987, 1992), and Markusen and Venables
(1998, 2000) for horizontal FDI models. For more a recent theory that incorporates firm
heterogeneity in evaluating the trade-off between exporting or investing abroad, see Helpman
et al (2004).
26 The reason is that if one country is much larger than the other economies of scale imply
that it is cheaper to produce in the large country and supply the small one through trade.
11
on factor prices while plant location is based on factor prices and market size,
an interaction term between both variables should be included in the empirical
specification of the model.27 Furthermore, trade costs in the host country encourage horizontal FDI, while trade costs in the source country restrict vertical
FDI, and investment restrictions in the host country discourage all forms of
multinational activity. Since trade costs in the host favor horizontal (but not
vertical) FDI, and horizontal FDI increases if factor endowments are similar,
CMM include in the specification an interaction between trade costs and the
squared endowment differences.28
The specification we use here is very close to Blonigen et al (2003). It
only differs in the variables we use to proxy trade costs, which are the average
levels of nominal tariffs between 1990 and 1997 from the World Bank. Also,
while Blonigen et al (2003) use the composite index from BERI to account
for investment costs, our proxy is the summary index of institutional quality
by Kaufmann et al (1999). We estimate the model in levels as Carr et al
(2003) and Blonigen et al (2003) do, without fixed effects. Clearly, in terms
of our qualitative results, columns 1 - 5 confirm our previous findings. For
all alternative measures of the time zone difference, there is a negative and
significant impact of time zones on FDI. In addition, the effects are economically
significant. For example, an increase in one hour of time zone difference decrease
on average the FDI stock by US$ 262 millions or alternatively, one more hour of
office hour overlap implies an increase of around US$ 318 millions. Interestingly,
distance is no longer significant, once we include the time zone variables. In line
with this result, when decomposing distance into latitude and longitude, only
the latter has a significant negative impact on FDI. In column 6, we estimated
the model in logs and confirm the significance of the time zone effect.
[Insert Table 5]
Finally, in the last column we include the volume of bilateral telephone traffic
used by Portes and Rey (2005) as a proxy for information frictions, given that
the time zone difference might be picking up part of the information effect.
There are several interesting results to comment on. First, the information
variable has a significant and positive effect on FDI, suggesting that, as in
the case of portfolio flows, this variable is also relevant for FDI. Second, the
time zone coefficient is still significant and negative, but the estimate is slightly
smaller. One possible explanation for this is that telephone traffic itslef may
be endogenous to the time zone difference. In this case, time zones would have
an effect through the volume of telephone traffic, but also a significant direct
effect.29 .
27 See
Carr et al (2001) for further discussion.
reason is that, if differences in factor prices are sufficiently large (for a given level of
trade costs), it will be more convenient to produce in the country with better comparative
advantage, and supply the other country through trade.
29 The coefficient of distance turns positive and significant, once we account for information
explicitly. This result may be reflecting the presence of horizontal FDI, which is increasing in
trade costs.
28 The
12
4
Extensions
In this section, we consider two extensions of our results. First, we repeat the
exercise for bilateral trade, rather than bilateral FDI stocks. While we expect
that time zones might be important for trade as well, we expect the impact to
be smaller than that in the case of FDI. The reason is that trade transactions
are not as demanding in terms of real time interaction between the parties as
is generally the case for FDI. In the second extension, we study the impact of
time zone differences as it evolves over time. This second extension is perhaps
more fundamental, as it may provide some clues regarding the channel through
which time zone differences matter. We expect the impact to change over time,
as the development of communications and Internet technologies facilitate long
distance real time interaction.
4.1
Time zones and bilateral trade
The OLS estimates for a traditional gravity model using trade as our dependent
variables are presented in Table 7. In our specification, we include the product
of GDPs (in logs), the product of GDP per capita (in logs), common language,
adjacency, colonial links, a RTA dummy and distance.30
In the first column of Table 6, we report the results of the basic gravity
equation, without the time zones variable. The effect of distance is smaller
than that found for FDI, but roughly consistent with the impact found in the
empirical trade literature. An increase in distance of 1 percent reduces bilateral
trade by approximately 0.9 percent. In column 2 we include our basic time zone
difference variable. The coefficient for the time zone difference is negative and
significant. In columns 3 and 4, we show that the impact of time zone differences
remains significant when we use either of our alternative definitions of time
zones. However, the impact of time zones for trade is somewhat smaller than
the impact on FDI. While an additional hour of time difference reduces bilateral
FDI by between 17 and 26 percent, the impact on trade ranges between 7 and
11 percent, depending on the specification used. Second, and not surprisingly,
distance itself continues to be significant in all cases after controlling for time
zone differences. In column 5 and 6, we analyze the decomposition of distance
in latitude and longitude. While the first regression shows that the longitude
effect is significantly greater than the latitude distance, the last column of Table
6 shows that this difference is mainly driven by the time zone effect. Once we
account explicitly for this effect, the effects of latitude and longitude are both
negative and significant, and not statistically different. In summary, time zones
also matter for trade, although the effects are quite a bit smaller than those on
FDI.31
30 The data on trade are from the IMF Direction of Trade Statistics database for 1999. In
our final sample, there are no observations in which trade takes a value of zero. Thus, we
consider just the log of exports plus imports. We also include host and source country effects,
following the suggestion by Anderson and van Wincoop (2003) to account for “multilateral
resistance”.
31 However, Melitz (2004) has proposed an alternative explanation for why latitude matters
13
[Insert Table 6]
4.2
Evolution of the time zone effect over time
We now turn our attention to the evolution over time of the time zone effect.
One of the reasons to do this exercise is to study the impact that the development - or widespread adoption - of new communications technologies might
have on the relative cost of doing business in different locations. Over the period under study, technologies such as the Internet or videoconference became
close substitutes to face-to-face communication, allowing cheaper and more fluid
interaction in real time among people in distant locations. While the Internet
dates from the early 1970s, it was not until the beginning of the 1990s that its
use - especially for business activities - spread widely around the world.32 Similarly, while technologies for teleconference and videoconference were available
before the 1990s, the introduction of ISDN telephone lines in the early 1990s
substantially increased their reliability, and reduced their cost. But as important as these technologies are to reduce the cost of real time interaction, they
cannot overcome the problem on which we focus here: if time zones are very
different, one or both parties in the interaction will have to work beyond regular business hours. Thus, these technologies should have differential effects on
transaction costs, reducing those among North-South parties more than those
involved in East-West interaction.
In order to investigate the impact of time zones over time, we estimate equation (1) using panel data from 1988 through 1999 and include in the regression
the interaction of the time zone difference with each of the year dummies. In
this way, the coefficient of our variable of interest is allowed to change over
time.33 We restrict our analysis to those country pairs that presented data for
the whole sample, to make sure that the results are not driven by changes in the
sample resulting from changes in investment patterns over time. In addition, in
our preferred specification we include source-year and host-year fixed effects, in
order to account for changing unobserved country characteristics over time, as
well as year dummies to account for the trend of overall increase of FDI during
the period. Also, given that our dependent variable is a stock and therefore
autocorrelated, we use the Prais-Winsten estimator to account for this feature.
Since we are already controlling for distance, a differential impact of communications technologies as discussed above would result in an increase over time in
the yearly time zone coefficients.
In Figure 2, we plot the point estimates for four different specifications of
our model. The OLS cross-section estimates are the estimated coefficients that
less than longitude with regards to trade. In some cases, differences in latitude are associated
with differences in comparative advantages, and therefore increase trade.
32 Hobbes internet timeline at www.zakon.org/robert/internet/timeline identifies the year
1991 as the year in which the World Wide Web was released by CERN (the European Organization for Nuclear Research), and the year 1993 as the year in which “business and media
really take notice of the internet.”
33 Observe also that the dummies for capital taxation treaties, BITs, and RTAs are timevariant.
14
result from estimating year-by-year equation (1). The OLS pooled results refer
to the estimates of the interaction of each year dummy with the time zone difference, including source and host country dummies, as well as year dummies.
Prais-Winsten with FE, reports the estimates of the pooled model, but correcting for autocorrelation. Finally, Prais-Winsten FE-time allows the source
and host country effects to vary year-by-year. Clearly, independently of the
estimation method, the point estimates show a clear trend of a stronger time
zone difference effect over time. This trend is especially pronounced in the early
1990s, although the impact continues to change at least until 1996.34
[Insert Figure 2]
In Table 7 we test whether the change over time in the yearly time zone coefficients is statistically significant and find significant evidence that the impact of
time zones has been decreasing during the early 1990s until around 1995/1996.
These results are consistent with the hypothesis that technological change is not
likely to do away with the transactions costs imposed by time zone difference.
If anything, technological change may increase the importance of this factor for
the location decisions of multinational businesses.
[Insert Table 7]
5
Conclusions
We have examined the effects of time zone differences on the location of FDI
around the world. The results show that time zone differences between the
source and host countries have a negative impact on bilateral FDI. This impact is
both statistically significant and economically important. Furthermore, once we
control for the time zone effect, the coefficient of the distance between the source
and the host is significantly reduced and under some specifications no longer
significant. This indicates that in the case of FDI an important component
of distance is the East-West component, since transaction costs in activities
that require real-time interaction between a firm’s headquarters and its foreign
affiliates are increasing in this dimension. Our results are robust to different
measures of time zone differences, as well as different estimation procedures.
These results suggest that empirical research on bilateral FDI should account
for this effect in order to obtain consistent estimates for their parameters of
interest. They may also be helpful to understand the patterns of competition to
attract FDI. Specifically, host countries would be more exposed to competition
from other countries in proximate time zones and less exposed to competition
from potential host countries in more distant time zones, other things equal.
34 It is worth mentioning that we also estimated the model using the Arellano-Bond GMM
method. Although the inclusion of the lagged dependent variable makes it harder to interpret
the results, we still find a significant and negative coefficient for the years 1991, 1995, 1996,
1997 and 1998. The results are available upon request.
15
Finally, our results suggest that the problem posed by time zone differences
has become more relevant over time, with the introduction and widespread
dissemination of new information technologies. This link between technology
and the impact of time zones is more than mere speculation. It is a pattern
that we observe more and more in everyday life, and is supported by numerous
examples: like that of an acquaintance, who works for a business located in
Washington DC, but telecommutes from Buenos Aires; or that of the chief
economist for Latin America at the World Bank, who is currently based in
Bogotá. These cases would have been unthinkable just ten years ago. And it
would be hard to imagine the World Bank’s chief economist for Asia working
from Bangkok. Interaction with the staff would be, literally, a nightmare.
16
References
Acemoglu, Daron, Johnson, Simon, and Robinson, James A., 2001. The Colonial Origins of Comparative Development. American Economic Review, 91 (5),
1369 - 1401.
Anderson, James E., and van Wincoop, Eric, 2003. Gravity with Gravitas:
A Solution to the Border Puzzle. American Economic Review, Vol. 93, No. 1,
170 - 191.
Barro, Robert J. and Lee, Jong-Wha, 2000. International Data on Educational
Attainment Updates and Implications. National Bureau of Economic Research
Working Paper 7911, September 2000.
Blonigen, Bruce A., Davies, Ronald B., and Head, Keith, 2003. Estimating the
Knowledge-Capital Model of the Multinational Enterprise: Comment. American Economic Review, 93 (3), 980 - 994.
Carr, David L., Markusen, James R. and Maskus, Keith E., 2001. Estimating
the Knowledge-Capital Model of Multinational Enterprise. American Economic
Review, 91 (3), 693 - 708.
Carr, David L., Markusen, James R. and Maskus, Keith E., 2003. Estimating
the Knowledge-Capital Model of Multinational Enterprise: Reply. American
Economic Review, 93 (3), 995 - 1001.
Daude, Christian, Levy-Yeyati, Eduardo, and Stein, Ernesto, 2003. Regional
Integration and the Location of FDI, Working Paper 492, Research Department,
Inter-American Development Bank, Washington, DC.
Espinosa, J. Alberto,and Carmel, Erran, 2005. The Impact of Time Separation
on Coordination in Global Software Teams: a Conceptual Foundation. Journal
of Software Process Improvement and Practice, forthcoming.
di Giovanni, Julian, 2005. What Drives Capital Flows? The Case of Crossborder M&A Activity and Financial Deepening. Journal of International Economics, 65, 127 - 149.
Eichengreen, Barry and Irwin, Douglas A., 1995. Trade Blocs, Currency Blocs
and the Reorientation of Trade in the 1930s. Journal of International Economics, 38 (1-2), 1 - 24.
Eichengreen, Barry and Irwin, Douglas A., 1997. The Role of History in Bilateral Trade Flows, in: Jeffrey Frankel (Ed.), The Regionalization of the World
Economy, University of Chicago Press, Chicago, IL, 33 - 57.
17
Egger, Peter, and Pfaffermayr, Michael, 2004. The Impact of Bilateral Investment Treaties on Foreign Direct Investment. Journal of Comparative Economics
32, 788 - 804.
Frankel, Jeffrey, with Stein, Ernesto H. and Wei, Shang-Jin, 1997. Regional
Trading Blocs in the World Economic System, Institute for International Economics, Washington DC.
Gallup, John Luke, Sachs, Jeffrey and Mellinger, Andrew, 1999. Geography
and Economic Development. International Regional Science Review, 22 (2),
179 - 232.
Greene, William H., 1981. On the Asymptotic Bias of the Ordinary Least
Squares Estimator of the Tobit Model. Econometrica, 49 (2), 505 - 513.
Hall, Robert E. and Jones, Charles I., 1999. Why Do Some Countries Produce So Much More Output per Worker than Others? Quarterly Journal of
Economics, 114 (1), 83 - 116.
Helpman, Elhanan, 1984. A Simple Theory of International Trade with Multinational Corporations. Journal of Political Economy, 92 (3), 451 - 471.
Helpman, Elhanan and Krugman, Paul, 1985. Market Structure and Foreign
Trade, Cambridge, MA, The MIT Press.
Helpman, Elhanan, Melitz, Marc J., and Yeaple, Stephen R., 2004. Export
versus FDI with Heterogeneous Firms. American Economic Review, 94 (1), 300
- 316.
Horstmann, Ignatius and Markusen, James, 1987. Strategic Investments and
the Development of Multinationals. International Economic Review, 28 (1),
109-121.
Horstmann, Ignatius and Markusen, James, 1992. Endogenous Market Structures in International Trade (Natura Facit Saltum). Journal of International
Economics, 32 (1-2), 109 - 129.
Kamstra, Mark, Kramer, L.A. and Levi, M., 2000. Losing Sleep at the Market:
The Daylight-Savings Anomaly. American Economic Review, 90 (4), 1005 1011.
Kaufmann, Daniel, Kraay, Aart; Zoido-Lobaton, Pablo, 1999. Governance Matters, World Bank Policy Research Working Paper 2196, Washington DC, Worldbank.
Lane, Philip, and Milesi-Ferretti, Gian Maria, 2004. International Investment
18
Patterns. IMF Working Paper WP/04/134, July.
La Porta, Rafael, Lopez-de-Silanes, Florencio, and Shleifer, Andrei, 2005. What
Works in Securities Laws? Journal of Finance, forthcoming.
La Porta, Rafael, Lopez-de-Silanes, Florencio, Shleifer, Andrei, and Vishny,
Robert W., 1998. Law and Finance.Journal of Political Economy 106, 1113 1155.
Leamer, Edward and Levinsohn, James, 1995. International Trade Theory: The
Evidence, in Gene M. Grossman and Kenneth S. Rogoff, (Eds.), The Handbook
of International Economics: Vol. III, North Holland, Amsterdam ; New York,
1339 - 1394.
Loungani, Prakash, Mody, Ashoka, and Razin, Assaf, 2002. The Global Disconnect: The Role of Transactional Distance and Scale Economies in Gravity
Equations. Scottish Journal of Political Economy, 49 (5), 526 - 543.
Markusen, James, 1984. Multinationals, Multi-plant Economies, and the Gains
from Trade. Journal of International Economics, 16 (1), 205 - 226.
Markusen, James, 1997. Trade versus Investment Liberalization. National Bureau of Economic Research Working Paper 6231, October.
Markusen, James, 2001. Contracts, Intellectual Property Rights, and Multinational Investment in Developing Countries. Journal of International Economics,
53 (1), 89 - 204.
Markusen, James and Venables, Anthony, 1998. Multinational Firms and the
New Trade Theory. Journal of International Economics, 46 (2), 183 - 204.
Markusen, James and Venables, Anthony, 2000. The Theory of Endowment,
Intra-Industry, and Multinational Trade. Journal of International Economics,
52 (2), 209 - 234.
Martin, Philippe and Rey, Helene, 2004. Financial Super-markets: Size Matters
for Asset Trade. Journal of International Economics 64, 335 - 361.
Melitz, Jacques, 2004. Geography, Trade and Currency Union, in Volbert
Alexander, George von Furstenberg and Jacques Mlitz, (eds) Monetary Unions
and Hard Pegs: Why, Effects on Trade, Financial Developments and Stability,
Oxford University Press.
OECD, 1999. OECD Benchmark Definition of Foreign Direct Investment, Third
Edition, mimeo. Organisation for Economic Co-operation and Development,
Paris, France.
19
Portes, Richard and Rey, Hlene, 2005. The Determinants of Cross-Section Border Equity Flows. Journal of International Economics, 65, 269 - 296.
Rauch, James E., 1999. Networks versus Markets in International Trade. Journal of International Economics, 48 (1), 7 - 35.
Rose, Andrew, 2000. One Money, One Market: The Effect of Common Currency on Trade. Economic Policy, 15 (30), 7-46.
Stein, Ernesto H. and Daude, Christian, 2002. Institutions, Integration, and
the location of Foreign Direct Investment, in: New Horizonts of Foreign Direct
Investment, OECD Global Forum on International Investment, Paris, France,
OECD.
Wei, Shang-Jin, 1997. Why is Corruption So Much more Taxing than Tax?
Arbitrariness Kills. National Bureau of Economic Research, Working Paper
6255.
Wei, Shang-Jin, 2000. How Taxing is Corruption to International Investors?,
Review of Economics and Statistics, 82 (1), 1 - 11.
20
Appendix
FDI Data
The OECD International Direct Investment Statistics database presents
some problems regarding the reporting of zeros and missing values. In particular, source countries are not required to report values for hosts for which
FDI is zero. For this reason, the data on bilateral FDI stocks required some
adjustments. We used the following criteria to distinguish the zero values from
the true missing observations:
- if a source country reports missing values to a particular host country, but
had reported positive values during a previous year, then those missing
values were considered to be truly missing.
- if a source country reports missing values to a given host country, but the
OECD dataset on bilateral flows of FDI shows that there was FDI activity
from that source to that host in previous years, then the missing values
are considered to be truly missing.
- if a source country starts reporting bilateral stocks and flows of FDI to
a given host in the same year, and the first positive value for the stock
is significantly larger than that of the flow, we considered all previous
observations for that country pair to be truly missing.
- if neither of the three previous conditions applies, and a source reports
positive values of FDI to some hosts, but missing values to others, those
missing values are considered to be zeros. As an example, if Germany
reports positive values to Chile, but missing values to Argentina (and
there is no evidence, either from previous stocks or flows or the comparison
between them, that there was some prior bilateral multinational activity)
then the missing values reported to Argentina are considered to be zeros.
There are some exceptions to the last criteria, however. In particular, if
a source country, having reported stocks to some countries in the sample in
previous years, suddenly begins reporting FDI to a block of new host countries
at the same time, we used our discretion to determine whether the previous
missing observations should be truly missing, or zeros. As an example, in 1993
Netherlands began reporting positive values to 26 host countries for which each
of the previous observations had been coded as missing. We determined that
such an abrupt change was more likely to reflect a change towards more detailed
information disclosure rather than a genuine change in investment. Thus, all the
missing values prior to 1993 were coded as truly missing. To provide an example
that goes the other way, several source countries began reporting positive values
of bilateral stocks to the transition economies, right after the fall of the Berlin
Wall. In this case, we considered the previous missing values as zeros, since we
determined that they were more likely to reflect the absence of investments.
21
Table A.1 Time Zones, Latitudes and Longitudes
Country
City
Algeria
Argentina
Australia
Austria
Belgium-Luxembourg
Brazil
Bulgaria
Canada
Chile
China
Colombia
Costa Rica
Czech Republic
Denmark
Egypt
Finland
France
Germany
Greece
Hong Kong
Hungary
Iceland
India
Indonesia
Iran
Ireland
Israel
Italy
Japan
Korea
Kuwait
Libya
Malaysia
Morocco
Mexico
Netherlands
New Zealand
Norway
Panama
Philippines
Poland
Portugal
Romania
Algiers
Buenos Aires
Canberra
Vienna
Brussels
Brasilia
Sofia
Toronto
Santiago
Shanghai
Bogotá
San José
Prague
Copenhagen
Cairo
Helsinki
Paris
Berlin
Athens
Hong Kong
Budapest
Reykjavik
New Delhi
Jakarta
Tehran
Dublin
Jerusalem
Rome
Tokyo
Seoul
Kuwait
Tripoli
Kuala Lumpur
Rabat
Mexico
Amsterdam
Wellington
Oslo
Panama
Manila
Warsaw
Lisbon
Bucharest
Time
City
Min.
Time
1
-3
10
1
1
-3
2
-5
-4
8
-5
-6
1
1
2
2
1
1
2
8
1
0
5.5
7
3.5
0
2
1
9
9
3
2
8
1
-6
1
12
1
-5
8
1
0
2
22
Max. Time
8
10
-3
-5
-5
-8
7
9
-6
-8
Latitude
36 º
34 º
35 º
48 º
50 º
15 º
42 º
43 º
33 º
31 º
4º
9º
50 º
55 º
30 º
60 º
48 º
52 º
37 º
22 º
47 º
64 º
28 º
6º
35 º
53 º
31 º
41 º
35 º
37 º
29 º
32 º
3º
34 º
19 º
52 º
41 º
59 º
8º
14 º
52 º
38 º
44 º
46 '
36 '
18 '
13 '
49 '
46 '
41 '
38 '
27 '
13 '
37 '
55 '
4'
40 '
3'
10 '
51 '
31 '
58 '
8'
30 '
8'
36 '
10 '
40 '
19 '
46 '
53 '
40 '
33 '
19 '
52 '
9'
1'
25 '
22 '
16 '
54 '
58 '
37 '
15 '
43 '
26 '
Longitude
N
S
S
N
N
S
N
N
S
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S
N
N
N
N
N
N
N
N
N
N
N
N
S
N
N
N
N
N
N
3º
58 º
149 º
16 º
4º
47 º
23 º
79 º
70 º
121 º
74 º
84 º
14 º
12 º
31 º
24 º
2º
13 º
23 º
114 º
19 º
21 º
77 º
106 º
51 º
6º
35 º
12 º
139 º
126 º
47 º
13 º
101 º
6º
99 º
4º
174 º
10 º
79 º
120 º
21 º
9º
26 º
2' E
22 ' W
5' E
22 ' E
19 ' E
54 ' W
18 ' E
22 ' W
38 ' W
28 ' E
5' W
4' W
25 ' E
34 ' E
15 ' E
56 ' E
20 ' E
22 ' E
43 ' E
5' E
4' E
55 ' W
13 ' E
49 ' E
25 ' E
15 ' W
13 ' E
30 ' E
46 ' E
59 ' E
58 ' E
10 ' E
42 ' E
50 ' W
8' W
53 ' E
46 ' E
45 ' E
31 ' W
58 ' E
1' E
8' W
6' E
City
Saudi Arabia
Singapore
Slovakia
Slovenia
South Africa
Spain
Sweden
Switzerland
Thailand
Turkey
Ukraine
United Arab Emirates
United Kingdom
United States
Venezuela
Riyadh
Singapore
Bratislava
Ljbljana
Pretoria
Madrid
Stockholm
Bern
Bangkok
Ankara
Kiev
Abu Dhabi
London
Chicago
Caracas
Time
City
Minimum Maximum
Time
Time
3
8
1
1
2
1
1
1
7
2
2
4
0
-6
-4
-5
Latitude
24 º
1º
48 º
46 º
25 º
40 º
59 º
46 º
13 º
39 º
50 º
24 º
51 º
-8 41 º
10 º
38 '
18 '
9'
3'
43 '
25 '
19 '
57 '
43 '
55 '
26 '
28 '
31 '
50 '
32 '
Longitude
N
N
N
N
S
N
N
N
N
N
N
N
N
N
N
46 º
103 º
17 º
14 º
28 º
3º
18 º
7º
100 º
32 º
30 º
54 º
0º
87 º
66 º
46 ' E
50 ' E
7' E
30 ' E
13 ' E
42 ' W
4' E
26 ' E
30 ' E
51 ' E
31 ' E
22 ' E
6' W
40 ' W
55 ' W
0
50
Frequency
100
150
200
Country
0
5
10
Time Zone Difference
Figure A.1 – Frequency Distribution of Time Zone Differences
23
15
TABLE 1 – Summary Statistics of main variables
Variable
Time zone difference
FDI stock (avg. 97 - 99)
Log(1+FDI Stock)
Sum GDPs (log)
Absolute difference GDP (logs)
Common language
Adjacency
Colonial Links
Common legal origin
RTA
Capital tax treaty
BIT
Distance (miles)
Log(Distance)
Latitude distance
Longitude distance
Log(Latitude distance)
Log(Longitude distance)
Absolute skill difference
Average nominal tariff (source)
Average nominal tariff (host)
Institutions
Tariffs host x squared diff. skills
Minimum time zone difference
Office hours overlap
Log(Telephone traffic)
Obs
1140
1046
1046
1120
1026
1140
1140
1140
1000
1064
1140
1140
1140
1140
1140
1140
1140
1140
980
1133
981
1140
901
1140
1140
625
Mean
4.314035
2949.295
3.983091
24.67155
0.796939
0.074561
0.037719
0.007895
0.221
0.178571
0.65614
0.007895
4261.451
7.945225
2000.694
3202.55
7.017266
7.51407
3.149122
5.800247
8.239919
0.56373
163.1212
4.140351
3.960088
3.44302
24
Std. Dev.
3.585109
12367.6
3.505843
2.128342
0.676584
0.262798
0.1906
0.08854
0.415128
0.383173
0.475203
0.08854
3202.145
1.041901
1902.291
2434.732
1.275867
1.325737
2.230307
1.865112
7.456723
0.770848
337.975
3.55152
3.171279
1.704421
Min
0
0
0
18.21777
0.000137
0
0
0
0
0
0
0
35.22688
3.561809
0
3.16144
0
1.151028
0.01
1.66
0
-1.46259
0
0
0
-1.20397
Max
12
202471.1
12.21836
31.46873
2.651846
1
1
1
1
1
1
1
12329.42
9.419744
7278.747
9269.531
8.892714
9.134488
10.36
11.8
45.66
1.712496
3856.265
12
8
8.867428
TABLE 2 – Baseline Regressions
Dependent Variable
Estimation Method
GDP Sum
GDP per capita
Absolute
Difference
Common Language
Adjacency
Colonial Links
Common Legal
Origin
Regional Trade
Agreement
Capital Tax
Agreement
Bilateral
Investment
Agreement
Distance
Time Zone
Difference
Observations
Censored
Observations
R-squared
Log-pseudo
Likelihood
Source Effects
Significance Test
Host Effects
Significance Test
(1)
Ln(1+FDI)
OLS
(2)
Ln(FDI)
TOBIT
(3)
Ln(1+FDI)
OLS
(4)
Ln(FDI)
TOBIT
(5)
Ln(1+FDI)
OLS
(6)
Ln(FDI)
TOBIT
1.023
(0.034)**
-0.422
(0.120)**
1.541
(0.073)**
-0.839
(0.243)**
1.318
(0.108)**
-1.381
(0.399)**
2.031
(0.211)**
-2.213
(0.783)**
1.353
(0.110)**
-1.472
(0.399)**
2.123
(0.207)**
-2.439
(0.784)**
1.6379
(0.246)**
-0.190
(0.259)
1.2825
(0.501)*
0.5358
(0.170)**
0.010
(0.199)
0.520
(0.187)**
1.843
(0.882)*
2.108
(0.423)**
-1.151
(0.464)*
2.104
(0.872)*
1.156
(0.327)**
-0.006
(0.350)
2.180
(0.403)**
2.894
(1.548)#
0.243
(0.257)
-0.066
(0.302)
0.890
(0.361)*
0.856
(0.137)**
0.173
(0.213)
0.284
(0.207)
1.149
(0.707)#
-0.154
(0.474)
-1.267
(0.542)*
1.440
(0.741)#
1.638
(0.269)**
0.104
(0.407)
1.812
(0.428)**
1.659
(1.290)
0.212
(0.253)
0.261
(0.317)
0.944
(0.360)**
0.910
(0.138)**
0.084
(0.215)
0.130
(0.204)
1.034
(0.689)
-0.202
(0.465)
-0.577
(0.570)
1.541
(0.716)*
1.758
(0.266)**
-0.092
(0.411)
1.476
(0.423)**
1.425
(1.274)
-0.957
(0.077)**
867
-
-1.551
(0.159)**
867
202
-1.115
(0.103)**
867
-
-1.899
(0.204)**
867
202
-0.643
(0.169)**
-0.141
(0.038)**
867
-
-0.881
(0.325)**
-0.303
(0.073)**
867
202
0.67
-
-2036.70
0.82
-
-1792.71
0.83
-
-1781.91
-
-
12.13**
135.60**
12.50**
147.74**
-
-
9.88**
368.84**
10.68**
374.26**
Note: GDP Sum and Distance are in logs. White-corrected robust standard errors are in parentheses. **, *, # denote statistical
significance at 1%, 5% and 10% levels, respectively. Source and host country tests are F-tests and chi-squared test for OLS
regressions and TOBIT regression, respectively.
25
TABLE 3 – Robustness Checks to different measures of Time Zone Differences and North –
South Sample
(1)
ALL
(2)
ALL
(3)
ALL
(4)
ALL
(5)
North South
(6)
North South
2.160
(0.205)**
-2.533
(0.780)**
-0.214
(0.470)
-0.511
(0.578)
1.318
(0.662)*
1.816
(0.267)**
-0.184
(0.417)
1.474
(0.422)**
1.419
(1.267)
-0.683
(0.355)#
-
1.949
(0.202)**
-2.303
(0.773)**
-0.094
(0.481)
-0.662
(0.562)
0.988
(0.726)
1.866
(0.272)**
0.469
(0.419)
1.790
(0.439)**
1.349
(1.451)
-
2.089
(0.206)**
-2.468
(0.784)**
-0.179
(0.464)
-0.304
(0.556)
1.479
(0.723)*
1.850
(0.265)**
-0.004
(0.413)
1.453
(0.424)**
1.335
(1.311)
-
0.604
(0.462)
-6.549
(3.033)*
0.681
(0.822)
-0.780
(1.106)
0.224
(1.025)
0.953
(0.406)*
-3.379
(1.986)#
2.929
(0.701)**
0.343
(1.302)
-
0.583
(0.423)
-5.556
(2.873)#
0.656
(0.779)
0.346
(1.187)
0.506
(0.934)
1.110
(0.403)**
-5.110
(2.062)**
2.501
(0.678)**
0.440
(1.114)
-
Time Zone Difference
2.104
(0.209)**
-2.336
(0.785)**
-0.115
(0.462)
-0.742
(0.567)
1.562
(0.723)*
1.762
(0.267)**
-0.024
(0.411)
1.475
(0.423)**
1.414
(1.256)
-1.041
(0.303)**
-
-
Minimum Time Zone
Difference
Office Hours Overlap
-0.266
(0.067)**
-
-
-
-0.384
(0.067)**
-
-
-0.448
(0.265)**
-
-
-
-
-
-0.262
(0.143)#
-0.227
(0.208)
867
202
-0.305
(0.284)
-1.796
(0.233)**
473
164
-0.427
(0.408)
-0.711
(0.127)#
473
164
-1783.531
0.02
-885.624
13.38**
-877.043
0.31
Sample
GDP Sum
GDP per capita Absolute
Difference
Common Language
Adjacency
Colonial Links
Common Legal Origin
RTA
Capital Tax Agreement
BIT
Distance
Latitude Distance
-
0.393
(0.089)**
-
Longitude Distance
-
-
867
202
867
202
-0.332
(0.150)*
-1.021
(0.143)**
867
202
-1782.526
-
-1780.14
-
-1802.576
9.77**
Observations
Censored Observations
R-squared
Log-pseudo Likelihood
Test Latitude = Longitude
(chi – squared[1])
Note: TOBIT estimates. GDP Sum and Distance are in logs. White-corrected robust standard errors are in parentheses. **, *, # denote
statistical significance at 1%, 5% and 10% levels, respectively. All regressions include source and host country fixed effects, which
are not reported.
26
TABLE 4 – Non-linearities and Spatial Correlation Regressions
Estimation
Method
GDP Sum
GDP per capita
Absolute
Difference
Common
Language
Adjacency
Colonial Links
Common Legal
Origin
RTA
Capital Tax
Agreement
BIT
Distance
Distance squared
Time Zone
Difference
Time Zone
Difference > 6
hours
Observations
Censored
Observations
R-squared
Log-pseudo
Likelihood
Variance Ratio
Spatial
Correlation Test
(chi-squared[1])
(1)
TOBIT
(2)
TOBIT
2.123
(0.207)**
-2.443
(0.787)**
2.132
(0.206)**
-2.476
(0.784)**
-0.199
(0.466)
-0.606
(0.593)
1.541
(0.716)*
1.756
(0.265)**
-0.085
(0.414)
1.476
(0.422)**
1.423
(1.275)
-1.131
(2.324)
0.018
(0.171)
-0.309
(0.010)**
-
-0.156
(0.467)
-0.570
(0.573)
1.534
(0.717)*
1.748
(0.266)**
-0.165
(0.419)
1.453
(0.421)**
1.457
(1.258)
-0.812
(0.327)**
-0.311
(0.073)**
-0.676
(0.431)
867
202
867
202
1781.899
-
1780.632
-
Note: GDP Sum and Distance are in logs, expect Distance in regression (2). White-corrected robust standard errors are in parentheses.
**, *, # denote statistical significance at 1%, 5% and 10% levels, respectively. Regressions (1) – (4) include source and host country
fixed effects, which are not reported. Regressions (6) – (7) do not include source fixed effects.
27
TABLE 5 – CMM Model OLS Estimation
Dep. Variable
Variables
Sum GDPs
Sq. Diff. GDPs
Absolute
Difference Skills
Abs. Diff. Skills
x Sq. Diff.
GDPs
Tariffs Source
Tariffs Host
Institutions Host
Tariffs Host x
Sq. Skill
Difference
Distance
Time Zone
Difference
Minimum Time
Zone Difference
Office Hours
Overlap
Latitude
Distance
Longitude
Distance
Telephone
Traffic
Constant
Observations
R-squared
(1)
FDI Stock
Levels
0.005
(0.001)**
-1.91e-10
(1.18e-10)
(2)
FDI Stock
Levels
0.005
(0.001)**
-1.91e-10
(1.18e-10)
(3)
FDI Stock
Levels
0.005
(0.001)**
-1.93e-10
(1.18e-10)
(4)
FDI Stock
Levels
0.005
(0.001)**
-1.87e-10
(1.18e-10)
(5)
FDI Stock
Levels
0.005
(0.001)**
-2.03e-10
(1.18e-10)#
(6)
ln (1+FDI)
Logs
2.171
(0.059)**
-0.198
(0.017)**
(7)
ln(1+FDI)
Logs
0.980
(0.156)**
-0.119
(0.021)**
769.925
(236.467)**
-0.0003
(0.0001)**
758.198
(236.443)**
-0.0003
(0.0001)**
751.141
(234.015)**
-0.0003
(0.0001)**
758.940
(237.036)**
-0.0003
(0.0001)**
748.136
(234.856)**
-0.0003
(0.0001)**
0.013
(0.063)
0.042
(0.023)#
-0.127
(0.084)
0.055
(0.027)*
-223.000
(99.428)*
-31.366
(40.526)
1641.797
(371.563)**
-1.352
(1.199)
-235.963
(99.730)*
-28.330
(40.576)
1683.809
(369.148)**
-1.291
(1.194)
-250.085
(100.374)*
-24.634
(40.987)
1675.697
(374.014)**
-1.329
(1.181)
-233.759
(99.747)*
-22.081
(40.481)
1629.129
(373.174)**
-1.439
(1.185)
-224.723
(98.205)*
-21.996
(41.851)
1512.737
(366.776)**
-1.259
(1.205)
-0.461
(0.036)**
0.029
(0.017)#
0.823
(0.132)**
-0.001
(0.0003)**
-0.466
(0.047)**
0.060
(0.017)**
0.561
(0.175)**
-0.001
(0.0003)**
-0.423
(0.090)**
-
-0.065
(0.118)
-
-0.161
(0.145)
-
-
-
-0.165
(0.154)
-262.622
(142.724)#
-
-
-
0.326
(0.165)*
-0.104
(0.049)*
-
-
-
-380.389
(133.813)**
-
-0.361
(0.142)**
-0.129
(0.040)**
-
-
-
-
-
-
-
318.264
(163.809)**
-
-
-
-
-
-
-
-
-
-
-
-
-
0.148
(0.129)
-0.724
(0.140)**
-
-
0.827
(0.088)**
-2187.228
(1588.652)
-2203.413
(1589.491)
-2134.856
(1589.699)
-4638.074
(1816.966)*
-2171.961
(1594.545)
-18.704
(1.363)**
-10.127
(1.989)**
827
0.317
827
0.318
827
0.320
827
0.319
827
0.321
827
0.680
493
0.675
Note: White-corrected robust standard errors are in parentheses. **, *, # denote statistical significance at 1%, 5% and 10% levels, respectively.
28
TABLE 6 – Trade Gravity Model Estimates and Time Zone Differences (OLS)
Minimum Time Zone
Difference
Office Hours Overlap
-
(2)
0.835
(0.032)**
0.319
(0.050)**
0.361
(0.077)**
0.492
(0.124)**
0.996
(0.182)**
0.245
(0.087)**
-0.666
(0.072)**
-0.069
(0.015)**
-
-
-
-0.070
(0.014)**
-
Latitude Distance
-
-
-
0.110
(0.016)**
-
Longitude Distance
-
-
-
-
988
0.932
-
988
0.935
-
988
0.935
-
988
0.936
-
Product of GDPs
Product of GDPs per capita
Common Language
Adjacency
Colonial Links
RTA
Distance
Time Zone Difference
Observations
R-squared
F-Test Latitude = Longitude
(1)
0.825
(0.033)**
0.288
(0.049)**
0.365
(0.081)**
0.303
(0.119)*
0.958
(0.184)**
0.283
(0.087)**
-0.898
(0.045)**
-
(3)
0.834
(0.033)**
0.315
(0.049)**
0.383
(0.076)**
0.477
(0.123)**
1.007
(0.182)**
0.252
(0.088)**
-0.673
(0.069)**
-
(4)
0.849
(0.032)**
0.336
(0.048)**
0.363
(0.074)**
0.562
(0.123)**
0.959
(0.177)**
0.217
(0.088)*
-0.559
(0.073)**
-
(5)
0.777
(0.035)**
0.278
(0.049)**
0.466
(0.086)**
0.683
(0.147)**
0.689
(0.206)**
0.472
(0.104)**
-
(6)
0.810
(0.033)**
0.339
(0.051)**
0.408
(0.078)**
0.842
(0.134)**
0.940
(0.201)**
0.321
(0.096)**
-
-
-
-
-0.139
(0.014)**
-
-
-
-0.167
(0.033)**
-0.430
(0.045)**
988
0.9179
18.91**
-0.141
(0.026)**
-0.147
(0.052)**
988
0.928
0.01
Note: All regressions are OLS estimations and include source and host country dummies, not reported. White-corrected robust
standard errors are in parentheses. **, *, # denote statistical significance at 1%, 5% and 10% levels, respectively. Dependent variable
is the log of exports plus imports between host and source in 1999.
29
Table 7 – Equality Tests of Time Zone Effect over Time
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
1988
1.32
3.57#
9.06**
11.89**
12.51**
15.39**
15.64**
19.8**
22.49**
22.03**
19.92**
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
2.18
7.73**
10.43**
10.98**
13.75**
13.93**
17.92**
20.47**
19.82**
17.79**
5.85*
8.32**
8.77**
11.35**
11.46**
15.15**
17.49**
16.86**
14.75**
2.24
1.68
3.48#
2.93#
5.38*
6.55*
5.26*
3.39#
0.06
1.46
1.21
3.29#
4.38*
3.27#
1.78
1.88
1.28
3.73*
4.95*
3.56#
1.82
0.05
1.82
2.96#
1.76
0.55
4.32*
5.04*
2.28
0.53
1.26
0.19
0.08
0.18
1.02
0.99
Note: Chi-squared (1) tests. **, *, # denote statistical significance at 1%, 5% and 10% levels, respectively. Estimated effects are
interactions of time zone differences with year dummies. The dependent variable is Log (1 + FDI). The econometric model includes
the controls specified in equation (1), source-year and host-year dummies, year dummies. The estimation method is Prais-Winsten to
correct for autocorrelation.
30
3
2
Estimated Dummy Coefficient
1
0
1
2
3
4
5
6
7
8
9
10
11+12
-1
-2
-3
-4
-- 95% confidence interval — point estimate
-5
-6
Time Zone Difference (hours)
Figure 1 – Estimated Dummy Coefficients by Time Zone Difference
0
-0.05
-0.1
-0.15
-0.2
-0.25
1988
1989
1990
1991
1992
1993
1994
1995
OLS cross section
OLS pooled
Prais-Winsten FE - time
95% confidence interval for PW FE-time
1996
1997
Prais-Winsten FE
Figure 2 – Time Zone Difference Effect over Time
31
1998
1999