Microsoft Word apsa paper VII jen doc

Download 133,37 Kb.

Pdf ko'rish

bet	5/8
Sana	30.12.2021
Hajmi	133,37 Kb.
	#93845

1 2 3 4 5 6 7 8

Bog'liq
Tobin and Busch

r

z

d

Where equation (2) includes 1) the number of BITs in force in the host country (minus

the BIT with the partner if it exists), 2) the square of that term in order to understand if

the relationship between BITs and PTAs is in fact nonlinear and 3) the interaction be-

tween the sum of BITs squared and the individual investment treaty to understand if the

effect of an individual country-pair BIT on PTA formation is non-linearly affected by

increased BIT formation more generally.

Variables

As the focus of our analysis is on the formation of a PTA between specific coun-

tries, based on whether that dyad has a BIT in place, our unit of analysis is the country

dyad-year. We analyze only those country dyads where the host is a developing country

and the partner is developed one. We do this for two reasons. First, PTAs signed be-

tween two poorer states typically have different origins, and effects, than those between

other pairings, as the theoretical and empirical literatures on the subject make abundantly

clear (see above). Second, and perhaps more important, BITs between developing coun-

tries are quite different from those signed between a developed and a developing country.

Until the 1990s, developing countries rarely signed investment treaties with each other.

Currently, BITs between two developing countries, while growing in popularity, are often

signed for reasons other than attracting investment flows. Developed countries are ex-

cluded as host countries from the analysis, since wealthy states never sign BITs with each

other.

1

For example, in 2004, the four largest foreign investors in the US were Britain,

France, the Netherlands and Japan, yet the United States does not have (nor does it plan

to negotiate) a BIT with any of these countries. In short, states with confidence in each

other’s investment environments do not conclude BITs, and our goal is to understand

how efforts to increase this level of confidence, through an investment treaty, might set

the stage for subsequently signing a PTA.

Our dependent variable, PTA, is coded 1 in the year that a trade agreement be-

tween dyad partners went into force and each subsequent year, and 0 otherwise. New

countries entering an existing PTA are scored 1 in the year that the host country enters

the PTA, and 0 in prior years. We include all types of PTAs (for example, free trade

agreements, customs unions, common markets and economic unions) that were notified

to the World Trade Organization (WTO) or made publicly available by the partner states.

We measure BITs in two ways. First, our BIT variable is measured 1 in the year

that an investment treaty between dyad partners takes effect and each subsequent year,

and 0 otherwise. Second, our sum of BITs variable is equal to the count of BITs in effect

One prominent exception to this rule is NAFTA which includes both the United States and Canada, and

contains a BIT as part of its chapter on investment.

between the host country and all OECD countries in each time period, minus the BIT be-

tween the dyad partner, if one exists. Any PTA between a dyad that includes an invest-

ment chapter equivalent to an investment treaty is also counted as a BIT.

To assess the relationship between BITs and PTAs, we must also control for other

factors that are likely to lead to trade agreements between a developing and a developed

country. Since our unit of analysis is the country-pair, we control for factors that are both

dyad-specific and host-specific. Krugman (1991) and Frankel et al. (1995) show that

greater distance between countries indicates increased transportation, transaction, and

contracting costs, and thus should lower the likelihood of trade flows and, as a result, a

PTA. Thus, the greater the geographic distance between countries, the lower the prob-

ability that the country pair will form a PTA. The variable DISTANCE measures the

geographic distance between dyad partners’ capital cities, following the great circle for-

mula, as per Mayer and Zignago (2006), which uses the relevant latitudes and longitudes.

A number of authors have speculated that PTAs may be as much a response to

trade as a source of it (Lawrence 1998). We account for this by including a measure of

trade flows between country pairs, the idea being that the greater this volume of com-

merce, the higher the probability that the dyad will form a PTA. TRADE measures ex-

ports flowing from the host country to the partner country in millions of constant US dol-

lars, lagged two years.

The data are from Gleditsch (2002). It is a compilation of the

IMF annual time series and the World Export Data (Faber and Nierop 1989).

Baier and Bergstrand (2004) find that PTAs are more likely between country-

pairs when those countries have the economic characteristics that theory predicts would

We also ran our estimations with exports from the partner country to the host country as well as the sum

of exports in both directions—neither resulted in significant changes in our results.

enhance trade between those countries. Specifically, countries with larger and more simi-

lar economies and factor endowments are more likely to enter into PTAs. Countries with

large domestic markets have less need to access foreign markets, and similar countries

tend to trade more with each other. We account for these considerations with a measure

of the size and purchasing strength of the economy, GDP per capita, and measures for

differences in the size of the economy and available factor endowments. The variable

SKILL is the difference between the percentage of the labor force in the partner and host

country that have a tertiary education (available from World Development Indicators).

INCOME, a measure of the difference in the size of the dyad partners’ economies, is

measured as the difference in GDP per capita between the partner and host country two

years prior (available from World Development Indicators).

Finally, in terms of dyad characteristics, Mansfield (1993) and Gowa (1994) theo-

rize and find that countries involved in alliances are more likely to form PTAs because

they can better “internalize” the security externalities of trade. Thus, dyads involved in a

political-military alliance are expected to be more likely to form PTAs. The variable

ALLIANCE is from the Correlates of War Formal Alliance data (Gibler and Sarkees

2004). It identifies all formal alliances between dyads in each year of the dataset.

In addition to these dyadic factors, several country-level characteristics help to de-

termine the formation of PTAs. Mansfield et al. (2002) show that as countries become

more democratic, leaders gain from forming trade agreements. Specifically, PTAs offer a

credible signal for political leaders to demonstrate their policy choices over trade. Thus,

we would expect that as a country becomes more democratic, its probability of forming a

PTA increases. As a measure of democracy, we use the ratings from the Polity IV project

two years prior to the current period (Marshall and Jaggers 2004). This measure is the

difference between a country’s democracy and autocracy score, both on a ten point scale,

so that the resulting variable ranges from -10 to 10

. POPULATION is the log of total

population measured in the middle of the year, available from the World Development

Indicators.

Similarly, Mansfield and Reinhardt (2003) show that countries that are members

of the WTO are more likely to enter into PTAs. The authors argue that the uncertain na-

ture of the multilateral negotiations drives countries into smaller and regional trading

blocs in order to improve their bargaining position in trade rounds, and to protect them-

selves from disruptions and disputes. Thus, we would expect that host country members

of the WTO/GATT are more likely to form PTAs. WTO equals one in the year of acces-

sion to the GATT or WTO and each subsequent year, and 0 otherwise. Membership data

are available from the WTO

.

Finally, AFRICA, LAC, APAC are regional dummies for Africa, Latin America

and the Caribbean, and Asia-Pacific generally, with Eastern and Central Europe serving

as the excluded category.

Our data cover a balanced panel of 132 low- and middle- income host countries

and 23 OECD partner countries from 1960 to 2004. To test our models of PTA forma-

tion, we use a logistic regression including robust standard errors clustered by country so

that multiple observations for each country are deemed to be independent. Four impor-

tant issues exist within our dataset that must be dealt with in order to estimate the model

Following imputation the polity variable ranges from -28 to 26.

http://www.wto.org/english/thewto_e/whatis_e/tif_e/org6_e.htm

using a logistic regression technique: serial correlation, rare events, endogeneity, and a

high degree of missing data. We deal with each of these issues in turn.

First, it is clear from the issues that we are studying that our dataset exhibits tem-

poral dependence. Thus, running a standard logistic regression analysis would result in at

least incorrect estimations of our standard errors, and at worst biased estimates of our co-

efficients. To correct for the possibility of autocorrelation, we follow Beck et al. (1998)

and include a natural spline function

—of the number of years from the beginning of our

data coverage—to test and account for the possibility of temporal dependence

. Tests of

time dependence proved an issue, and thus we include a natural spline with 5 knots in

each of our estimations (these results are not reported to conserve space).

Second, King and Zeng (2001) demonstrate that finite sample bias exists when the

number of events being analyzed is small or unbalanced. Our dataset is not small, but it

is unbalanced, since far fewer dyads have entered into PTAs than have not. In our analy-

sis, the full dataset includes fewer then 2 percent of cases where our dependent variable is

equal to 1. To account for this bias, we use a rare events correction for our logistic re-

gression.

Third, there exists the possibility that BITs are endogenously determined. That is,

the existence of a BIT may be influenced by factors similar to that influencing the forma-

We use Newson’s (1994) bspline program to calculate natural splines and knots.

Beck et al. (1998) show that problems stemming from temporal dependence can be solved by including

dummy variables for each time period in the data set. These time variables would be equivalent to a base-

line hazard—or the probability for a respective country when all independent variable values are equal to

zero. One problem with including each year dummy is that the baseline hazard jumps around from year to

year—although we would expect changes over time to be smooth, rather than jagged. To smooth the base-

line hazard function, we employ natural splines which are basically smooth functions of our time dummies.

Employing these splines ameliorates any issues of time dependence in our model, while maintaining a

smooth baseline hazard rate.

Results do not vary based on the inclusion of a natural spline versus the inclusion of year to year dummy

variables.

tion of a PTA. Unfortunately, good instruments for BITs are elusive. Thus, similar to

the literature on BITs we account for the possibility of endogeneity by lagging our BIT

variable by 2-years. Lagging the BIT variable controls for endogenous BIT formation if

past BITs are determined by different factors than are current PTAs.

Data on many variables were unavailable for some country years. If we use list-

wise deletion to remove all observations with missing data, our sample size would be re-

duced by more than 100 countries, most of which fall into the lowest-income category.

Any inferences made from an analysis using list-wise deletion would have been ineffi-

cient. More important, because the missing values are not missing at random, our infer-

ences would be biased. The data generation process for our dataset falls into the category

of missing at random (MAR). That is, although the lowest income countries are more

likely to have missing data than higher income countries, we are able to predict this dif-

ference from other variables in our dataset. Therefore, deletion could introduce substan-

tial bias into our study (King et al. 2001). To deal with this, we use the Amelia (Honaker

et al. 2003) program for multiple imputation to replace missing data with the best predic-

tions of the data.

Amelia uses the known values of certain variables and correlations

across independent variables to generate five datasets with unique values for missing ob-

servations (King et al. 2001). To reflect the level of uncertainty of the estimated values,

we then use the program Clarify (Tomz et al. 2003) to combine the estimated results from

each of the five estimations.

Finally, we performed a number of tests to be sure that our results were not un-

duly affected by any outlying observations, and eliminated any small island nations with

In future versions of the paper we also plan to instrument for BITs and run the analsyis in two-stages.

Listwise deletion estimates are available from the authors. Results did not change substantially, but are

not included because of potential bias.

population below 1 million. This leaves us with data on 132 countries over the period

1960-2004. A complete list of countries included in our study is in appendix B.

IV. Results

In the results that follow, we find considerable support for all three of our hy-

potheses. Specifically, we find that a BIT between a host and partner country increases

the predicted probability that the dyad will form a PTA. We further find that as a country

enters into more BITs with developed countries, its predicted probability of forming a

PTA with a partner increases. However, both of these positive findings are contingent on

the number of BITs that a host country has overall. That is, as a developing country en-

ters into greater numbers of BITs generally, the positive impact on PTA formation of

having an individual BIT with a partner country, and having more BITs overall, actually

falls.

Table 1 reports the results of equations (1)–(2). In column (1) of Table 1, we see

the effect of a simplified model where we only examine how a dyadic BIT affects the

probability of PTA formation. We find that a BIT between a dyad has a positive effect

on the probability of signing a PTA, but the effect is not significantly different from 0.

The positive effect of a dyadic BIT on PTA formation remains positive throughout our

specifications and becomes significant when we include the number of BITs that a coun-

try has signed with other developed countries. In other words, the relationship between a

dyadic BIT and PTA formation is highly dependent on the number of BITs that a host

country has more generally.

In column (2), the individual BIT between the dyad has a positive impact on the

probability of forming a BIT, but the sum of BITs with other developed countries is nega-

tive (though insignificant). This is suggestive of our first hypothesis that, if a developing

country has a BIT with a developed one, this pair is more likely to sign a PTA. Column

(3) probes our hypotheses that the dyadic BIT and sum of BITs are contingently related

to PTA formation. In column (3), the dyadic BIT has a strong, positive influence, while

the sum of BITs with wealthy states is positive but insignificant (though it is jointly sig-

nificant with its interaction term). This endorses our second hypothesis that, if a develop-

ing country has BITs with other developed countries, this pair is more likely to sign a

PTA.

From the coefficient estimate on the square of the sum of BITs term, however, we

also see that this positive effect is not linear. That is, initially entering into BITs with

other developed countries has a positive effect on the formation of a North-South PTA.

However, after about 5 BITs, this positive effect begins to decline. (See Appendix C for

a list of each of our included countries and the number of BITs they have with developed

countries.) The average number of treaties for our sample is 6.5, and many have more.

Indeed, countries like Uganda, Algeria, and Rwanda are nearing the point where signing

additional BITs will actually reduce their probability of forming additional PTAs with

other developed countries. More worrisome still, a great many countries in our sample

are already well past that point.

Returning to column (3), the negative coefficient estimate on the interaction term

(

BIT-Dyad* BITs-SUM

) lends support to hypothesis 3, that the individual positive effect

of a country-pair BIT is weakened as a country enters into more BITs with other devel-

oped countries. Although the probability of forming a PTA between a dyad-pair will al-

most always remain positive, the positive effect diminishes considerably as a country en-

ters into greater numbers of BITs with developed countries aside from its PTA partner.

To better understand the magnitude of our effects, we derive the predicted prob-

ability of a dyad forming a PTA at each level of the sum of other developed country

BITs, holding all other variables in the model constant. Figure 2 uses the results of col-

umn (3), and holds all countries at the average values of their samples to show the pre-

dicted probability results of the interaction term graphically. Here, the predicted prob-

ability of PTA formation is always positive, but it decreases drastically as a country rati-

fies more investment treaties. This is a graphical demonstration of our third hypothesis

that, as a developing country accumulates a greater number of BITs with other developed

countries, it will be less likely to sign a dyadic PTA.

We are also interested in how these probabilities change from the benchmark case

in which a developing country has no BITs at all. Table 2 takes this up, comparing this

benchmark case to a country with 1 BIT, and to a country with 7, which is roughly the

average in our sample. With no other investment treaties in place, the predicted probabil-

ity of forming a PTA increases from 0.10 to 0.44 when a given dyad enters into a BIT.

This probability decreases, however, as a country enters into BITs with other developing

countries, falling by over half to 0.21 in the case of 7 BITs, our sample average.

To further tease out these substantive effects, we randomly selected a country

from each quartile of BITs in the sample, and used their current characteristics to predict

the probability of them forming a PTA with a developed country. Table 3 lists each of

the four countries, and their odds of getting a trade agreement with and without a BIT in

place. For Colombia, Chad and Thailand, the probability of forming a PTA is higher

when these countries anchor a dyad that has a BIT, versus no BIT. However, for Bangla-

desh, the country with the highest number of BITs, the likelihood of signing a PTA actu-

ally decreases when it participates in a dyad that has a BIT, as opposed to no BIT. These

results lend further to support to all three of our hypotheses. When a country has a BIT

in place with a partner country, the predicted probability of forming a PTA is always

positive. However, as we move into quartiles with greater numbers of BITs, the differ-

ence in predicted probabilities shrinks, and in the highest quartile, at least for our exam-

ple, Bangladesh has a higher predicted probability of forming a PTA without a BIT with

a partner-country than with one.

It is also interesting to note the direction and significance of the control variables

included in our models. Although coefficient estimates from logistic regressions are not

highly informative, we are able to get some idea of the relationship and degree of signifi-

cance of these variables on the formation of a PTA. The level of trade between the host

and partner country has the expected positive impact on the formation of a PTA, though it

is not statistically significant in any of the equations. Geographic distance is, as ex-

pected, negatively signed; as countries are farther apart geographically, the probability of

the dyad forming a PTA goes down. Distance is significant throughout all of our specifi-

cations. Similarly, income difference is significant and negative, but its impact on PTA

formation is minor. The level of skill difference within the dyad has a positive sign,

which is surprising, as we would expect that, as the skill difference between countries

grows, the probability of a PTA decreases. However, this variable is not significant in

any of our models, so the result does not merit much scrutiny. A political-military alli-

ance between the dyad has the predicted positive sign and becomes minimally significant,

but only once we add our BIT interaction terms.

Turning to host country characteristics, we see that the level of democracy actu-

ally has a negative impact on PTA formation. That is, as a country becomes more de-

mocratic, its likelihood of forming a PTA actually decreases. Again, this variable is

never significant in our estimations, perhaps because of the way our sample is con-

structed (i.e., we only examine developing countries as hosts). Similarly, membership in

the GATT/WTO has the expected positive sign, but is never significant in our estima-

tions. Finally, population is positive and significant throughout our estimations, indicat-

ing that the larger the host country market, the more likely a dyad is to form a PTA.

We ran a number of checks on the robustness of our results. First, we eliminated

all variables that were not significant in the final estimation of the model. Column (4) in

Table 5 presents the results. Here, eliminating theoretically important but statistically

insignificant variables had no discernable impact on our results. Second, we created a

“shortened” dataset by removing any dyad from the sample as soon as it entered a PTA.

Column 6 reveals the robustness of our results to this test.

Next, we included on the right hand side of the equation two variables to account

for any PTAs that the host country has with other developed countries. After all, if our

congestion argument about having too many BITs is reasonable, the same logic should

hold with respect to PTAs as well, and thus we should test for a curvilinear relationship

here as well. We thus include a variable equal to the sum of the host country’s PTAs in

place with other developed economies, and the square of this variable.

Column (6) pre-

Mansfield and Reinhardt (2003) test for the formation of a PTA within GATT/WTO members by squar-

ing a measure of the density of PTAs. They theorize that as members of GATT/WTO form PTAs, those

sents the results of this specification. Not only are our earlier results on BITs robust to

this, but the coefficient estimates on PTAs are interesting in and of themselves. To be

sure, the sum of PTAs variable has a positive, significant impact on dyadic PTA forma-

tion, but the square is significant and negative. This says that forming additional PTAs

has a positive impact up to a point—precisely nine—after which the impact declines. In

short, our argument about congestion is just as relevant to trade agreements as it is to in-

vestment treaties, the implication being that more of either type of institution is not better

in explaining the formation of PTAs.

Finally, we ran two additional tests whose results we do not include because of

their similarity to the results in column (3) of Table 1 and to conserve space. We re-ran

equation (2) including the partner country characteristics that were originally included for

the host country. The results were insignificantly different from those shown in column

(3) of Table 1. Further, because of the high degree of collinearity between the BIT vari-

able and the interaction term as well as the BIT sum variable, its square and the interac-

tion term, we re-ran equation (2) using mean-centered versions of the interaction term

and BIT sum square term. Again, the results—while slightly moderated—were not sig-

nificantly different from the results presented in column (3) of Table 1.

V. Conclusion

Developing countries looking to embrace more investment and trade have several

tools at their disposal, BITs and PTAs being among the most notable in this regard. Pol-

icy-makers and academics have generally argued that, with respect to both of these in-

left uncovered will be pressured to enter trade blocs. However, after enough dyads enter PTAs, few will be

left to join new trade blocs and thus the relationship will begin to decline.

struments, more is better. After all, the logic goes, if a developing country wants to ad-

vertise that it is “open for business,” what better way to do this than by joining as many

institutions that tie its hands as possible. We show that this line of thinking needs to be

substantially revised.

For developing countries, having a BIT in place with a rich state does, in fact, im-

prove the odds of getting a trade agreement with this country. In this sense, investment

treaties are not just cheap talk, but in fact credible mechanisms that can prime the pump

for the deeper, and often reciprocal, obligations required to sign trade agreements.

However, having more BITs with other partners actually erodes that effect. Ac-

cording to our results, the impact of a BIT with a partner on forming a PTA is conditional

on the number of BITs that a host country has with others. In our study, the tipping point

is five BITs, which is roughly two fewer than the average developing country has ratified.

Put another way, many poorer states are making it more difficult for themselves to secure

trade agreements with richer ones by signing on to additional investment treaties. The

implication is that BITs should be more selectively targeted at those countries from

which poorer states might wish to solicit a PTA.

More generally, our results speak to the concern that overlapping institutions can

complicate international affairs, and that too many institutions may be worse in this re-

gard than none at all (Busch 2007). Students of regionalism bemoan the ever-expanding

“spaghetti bowl” of PTAs, and the myriad obligations that can come into conflict across

these pacts. Likewise, critics of BITs lament that the relative ease with which they are

negotiated may compromise the signal sent by their ratification. Our results shed new

light on both of these worries: having many BITs can undermine the likelihood of getting

a PTA, just as having multiple trade agreements can have this effect. The competition for

footloose capital is undoubtedly seeding this proliferation of BITs, but the message of our

paper is that, at least with respect to signing PTAs, a BIT may be better than a lot.

Bagwell, Kyle, and Robert W. Staiger. 2001. "Reciprocity, Non-Discrimination, and

Preferential Agreements in the Multilateral Trading System." European Journal

Download 133,37 Kb.

Do'stlaringiz bilan baham:

1 2 3 4 5 6 7 8