WPTC 04-04

I '-ional Agricultural Trade and Policy Center

FOREIGN DIRECT INVESTMENT AND EDUCATION

INVESTMENT IN DEVELOPING COUNTRIES

By

Nobuyuki Iwai, Stanley R. Thompson, Priyodorshi Banerjee

WPTC 04-04 October 2004

WORKING PAPER SERIES

'V

i~fr

UNIVERSITY OF

FLORIDA

Institute of Food and Agricultural Sciences

INTERNATIONAL AGRICULTURAL TRADE AND POLICY CENTER

THE INTERNATIONAL AGRICULTURAL TRADE AND POLICY CENTER

(IATPC)

The International Agricultural Trade and Policy Center (IATPC) was established in 1990

in the Institute of Food and Agriculture Sciences (IFAS) at the University of Florida

(UF). The mission of the Center is to conduct a multi-disciplinary research, education and

outreach program with a major focus on issues that influence competitiveness of specialty

crop agriculture in support of consumers, industry, resource owners and policy makers.

The Center facilitates collaborative research, education and outreach programs across

colleges of the university, with other universities and with state, national and

international organizations. The Center's objectives are to:

* Serve as the University-wide focal point for research on international trade,

domestic and foreign legal and policy issues influencing specialty crop agriculture.

* Support initiatives that enable a better understanding of state, U.S. and international

policy issues impacting the competitiveness of specialty crops locally, nationally,

and internationally.

* Serve as a nation-wide resource for research on public policy issues concerning

specialty crops.

* Disseminate research results to, and interact with, policymakers; research, business,

industry, and resource groups; and state, federal, and international agencies to

facilitate the policy debate on specialty crop issues.

Foreign Direct Investment and Education Investment in Developing Countries

By

Nobuyuki Iwai, Stanley R. Thompson, and Priyodorshi Banerjee*

We introduce a model to explain the economic rationale for the observed policy

combination of a developing country (hosting foreign direct investment (FDI) through

education investment (EDI)) and the interest of a multinational corporation (MNC) in the

local labor quality when it contemplates FDI. Information on local labor is the source of a

more efficient contract for the MNC with local labor, and the local government can

benefit both agents through EDI, FDI, and information sharing. This strategy set is likely

to be used by a country in the early stage of economic development. The education level

chosen by the local government, however, will be higher than that which maximizes the

welfare of local labor. In that sense, the government has the incentive to benefit itself and

the MNC at the expense of local labor.

SIwai is at the University of Florida; Thompson and Banerjee, are at The Ohio State University.

Foreign Direst Investment and Educational Investment in Developing Countries

1 Introduction

The role of informational asymmetries occupies a central position in the theory of

foreign direct investment (FDI). The existing literature has focused on the choice

between FDI and licensing to a local firm in an environment where a multinational

corporation (MNC) sells in the host-country (local) market (Ethier 1986, Ethier and

Markusen 1996, and Horstmann and Markusen 1996). These studies have focused on the

advantage of internalizing information about technology and local demand. When

considering FDI in developing countries, however, the local market is often of limited

importance. MNCs largely use these production centers as an export base to

industrialized-country markets, as evidenced by high export-to-local-sales ratios (see the

evidence below and also Hayami 2001 and McMilan et al. 1999). As such, two additional

considerations absent from the literature are important: information about local labor

quality and the role of governments (local or national).1

A compelling reason for FDI by an MNC is the availability of relatively cheap skilled

labor, making production operations within a developing country profitable for the MNC

(Wakasugi 1996). The payoff from investing in a developing country is thus dependent

on the quality of the local labor force. Since an MNC typically has imperfect knowledge

about the local labor quality, the MNC can gain if the local government shares

information about its labor quality.

This paper provides an economic rationale for the observed policy combination

whereby developing-country governments host FDI by investing in education and sharing

1 This is true, especially for FDI in Southeast Asian countries. See Collins and Bosworth (1996), and

Bloom et al. (1998) for evidence on East and Southeast Asian countries' outward oriented economic policy

and promotion of education.

information (OECD 20022, United Nations 1999, McMilan et al. 1999). In our

benchmark model, the MNC makes FDI in a developing country and offers contracts to

local laborers. Precise information on local labor quality will lead to a more efficient

contract, which is the source of information rent. Since the local government has superior

information about the quality of the labor force, it exploits this advantage by sharing

information with the MNC and taxing its profits. Furthermore, the local government has

an incentive to use these tax revenues for education investment (EDI) because this raises

the value of information and, thus, the profits of the MNC. The point at which the

developing-country government starts investing in education we term the "take-off

point. The behavior of the take-off point is an important contribution of our work.

This paper adds to the literature on the relationship between FDI and local

government policy (Horstmann and Markusen 1987, Glass and Saggi 1999, Glass and

Saggi 2002). Glass and Saggi (2002) analyze direct subsidies to an MNC by a

developing-country government. In their model, the MNC possesses a superior

technology compared to the local firms. The local firms can acquire the superior

technology by hiring away the MNC workers, but the MNC can prevent this by paying a

wage premium. Then the welfare-maximizing government may induce FDI through the

direct subsidy aiming either for technology transfer or wage increases. In Glass and Saggi

(1999), the impact of government policy is studied in a model which examines how FDI

changes the distribution of wages and profits between host and source country with

oligopolisitic industries. Horstmann and Markusen (1987) study whether an MNC

interested in selling to a foreign market prefers to export, license production or invest

2 OECD (2002) emphasizes importance of the role of local government by concluding "The major impact

of FDI on human capital appears to have occurred not so much through the efforts of individual MNEs as

from government policies designed to attract FDI via enhanced human capital." (p.122)

directly in a reputation based model. The local government has the incentive to induce

licensing production because it leads to higher consumer surplus and licensee profit

which otherwise accrues to the MNC.

None of these papers address the possibility that the host country government can

have private information which is valuable to the MNC or study the information sharing

and education investment policies. Further, by making the government one of the players

which may either choose to maximize its own net revenue or the welfare of the local

labor,3 we more closely reflect reality in many developing countries, in that the

government may place a higher priority on its net revenue over the welfare of local labor.

Above all, we are the first to model the relationships among FDI, EDI, and the rents

accruing from labor quality information.

The rest of the paper is as follows. In section 2, evidence is presented on the

relationship between FDI and EDI in developing countries. We present some evidence

showing that, relative to the local output market, the local labor market is more important

for FDI in developing countries. Section 3 presents the basic structure of the model. In

section 4, the MNC decision problem is analyzed. In section 5, we investigate the policy

of a local government that chooses a level of education and a tax rate to maximize its net

revenue. In section 6, the welfare effect of the policy is illustrated. We show that the local

government policy can benefit both itself and the MNC at the expense of local labor

welfare. We also study how government policy will change when the welfare of local

labor is considered. Finally, conclusions follow in section 7.

3 In section 6, it is shown that the policy which maximizes net revenue and that which maximizes welfare

of local labor, is quite different in our model. We further study the case in which the government

maximizes the sum of these.

2 Evidence

A positive relationship between FDI and EDI in developing countries is shown in

Figure 1. Here we depict the average annual growth rate of FDI inflow and that of public

spending on education between 1980 and 2000 for 20 countries.4 The correlation

coefficient between the two annual growth rates is 0.45, one-sided p-value less than

0.025.

15

Banglades

Brazil Philippines

10 Jamaica* Bolivia Thailand

D0 Ivory Argentina

Coast

5 Pakistan

Costa *Trinidad *

osta Paraguay

Rica Kenya *Tunisia

O Malaysia

0 Togo Panama

Guatemala

5 Iran

LL -5 -

Niger *

-10 I I I I

-4 -2 0 2 4 6 8

Public Spending on Education (% GDP)

Figure 1. The average annual growth rate of FDI inflow and public spending on

education (both as a percent of GDP).

Although not shown, we calculated the average growth rate of secondary school

enrollment ratio for the same 20 countries over the same period. The correlation

coefficient with this measure of education investment and FDI is 0.33, one-sided p-value

less than 0.10. Again, higher EDI tends to host more FDI.

4 Per capital GDP in all 20 countries was less than $10,000 in year 2000. FDI inflow data were from

UNCTAD (l '14). All other data are taken from World Bank (2003).

For many industries, the importance of the local labor market relative to the local

output market is dramatic. This is typical in East Asia. In 1995, for instance, 27.4 percent

of laborers in the Japanese electrical products industry lived abroad, while for the

automobile and nonferrous metal industries these numbers were 32.3 and 36.5 percent,

respectively. In turn, FDI can be very important for the local labor market. The share of

employment within the electrical products industry held by Japanese firms was 38.8

percent in Singapore, 37.3 percent in Indonesia, 38.0 percent in the Philippines, and 28.2

percent in Malaysia. For the automobile industry the employment shares were 34.3

percent in Indonesia and 30.9 percent in Malaysia.5

The data above show that the local labor market is of considerable importance in the

context of FDI in developing countries. What about the local-goods market as a source of

revenue for the MNC? For Japanese firms within the ASEAN 4 (Indonesia, Thailand, the

Philippines and Malaysia), the proportion of exports in total sales in 1996 was 79.0

percent for the electrical products industry, 57.6 percent for the precision machinery

industry, and 54.0 percent for the textile industry (Fukao and Amano 1998). For all

industries, 18.1 percent of the total exports from the Philippines in 1996 was by Japanese

firms. For Malaysia and Thailand, total exports by Japanese firms were 17.7 and 17.3

percent, respectively (MITI 1999). These data suggest that the local output market is a

less important revenue source than the export market.

3 Model Description

Information about the quality of local labor plays an important role in our model. The

local government has superior information on local labor quality (composition of the

5 All of these are data for 1995, cited from Fukao and Amano (1998).

skilled labor) over the MNC seeking to set up a production site in a developing country.6

Because higher tax revenue flows from higher MNC profits, the local government takes

advantage of the MNC's interest in labor quality information by investing in education

and revealing it (in another context, the MNC has incentive to inquire about the

information). This strategy set (FDI, EDI, and information sharing) makes both agents at

least as well off as without it. However, the education level chosen by the local

government will be higher than that which maximizes the welfare of local labor. Hence,

the local government would increase its own net revenue and the MNC's profit at the

expense of local labor.

The sequence of actions in the basic model is as follows. First, the MNC makes its

decision to enter based on the probability distribution of labor quality. Without other

information it cannot compute the probability of positive EDI, so it enters if expected

profit is non-negative. Upon entrance, the local government chooses the level of

education, and reveals the information on education and tax rates. Then, the MNC

implements a principal-agent contract with local labor. Finally, the local government

taxes the profit of the MNC. This sequence is shown as Figure 2.

I I I I I >

MNC Government Discloses Principal Taxes

Decides Chooses Information on Agent Profit

Entrance Level of Education Contract

Education and Tax Rates n ith Local

and Tax Rates Labor

Figure 2. Sequence of actions in the basic model

6 We assume that the MNC does not have the option to license the local firm due to the high merit of

internalizing technological information.

The local government can contact the potential entrant but will incur costs of search

and access. The local government avoids this extra cost if the MNC enters the region

without this "before-entrance-contact." However, if the expected MNC profit is negative

but the local government still can profitably host FDI, the government will contact the

potential entrant even with the extra cost. Then the sequence of actions is changed as in

Figure 3; we call this case "FDIinducingEDL "

I I I I I >

Government Discloses MNC Principal Taxes

Chooses Information on Decides Agent Profit

Level of Education Entrance Contract

Education and Tax Rates i ilh Local

and Tax Rates Labor

Figure 3. Sequence of actions when the local government engages in "FDI inducing

EDI. "

4 MNC Decisions

4.1 Principal-Agent Contract

Consider an MNC setting up a factory in a developing country and implementing a

principal-agent contract with local labor. We first analyze the case with precise

information of labor quality, and then the case with only the probability distribution of

labor quality.

The firm sells the product competitively as a price taker in the global market; thus, it

can sell as much of the product as it wants at p = 1 (price is normalized to one). The firm

is owned by shareholders with globally diversified portfolios; the global capital market

affords complete hedge against the risk of operation at this developing country site. This

assumption results in risk-neutrality of the firm with respect to profit from this site.7

Within the developing country, the labor supply for the specific industry is fixed (also

normalized to one).8

There are two types of labor (skilled and unskilled). We normalize the reservation

utilities for each type of labor to zero. While a laborer knows his own type, outsiders

cannot distinguish an individual's laborer type.9 We can imagine the proportion of skilled

labor P (0< P < 1) to be the indicator of local labor quality. We assume that P is

increasing in the level of education. Thus P can also be understood as a measure of the

level of education.

In keeping with the terms skilled and unskilled, we assume that skilled labor suffers a

lower disutility of effort. If the MNC knew the real value of P, it would implement a

screening contract. The principal-agent problem is given by

max P(x w) + (1- P)(x, w)- F

x1,x2,W1,W2

s.t. w c,(x,)> 0

w, c, (x2) > 0

w, c (x,) > w2 c, (x,)

w2 c (x2) > w1 c (xI),

where, i= type of labor (1 if skilled, 2 if unskilled), x, > 0 is the production level of

labor type i, w, is the wage of labor type i, c, is the disutility from working for labor type

i, and F>0 is a fixed setup cost for the MNC. The constraints are standard; the first two

are the participation constraints guaranteeing each type of labor their reservation utility,

7 Ethier (1985) used a similar assumption to have risk-neutrality of a firm. In the following adverse

selection type setting, no other agent faces a risk.

8 Alternatively we can assume that there is an unpaid local resource whose volume is fixed. We can also

assume that an internal resource, such as managers sent from the source country, is limited.

9 We assume laborers cannot engage in ex-ante signaling.

and the next two are the incentive compatibility constraints ensuring that a laborer of type

i prefers the contract (x,, w,). We impose the following assumption:

Assumption 1: c (x) is continuous and twice differentiable for x > 0. c (0) = 0, c'(0) = 0,

and for x > 0, 0 < c,(x) < c2(x), 0 < cA'(x) < c2'(x) and 0 < c,"(x) < c2"(x) 10

Solving this problem for a given P leads to the first-order conditions (Mas-Colell et

al. 1995),

w = c1(x) + 2 -c1(x2), (1)

2 2 (x2), (2)

c,'(x) =1, (3)

c2'(2) = 1- P + Pc'(2). (4)

The MNC's problem yields the optimal strategy vector (x1(P), w,(P), x2(P),

2(P) ).11 Let rT(P!) denote optimal profit for the MNC when P P. Further we define

a,(P)=x,(P)-w,(P), and b(P) =x(P)-w,(P)-(x2(P)-w,(P)). a,(P) is the profit

from a contract with a laborer of type i, and b(P) is the extra profit earned from a

contract with a skilled laborer over that with an unskilled laborer. From equations (1)

10 These are standard conditions; disutility is increasing and convex in effort, and unskilled labor suffers a

larger disutility than skilled labor. The assumption that c, (0) = c '(0) = 0 is unnecessary but simplifies

calculations.

1 Equations (1) through (4) express the solutions even for P=0 and P= cases. When P=0, they yield

' (x2) = 1, W2 = C2 (x2). When P= 1, they yield c '(x,) = 1, w, = c, (x,). These are the identical

solutions as in the principal-agent contract with one kind of labor. From Assumption 1, w, and x, are

continuous.

through (4) and Assumption 1, we have a2(P)> 0, a2'(P)<0, b(P) > 0 and b'(P) 0

(formal proofs are given in Appendix I). Furthermore, we have the following lemma:

Lemma 1: H*(P) is increasing and convex in P .

Proof: See Appendix II.

Next, we investigate the case in which MNC does not know the exact value of P.

However, the distribution is common knowledge. Let f(P) be the probability density

function of P, and let P (0 < P < 1) be the mean of the random variable.12 Knowing this

probability distribution, the MNC seeks to design the contract to maximize expected

profit. It is easy to show that the optimal contract involves the same first-order conditions

(equations (1) through (4)), but uses P instead of P. This means that the MNC sets the

target value to P.

Given the target value, we denote (x,(P), w, (P), x2 (P) ,w2(P)) as the strategy vector

chosen by the firm. This strategy successfully distinguishes skilled from unskilled labor

regardless of the real value of P, implying that profits earned from skilled labor and from

unskilled labor are fixed. The MNC can compute its profit from this contract for each real

value of P. Profit from this strategy can be expressed as

H(P, P, F)= P[x (P) w,(P)]+(- P)[x2(P) w2(P)]- F

(5)

=a2 (P) + b(P)P F, (5)

12 One way of thinking about this is that the MNC is planning to enter a province or state in a developing

country. While it may know the probability distribution of the labor quality for the entire country, it is

uncertain about the exact labor quality for a particular province.

where a2(-) and b(-) are previously defined. Equation (5) gives the reservation profit of

the MNC when it does not know the real value of P. Since b(P)> 0, the reservation

profit is increasing and linear in P (see Figure 4).13 If the firm knows the real value ofP,

the optimal profit is

H (P,F) = a(P)+ b(P)P- F. (6)

The gain to the firm from knowing the real value of P is therefore the difference between

equations (6) and (5). We call this difference information rent and denote it as R(P, P) .

R(P, P) = [a2(P) + b(P)P] [a2(P) + b(P)P]. (7)

From Lemma 1, optimal profit H*(P,F) is increasing and convex in P. Reservation

profit I(P,P,F) is increasing and linear in P. These two are equal at P =P, which

means that, if the expected value P is equal to the real value P, information rent is zero

and, there is no loss of efficiency. At any other point H*(P,F) > I(P,P,F) Reservation

profit is shown in Figure 4 as the tangency line at P to the optimal profit. The vertical

distance between the two is information rent which is shown in Figure 5.14 Figure 5 also

shows that information rent weakly increases for any P larger than, or equal to, P as the

expected value P decreases. Formally, we have

LR(P,P)

Lemma 2: -- > 0 for P >P. (8)

OP

Proof: See Appendix III.

13 a2(P) F is the intercept and b(P) is the slope of reservation profit in Figure 4.

14 In Figure 4, information rent is larger on the right side of P than on the left side. But this is just because

of the way the figure is drawn. The relative size could be opposite, and it does not affect our analysis.

optimal

profit (P)

reservation

profit (P,P)

O P

Figure 4. Optimal profit and reservation profit.

information

Irent

R(P,P)

O PP 1

Figure 5. Information rent. Decrease in P weakly increases information rent for any

level of P larger than, or equal to, P .

A downward shift of P shifts the point of tangency between Yl*(P,F)and

YI(P,P,F) to the left in Figure 4. This shift weakly increases the information rent at any

point to the right of, or on, P.

4.2 Entrance Decision

The entry point of the MNC can now be defined, given that the real level of local

labor quality (P) is unknown to the MNC. In this case the entry decision depends on

whether or not its expected reservation profit is non-negative.15 From equation (5)

expected reservation profit is given as,

E{f(P, P,F)} = a2(P) + b(P)P- F.

We can use Lemma 1 to show that the expression above is increasing and convex in P.

In Figure 6, the expected reservation profit for a given F is depicted as an increasing

convex curve. Let Pe denote the value of P which makes expected reservation profit

equal to zero.

$

expected

reservation

profit

0 1

o P 1P

Figure 6. Expected reservation profit. MNC enters the developing country if P > Pe(F),

where P(F) > 0.

15 Without any other information than the distribution of labor quality, the MNC cannot compute

probability of a positive EDI.

Any value of P greater than Pe will yield a positive expected reservation profit. As long

as P is higher than Pe(F) the MNC will enter the developing country. As the setup cost

increases, the expected reservation profit decreases; Pe(F) shifts to the right, and the

range of P, which allows the MNC to enter, becomes smaller. In summary, the MNC's

decision to enter depends on the values of P and F in the following way:

Proposition 1: The MNC's incentive to enter is increasing in P, and decreasing in F. It

enters if P > PT(F), where Pe'(F) > 0; otherwise, it does not.

Having defined the MNC's entry problem when it is uninformed about the real value

of labor quality, we now analyze the local government's information sharing and

education investment policy.

5 Local Government Policy

5.1 Education Investment

As shown above, the MNC can gain from knowing the labor quality P (i.e., the true

probability that a randomly selected local laborer is skilled). The local government can

take advantage of that interest by sharing information with the MNC in exchange for tax

revenues. Furthermore, the government can use these revenues to invest in education

(EDI).16 In this section, we shall study the behavior of a government that is assumed

solely to care about its own net revenue (i.e., the government is assumed to maximize tax

16 Tax revenue is used for educational investment before production. One can think of the government as

having access to credit at a zero interest rate. The introduction of a (small) positive interest rate does not

change our results qualitatively.

revenues less education cost). In section 6.3 we study the behavior of a government that

also cares about the welfare of local labor.17

We begin by assuming relatively high expected labor quality and relatively low setup

cost, so that the expected reservation profit is positive and the MNC enters without

information sharing. However, if expected labor quality is low and setup cost is high, the

expected reservation profit is negative. Then the MNC enters only if the government

induces it with EDI and information revelation. Such a situation has important welfare

implications, and it is investigated as the "FDIinducingEDI" discussed in section 6.2.

The optimization problem of the local government, given FDI, takes the following

form:

max[tH (P,F)- [ C(P)dP]

P,t Po

subject to

(1 t)H*(P,F) > H(P,P, F),

tf*(P,F)- C(P)dP >0,

where, t is the tax rate on the MNC profit, P is the portion of skilled laborer after EDI, PO

is the portion of skilled labor before EDI, C(P) is the marginal cost of education,

n*(P,F) is optimal profit of the MNC knowing the real value of P, and H(P,P,F) is

reservation profit for the MNC.

This constrained optimization problem has a double principal-agent structure. The

local government is the principal to the MNC, and the MNC is the principal to local

labor. Given the principal-agent strategy of the MNC against local labor, the local

17 We assume that the only revenue source for the government is taxes from the MNC. We do not look at

the possibility of income taxes on local labor. One can think of the developing country in question as a

country where it is difficult, either for reasons of low prevalent incomes, or because of administrative

difficulties, to levy income taxes.

government chooses the tax rate and the education level to maximize tax revenue less the

cost of education. The first constraint guarantees that the after tax profit of the MNC is

larger than, or equal to, the reservation profit. The second constraint guarantees the local

government tax revenue less the cost of education is non-negative.

Also, note that we simply assume that the government tells the MNC the true value of

P. However, this truth-telling can be made the optimal strategy with the following

assumptions often made in persuasion games (Grossman 1981, Milgrom 1981). First, the

principal (government) cannot reveal false information. Second, an agent (MNC) has

rational expectations about the strategy of the counterpart. In our case, we can show that

the government cannot increase its revenue by misleading the MNC to believe the value

of P, which is higher than the true value. Then, an MNC with rational expectations

should expect that the true P is the maximum value in the set presented by the

government. So, there is no incentive for the government to tell a lie.

From the first constraint, we have

n*(P,F) n(P,P,F)- tH*(P,F) = 0.

> R(P, P) tn* (P, F) = 0,

where R(P,P) is the information rent defined earlier, which does not depend on setup

cost (F). Substitution of this into the objective function and the second constraint results

in the following local government optimization problem:

P

max[R(P, P) C(P)dP] (9)

Pt P o

s.t. R(P,P)- tn*(P,F) = 0, (10)

(PP) c(P)dP (11)

R(P,P) C(P)dP > 0. (11)

fP'O

The local government chooses P so as to maximize the information rent less the cost

of education. It then sets the tax rate so that the after tax profit of the MNC is equal to

reservation profit at P.

From Figure 5, we know that the marginal information rent is negative at any point

P

P. Also, marginal information rent is increasing

everywhere. Marginal information rent (MR) and marginal education cost (MC), which is

assumed to be constant, are shown in Figure 7. Because MR is increasing everywhere

while MC is constant, marginal net revenue for the local government is increasing

everywhere.18

MR

MC

1P

Figure 7. As long as P, < P, there is no

the government immediately jumps to P

incentive for EDI; however, once P, equals P,

=1. This point (P) is the take-offpoint.

The conventional maximization rule (MR=MC) does not apply here. Rather, the

optimal value of P is 1 or 0, depending on at which point the information rent less the

18 The assumption of constant marginal cost (MC) is not essential, but is made for expository simplicity.

Changing this assumption does not affect our arguments. Neither of the following two cases changes the

arguments at all: decreasing MC or increasing MC, which intersects with MR only once. When increasing

MC intersects with MR twice, the second intersection becomes the destination of the jump-process

(explained later) instead of P=l. As long as MC is positive everywhere, we have the jump-process because

MR is negative when P is low.

cost of education is greater. But, since we assume that the initial value of P cannot

decrease, either P=l or the initial level P, will be chosen.

Suppose that a developing country has the initial value P, = P, so that the area B+C

=A+B+D in Figure 7. The area B+C corresponds to the total information rent at P=l;

A+B corresponds to the total education cost required to reach P=l from P, and; D

corresponds to the information rent earned by retaining the initial level of education P,.

Thus, the government is indifferent between choosing the initial level P, and P=l. Any

local government whose initial P, is lower than P, has no incentive to invest in education

because its cost exceeds the rent it can earn. Any local government whose level of initial

P, is higher than P, has the incentive to choose P=l. As long as P, < P there is no

incentive for EDI; however, once P, equals P, the government immediately jumps to

P=I. This sudden change is because the marginal net revenue of the government is

increasing everywhere, and it has no ability to reduce the value of P. We term this point

(P ) the take-offpoint.

An interesting property of take-off point P, is that it decreases as the expected labor

quality P decreases. As shown in Figure 5, marginal rent to the right of P increases as

P decreases. Consequently the take-off point P, moves to the left because the cost of

reaching P=l from old P, is now less than rent earned. This means that as the expected

labor quality P goes down, the local government has the incentive to jump to P= from a

lower level of P,. This situation is shown in Figure 8 in which the shift of expectation

P to P' causes a leftward shift of MR to MR' and P, to P'.

MR' -.'RVMR

O 1- 1

Figure 8. The take-offpoint (P,) decreases as the expected labor quality (P) decreases.

The take-off point is an increasing function of marginal cost (C(P)). In Figure 7, an

increase in marginal cost will shift MC up without changing MR. Total education cost is

now greater than total rent at P,. So, P, must move to the right. There will be no

incentive to invest in education regardless of the value of P, when MC intersects with

MR at P=1.

Setup cost does not shift the take-off point. This is because it neither changes the

value of information rent nor the of education cost. But, setup cost does affect the entry

decision of the MNC, and it has important welfare implications (investigated in section

6.2). So far, given the principal-agent contract of the MNC, the local government jumps

to the maximum level of education at the take-off point. The lower the expected level of

education and the cost of education, the lower the take-offpoint.

Proposition 2:

The local government has no incentive for EDI whenP, < P,; however, once P, equals P,

it immediately jumps to P=l. P, is increasing in P and C(P).

This jump process shows some similarity to the popular multiple equilibria problem

of economic development. Considerable literature recognizes the deadlock in the multiple

equilibria of developing countries (Murphy et al. 1989, Becker et al. 1990). In Becker et

al. (1990), there is little incentive to invest in education during the early stage of

economic development so that the economy is trapped in the lower equilibrium. But, our

model shows the opposite results are possible.

Suppose that a region within a developing country hosts an MNC, where, the

probability distribution of the labor quality within the entire country is known, but the

exact labor quality of the region is not known. Then, the host country (or regional)

government has an incentive to reach the maximum level of education from the take-off

point. Besides, the take-off point gets lower as the known expected labor quality of the

country gets lower.19 The lower the expectation (country in early stage of economic

development) is, the lower is the take-off point. This shows the distinct possibility of the

jump occurring during the early stage of economic development, and may shed light on

considerable differences in economic development among regions within a country.

These remarkable characteristics of our model come from the role of labor quality

information. In the early stage of economic development the expected labor quality is

low. The higher the real labor quality relative to expected quality, the higher the value of

19 Note, however, that expected labor quality has to be at least as high as P to host FDI. If not, the

government has to engage in "FDI inducing EDI" which is analyzed in section 6.2.

information. When the value of information transacted between the government and

MNC is high, the former can increase its share of the profits, and has a stronger incentive

to invest in education.

5.2 Tax Policy

Now we investigate the tax policy of the local government. Repeating equation (10),

the tax rate t is adjusted so that the local government can earn information rent. The tax

revenue is shown as

trI*(P,F) = r*(P,F) I(P, P,F) > 0.

This is non-negative because the optimal profit is at least as high as the reservation profit

(see section 4.1). If the initial PO is located on, or to the right of, the take-offpoint P, the

local government will increase education level to P=I. So the tax revenue is

tr*(P,F) = r*(1,F) HI(1,P,F) > 0.

This is positive because we assumed 0 < P < 1. If the initial P1 is located to the left of the

take-off point P], the local government retains the initial level of education with the tax

revenue

trI(P, F) = H(P, F) H(P,P,F) > 0.

This is equal to zero when Po =P. With any other value of P1, the above equation is

positive. In summary, tax revenue of the local government is non-negative; it is zero only

when P = P, given P < P.

6 Welfare Analysis

In this section we investigate the welfare effects of FDI, EDI, and information sharing

between the government and the MNC. There are two kinds of welfare gains from this

strategy set. The first gain is from EDI and a more efficient local labor contract, given

that the MNC enters the country anyway (efficiency gain effect). The second is the

welfare gain when the MNC enters the region due to EDI and information sharing, which

does not enter without these ("FDI inducing EDI" case). We also investigate the policy

effects on the welfare of local labor.

6.1 Efficiency Gain Effect

Here we retain the assumption that the MNC enters the region with, or without,

information sharing. The optimization problem given in section 5.1 satisfies the incentive

compatibility for both the local government and the MNC; the former earns non-negative

tax revenue less education cost, and the latter earns after tax profit greater than, or equal

to, its reservation profit. So, neither agent has an incentive to deviate from its own

strategy, given that of the other (Nash equilibrium). Because the local government has

perfect information, it can make a Pareto efficient contract with the MNC. The problem is

to determine the situation in which each agent is strictly better off than without this

strategy set.

From the tax policy analysis in section 5.2 the local government earns positive tax

revenue except when P,=P, given P, < P,. Tax revenue is always greater than education

cost because the local government can choose zero education cost simply by retaining the

initial level. We can conclude that the local government is strictly better off except when

P,= P, given P, < P,.

Can the MNC increase profits by acquiring information about local labor quality?

From equation (10), the after tax MNC profit is always equal to reservation profit at P, so

that (1- t)H(P,F)= H(P,P,F). Without information sharing, MNC profit would be

the reservation profit for the initial P, which is given by H(P, P, F). From equation (5)

the reservation profit is a linear increasing function of P, so that H(P,P,F) >I(Po,,P,F)

if, and only if, P >P,. In other words, the MNC is strictly better off whenever the local

government is induced to invest in education (the case in which the initial P, is larger

than, or equal to, P,). With positive EDI both agents are strictly better off. We summarize

our findings:

Proposition 3:

The strategy set makes both the local government and the MNC at least as well off as

without it; the former is strictly better off except when P,=P, givenP,

latter is strictly better off when P > P .

6.2 FDI inducing EDI

From Proposition 1, when the expected labor quality is low, it is less likely for the

MNC (without information sharing) to enter the developing country. On the other hand,

from Proposition 2, when the expected labor quality is low, it is more likely that the local

government will invest in education, given FDI. This inconsistent behavior of the two

agents results in failed opportunities for both parties. In the early stage of development

(the lower expected labor quality), the local government has a stronger incentive for EDI,

given FDI, but the MNC will likely not want to invest (FDI) in the region. Therefore, the

local government has incentive to have EDI, contact and reveal the information to a

potential entrant (even with extra cost) that, otherwise, would not enter the region. We

call this action of the government "FDI inducing EDI."

In order for the FDI inducing EDI to be realized three conditions must be met. First,

the MNC does not have FDI without information sharing (negative expected reservation

profit). Second, the local government can gain by investing in education (EDI), revealing

labor quality information to the MNC, and guaranteeing the MNC reservation profit.

Third, the MNC provides FDI when informed of the tax rate and the improved labor

quality due to EDI (positive reservation profit after EDI).20 These conditions are given as

H(P,P,F) < 0 a2(P)+b(P)P F <0, (12)

H*(1,F)- n(1, P,F) > (1 P)C + A

(13)

< a2(1) + b(1)- [a2(P) + b(P)] > (1 P)C + A,

n(l, P, F) > 0 <> a2 (P) + b(P) F > 0, (14)

where A is the access cost of the local government to a potential entrant. Definitions of

a2(.) and b(-) were previously provided. Combining equations (13) and (14)

a2(1) + b(1) F (1 P)C A > a2(P) + b(P) F > 0. (15)

The first inequality of equation (15) is more easily satisfied, as the expected education

level (P), the marginal cost of education (C), and access cost (A) are lower, and the

initial level of education (Po) is higher. Also, the inequality of equation (12) is more

20 There are cases in which the local government can induce FDI by simply revealing information.

However, we focus on the case with positive EDI because the relation between EDI and FDI is the theme

of this paper.

easily satisfied as P is lower. Hence, as long as the second inequality of (15) is satisfied,

the lower P is, the easier the conditions are satisfied.

However, when P is too low, a negative reservation profit results for the MNC even

after positive EDI. This is because the lower P is, the higher is the share taken by the

government. The allowable range of P depends on the value of setup cost (F).

Combining (12) and (15), we have

a, (P) + b(P)P < F < a (P) + b(P) .

First, note that a2(P)+ b(P)P is increasing and convex in P, and a(P) + b(P) is non-

decreasing in P .21 The former is smaller than, or equal to, the latter. The allowable range

of F gets greater, as the difference between these, b(P)- b(P)P, gets greater. This is

positive for P <1, but is getting smaller when P is sufficiently high. It asymptotically

approaches zero as P approaches one. Hence, FDI inducting EDI is more common for

the government with relatively low P .

This claim can be illustrated in much-exaggerated measure by assuming commonly

used disutility functions: c,(x)= x /y,c2(x)= kx ly, where k>1 is the productivity

difference indicator, and >1 is the common elasticity of disutility.22 Then we have

-1- 1 --P 1-P -1 1 -

b(P)-b(P)P +1 01-P,

7/ k-P k-P Y\

21 Increasing and convexity of a, (P) + b(P)P are directly from Lemma 1. Non-decreasing

dx dx

a,(P) + b(P) is from a2'(P)+ b'(P) = (c,'(x,)- c,'(x2)) where < 0.

dP dP

22 Disutility function for each type cannot have different elasticity because it would violate the conditions

of c, (x) < c,(x),0 0.

1-P 1

Since k -P >1-P > O,we have 0 <- <1. Also from y >1, 1- > It is easy to

k-P Y

1-P

see that the expression in the curly bracket is decreasing in so that it approaches

k-P

1-P

zero as approaches one. Therefore, we have

k-P

{L +1 (-l-P) > 0.

SYlk-P_ Lk-P Y

The change of allowable range with P is given by

8[b(P)-b(P)P]- 1 ~k-1 1'l-PF rL -P 1-F -+1- 1

OP ()-1) k-P k-P k-P k-P 7

k-l 1-P 1 kk- I _

Since 0< <1 and 0 < <1, -- is negligible when y

k-P k-P (-1)Lk-P k-P

8(b bP)

is sufficiently high. If this is the case, we have < 0, which means that the

OP

allowable range off gets larger as P gets smaller. We illustrate the case with the

quadratic disutility function (y = 2, k = 2) in Figure 9. The shaded area in Figure 9 is the

allowable range of P and F to realize "FDI inducting EDI."

F

0.5

a2 + b

0.4

0.3 a2 +bP

0.2 P

0.0 0.2 0.4 0.6 0.8 1.0

Figure 9. The allowable range of P and F with quadratic disutility function.

Note that the allowable range ofF is decreasing in P everywhere. Also from Figure

9, we can see that the allowable range of P is wider when F is relatively low. This comes

from the convexity of a2 (P) + b(P)P .

In summary, the FDI inducing EDI is more common when the country is in an early

stage of development (low expected labor quality); its initial labor quality is high; and its

education cost, access cost, and MNC's setup cost are low.

6.3 Local Labor Welfare

In this section we investigate the effect of the strategy set on local labor welfare.

First, we show the case in which the strategy set benefits local labor. Second, we analyze

how government policy will change if it tries to maximize the welfare of local labor.

When there is neither information sharing nor EDI, the MNC uses P as the target

value as shown in section 4. The total labor surplus (W,) without the strategy set depends

on the initial and expected labor quality as

W(P0,P)= P [1 (P) -c, (x,(P))] + (1 P,)[w(P) c (x1(P))].

Setting the target value to P in equation (2),

W (Po, P) = P0 [w,o(P) c (x ())].

With the strategy set, the total labor surplus (W) becomes

W(P) = P[w, (P) c, (x,(P))].

Again we used equation (2). First, we show that positive EDI is always welfare

deteriorating for local labor.

Whenever the local government has positive EDI it has an incentive to go to the

maximum level (P =1). If there are only skilled laborers, the MNC's strategy becomes

max(x, w1)

s.t. w, c,(xi) > 0

-> FOCs w, = c,(x,), c,'(x) =1.23

This means that skilled labor surplus is zero. Therefore, when there are only skilled

laborers, all labor surplus disappears so that W(1) = 0. On the other hand, the total labor

surplus without the strategy set is positive because skilled labor has a positive surplus

with 0

the local labor benefits from the strategy set are limited to those without EDI (i.e.,

P > P).

23 The same result is derived by setting P= in equations (1) through (4).

24 From Assumption 1 and equations (1) and (2), we can derive w, c, (x) = w2 c,(x2) > 0 if 0< <1.

When the government retains the initial level of labor quality, the total labor surplus

with the strategy set becomes

W(P,) = PO( (PO)-c1 (Xi(PO))

Then the difference in total labor surplus between with and without the strategy set

now can be expressed as

W(Po) Wo(PoP) = Po[(w1 (PO) c, (x, (P))) (w1 (P) c, (x,(P)))].

From equations (1), (2), and (4) we can derive (w, (P) c (x,(P))) is non-increasing

in P. This leads us to that W(P) Wo(Po,P) > 0 only if PO

strategy set benefits the local labor only if P, < min P,}. Local labor is worse off in all

other cases. This is a very restrictive condition compared to the government and MNC,

both of which are always at least as well off as without the strategy set. Besides, both are

strictly better off in the case of positive EDI, while the local labor is strictly worse off

with a positive EDI.

How will government policy change if it tries to maximize the welfare of the local

labor? This optimization problem is the same as shown in section 5, except for the

objective function. Now the government's objective is to maximize the total welfare of

local labor, so that

max[P(w, c(x)) + (1- P)(w, -c,(x2))] (9')

P,t

s.t. R(P,P) tH *(P,F) = 0, (10')

R(P, P) C(P)dP >0. (11')

As previously shown, when there is only one kind of laborer (P=O or 1), all labor

surplus disappears. However, when O

because skilled labor has a positive surplus. Because the total labor surplus is a

continuous function of P, there must be 0

We denote this optimal P as P*.

First, if the initial level of education is higher than, or equal to, the optimal point P*,

there is no incentive for education investment. Because further education investment will

result in reduced welfare of local labor, the government will maintain education at the

initial level.

Second, if the initial level of education is lower than P*, we can define the take-off

point again. Before the take-offpoint, the government has no incentive for EDI. But, once

it reaches the take-off point, the government incentive is to jump to P*. As before, the

take-off point is increasing in the marginal cost of education and the expected labor

quality.

There are some important changes from the net revenue maximizing government.

Under the current policy, the government jumps to P* which is strictly lower than one.

Because it jumps to the lower point than before, the budget constraints for education

mandate the take-off point to be higher than before.25 So, both the range where the jump

process occurs and the distance of the jump are smaller than for net revenue maximizing

government.26 Any value of P greater than P*, which happens whenever net revenue

maximizing government has EDI, damages the welfare of local labor. In that sense,

incentives are provided to itself and to the MNC at the expense of local labor. This result

is in clear contrast to the conventional belief that skill building in the host country must

25 Remember that marginal information rent is increasing everywhere while marginal education cost is

constant. Hence, the higher the destination of jump process is, the lower is the take-oif point.

26 When the government tries to maximize the sum of labor surplus and net tax revenue, the chosen level of

education would be between P and 1.

always benefit local labor (United Nations 1999). Finally, we summarize these results as

Proposition 4.

Proposition 4: Local labor benefits from the strategy set only if P, < minP,P }. The

take-off point becomes higher and the distance of the jump becomes shorter when the

local government tries to maximize the local labor welfare.

7 Conclusion

We have developed a model explaining the observed policy combination of a

developing country (hosting FDI and investing in education) and the interest of the MNC

about the quality of the local labor force when it contemplates FDI in a developing

country. Information on local labor is the source of a more efficient local labor contract

for the MNC. The local government has an incentive to invest in education (EDI) because

it increases both its net tax revenue and MNC profit.

The important implication of EDI by the local government is that it suddenly jumps to

the maximum level of education when it reaches the take-off point. However, the local

government does not invest in education before it reaches the take-off point. An

interesting finding is that the take-off point becomes lower as the expectation of

education level decreases. This means that a country in an early stage of development has

the incentive to take a larger leap, which heretofore has not been considered in existing

multiple equilibria models.

The strategy set (FDI, EDI, and information sharing) has two positive welfare effects.

First, EDI and information sharing contributes to a more efficient contract between the

MNC and local labor, given the MNC has FDI in the country. Second, with EDI and

information sharing, the local government can induce an MNC to have FDI in the

country, which otherwise would not occur. This "FDI inducting EDI" is more likely to

occur for a country in the early stage of economic development.

The strategy set makes both the government and the MNC at least as well off as

without it, but it benefits local labor only under very restrictive conditions. Furthermore,

when the local government invests in education, it must over-invest to the level that

deteriorates the local labor welfare. So, the government has the incentive to benefit itself

and the MNC at the expense of local labor. This result is in clear contrast to the

conventional belief that skill building in a developing country always benefits local labor.

Appendix I.1

Proof of a2 (P) > 0, b(P) > 0.

From Assumption 1, equation (2) must pass through the origin ((x2, w) = (0,0)).

Equation (2) is shown as the w2 = c2(x2) curve in Figure A. 1. The iso-profit line

(w2 = x2 -a2) has slope of one and passes through a point on the w2 = c2 (x) curve.

From the incentive compatibility constraints we have c2(x1) c(x2) > c(x,) c(x2) .

This, combined with Assumption 1, indicates that x, > x2. From equation (3), and

x, > x,, we have

1- '(x2) >0. (A. 1)

Substituting this into equation (4) yields

1-c2'(x) >0. (A. I. 2)

Hence, slope of the w2 = c2(x2) curve must be less than, or equal to, one at (x2,w2). This

and Assumption 1 indicate that slope of the w2 = c2(x2) curve is less than one all the way

up to (x2, w2). Then the w-intercept of the iso-profit line (w2 = x2 a2) is non-positive so

that a2 > 0.

From participation constraints and equation (1), we have w, -c(x,)= w2 -c,(x2) > 0.

This means that a skilled laborer has at least the reservation utility, and is indifferent

between (w,,x,) and (w2,x2). In Figure A. 1 we draw the skilled laborer's indifference

curve (w = U, + c,(x,) with a constant U > 0 ) passing through (wx,2).

When x2 > 0, the skilled laborer's indifference curve is less steep than that of

unskilled laborer at (w,,x2) from Assumption 1. From equation (3), the iso-profit line

(w = x, a) is tangent to the indifference curve on (w1, x). From Figure A. 1, it is clear

that a, > a2. This means that b > 0. When x2 = 0, we have c'(x2) = c2'(x) = 0 from

Assumption 1. Then equation (3), which is c1'(x,) = 1, means that x, > x2 = 0. This leads

to a, > a2 = 0, and b > 0.

2 c2(x2) w =U, +c(x,)

w2 =X2

0x

a/2 /W1 = x1 a7

a,

Figure A. 1 Indifference curves and iso-profit lines for a givenP .

Appendix 1.2

Proof of a2'(P) < 0, b'(P) > 0.

By definition a2 (P) = x2 (P) w(P). Substituting equation (2) yields

a2(P) = x2(P) c2(x2 (P)) Differentiating with respect to P yields

a2'(P) = (1- c'(x2)) d (A. I. 3)

dP

Differentiating equation (4) with respect to P yields

d2= [c2 (x2) Pc" (x2)] -1[c(x2) 1]

dP

From Assumption 1, c2"(2) -Pc,"(x2) > 0. Substituting this and equation (A. I. 1) into

dx dx

the above yields -d < 0. Substituting x2 < 0 and equation (A. I. 2) into equation (A. I.

dP dP

3) yields a2'(P)< 0.

By definition, b(P) = x(P)- w(P)-(x(P) w2(P)). Using equations (1) and (2),

b(P) = x (P) x2 (P) c, (x (P)) + c (x2 (P)). Differentiating with respect to P yields

b'(P)= (1-cl'(x)) -(1 -c'(x2)) 2

dP dP

Using equation (3), 0. Substituting this into the above equation yields

dP

b'(P) = -(1-c1'(x2)) d

dP

dx

Substituting -- < 0 and equation (A. I. 1) yields b'(P) > 0.

dP

Appendix II

Proof of Lemma 1.

First, we show that I*(P!,F)l> I (iP',F) for P > P> Denote (x1, x2, 2) as the

strategy vector for P, and (x t,w x', ,') as the strategy vector for P'. Then, we have

the following relation:

P[ w] + (1 P)[x2 w2, ]> P[x1 w] + (1 P)[x~ w

>Pt[x w]+(-P )[x2 ].

This means that H*(P,F) > H(Pt,F). The first inequality is from the optimization at

P, and the second inequality is from the fact that P > Pt and that xt, wi > x3 wi

(proof in Appendix I. 1). Now we show that the profit function is convex in P.

For P, P, and P" 2P + ( 2)P' such that 0 < A < 1, let the strategy vectors be

(xY, w1, x,w2), (x', ', ,x2', w2') and (x", 1" ,x2 w,"), respectively. We therefore

have the following results:

[P(1 )+ (1 -P)(x, w)] > [P( "-w")+ (1 P)(x'"-2")].

(1 A)[P'(,'-1,') + (1 P')(x,'-w,')] > (1 A)[P'(x1,"-,") + (1- P')(x,"-,, ")].

Summing up each side of the inequalities, we can compute

An*(P,F) + (1 )In*(P',F) 2 n*(AP + (1 A)P', F).

Therefore, the profit function is convex in P .

Appendix III

Proof of Lemma 2.

The effect of a decrease in P on information rent is given by

OR(P,P) _a(* (P,F) (P, P, F)) (P,P,F)

8P 8P 8P

= -Pw,'(P) + (1 P)(x,'(P) w'(P)).

In deriving the result above, we used equation (3). Using equations (1) and (2), the above

equation is transformed as follows:

OR(P,P) dx2 dx

= -P[c '(x,)- c'(x)] + (1-P)[1- c'(x)]

OP dP dP

Using equation (4) we can compute

aR(P,P) dx-

R(P -(P-P)(1-c,'(x,)) ~>0, for P >P.

OP dP

In deriving the final inequality, we used 1-c,'(x,) 0 from equation (A. I. 1), and

dx < 0 from Appendix I. 2.

dP

References

Becker, Gary S., Murphy, Kevin M., Tamura, Robert. "Human Capital, Fertility, and

Economic Growth," Journal of Political Economy 98 (1990): 12-37.

Bloom, David E., Sachs, Jeffrey D., Collier, Paul, and Udry, Christopher. "Geography,

Demography, and Economic Growth in Africa," Brookings Papers on Economic

Activity 1998(2) (1998): 207-295.

Collins, Susan M., Bosworth Barry P. "Economic Growth in East Asia: Accumulation

versus Assimilation," Brookings Papers on Economic Activity 1996(2) (1996): 135-

191.

Ethier, Wilfred J. and Markusen, James R. "International Trade and Labor Migration,"

American Economic Review 75 (1985): 691-707.

Ethier, Wilfred J. "The Multinational Firm," Quarterly Journal of Economics 80 (1986):

805-833.

Ethier, Wilfred J. and Markusen, James R. "Multinational Firms, Technology Diffusion

and Trade," Journal ofInternational Economics 41(1996): 1-28.

Fukao, Kyoji. and Amano, Tomofumi. "Direct Investment Abroad and "Hollowing Out"

of Manufacturing Industry," The Economic Review 48 (1998): 259-276.

Glass, Amy Jocelyn and Kamal Saggi. "FDI Policies under Shared Factor Markets,"

Journal ofInternational Economics 49(2) (1999): 309-32.

Glass, Amy Jocelyn and Kamal Saggi. "Multinational Firms and Technology Transfer,"

Scandinavian Journal of Economics 104 (2002): 495-513.

Grossman, Stanford J. "The Informational Role of Warranties and Private Disclosure

about Product Quality," Journal of Law and Economics 24 (1981): 461-483.

Hayami, Yujiro. Development Economics: From the Poverty to the Wealth of Nations.

2nd edition. Oxford University Press (2001): 163-168.

Horstmann, Ignaius J. and Markusen, James R. "Licensing versus Direct Investment: A

Model of Internalization by the Multinational Enterprise," The Canadian Journal of

Economics 20 (1987): 464-481.

Horstmann, Ignaius J. and Markusen, James R. "Exploring New Markets: Direct

Investment, Contractual Relations and the Multinational Enterprise," International

Economic Review 37 (1996): 1-19.

Mas-Colell, Andreu, Winston, Michael D. and Green, Jerry R. Microeconomic Theory.

Oxford University Press (1995): 492-501.

Murphy, Kevin M., Shleifer, Andrei, and Vishny Robert W. "Industrialization and the

Big Push," Journal of Political Economy 97 (1989): 1003-1026.

McMilan, Margaret, and Pandolfi, Selina, and Salinger, Lynn B. "Promoting Foreign

Direct Investment in Labor-Intensive, Manufacturing Exports in Developing

Countries" Consulting Assistance on Economic Reform II Discussion Paper No. 42,

(1999).

Milgrom, Paul R. "Good News and Bad News: Representation Theorems and

Applications," Bell Journal of Economics 12 (1981): 380-391.

MITI (Japan Ministry of International Trade and Industry), Kaigai Jigyou Katsudou

Doukou Chousa Gaiyou (Survey of Trends in Japanese Business Activities Abroad).

Okurashou Insatsukyoku, (1999).

UNCTAD. Foreign Direct Investment Database. Available at

www.unctad.org/Templates/Page.asp?intItemID= 1923&lang= 1. (2003).

OECD. Foreign Direct Investment for Development: Maximising Benfits, Minimising

Costs. OECD, (2002).

United Nations. World Investment Report: Foreign Direct Investment and Challenge of

Development. New York: United Nations, (1999).

Wakasugi, Ryuhei. Kokusai Keizaigaku (International Economics). Iwanami Syoten,

(1996): 161-172.

World Bank. World Development Indicators 2003 CD-ROM edition. World Bank,

(2003).