Wake up and smell the ginseng: international trade and the rise of incremental innovation in low-wage countries
Wake Up and Smell the Ginseng:
International Trade and the Rise of Incremental Innovation
in Low-Wage Countries
, Universidad Carlos III
University of Toronto
First version, August 2005
This version, January 2009
Increasingly, a small number of low-wage countries such
as China and India are involved in incremental innovation. That is,they are responsible for resolving production-line bugs and suggest-ing product improvements. We provide evidence of this new phe-nomenon and develop a model in which there is a transition fromold-style product-cycle trade to trade involving incremental innovationin low-wage countries. The model explains why levels of involvementin incremental innovation vary across low-wage countries and acrossfirms within each low-wage country. We draw out implications forsectoral earnings, living standards, the capital account and, foremost,international trade in goods.
Key words: international trade, low-wage country innovationjel classification: f1
∗We are thankful to Joel Blit for research assistance and to Pol Antràs and Elhanan Helpman for helpful comments
on an earlier draft. Both authors thank the Canadian Institute for Advanced Research (cifar) for its tremendoussupport. Funding from the Social Sciences and Humanities Research Council of Canada (sshrc), Spain's Ministeriode Educación y Ciencia (sej2006–09993), the Comunidad de Madrid (prociudad-cm), and the Centre de Recerca enEconomia Internacional is gratefully acknowledged.
‡Madrid Institute for Advanced Studies (imdea) Social Sciences, Antiguo pabellón central del Hospital de Canto-
blanco, Carretera de Colmenar Viejo km. 14, 28049 Madrid, Spain (e-mail: website:
§Rotman School of Management and Department of Economics, University of Toronto, 105 Saint George Street,
Toronto, Ontario m5s 3e6, Canada (e-mail: website: Also affiliated with the National Bureau of Economic Research (nber).
When the auto parts giant Delphi Technologies first set up shop in Chihuahua, no one predictedthat the Mexican shop-floor engineers at this low-wage affair would soon be introducing minorproduct changes that would slash product defect rates. While the cumulative effects of theseproduct changes are large, no single innovation is pathbreaking. Even the most sophisticatedinnovations — those that actually generate patents — are just better mouse traps that increment-ally improve on existing auto parts technology. For example, most Delphi-Chihuahua patentsimprove on the control systems of minor moving parts. These patents are examples of incrementalinnovation, Rosenberg's unsung hero of modern economic growth. This paper is about therise of incremental innovation in a select group of low-wage countries, how it is allowing thesecountries to export increasingly sophisticated new goods, and why it is leading to a shift in worldtrade patterns.
Incremental innovation in low-wage countries is not part of the lexicon of international trade
theory. Instead, we are glued to product-cycle theory in which products and pro-cesses are developed and standardized in rich countries before being moved offshore to low-wagecountries. (See also and the more sophisticated approaches of and Increasingly, though, the first location of production formoderately sophisticated goods is taking place in a handful of low-wage countries such as China.
This implies that the many bugs that plague new goods and their production processes are partlyand sometimes wholly resolved by local managers and engineers. That is, a handful of low-wagecountries are becoming engaged in incremental innovation. If incremental innovation is indeed theunsung hero of modern economic growth, then the long-term implications of this development aresignificant.
From the perspective of a rich-country firm, there are costs and benefits of involving low-wage
countries in incremental innovation. One key benefit is that it frees up valuable innovationresources at home so that they can be focused on ‘big ideas' innovation. In addition, low-wagecountry involvement in incremental innovation allows the firm to locate production in a low-wage country even before products and processes are fully developed and standardized: localengineers are used to help complete the process of standardization. By locating a good's firstproduction line in a low-wage country, the process of lowering labour costs is telescoped forward.
Another benefit is that in a world of complex foreign supply chains, a firm that involves localsuppliers in incremental innovation can insist that each supplier deliver continual product andprocess upgrades. Being directly engaged in the production process, suppliers can come up withimprovements on the shop floor that would have been more costly for the firm to identify fromhead office. These additional improvements allow firms such as Delphi to stay a hair's breadthahead of the competition.
Against these benefits are some significant drawbacks of involving local suppliers in incre-
mental innovation. First and foremost, parts suppliers from low-wage countries typically producecomponents for complex, interdependent systems in which an incremental improvement in onecomponent is not effective unless other components are also modified. This interdependence
means that parts suppliers do not internalize all of a firm's innovation costs. In the simplest case,when a firm asks a parts supplier to improve a component, the solution will usually entail residualincompatibilities
with other components of the system, thus forcing the firm to incur the additionalexpense of bringing other components into line.
We also model a second and more familiar cost of involving a local supplier in incremental
innovation. Once involved, the supplier acquires information and expertise that can be usedoutside the relationship. This makes it necessary for the firm to offer expensive incentives aimedat encouraging the supplier to stay within the relationship.
The modelling core of this paper is about the firm's choice of whether or not to involve a sup-
plier in innovation. Nothing in the formal set-up of the model is specific to incremental innovationas opposed to more substantial innovation. Thus, the reader should feel free to interpret our modelin terms of innovation more broadly defined. Our reading of the available evidence, however,persuades us that innovation in low-wage countries is to date dominated by incremental productand process innovation. We embed the firm's choice of whether or not to involve a supplier ininnovation into a general equilibrium model. Each firm in the model chooses where to locateproduction among a set of low-wage countries and whether or not to involve locals in innovation.
The general equilibrium choices of firms aggregate to yield predictions about the extent to whicha low-wage country will attract rich-country firms and engage in incremental innovation. Wethen build on these results to generate a rich set of predictions about international trade in goods,earnings and living standards and, to some extent, the capital account.
Evidence on incremental innovation in rich countries is abundant e.g., and Systematic evidence for developing countries is only just emerging e.g., Fortunately, there are two observable activities that are tied into incremental innova-tion. First, incremental innovation is most intense during the early stages of the production of newgoods, when product bugs are fixed and production processes optimized. This is why learningcurves are steepest by far in the earliest stages of production (e.g., and
Thus, one way to track incremental innovation is to track where new goods are produced.
Second, while incremental innovation is about minor cumulative improvements to products andprocesses, occasionally it leads to patents. Thus, a second way to track incremental innovation isto track patents developed by residents in low-wage countries for rich-country corporations.
Some evidence on the first location of production of new goods can be culled from the business
strategies of contract manufacturers. The industry is dominated by six giants. Typical of these is
1 Anecdotal evidence on more substantial innovations in developing countries frequently appears in the press. For
example, 3com's 8800 high-end network switch and Nokia's 6108 handset were developed in China. The predicts that by the end of 2006 China will be second only to theus in R&D expenditures. However, the incremental innovations on which our paper focuses are rarely the stuff of newsreports or formal R&D expenditures.
Sanmina-sciSanmina-sci reports that 54 of its 159 plants are exclusively engaged in producingnew goods and developing new products. The top locations for these plants are (number of plantsin parentheses): us (15); Canada (5); China, Hungary and Sweden (4 each); Finland, Germany andIsrael (3 each); and the uk and Mexico (2 each). What stands out from this list is the appearance ofthree lower-wage countries: China, Hungary and Mexico.
For more systematic evidence showing that this Sanmina-sci example generalizes we turn to
trade data. Following we can identify where new goods are first pro-duced using us import statistics. The United States classifies imports using 10-digit HarmonizedSystem (hs) codes, which currently cover over 17,000 products and are regularly updated to incor-porate new goods. A 10-digit hs good is ‘new in year t
' if it was first imported into the us in yeart
. For example, the United States first imported 20-page per minute laser printers (hs 8471923600)in 1994, primarily from Japan. For another example, the us first imported machinery for making
dvds (hs 8477104000) in 1995, with two-thirds of imports by value coming from Germany, theNetherlands and
Table shows that production of new goods is beginning to happen in low-wage countries. The
table reports statistics for innovative new goods. Innovative goods are defined as goods in sectorsidentified by the as being medium-high-technology and high-technology sectors i.e., chemicals (including pharmaceuticals), machinery (including electronics),transportation equipment, and instruments. In the 1990-92 period, the us imported 694 innovativenew goods. Of these, 472 or 68% were exported by Japan to the us (see column 1). A decadelater, of the innovative new goods that first appeared in the 2000-02 period, only 55% of thesewere exported by Japan to the us (see column 2). The decline of 13% is statistically significant(t
= −15.51). In contrast, only 17% of innovative new goods arrived from China in 1990–92,whereas 48% arrived from China in 2000-02. This is an increase of 30 percentage points (t
The other big gainers among poor and middle-income countries were Malaysia, Thailand, India,Mexico and the
Changes in the number of new goods could be an artifact of changes in us administrative pro-
cedures used to create hs codes. For example, if these procedures became easier over time then allcountries would experience an upward trend in the number of new goods. One way of controllingfor a common trend associated with changed administrative procedures is to double-differencethe data. For example, one could compare China's change of 30% with the Japanese change of
2 The six giants are (employment in parentheses): Hon Hai (167,000), Flextronics (92,000), Jabil Circuit (61,000),
Solectron (57,000), Sanmina-SCI (53,000), and Celestica (42,000). We focus on Sanmina-SCI because it alone providesinformation about what each of its plants does in each country. Sanmina-SCI appears typical in that its sales per workeris $201,000, which is in the middle of the $166,000–$244,000 range for the six giants.
3 Both 20-page per minute laser printers and machinery for making dvds were incremental innovations over earlier
products (i.e., slower laser printers and other plastic-moulding machinery). Following those incremental innovations,an hs code was created specifically for them. In the case of laser printers, the key is that fast printers (i.e., 20 pages perminute) only appeared in 1994 with the introduction of the Xerox 4520ps and DataProducts lzr2080 printers. Likewise,the hs code for dvd machinery was introduced the same year that an international agreement on standards for dvdswas reached. The first dvd players were brought to market the following year.
4 Data on us imports at the 10-digit hs level are subject to mis-classification by customs officials and jobbers, which
result in minuscule amounts of exports of most goods by many countries. To eliminate such errors, a country is onlycounted as having exported a new good to the us if it accounted for at least 1% of us imports of that particular newgood. Had we not done this, the statistics in table would have been even more supportive of our claims.
% of New Goods Imported by the United States from Each Country
: Authors' calculations based on data from us Bureau of the Census.
: This table reports statistics on goods that were first imported by the United States either in the period 1990–92or the period 2000–02. Only new goods in medium-high-technology and high-technology sectors are included chemicals (including pharmaceuticals), machinery (including electronics), transportationequipment, and instruments. Column 1 states that of the innovative new goods first imported by the United States in1990-92, Japan was an exporter 68% of the time. Column 1 sums to more than 100% because most new goods areimported into the United States from several countries. Column 5 states that of the innovative new goods first importedby the United States in 1990-92, 34% by value was imported from Japan. Column 5 would sum to 100% if all us tradingpartners were listed.
Table 1: Origin of us Imports of Innovative New Goods
−13%. The difference of 43% has a t
-statistic of 14.34. Comparing Japan with each of the otherpoor countries in table the t
-statistic for each double difference is never less than 4.97. Thus, ourconclusion that some poor countries are increasingly the location of first production of innovativegoods does not appear to be an artifact of administrative procedures for introducing new hs codes.
In addition to showing where early-stage production takes place, table displays two other
features that will be outcomes of our model. First, the list of countries is short. All other middle-income and poor countries never exported substantial numbers of innovative new goods. We haveincluded Indonesia in table in order to make this point. Even though a considerable amount ofmanufacturing takes place in Indonesia, it is not the location of first production of innovativegoods. First location of production is confined to only a handful of countries. Second, new goodsare often exported to the us by several countries at once. This would not happen in a quality-ladder model where only a single producer survives e.g., In ourmodel, any firm that has developed a new good can bring it to market provided that it successfullyengages in incremental innovation. Our focus is not on developing new goods, but on incrementalinnovation.
Columns 5–7 of table provide an alternative view of the same phenomenon. They report the
share of us imports of innovative new goods that originated in the indicated country by valueinstead of by number. For example, Japan accounted for 34% of us imports of innovative new
400 Number of Patents (Line)
Patents Relative to Japanese
Number of Patents (Line)
Patents Relative to
Japanese Patents (Bars)
: Authors' calculations from the United States Patent and Trademark Office website
: Lines (right-hand scale) indicate the total number
of us-owned patents with at least one inventor who is a residentof the indicated country. Bars (left-hand scale) scale the lines by the number of us-owned patents with at least oneinventor who is a resident of Japan.
Figure 1: Share of us-Owned Patents with a Foreign Inventor
goods in 1990–92, but only 9% in 2000–02. In contrast, China's share rose from 0% to 5% in justa decade. Thus, the conclusion that low-wage countries are beginning to be the first location ofproduction comes both from frequency data and trade volume data. Table leaves little doubtthat some low-wage countries are now the location of early stages of production and hence of theincremental innovation that is tied to early-stage production.
A second source of evidence on incremental innovation is suggested by our introductory ex-
ample of Delphi patents generated by Mexican engineers at Delphi's Guadalajara facility. Thissource is the us patent database For each patent, the database listswhether the patent is owned by a us entity (such as Michigan-based Delphi) and whether any ofthe inventors reside in a low-wage country. We identified for analysis each us patent that (a
) isowned by a us entity i.e., has a us assignee and (b
) includes a resident
of a foreign country in itslist of inventors. Perusal of detailed patent applications for this group of patents suggests that theDelphi example is common for patents with an inventor from a low-wage country. That is, many
of these patents do indeed capture incremental innovation.
The top left panel of figure tracks all such patents with at least one inventor that resides in
China. (These Chinese residents invariably have Chinese names, which suggests they are not usengineers working in China.) The solid line gives the absolute number of such patents. To facilitateinterpretation of the line we also scale it by the corresponding number for Japan i.e., the numberof us patents with a us assignee and an inventor residing in Japan. This is shown in the bars andthe left-hand axis. Scaling does two things. First, it controls for the explosion of patenting activityin the us since the mid-1990s. Second, it gives a clearer sense of the level of Chinese innovativeactivity: by 2006 it was almost half that of Japan, a country that is among the most innovative onthe planet.
Three features stand out. (1) In recent years there has been an explosion of us-owned patents
involving Chinese and Indian inventors. While in 1981 the Chinese bar was zero, in 2006 it was
0.46 which indicates that this measure of Chinese innovative activity is at 46% of Japanese Similarly, 2006 Indian activity is at 47% of Japanese (2) In contrast to China and India, manyother low-wage countries have barely been touched by this tide of incremental innovation. Eventhe combined numbers for Indonesia, the Philippines and Thailand are small. As with us importsof innovative new goods, there are only a handful of low-wage countries appearing frequently inthe us patent database. (3) Since 2002, patenting has plateaued in Mexico (and some other popular
fdi destinations in Eastern Europe and Latin America) as companies such as Delphi have shiftedproduction to China. This is an important general equilibrium effect that we will be modelling.
The patent database does not distinguish between incremental patents and ‘fundamental' (non-
incremental) patents. A casual examination of the patents underlying figure suggests that theChinese patents are typically incremental whereas the Japanese patents are a mix of incrementaland fundamental. Thus, had we been able to look only at incremental patents the figure surgewould be even more pr
Summarizing, us import and patent data both provide consistent and systematic evidence that a
small number of low-wage countries are increasingly engaged in incremental product and processinnovation. We will show that this has important implications for the rapidly evolving pattern ofworld trade.
5 Another way of thinking about scaling by Japan is that it controls for size as measured by gdp. Japanese gdp is the
same as that of India and about half that of China. Hence, the Chinese and Indian numbers are large even relative togdp. Further, the Chinese and Indian numbers have been growing rapidly relative to gdp. Over the 2000–2006 periodthese patents have grown three times faster than gdp in both countries. Finally, the trends in figure are almost identicalwhen scaling is by total us patents rather than Japanese-inventor patents.
6 The fact that patent numbers for India are virtually identical to those for China is surprising given India's low values
in table The explanation is that Indian patents are mostly related to services.
7 We have tried to identify incremental patents by equating them with process innovation. This does not work. For
one, it is surprisingly hard to identify process innovations. For example, in the Canadian innovation survey used byfirms are explicitly asked to identify which of their innovations are process innovations. Yethalf of the firms report that they cannot: their product innovations are inseparable from their process innovations. Foranother, many incremental innovations are in fact product innovations. In our introductory Delphi example, electricdoor locks are made more reliable by improvements to solenoids. This is an incremental product innovation (wherethe product is solenoids or door locks or autos). In short, it is difficult to equate incremental innovation with processinnovation.
This paper has many touchstones with the existing international trade literature. product-cycle model and its dynamic Ricardian variants (e.g., and either assume or predict that innovation occurs exclusively inrich countries. They rule out innovation in low-wage countries. In the absence of local innovation,technologies diffuse to low-wage countries via such channels as imported machinery, fdi, scientificjournals, technology licensing and theft A fewpapers (e.g., allow agents in low-wage countries to activelyinvest in acquiring knowledge. However, this knowledge acquisition is just reverse engineering ofproducts originally developed and produced in rich countries.
Our paper also fits into the literature on incomplete contracts and trade. However, in order to
focus on our novel contribution we sidestep the two most important questions addressed by thisliterature. The first is about the choice between vertical integration and outsourcing in generalequilibrium (e.g., The second is about the choice between sourcing inputs from the North or fromthe South (e.g., We do not tackle these two questions, choosinginstead to focus on the conditions that promote incremental innovation in low-wage countries.
Our starting point is that a Northern firm has already decided to locate production in the South.
We model the firm's choices about (i
) which of several low-wage countries to enter and (ii
) whetheror not to involve a local agent in incremental
Notice that we are silent on whether we are dealing with fdi or offshore outsourcing. What
matters to us is only whether incremental innovation is being done. However, we will be as-suming that if residual incompatibilities within the match are large, the principal takes control ofincremental innovation. This is appropriate in a long-term relationship i.e., in an fdi relationship.
In a short-term offshore outsourcing relationship it would be more appropriate to assume thatthe principal walks away from relationships with high residual incompatibilities. In the real worldthere is frequently a medium-term relationship, namely, that the principal takes a minority positionin the agent's firm. Our assumption is appropriate for such medium-term relationships as well.
Our model has a second cost of doing incremental innovation in low-wage countries, namely,
the possibility that a local parts supplier could take the knowledge gained from incrementalinnovation and use it outside of the relationship. This means that our paper is also related towork on contract enforcement and trade. See the seminal work of aswell as and Weak contract enforcement is one of several possiblereasons why a local parts supplier can use proprietary knowledge outside of the relationship.
8 Several papers combine both questions in order to deal with the choice between integrated home production,
domestic outsourcing, fdi, and offshore outsourcing, or some subset of these (e.g., and also who integrates this choice in a product-cycle model). and review this literature.
9 In an earlier draft of this paper there was a third choice, namely, whether to delegate control
of incremental innovation decisions to the local agent. This aspect of the paper built on our earlier closed-economy workon organizational choice with incremental innovation which in turn built on This choice and the common elements with have been eliminated in the current versionof the paper in order to simplify the analysis.
are interested in the choice between exporting and producing abroadfor goods whose production has already been standardized. Producing abroad allows the firmto benefit from lower wages, but at a cost: it also allows the local agent to passively
absorb thefirm's technology and steal it. This is an important insight. In contrast, we are interested in active
participation of local agents in the incremental innovation of new goods.
There are also some interesting comparisons between our work and that of
and Their models elegantly focus on search and matchingfrictions associated with imperfect information in a single foreign market. In contrast, our modelfocuses on agency problems associated with incremental innovation in multiple foreign markets.
They are interested in how imperfect information reduces trade between the domestic country andthe foreign country whereas we are interested in the implications of incremental innovation for thechoice between foreign locations. There is however a similarity. Both business networks and agentinnovation increase the value of matching in a foreign country. This makes foreign countries moreattractive places in which to produce.
The paper is organized as follows. Section sets up the model. Section analyzes the de-
cision by a single rich-country firm about whether to involve her low-wage country partner inincremental innovation. Sections and embed this choice in a general equilibrium internationaltrade framework with multiple rich-country firms and alternative low-wage country locations.
This yields predictions about the extent to which different low wage countries will engage inincremental innovation and attract rich-country firms, as well as predictions about internationaltrade in goods, the capital account, earnings and living standards. Section concludes.
We have in mind a situation in which a us firm with an existing product has decided to produce it inChina. Changes in market conditions and available technologies force the us firm to incrementallyimprove the product or its production process so that it remains profitable. We refer to theAmerican firm as the principal, denoted by ‘she' or a subscript p
. The firm has a Chinese partner,who we refer to as the agent, denoted by ‘he' or a subscript a
. We are not concerned here with howthe us principal came to have the asset that allows her to produce (although we will discuss freeentry in section or why she has decided to produce abroad. Nor are we concerned here withthe mode of entry into China. (As discussed above, it is easiest to think of the Chinese partneras the senior manager or engineer of a us-owned subsidiary.) Our analysis focuses on the extentto which American firms involve the managers and engineers of Chinese plants in the innovationprocess. This will provide the building blocks for our general equilibrium analysis of incrementalinnovation and international trade.
The American principal can develop the incremental innovation in the United States and ask theChinese agent merely to implement it. We call this ‘Principal Innovation' and denote it by super-script PI
. Alternatively, the American principal can assign the role of developing the incremental
innovation to her agent in China. We call this ‘Agent Innovation' and denote it by superscript AI
Developing the incremental innovation requires ‘creative' effort either from the principal underprincipal innovation or from the agent under agent innovation. Let ei
be the innovation effort ofi
) where ei
∈ [0,1]. Innovation effort level ei
leads to a successful incremental innovationwith probability ei
and to no innovation with probability 1 − ei
Incremental innovations designed by the Chinese agent are different from those designed by the
American principal. In particular, interactions between interdependent components of a complexsystem imply that an incremental innovation designed by the Chinese agent creates some residualincompatibilities for the American principal. The principal must then exert some ‘debugging'effort to sort out these residual incompatibilities. The extent of residual incompatibilities is para-meterized by r
∈ [0,1]. We assume that debugging effort by the principal is the only character-istic distinguishing one incremental innovation from another. Every successfully developed anddebugged incremental innovation allows production of one unit of output. Each unit of outputyields the same total profit π
is endogenously determined in general equilibrium). Forsimplicity, no additional inputs are required other than those already provided by the principaland agent.
Both the creative effort ei
required to develop an incremental innovation and the debugging ef-
fort required to sort out residual incompatibilities r
are costly because they eat into an individual'sunit endowment of leisure. Leisure for the principal and the agent are:
1 − e
under principal innovation,
1 − r
under agent innovation,
under principal innovation,
1 − ea
under agent innovation.
Preferences are Cobb-Douglas over consumption and leisure and homothetic over all goods. Thus,indirect utility for economic actor i
's income li
's leisure and P
is the relevant price index. In the appendix we show thatnone of our propositions depend on the values of α
, so in the main text we set α
= 1,which yields particularly simple expr
Residual incompatibilities in a developing-country context
When a firm looks for a supplier in a low-wage country, a notable feature is the scarcity of inform-ation about whether the supplier can meet the expectations of the firm. Thus, the firm and thesupplier do not know r
until after they have been matched for some time and committed resources
10 Setting α
= 1 has the advantage of making our core expressions linear in r
. Otherwise they would be linear
in an increasing transformation of r
, 1 − (1 − r
. Thus, all of our propositions also hold with β
6= 1. As will becomeobvious below, the only modification needed when
6= 1 is that r
≡ 1 − 1
must be redefined as
≡ 1 − β
. See the appendix for details.
to the relationship. Evidence on this abounds. At the most general level, argue that the inaccessibility of information about enterprises in low-wage countries is the primaryreason for underdevelopment — it makes investment in a relationship with developing-countryenterprises too risky. At a more micro level, studies Chinese and Indian first-tiersuppliers to car makers such as gm. In an industry where the main purpose of incrementalinnovation is to reduce product defect rates (as in our Delphi example), 45% of Chinese suppli-ers and 57% of Indian suppliers fail to deliver the defect rate improvements needed to achieveinternational best-practice targets. It is common for a firm to take 3 years to figure out whether itssupplier can achieve these targets i.e., three years before r
is known with certainty. Indeed, describes an interview with a firm that took 5 years to arrive at this conclusion. At the endof this process of discovery the firm must make a decision about whether to leave incrementalinnovation in the hands of the supplier (agent innovation in our model) or whether to take controlof incremental innovation (principal innovation in our model).
This discussion suggests three important features that we must model. (1) There is wide
heterogeneity in the residual incompatibilities r
created by different Chinese suppliers. (2) In anymatch there is uncertainty about the extent of the residual incompatibilities, uncertainty which cantake years to resolve. Restated, r
may be unknown for a long time. (3) As a consequence of thisdelay, it is often better for a us firm in a high-r
match to take control of incremental innovationrather than to search for and switch to another supplier.
We formalize these three features in a stylized way. Each supplier is characterized by a residual
. There is supplier heterogeneity that is captured by a cumulative distributionfunction F
) over residual incompatibilities r
. The firm is matched with a supplier that is ran-domly drawn according to F
. We completely telescope the process of discovering r
is learnedimmediately after matching. What is important for this paper is that if a firm finds itself in a badmatch (i.e., a high-r
match) it is more profitable to take control of innovation than to incur the costsof searching for a new supplier and bringing him up to speed on the project. We make this extremeby assuming the former always happens. Thus, after r
is learned the firm must stay with thesupplier and choose one of two organizational forms (agent innovation or principal
We now turn to aspects of contractual incompleteness related to residual incompatibilities. Con-
tracts are incomplete in the sense that they cannot be contingent on innovation effort. The agent(Chinese supplier) has no liquid assets so the principal (us firm) cannot ask for lump-sum transfersfrom the agent. The contract only specifies what the payment to the agent will be conditional onwhether or not there is production and whether or not the agent will be involved in incrementalinnovation. After the contract is signed, r
is revealed and the principal decides whether or not tooffer the agent the choice of getting involved in innovation.
One can imagine that if the agent were involved in innovation his payment would be condi-
tional on residual incompatibilities r
. However, there are two features of incremental innovationin developing countries that lead us to rule out r
-contingent contracts. First, residual incompat-
11 In contrast, when there is a bad match in the models of and
then the principal abandons the match and returns to the domestic market. This leads to important insights about howthe level of international trade is affected by business networks and information flows. It also leads to novel insightsabout trade and inequality.
ibilities will be difficult to verify by an outside party. Second, to be able to contribute to the in-cremental innovation, the agent needs confidential technical and/or marketing specifications fromthe American firm. Much of this information will typically be non-codifiable information passedon by us managers and engineers to their Chinese counterparts. This information will almostcertainly be valuable outside the relationship. Thus, once the Chinese agent has this information,the American principal will have to pay an earnings premium in order to retain the Chinese agent.
This premium will not necessarily reflect the size of the residual incompatibilities created by theagent's innovation, but rather the value of the information that the agent has acquired by beinginvolved in innovation.
We capture this by making the payment to an agent who is involved in innovation independent
. We express this payment as a fraction λ
of profits π
. Since, conditional on production, π
is exogenous to the relationship, the issue is what determines λ
. We think about this as follows.
Once the agent has come up with an incremental innovation the agent could walk away fromthe relationship and produce on his own. However, there is a surplus if the principal and agentmaintain their relationship. The outcome of bargaining over this surplus can be described as anex-post sharing rule whereby the agent receives a share λ
of profits and the principal a share 1 − λ
For instance, suppose the principal and agent successfully develop an incremental innovation. Ifthey produce together, joint profits are π
. If they go their separate ways, the principal obtains afraction γp
of the total profits of joint production, net of the additional costs she must incur to beable to produce without the agent. Likewise, the agent obtains a fraction γa
of the total profitsof joint production, net of the additional costs he must incur to be able to produce without theprincipal. The surplus in the relationship is π
(1 − γp
), which we assume is If, forexample, we use the Nash bargaining solution with equal weights to allocate the surplus, whenthe agent is involved in incremental innovation he gets 1 (
2 1 + γa
while the principal keeps
2 1 − γa
. In this case, under agent innovation ex-post bargaining leaves the agent with a
share of profits λ
2 1 + γa
). This determines λ
. The key point is that λ
is independent of
If the agent is not involved in innovation he does not create any residual incompatibilities so
that his payment is also independent of r
. We express this payment as a wage w
It follows that income for the agent and principal, conditional on production, is
under principal innovation,
under agent innovation,
3. The choice over the agent's involvement in innovation
The level of agent involvement in innovation will depend on the earnings premium λπ
and onthe importance of residual incompatibilities r
. This section describes exactly how.
12 Note that at this point the effort levels affecting leisure have already been incurred and therefore the surplus
depends only on profits within and outside the relationship.
13 We actually do not even need that λ
is independent of r
. See footnote below.
Under principal innovation, with probability ep
an incremental innovation is developed by theprincipal which yields incomes ya
. With probability (1 − ep
) no incrementalinnovation is developed and ya
= Substituting this and into yields the followingexpected utility levels for the principal and the agent:
The problem for the principal is to choose ep
in order to maximize EUPI
, which yields ep
Substituting this equilibrium effort level into and yields equilibrium expected utility levels:
Under agent innovation, substituting and into yields the following expected utility levelsfor the principal and the agent:
The problem for the agent is to choose ea
in order to maximize EUAI
, which yields ea
Substituting this equilibrium effort level into and yields equilibrium expected utility levels:
The principal's choice
The principal chooses her preferred organizational form — principal innovation or agent innova-tion.
(Extent of the agent's involvement in innovation) The principal prefers agent innovation
to principal innovation if and only if
≡ 1 −
14 The principal is protected by limited liability, so if there is no incremental innovation and thus no production the
principal does not need to pay the agent's wage.
The principal prefers agent innovation to principal innovation if and only if EUAI
is given by equation and EUPI
Recall that agent innovation requires costly effort on the part of the agent. The agent is willing
to incur this effort because he receives an earnings premium or ‘bonus' that makes it worthwhile.
That is, the agent's earnings with innovation (λπ
) are sufficiently higher than earnings withoutinnovation (w
). When we introduce the full general equilibrium model, we establish conditions onprimitives which ensure that
From equations and this ensures that the agent prefers agent innovation over principalinnovation when this is offer
4. General equilibrium
We now embed our previous discussion into a general equilibrium model. For concreteness, leteach principal be a us auto designer or manufacturer who has decided to produce auto parts (orautos for short) in a low-wage country. Since we are not interested in the choice between producingin high-wage versus low-wage countries we assume that all auto production takes place in low-wage countries. For clarity we consider only two low-wage Recall from table andfigure that while China is increasingly involved in incremental innovation, Indonesia is not. Withthis in mind we call the two countries China and Indonesia.
There are two additional sectors: apparel and fine-chemical drug manufacture. Apparel is only
produced in China and Indonesia and serves as an alternative occupation for agents. This willbe needed to endogenize the number of agents in each country. Drugs are only produced in theUnited States and serve as an alternative occupation for principals. This is needed to endogenizethe total number of principals.
Apparel is a homogenous good produced under perfect competition with raw labour i.e.,
without innovation effort ei
. Apparel production is subject to diminishing returns to labour —think of this as capturing a fixed factor such as land. To avoid scale effects associated with the
15 Is there any additional scope for renegotiation between the principal and the agent? Consider first the case where
. The agent prefers to be involved in innovation (because this will allow him to steal the critical technology) andso might be willing to make a lump-sum payment to the principal. We have ruled this out by assuming that the agent isliquidity constrained. (Even if he had the cash, it would be an odd model if the agent were allowed to pay the principalfor the opportunity to steal the principal's technology.) Now consider the case where r
. Both parties strictly preferagent innovation to principal innovation. We have already assumed that the surplus generated by agent innovation issplit according to a Nash bargain with equal weights, which leaves the agent with a share λ
= (1 + γa
)/2 of profits.
This assumes that what the agent and principal get in the event of bargaining breakdown (γaπ
is independent of r
. This independence is not necessary for our results. In particular, if one thinks that agents in low-r
relationships also tend to have better outside options (higher γa
) then one could allow γa
to be decreasing in r
of our propositions would be altered by this provided that γa
does not decrease too rapidly in r
. (If it decreases too
rapidly then the principal perversely likes high-r
agents — their bargaining power is so weak that it dominates anyinnovation-related issues.) To see this more formally, consider figure The key restriction we require on how γa
is that neither EUAI
are increasing in r
. To recap, one could extend the model to allow for more
sophisticated renegotiation while still retaining all of our results.
16 As will become clear, it requires virtually no change in our analysis to allow for any finite number of low-wage
agent countries and high-wage principal countries.
fixed factor we assume that China and Indonesia have the same technology and are the same size,each having a workforce of size L
. We denote the (endogenous) number Chinese nationals whochoose to become agents in the auto sector by m
, so that L
is employment in the Chineseapparel sector. We choose apparel as the numéraire so that its price is unity. Let wA
) bea function denoting the marginal product and wage of labour in the apparel sector when L
workers are employed in the apparel sector. By diminishing returns to labour w
< 0. Asteriskswill denote Indonesian variables. In particular m
∗ is the endogenous number of Indonesian agentsand wA
∗) is the wage in the apparel sector. All our results refer to diversified equilibria,that is, equilibria in which both China and Indonesia produce both autos and apparel. (See theappendix for a discussion of what is required for this to be the case.)
The endogenous number of agents m
∗ must equal the endogenous number of us principals
. The principals' alternative occupation is drug manufacture which is set up analogously to
apparel. Since the size of the United States plays absolutely no interesting role we also set the uspopulation to L
. Let wD
) be the marginal product of labour in the drug sector when L
workers are employed in it. By diminishing returns to labour w
We next turn to returns in the auto sector. Recall that in the partial equilibrium model there
is a cut-off ¯r
such that when r
is below ¯r
the agent is engaged in incremental innovation andwhen r
is above ¯r
the principal takes charge of incremental We now assume thatthere is heterogeneity in the value of residual incompatibilities r
across principal-agent matches.
This generates heterogeneity in organizational forms and agents' involvement in Tomodel heterogeneity we assume that r
is randomly drawn from a distribution with support [0,1].
∗ be the cumulative distribution functions of r
in China and Indonesia, respectively.
∗ are assumed differentiable. We focus our analysis on the case in which agentinnovation and principal innovation co-exist in one and possibly both low-wage countries. Thisrequires that the cut-off ¯r
defined by equation lies strictly between 0 and Principals andagents must decide whether to enter the auto sector and, in the case of principals, whether to enterChina or Indonesia. Only after this decision is made are residual incompatibilities within matchesdiscovered. Taking expectations over r
, expected utility from entering the Chinese auto sector isgiven by
The Indonesian equivalent of equation replaces F
We can now state the conditions for factor-market clearing. Consider equation from the
perspective of a Chinese agent. Substituting in the values of EUAI
given respectively by
depends on the earnings of agents under different organizational forms, something which we now endogenize.
18 This heterogeneity has the flavour of where exporters and non-exporters coexist because of cross-plant
differences in productivity. Likewise, use Melitz's productivity heterogeneity to generatethe coexistence of integrated home production, domestic outsourcing, fdi and offshore outsourcing.
19 The appendix discusses parameter restrictions required to simultaneously ensure both this and the agent participa-
tion constraint in equation
and yields expected ex ante
utility for a Chinese national from being in the auto sector:
) = 1
)) + 1 [1
where, from proposition
) = 1 − 1
. The Indonesian equivalent of equation
∗ and w
∗ to yield EUa
∗). If China produces both apparel and autos
then its nationals must obtain the same ex-ante
expected utility whether they work in the apparel
sector or are agents in the auto sector. Utility from working in the apparel sector is wA
Chinese agent is ex-ante
indifferent between working in the apparel and auto sectors when
) = wA
The corresponding indifference condition for Indonesian nationals is
∗) = wA
These two equations are central to our
Turning to the principals' occupation and location decisions, each American principal must
employ either a Chinese or Indonesian agent to produce one unit of autos. In choosing between
locating in China and Indonesia, each principal compares ex ante
expected returns of entering each
country. Given that all successfully developed and debugged innovations yield the same total
profit, from the point of view of a principal there are only two differences between China and
) the distributions F
∗ and (b
) auto sector wages, w
∗. Substituting the
values of EUAI
in equations and respectively, into yields expected ex ante
utility for an American principal from entering the Chinese auto sector:
) 1 (1
(1 − r
) + 1
[1 − F
Note that r
enters equation in two ways. From proposition r
determines whether each
agent is involved in innovation or not. Second, under agent innovation EUAI
the costs of debugging residual incompatibilities r
. The principal's corresponding return from
entering Indonesia is EUp
∗). If principals operate in both China and Indonesia, expectedex-ante
returns must be equalized across the two countries:
) = EUp
This equation is also central to our analysis.
20 Agents who choose the auto sector and get a low draw of r
better off than workers in the apparel sector,
whereas agents who get a high draw of r
worse off than workers in the apparel sector. As in models à lawe assume that after r
is realized it is not possible for workers with a bad draw to relocate tothe apparel sector. Alternatively, we could introduce the need to obtain a degree prior to working in the auto sector andfinding out the realization of r
. The earnings of agents would then have to be higher in order to compensate for the costof an education. For a high enough cost of education, even an agent in a high-r
match will be ex post
better off in autosthan in apparel.
In addition to becoming a principal in the auto sector, us nationals also have the choice of
working in the drug sector. A us national is ex-ante
indifferent between working in the drug andauto sectors when
) = PD
is the price of drugs.
This completes the discussion of factor markets. Agents are indifferent between autos and
apparel. Principals are indifferent between autos in China, autos in Indonesia, and drugs in theUnited States.
Finally, we need conditions for product market clearing. We make four additional assumptions.
(a) Demand for apparel and drugs slopes downwards. (b) Autos are differentiated goods with asymmetric imperfectly competitive market structure. (c) Each auto firm that ends up producingearns common profits π
that are decreasing in the number of producers. (d) Preferences areinternationally identical and homothetic. Assumption (a) together with diminishing returns tolabour in apparel and drugs ensure that entry into a low-wage country's auto sector increasesreturns in the local apparel sector and in the us drug sector, thus increasing the attractiveness ofthe alternative occupations of both local agents and us principals. Assumptions (b) and (c) ensurethat entry into auto production does not make further entry even more attractive. Together with(d), assumptions (b) and (c) are required only to determine equilibrium trade flows in corollary and some secondary results related to entry in proposition
While this structure may seem complicated, our key results about the location choices of Amer-
ican principals, local involvement in incremental innovation, wages and well-being are derivedfrom only three equations: and This is because all we need to know is howdifferences between F
∗ lead to differences in wages (w
∗) and differences in the locationof American principals (m
∗). All of our core results flow from these two inequalities.
5. International trade and incremental innovation
The difference between China and Indonesia
Our choice of country labels reflects the fact that both China and Indonesia are low-wage recipientsof substantial foreign investments. However, unlike Indonesia, China has become a prominentlocation for first production of new innovative goods (table and for inventors listed in patentsowned by us multinationals (figure
To bring the results into stark relief we allow for only one difference between China and
Indonesia. When a principal goes to China, she is less likely to face large residual incompatibilities.
Mathematically, we assume that F
∗ first-order stochastic dominates F
. This means that F
∗ isright-shifted relative to F
∗ 6 F
for all r
. To avoid the uninteresting possibility that F
∗ = F
we assume that F
∗ < F
= As is well known, first-order stochastic dominanceimplies R 1 u
) for any non-increasing function u
21 More formally, dF
when evaluated at r
Although we believe the assumption that F
∗ first-order stochastic dominates F
to be the obvious
one, it is worth reviewing the evidence for what the Chinese distribution looks like relative tocountries such as Indonesia, Thailand and the Philippines. First, relative to Indonesian engineers,Chinese engineers receive training that allows them to work more effectively with us engineers.
Part of this is the high quality of Chinese engineering schools. The Times Higher Education Supple-ment
places 2 Chinese engineering schools in the top 15 worldwide and another 6 in the top 100. Incontrast, no Indonesian, Thai or Philippine school makes this Also, there is a large number ofChinese nationals who graduated from us engineering schools and moved back to China. Amongforeign-born scientist and engineering students who are enrolled in us schools but have no firmplans to stay in the United States, 25% are from China whereas only 1% are from Indonesia,Thailand and the Philippines The large number of Chinese with us engineeringdegrees makes it easier to initiate contacts (credentialism) and communicate engineering solutions.
Second, Chinese engineers likely have better specific industrial training than their Southeast
Asian counterparts. They have been nurtured by the Chinese diaspora in Hong Kong, Taiwan,Singapore and the United States. This diaspora has invested heavily in bringing Chinese manu-facturing plants up to snuff. shows that the large ethnic Chinese and ethnic Indianresearch communities based in the United States have greatly facilitated knowledge diffusionand increased output in innovative sectors back in China and India. (Kerr uses data on ethnicinventor names appearing in us patents, data on foreign citations to these patents, as well as dataon migration and production patterns.) There is also significant evidence that Chinese plants areadopting Western management techniques which emphasize quality control and information flow.
This can be seen, for example, in the prevalence of iso 9001 certificates, a standard reference forquality management practices in business-to-business dealings. As of December 2003, China hada stock of almost 100,000 iso 9001 certificates compared to only 3,449 for Indonesia, Thailand andthe Philippines combined. Even controlling for differences in country size, this is a huge differencein
Third, China is a major fdi destination for us firms not just because of its low-wages, but also
because of the size of its internal market. This means that firms in China are producing for domesticconsumption, a fact that puts Chinese engineers in closer proximity to customers. For instance,Nokia designed its 6108 handset in Beijing to optimize its Chinese text messaging capabilities. Ithas since had several other handsets designed in Beijing for different markets. The size of China'smarket and the nascent sophistication of its consumers flips argument on its head.
Proximity to discerning consumers is a key driver of Vernon's argument for why innovation occursin rich countries. Now it is an argument for why incremental innovation occurs in China.
For these three reasons it is appropriate to assume that F
∗ first order stochastic dominates F
Note that all three reasons apply almost as much to India as they do to China.
22 Rankings are published in Times Higher Education Supplement
, 5 November 2004, available at
23 Authors' calculations based on data in and 24 Data are from
Wages as an equilibrating mechanism
The wage w
paid under principal innovation is the key equilibrating mechanism in our model. Ourfirst result discusses equilibrium wage differences in the auto sector between China and Indonesia.
(Cross-country wage differences) Suppose F
∗ FOSD F. Then in any diversified equilib-
rium, auto wages are higher in China than in Indonesia, w
Suppose, contrary to proposition that w
∗. Differentiating equation yields
= −[1 − F
By assumption, F
places positive mass in the neighbourhood
= 1 so that F
) < 1 and ∂EUp
< 0. Hence w
∗ implies EUp
) > EUp
(1 − r
) for r
) = 12
) = 14 P
(See equations and Note that up
) is a non-increasing function of r
and is strictly
decreasing on [0, r
). Since F
∗ FOSD F
, the definition of first-order stochastic dominance impliesEUp
) = R 1 u
) > R 1
) = EUp
∗). Strict inequality comes
from the fact that, by assumption, dF
, which means that there is a subinterval
on which both F
∗ < F
is strictly decreasing. Combining EUp
) > EUp
) > EUp
∗) implies EUp
) > EUp
∗), a violation of equation Hencew
The basic insight is straightforward. Consider the top panel of figure It plots the expected
utility for principals facing different values of residual incompatibilities r
in China (solid line)
and Indonesia (dashed line). That is, the solid line plots up
), which equation defines asEUAI
) to the left of r
) to the right of r
. The dashed line is the corresponding curve for
∗). Note that EUp
) and EUp
∗) are integrals over r
of these curves. A
core feature of the principal's problem is that her expected utility is decreasing in r
under agent in-
novation. This reflects the fact that the principal prefers working with an agent whose involvement
in innovation creates fewer residual incompatibilities. Since, as illustrated in the bottom panel of
figure Indonesia's distribution of residual incompatibilities F
∗ first order stochastic dominates
, if w
∗ then each principal would strictly prefer China over Indonesia. To ensure
that each principal is ex-ante
indifferent between locating in these two countries, lower wages are
needed in Indonesia to offset the higher expected residual incompatibilities.
Involvement of Chinese and Indonesian agents in innovation
We saw empirically in figure that Chinese agents are more involved in incremental innovationthan their Indonesian counterparts. We now show that the smaller expected residual incompat-ibilities created by Chinese agents imply that a higher fraction of Chinese agents is involved in
25 We are using the fact that r
) is defined to satisfy 12 (1 − λ
(1 − r
) = 14 (π
). Thus, the derivative of EUp
with respect to r
Figure 2: Principals' utility, agents' utility and distribution of residual incompatibilities
innovation. More remarkably, even among agents who create identical residual incompatibilit-ies, Chinese agents are more involved in incremental innovation than their identical
(General equilibrium involvement in innovation) Suppose F
∗ FOSD F. Then in
any diversified equilibrium r
) > r
∗). This implies the following. (1) Chinese agents have a higher
probability of being involved in incremental innovation than Indonesian agents. (2) Consider a Chinese
match and an Indonesian match that have identical residual incompatibilities r with r
∗) < r
Then only the Chinese agent will be involved in incremental innovation. The Indonesian agent will not be.
Further, the Chinese agent will be paid more and have higher utility than the Indonesian agent.
By equation ∂r
> 0. By proposition w
∗, so that r
) > r
Consider part (1). Since r
) > r
∗), first order stochastic dominance implies that F
more mass on the interval [0,r
)) than F
∗ places on the interval [0,r
∗)). That is, Chinese
agents have a higher probability of being involved in innovation. Consider part (2). The result
that only the Chinese agent will be involved in incremental innovation follows immediately from
proposition and r
∗) < r
). Regarding utility and income, agents' voluntary participation
in innovation implies that EUAI
) > EUPI
), i.e., that equation holds. Since w
) > EUPI
∗). Hence, EU AI
) = 1
> 12 P
∗), i.e., the Chinese agent has
higher utility. Consequently, λπ
∗, i.e., the Chinese agent is paid more.
Part (2) of proposition and to some extent part (1) operate through the endogenous general
equilibrium wage differences between China and Indonesia. Agents performing basic auto tasksare paid more in China than in Indonesia so that the additional monetary cost of including workersin incremental innovation is lower in China (λπ
∗). Thus, the maximum residualincompatibilities under which agents are involved in incremental innovation is also higher inChina: r
) > r
∗). Therefore, even agents who create identical
residual incompatibilities aremore involved in incremental innovation in China than in Indonesia. This is a general equilibriumeffect and it highlights that the cost of protecting intellectual property rights depends not only onformal institutions but also on income levels. Here formal institutions are part of what determines
, the share of profits that agents can capture by threatening to run away with the knowledgeacquired while involved in incremental innovation. We have assumed that λ
is the same inChina and Indonesia. Nevertheless, the lower wage for workers performing basic auto tasks inIndonesia means that Indonesian workers have to be paid a higher earnings premium not to tryto appropriate the profits from a successful incremental innovation. As a result, workers with avalue of r
that would have seen them involved in incremental innovation in China are kept out ofthe innovation process in Indonesia.
We could easily introduce institutional differences across countries by having λ
China and Indonesia. In that case, a higher λ
would reduce r
(see equation and, throughgeneral equilibrium interactions, also wages (by a similar argument to proposition and the shareof auto firms locating in the country (by a similar argument to proposition below). Thus coun-tries with weaker protection of intellectual property (higher λ
) are less involved in incrementalinnovation, have lower income, and attract less foreign direct investment.
The location of American principals
The better distribution of Chinese agents makes China a more attractive location than Indonesiafor the same wages. However, wages are lower in Indonesia and this offsets China's advantagefrom the point of view of principals. We now wish to see what the trade-off between a betterdistribution of agents in China and lower wages in Indonesia means for the relative number ofprincipals in each country. The next proposition states that China must end up with more thanhalf of all principals.
(Location of American principals) Suppose F
∗ FOSD F. Then in any diversified
equilibrium, more American principals locate in China than in Indonesia, m
We know from proposition that auto wages are higher in China than in Indonesia (w
∗). We first show this implies that auto-sector expected utility (before r
is know) is also higher in
China. We use this to show that apparel wages are also higher in China: wA
) > wA
By diminishing returns this implies L
∗ or m
∗. Differentiating equation yields
) = 1 [
1 − F
− 2 P
By assumption, F
places positive mass in the neighbourhood of r
= 1 so that F
) < 1. The term
in brackets, [λπ
], is EUAI
) − EUPI
) which is non-negative (otherwise the agent
would turn down the principal's request to assist in innovation). Since F
is a cumulative density
> 0. By equation ∂r
> 0. Hence, ∂EUa
> 0. By proposition
∗ so that EUa
) > EUa
) = 12 P
Note that, by equation ua
) is a non-increasing function of r
. Since F
∗ FOSD F
the definition of first-order stochastic dominance implies EUa
) = R 1 u
) = EUa
∗). Combining this inequality with EUa
) > EUa
) > EUa
∗). From equations and
) = EU
) > EUa
∗) = wA
) > wA
By diminishing returns to labour w
< 0. Hence m
(Apparel wages and utility) Suppose F
∗ FOSD F. Then in any diversified equilibrium,
wages in the apparel sector are higher in China than in Indonesia, wA
) > wA
∗), and expected
utility in both the apparel and auto sectors is higher in China than in Indonesia.
The basic insight works off the agent's indifference between the apparel and auto sectors.
Consider the middle panel of figure which is the agent's counterpart to the top panel. It plots
the expected utility for agents facing different values of residual incompatibilities r
in China (solid
line) and Indonesia (dashed line). That is, the solid line plots ua
), which equation defines
to the left of r
) to the right of r
. The dashed line is the corresponding curve
for Indonesia, ua
∗). Note that EUa
) and EUa
∗) are integrals over r
of these curves.
Since Chinese agents are paid more than Indonesian agents under principal innovation (w
∗), the Chinese profile lies above the Indonesian profile for r
). Moving left of r
)the profile jumps up because of the agent's participation constraint. This raises the Chinese profileeven higher above the Indonesian profile in the interval r
)]. Thus, Chinese agents arebetter off than Indonesian agents both because they have a higher profile and because the Chinesedistribution puts more weight on the higher outcomes to the left. Since an agent's returns mustbe equalized across the apparel and auto sectors, Chinese apparel wages wA
) must alsobe higher than Indonesian apparel wages wA
∗). In general equilibrium, for wages to behigher in China, there must be more agents and hence more American principals in China than inIndonesia i.e., m
∗. As a corollary, higher Chinese apparel wages also imply that utility in theapparel sector is higher in China than in Indonesia i.e., wA
. Finally,since agents are indifferent between sectors, expected utility in the auto sector must also be higherin China than in Indonesia.
Notice that a larger share of principals go to where agents are most expensive. This result never
occurs in standard product-cycle models where us principals always locate in the lowest-wagecountry. In our model us firms take into account wages and
the ability of local agents to participatein incremental innovation. In general equilibrium, there are sufficient additional principals inChina relative to Indonesia to make wages differ just enough to offset China's advantages inincremental innovation. These advantages stem from the lower average residual incompatibilitiescreated by Chinese agents and also from the lower general equilibrium earnings premium Chineseagents must be paid to prevent them from walking away. Higher wages and greater chancesof becoming involved in incremental innovation attract more Chinese nationals into the autosector, where they are better off than their Indonesian counterparts. Note also that the higherincome and welfare of Chinese agents are driven by more than just China's lower average residualincompatibilities. Open-economy general equilibrium wage adjustments are central. In fact, to usestandard international trade terminology, there is no conditional factor price equalization: agentswho create identical residual incompatibilities can end up involved in incremental innovation andearning a premium in China but not in Indonesia.
Production patterns and trade flows
We finally establish the pattern of world production and trade flows. There will be trade in goodsas well as a capital account transaction, namely, repatriated profits from us activities in China.
Since China has a larger number of American investments, China has more auto productionand less apparel production than Indonesia. These production patterns together with identicalhomothetic preferences and the assumptions we made about market structure imply the followingresult about equilibrium trade flows.
(Trade flows) Suppose F
∗ FOSD F. Then in any diversified equilibrium trade flows are
characterized as follows:
Autos: The United States imports more autos from China than from Indonesia. China and Indonesia
export autos to each other, but China is a net auto exporter to Indonesia.
Apparel: Indonesia exports apparel to the United States. If China also exports apparel to the United
States, it exports less than does Indonesia. If China does not export apparel to the United States, itimports apparel from Indonesia.
Drugs: The United States exports drugs to both China and Indonesia.
Capital Account: The United States runs a trade deficit that is financed with a capital account sur-
plus against both China and Indonesia. This capital account surplus comes about because principalsrepatriate to the
us their share of profits from auto sales in China and India.
6. From the product cycle to the rise of innovation in low-wage countries
In January 1992, China's Deng Xiaoping visited the nascent special economic zone of Shenzhenas part of his now famous Nanxun
or Southern Tour. His purpose was revolutionary — to praisethe efficiency of capitalist firms operating in this and similar zones. He announced the expansionof the export-processing zone program and the liberalization of the foreign investment regime toallow more foreign companies to operate in China. The rest is history. The new investment regimehas led to massive entry of foreigner entrepreneurs into Chinese manufacturing. In addition, asshown in table and figure Chinese manufacturing facilities are increasingly moving beyondthe production of mature goods and getting more and more involved in incremental innovation.
In this section, we examine the consequences of China's opening up to foreign investment.
Consider a world in which initially all American principals locate in Indonesia because they are
not allowed into China. Suppose now that Deng Xiaoping makes his Southern Tour and Americanprincipals are allowed to enter China. Once China opens up, what is the equilibrium distributionof American principals between China and Indonesia, to what extent are agents in each countryinvolved in innovation, and what are the consequences of this for income levels and trade patterns?
(The consequences of China's opening up) Suppose F
∗ FOSD F. Suppose that, starting
from an initial situation where all American principals locate in Indonesia because they are not allowed
into China, China opens up to American firms. Once American principals are allowed to enter China, in
any diversified equilibrium Indonesia ends up with fewer American principals and lower wages in both
sectors than before China's opening up to American firms. Relative to Indonesia, China ends up with more
principals and higher wages in both sectors. Chinese agents are involved in incremental innovation more
often than their Indonesian counterparts. Even agents creating residual incompatibilities r
))are involved in innovation in China but not in Indonesia.
Initially, Indonesia hosts all principals. After China's opening to American principals, by
proposition Indonesia hosts less than one half of all principals. If the total number of principals
has not increased, Indonesia must have fewer principals than before. We next show that the same
result holds even if the total number of principals increases following China's opening. Entry of
principals increases their earnings in the alternative occupation. To reestablish indifference of us
nationals between becoming principals and engaging in an alternative occupation, EUp
must be higher than before China's opening. By assumption, entry of principals lowers auto sector
. By equation this tends to reduce EUp
∗). Thus, the only way for EUp
to end up being higher than before China's opening is by having auto wages in Indonesia w
(recall that EUp
∗) is decreasing in w
∗). Recall that EUa
∗) is increasing in w
∗, so the
reduction of auto wages in Indonesia makes Indonesian agents worse off. By equation so does
the fall in profits. Thus, to keep Indonesian nationals indifferent between being agents in the auto
sector and working in their alternative occupation in the apparel sector, Indonesian apparel wageswA
∗) must also fall. Since w
< 0, this implies that the number of principals operating
in Indonesia m
∗ is lower after than before China's opening. The higher Chinese wages and the
greater Chinese involvement in innovation follow from propositions and
Prior to China's entry, the world looks close to a product cycle model: most innovation is
done in the United States and the production of the standardized good is done in Indonesia.
After China's opening up to American investments, most American firms locating in Indonesiaand some of those locating in China continue this product-cycle pattern of developed-countryinnovation followed by low-wage standardized production. However, many American firmsbegin involving their Chinese agents in incremental innovation. These were precisely the patternsimplied by our numbers on first locations of innovative new products and on patents in theintroduction.
Recall from figure that the rise of Chinese and Indian patenting has coincided with a plat-
eauing of or even a decline in patenting by Indonesia, the Philippines, Thailand and Mexico.
Proposition explains this. It states that the appearance of China on the world scene has a negativeimpact on investment in Indonesia. Notice that the proposition does not
state that China's entryinto world markets reduces Indonesian welfare. We have not tracked any of the traditional gainsfrom trade so it is still possible that Indonesia benefits from China's entry. Our main point issimply that these traditional gains from trade for Indonesia will be offset, at least in part, by thedeparture of American principals from Indonesia. This problem has been commented on by manyin the press who point out that many firms are moving operations to China from other low-wagecountries such as Indonesia and Mexico.
To our mind, a central feature behind the recent success of China and India in international marketshas been the ability of these countries to deliver shop-floor incremental innovation to foreignbuyers operating complex supply chains. Firms in rich countries need their suppliers to producehigh-quality goods — goods that are reliable, have low failure rates and incorporate the latestdemands of an ever-changing marketplace.
In the old product-cycle view, all innovation, including incremental innovation, is done in the
North. The Northern-designed factory is shipped to the South without any ensuing technicalproblems. However, the claim that all innovation is done in the North is no longer tenable. Weprovided the first systematic evidence on incremental innovation in low-wage countries usingdata on the location of first production for innovative new goods and data on patents developedby low-wage country inventors for us corporations.
To model this phenomenon we introduced the notion of residual incompatibilities. We showed
that if residual incompatibilities were above a threshold r
then the foreign agent did not engagein incremental innovation. Below the threshold there was incremental innovation. This resultedin a model in which the heterogeneity of residual incompatibilities within matches induced aheterogeneity of incremental innovation across countries and even within countries.
Differences across countries are due to differences in the distribution of residual incompatibil-
ities i.e., due to differences between F
∗. Improvements in information and communicationtechnologies will allow principals to better communicate with agents and reduce residual incom-patibilities everywhere. Thus, we expect to see a trend where product-cycle trade is replaced tosome extent by trade in which agents are involved in incremental innovation and new productsare first produced in low-wage countries. However, some countries such as China and India havedeveloped world-class engineering schools, send more students for training to the United States,have a more committed diaspora, emphasize standards and quality control, or get a head-start inworking with foreign multinationals by developing products catered to a particularly large localmarket. All of these developments shift the distribution of matches in these countries towardslower values of residual incompatibilities much faster, and serve to simultaneously attract moreforeign firms and to increase the proportion of those firms that involve locals in innovation relativeto other low-wage countries. As a result, the liberalizing of the international trade regime in Chinaand India has led to a vast inflow of foreign investments into these countries. This has led tothe growth of increasingly sophisticated, high-quality Chinese and Indian exports. It has also ledto problems for countries such as Indonesia and Mexico that were once the major recipients ofWestern fdi.
The rise of incremental innovation in some low-wage countries is a fact of rapidly growing
importance. This paper is the first to explain its implications for international trade. As we haveshown, the implications are significant — it's time to wake up and smell the ginseng.
Our analysis has focused on the richest possible case in which (1) agent innovation and principalinnovation can both arise for some values of r
, (2) agents willingly participate in innovationwhen asked, and (3) production remains diversified in both China and Indonesia. This appendixdiscusses what is required to ensure this. By proposition a principal chooses agent innovationover principal innovation if and only if r
. Since r
∈ [0,1], with r
< 0 a principal wouldprefer principal innovation for all r
, while with r
> 1 a principal would prefer agent innovation
for all r
. Thus, agent innovation and principal innovation are both possible preferred choices fora principal for some values of r
if and only if 0 < r
< 1, where r
is given by proposition Thisrequires 0 < 1 − 1 1−w
< 1. Agents always accept this choice if and only equation
so that λπ
. These two conditions put together reduce to 2λ
− 1 < w
2 . For this to be
satisfied, we must have 2λ
− 1 < λ
2, which simply requires λ
< 23. In addition, we need w
somewhere in the interval (2λ
− 1, λ
2 . From equations and we know that the agent's
is directly related to the wage in the alternative apparel sector wA
. Thus, by changingthe endowment of (apparel-sector) land we can always shift wA
so that w
lies in the required
interval. Finally, we have also focused on situations in which both China and Indonesia keep someproduction in the auto and the apparel sectors. This is akin to the usual restriction in trade modelsof being inside the cone of diversification. To ensure apparel production in China and Indonesia,we assume wA
(0) = ∞. To have auto production in both countries, we also require differencesbetween F
∗ not to be so extreme that principals would ex-ante
prefer locating in China evenwhen Indonesian wages are zero.
In this appendix we relax the assumption of α
= 1 in the utility function of equation Thenthe principal's expected utility with principal innovation that was described by equation mustbe changed to
Optimizing with respect to effort level ep
= 1/(1 + β
). Plugging this back into theprevious equation and simplifying yields
1 + β
The agent's expected utility with agent innovation that was described by equation must be
Optimizing with respect to effort level ea
= 1/(1 + β
The principal's expected utility with agent innovation that was described by equation must
1 − λ
The principal chooses agent innovation whenever EUAI
. From equations this
inequality reduces to
≡ 1 −
All the results of the paper go through with r
of equation redefined in this way.
Aghion, Philippe and Jean Tirole. 1997. Formal and real authority in organizations. Journal of
Antràs, Pol. 2003. Firms, contracts, and trade structure. Quarterly Journal of Economics
Antràs, Pol. 2005. Incomplete contracts and the product cycle. American Economic Review
Antràs, Pol and Elhanan Helpman. 2004. Global sourcing. Journal of Political Economy
Casella, Alessandra and James Rauch. 2002. Anonymous market and group ties in international
trade. Journal of International Economics
Ethier, Wilfred J. and James R. Markusen. 1996. Multinational firms, technology diffusion and
trade. Journal of International Economics
Feenstra, Robert C. and Andrew K. Rose. 2000. Putting things in order: Patterns of trade dynamics
and growth. Review of Economics and Statistics
Grossman, Gene and Elhanan Helpman. 1991a
. Innovation and Growth in the World Economy
Cambridge, ma: mit Press.
Grossman, Gene M. and Elhanan Helpman. 1991b
. Endogenous product cycles. Economic Journal
Grossman, Gene M. and Elhanan Helpman. 1991c
. Quality ladders and product cycles. Quarterly
Journal of Economics
Grossman, Gene M. and Elhanan Helpman. 2002. Integration versus outsourcing in industry
equilibrium. Quarterly Journal of Economics
Grossman, Gene M. and Elhanan Helpman. 2003. Outsourcing versus FDI in industry equilibrium.
Journal of the European Economic Association
Grossman, Gene M. and Elhanan Helpman. 2004. Managerial incentives and the international
organization of production. Journal of International Economics
Grossman, Gene M. and Elhanan Helpman. 2005. Outsourcing in a global economy. Review of
Harris, John R. and Michael P. Todaro. 1970. Migration, unemployment and development — 2-
sector analysis. American Economic Review
Helpman, Elhanan. 2006. Trade, fdi, and the organization of firms. Journal of Economic Literature
iso Central Secretariat. 2003.
iso Survey of
iso 9001:2000 and
iso 14001 Certificates – 2003
Geneva, Switzerland: iso Central Secretariat.
Johnson, Jean M. 1998. Statistical Profiles of Foreign Doctoral Recipients in Science and Engineering:
Plans to Stay in the United States
. Arlington, va: National Science Foundation, Division of ScienceResources Studies. nsf 99-304.
Kerr, William R. 2008. Ethnic scientific communities and international technology diffusion. Review
of Economics and Statistics
Krugman, Paul R. 1979. A model of innovation, technology transfer, and the world distribution of
income. Journal of Political Economy
Lileeva, Alla and Daniel Trefler. 2007. Improved market access to foreign markets raises plant-level
productivity. for some plants. Working Paper 13297, National Bureau of Economic Research.
Lucas, Robert E., Jr. 1993. Making a miracle. Econometrica
Markusen, James R. 2002. Multinational Firms and the Theory of International Trade
. Cambridge: mit
McLaren, John. 2000. ‘Globalization' and vertical structure. American Economic Review
Melitz, Marc J. 2003. The impact of trade on intra-industry reallocations and aggregate industry
Nahmias, Steven. 2005. Production and Operations Analysis
. Fifth edition. Toronto: McGraw-Hill
Nunn, Nathan. 2007. Relationship-specificity, incomplete contracts, and the pattern of trade.
Quarterly Journal of Economics
Nunn, Nathan and Daniel Trefler. 2008. The boundaries of the multinational firm: An empirical
analysis. In Elhanan Helpman, Dalia Marin, and Thierry Verdier (eds.) The Organization of Firmsin a Global Economy
. Cambridge, ma: Harvard University Press.
Organisation for Economic Co-operation and Development. 2006. Science, Technology and Industry
. Paris: Organisation for Economic Co-operation and Development Press.
Puga, Diego and Daniel Trefler. 2002. Knowledge creation and control in organizations. Working
Paper 9121, National Bureau of Economic Research.
Puga, Diego and Daniel Trefler. 2005. Wake up and smell the ginseng: The rise of incremental in-
novation in low-wage countries. Working Paper 11571, National Bureau of Economic Research.
Rajan, Raghuram G. and Luigi Zingales. 2003. Saving Capitalism from the Capitalists: Unleashing
the Power of Financial Markets to Create Wealth and Spread Opportunity
. Princeton: PrincetonUniversity Press.
Rauch, James and Vitor Trindade. 2003. Information, international substitutability, and globaliza-
tion. American Economic Review
Rosenberg, Nathan. 1982a
. Technological interdependence in the American economy. In Inside the
Black Box: Technology and Economics
, chapter 3. Cambridge: Cambridge University Press, 55–80.
Rosenberg, Nathan. 1982b
. Learning by using. In Inside the Black Box: Technology and Economics
chapter 6. Cambridge: Cambridge University Press, 120–140.
Spencer, Barbara J. 2005. International outsourcing and incomplete contracts. Canadian Journal of
Sutton, John. 2001. Rich trades, scarce capabilities: Industrial development revisited. Proceedings
of the British Academy
Sutton, John. 2004. The auto-component supply chain in China and India: A benchmarking study.
Processed, London School of Economics.
Trefler, Daniel. 2006. Service offshoring: Threats and opportunities. In Susan M. Collins and
Lael Brainard (eds.) Brookings Trade Forum 2005: Offshoring White-Collar Work
. Washington dc:Brookings Institution Press, 35–60.
us National Science Board. 2006. Science and Engineering Indicators 2006
. Arlington, va: United
States National Science Foundation.
Vernon, Raymond. 1966. International investment and international trade in product cycle.
Quarterly Journal of Economics
von Hippel, Eric. 1988. The Sources of Innovation
. New York: Oxford University Press.
Biochimie 91 (2009) 596–603 Contents lists available at Peculiar inhibition of human mitochondrial aspartyl-tRNA synthetaseby adenylate analogs Marie Messmer Se´bastien P. Blais , Christian Balg Robert Cheˆnevert Luc Grenier , Patrick Lagu¨e ,Claude Sauter Marie Sissler Richard Giege´ Jacques Lapointe , Catherine Florentz a Architecture et Re´activite´ de l'ARN, Universite´ Louis Pasteur, CNRS, IBMC 15 rue Rene´ Descartes, 67084 Strasbourg Cedex, Franceb De´partement de Biochimie et de Microbiologie, Centre de recherche sur la fonction, la structure et l'inge´nierie des prote´ines (CREFSIP), Faculte´ des sciences et de ge´nie, Universite´Laval, Que´bec, Canada G1V 0A6c De´partement de Chimie, Centre de recherche sur la fonction, la structure et l'inge´nierie des prote´ines (CREFSIP), Faculte´ des sciences et de ge´nie, Universite´ Laval, Que´bec,Canada G1V 0A6
An examination of existing trial data on the treatment of prostate cancer using external beam radiotherapy combined with hormone therapy. Professor L.J.S.Bradbury Ph.D. 1.0 Introduction. For low or intermediate risk prostate cancer patients, external beam radiotherapy is one of the common forms of treatment available to them. In the UK, this treatment is most commonly given in fractions of 2Gy with a total dose of 70Gy. This dose combination seems to have emerged over time as being a reasonable compromise between curative efficacy and the avoidance of too severe acute and long-term side effects. However, improvements in linear accelerator technology in the form of three-dimensional conformal beam radiotherapy and intensity modulated radiotherapy have led to a situation where higher doses can be delivered without increasing these side effects. In addition, it is clear that the use of hormone therapies in conjunction with radiotherapy also improves disease free survival rates. This paper discusses the evidence in favour of changing the present UK standard practice to a more aggressive radiotherapy regime but one with a shorter period of hormone use. Firstly, in sections 2 and 3, it is shown that the use of LHRH agonists like goserelin (i.e. Zoladex as its trade name) used in conjunction with external beam radiotherapy leads to very significant improvement in disease free survival. In particular, it is the use during and after the radiotherapy (concurrent and adjuvant use) that leads to the largest improvement and this improvement can be achieved by a period of treatment measured in months rather than the years previously thought necessary. There seems to be less benefit (if any) from the neo-adjuvant use of hormones but nonetheless it is probably worthwhile if only to reduce the prostate size to a more compact target. The reduction in risk of disease recurrence from the adjuvant use of hormones seems to be around the 50% level. The question of the relative merits of bicalutamide (i.e. Casodex as its trade name) as an alternative to goserelin is briefly discussed in section 4. Section 5 then discusses the effect of the overall radiotherapy dose on disease free survival and it is argued that there is a worthwhile gain to be achieved by increasing the dose to between 76 and 80Gy which should not lead to a significant increase in side effects provided either 3-D conformal or IMRT techniques are used. Compared with a 70Gy dose, this would lead to an improvement in the five-year disease free survival probability for an intermediate risk patient from around the 70% level to the region of 80% from the dose increase alone and a further improvement to around 90% from the use of adjuvant hormones. Some comments on possible synergy between hormones and radiotherapy are given in section 6 and the possible advantages of a lower level of radiation being applied to the whole pelvic region is discussed in section 7.