Working Paper What happened to foreign outsourcing when firms went online?

Transcription

1 econstor Der Open-Access-Publikationsserver der ZBW Leibniz-Inormationszentrum Wirtschat The Open Access Publication Server o the ZBW Leibniz Inormation Centre or Economics Hanley, Aoie; Ott, Ingrid Working Paper What happened to oreign outsourcing when irms went online? Kiel Working Paper, No Provided in Cooperation with: Kiel Institute or the World Economy (IW) Suggested Citation: Hanley, Aoie; Ott, Ingrid (2012) : What happened to oreign outsourcing when irms went online?, Kiel Working Paper, No This Version is available at: Nutzungsbedingungen: Die ZBW räumt Ihnen als Nutzerin/Nutzer das unentgeltliche, räumlich unbeschränkte und zeitlich au die Dauer des Schutzrechts beschränkte einache Recht ein, das ausgewählte Werk im Rahmen der unter nachzulesenden vollständigen Nutzungsbedingungen zu vervielältigen, mit denen die Nutzerin/der Nutzer sich durch die erste Nutzung einverstanden erklärt. Terms o use: The ZBW grants you, the user, the non-exclusive right to use the selected work ree o charge, territorially unrestricted and within the time limit o the term o the property rights according to the terms speciied at By the irst use o the selected work the user agrees and declares to comply with these terms o use. zbw Leibniz-Inormationszentrum Wirtschat Leibniz Inormation Centre or Economics

2 What happened to Foreign Outsourcing when Firms went Online? by Aoie Hanley and Ingrid Ott No June 2012

3 Kiel Institute or the World Economy, Hindenburguer 66, Kiel, Germany Kiel Working Paper No May 2012 What happened to Foreign Outsourcing when Firms went Online?* Aoie Hanley a and Ingrid Ott b Abstract The possibility to outsource over the internet should revolutionize oreign outsourcing, especially or services (UNCTAD, 2004). Our model describes materials and services input allocation rom domestic vs. oreign suppliers. Allocations change when irms outsource online due to access and competition eects. Using data or 99 irms who started outsourcing online in 2003 together with a control group (never outsourcing online) o over 682 Irish irms, we apply OLS and Propensity Score Matching with Dierence-in-Dierences to ind that percent o oreign services inputs growth arises rom online outsourcing. Keywords: International Outsourcing, Propensity Score Matching, Input Price Uncertainty, Input Demand JEL: L23 Aoie Hanley a Corresponding author: Institute or the World Economy / Hindenburguer 66 D Kiel, Germany phone: / Fax: / aoie.hanley@iw-kiel.de Homeage: Ingrid Ott b Karlsruhe Institute o Technology (KIT)/ Institute or the World Economy Geb , 2. OG Schlossbezirk 14 D Karlsruhe ingrid.ott@kit.edu Homepage: This version: 25 May 2012 * We would like to thank our sources at Forás or assistance with the data. Also Eckhardt Bode, Frank Bickenbach, Sebastian Braun, Holger Görg and Maik Heinemann, Jacques Mairesse and discussants at the 2010 Verein ür Socialpolitik conerence and Innovation: From Europe to China workshop at Kiel Institute or the World Economy, O course, all remaining errors are our own The responsibility or the contents o the working papers rests with the author, not the Institute. Since working papers are o a preliminary nature, it may be useul to contact the author o a particular working paper about results or caveats beore reerring to, or quoting, a paper. Any comments on working papers should be sent directly to the author. Coverphoto: uni_com on photocase.com 1

4 1 Background The introduction o the Internet was expected to revolutionize the way in which services are exchanged. Many services such as legal, inancial, R&D or technical services are traditionally exchanged at a distance but technical constraints, habits or customs had hitherto prevented this rom happening (p.148, UNCTAD, 2004; Bhagwati et al., 2004). These technical constraints have largely been removed with improvements in inormation technology, e.g. online payments, aster Web connections and better user interaces and visual cues. Economists are silent about the predicted eects or materials outsourcing. But should it automatically ollow that the outsourcing o international materials has remained unaected by the new possibilities o transacting online? Materials are ater all tangible and their quality can be more easily gauged at the time o purchase. Take the example o a textile manuacturer. A product like a chemical pigment can be identiied by its chemical ormula and other physical attributes. Thereore, the textile irm should have a deinite idea o the price-quality relationship when purchasing units o the dye. This is not the case or a consultant advising on the organization o the textile irm s inancial accounts. Here the price-quality relationship is more ambiguous, communications and troubleshooting are likely to be ongoing and only ex-post can the textile irm say with certainty that the service delivered value or money. It is by now generally accepted that outsourcing increased signiicantly over the last years, an increase arguably helped by the role o the internet in acilitating exchanges (UNCTAD, 2004; Bhagwati et al., 2004). The consensus among economists is that oreign services patterns are highly sensitive to this increased scope or arms length exchange. Amiti and Wei (2006) recorded growth rates o 6 percent in oreign services outsourcing as early as 2000, largely pre-dating the growth in online transactions. This growth was largely based on the low starting-levels or services in the economy vis-à-vis outsourced materials. Hijzen et al. (2010) recently show that outsourcing is so expensive in terms o transaction costs that, on average, in-house production is more eicient. Transaction cost economics spells out how technological advances such as ICT (Inormation and Communication Technologies) can reduce these transacting costs (Milgrom and Roberts, 1990; Grossman and Hart, 1986). The scope to streamline irm-supplier exchanges with increased internet use, makes outsourcing generally more attractive or irms looking to outsource oreign inputs. This is especially true or non-core inputs which garner only a small share o the irm s overall production cost (Acemoglu et al., 2010). 1 Abramovsky and Griith (2006) show how proit maximizing irms set out to minimize their outsourcing costs, costs which are minimized as the number o buyer-supplier online exchanges grows. Their estimations or a 1 Somewhat analogously, Bartel et al. (2005) show that what matters most in explaining outsourcing dierences between irms are changes in external technologies because the pace o change makes it more cost-eective or irms to outsource rom specialized outside irms. The increase in outsourced Sotware and Accounting services is 8 percent when ICT intensity increases by 3.8 percent. 2

5 cross-section o UK irms show that irms which use online outsourcing are 6 percent more likely to outsource oreign services than irms conducting no online outsourcing. The above studies show the predicted and actual eects impacts on outsourcing as transaction costs all. But there are other reasons to expect changes in the quantities o oreign outsourced inputs: namely the introduction o irms to a wider supplier pool. While advances in ICT help both domestic and oreign outsourcing, it is arguably the widening availability o international input suppliers (many more irms coupled with intensiied competition) that tips the balance towards oreign intermediates. Moreover, international suppliers take advantage o cheap labour which can be internalized in the input price (Lileeva and van Biesebroeck, 2008). The result: cheaper inputs rom oreign suppliers. But the appetite o irms or oreign inputs may have more to do with growth and capacity constraints than a preerence or oreign inputs as revealed in labour demand unctions by Falk and Koebel (2002). 2 I so, estimations must control or a irm s sales growth i anything meaningul can be said about overall impacts o ICT improvements on oreign outsourcing. Summing up: any reduction in transactions costs (e.g. online outsourcing) is expected to tip the balance towards oreign outsourced inputs. Other actors (i.e. widened supplier availability or better value or money rom oreign suppliers whose prices internalize cheaper overseas labour) can aect all domestic irms equally, even those that do not switch to online outsourcing. Moreover, economists predict positive impacts o internet improvements or quantities o oreign outsourced services, not materials. It is time to organize these stylized acts into a simple but coherent model. We stylize the input outsourcing set-up, allowing irms to grow and increase their input demand (oreign vs. domestic inputs) irrespective o whether they switch to internet procurement or not. Separate impacts o the internet or both internet switchers (treatment group) and non-switchers (control group respectively) are derived. In sum, our simple model o input demand describes our components o the oreign vs. domestic outsourcing decision; 1) dierences in domestic vs. oreign input prices internalizing cheaper oreign input, 2) uncertainty about whether this actor price correctly relects value or money o the input, 3) the irm s attitude towards risk and 4) supplier-purchaser transactions, o which (at least or materials) transport cost is a major component. Internet improvements increase the visibility o oreign services and material inputs. Nevertheless, uncertainty remains as to whether oreign services coner value or money at the listed price. 3 Moreover, although 2 The Falk and Koebel (2002) paper arguably pre-dates the boom in Internet procurement and it would have been interesting to observe whether the weak domestic/oreign substitution eect would be seen or post-2000 data. 3 The price-quality relationships or oreign outsourced services can be weak. To illustrate: consider the outsourcing o oreign services by the Cisco (US) rom the supplier Huawei (China). Houseman (2007) reports that the amount o work hours required by Huawei s engineers to develop the product is roughly double that required by Cisco engineers (purchasing irm), but because labor costs o Chinese engineers are dramatically lower than those o their American counterparts, the cost o R&D in China is about one-ith that in the United States (Houseman, 2007; p. 14). Accordingly, a potential outsourcer cannot easily benchmark the price o oreign vis-à-vis domestic services, due to productivity dierences. Some Indian IT suppliers are hiring US 3

6 the cost o transacting exchanges alls, transport costs remain or materials inputs. Thereore, while we expect the (optimal) ratio o oreign to domestic outsourced inputs to change with the switch to online outsourcing, the direction o this change is ambiguous. This is because uncertainty about value or money (aecting mostly service inputs) may be oset by transport costs (aecting mostly materials inputs). We then take our model to Irish data or a sample o 99 irms who started outsourcing online in 2003 together with a control group (never outsourcing online) o over 682 irms rom the EU harmonized ICT usage survey and the E-commerce survey. 4 The questions we tackle are (i) whether the growth in the oreign/total input ratio is higher or irms initiating online outsourcing and (ii) whether some o these growth dierences are attributable to the irm s switch to online outsourcing. By inputs, we mean both service and material inputs. This is the gap that our paper aims to ill. To do this we use a standard regression methodology as well as checking the robustness o our estimations using a Propensity Score Kernel Matching model with a Dierence-in-Dierences (DID) estimator. This latter robustness check helps eliminate variation in oreign outsourcing growth rates caused by other observable variables which co-determine the irm s move to online outsourcing (e.g. worker skills, irm size, labour productivity, ownership status or sector). Moreover, the combined DID methodology allows us to net out growth dierences arising rom common shocks (e.g. generally increasing the supply o traded services). 5 Applying OLS and Propensity Score Matching with Dierence-in-Dierences we ind that the move to online outsourcing is associated with a percent increase in oreign outsourced services. The eect on oreign materials outsourcing is either weak or insigniicant, depending on the methodology used. Our paper is set up as ollows. The next section describes our basic model. Then ollows a section outlining our methodology beore we describe our data. We report the results or our regressions beore concluding in a inal section. 2 Simple Model o Foreign Outsourcing Growth The model provides a stylization o an input outsourcing set-up that allows or growing input demand irrespective o whether or not irms switch to internet procurement. For simplicity, we move sotware programmers to improve the productivity o service delivery in a phenomenon known as reverse outsourcing. 4 Over 91 percent o Irish irms surveyed in the E-commerce survey used computers in 2003 though by 2004 this had jumped to almost 100 percent (Haller/ESRI, 2008). Between 2002 and 2004 there was a 32 percent increase in the number o irms oering services online. Our observation period was marked by the piloting o the E-commerce survey and considerable increases in online transactions or irms 5 See Girma and Görg (2007) 4

7 beyond the make-or-buy decision. 6 We irst derive the composition o oreign and domestically outsourced inputs as a result o optimal actor demand. Subsequently, we isolate the impact o the internet or both internet switchers (treatment group) and non-switchers (control group respectively) thereby ocussing on oreign outsourced service inputs as an application. Our argumentation is based on the interaction o several eects concerning demand or and prices o the oreign outsourced input: (i) Input demand is aected via a competition eect which is due to the mere existence o the internet as well as via an access eect which is exclusively based on internet usage. The latter thus only comes into play or those irms already using the internet or switching to internet usage during the considered period. (ii) Besides, oreign input procurement is accompanied by actor price uncertainty. Since inputs are new to the irms and irms are unclear as to the exact value they will get or their money, the input price or a given quality is then stochastic. It includes the expected actor price, the irms degree o risk aversion and transportation costs. I any o these components increases up to a signiicant point, this may even tip the scales against international outsourcing. Overall, the irms' demand or oreign outsourced inputs is co-determined by the interaction o the aorementioned eects. Our ollowing stylization o the input outsourcing set-up allows irms to increase their input demand irrespective o whether they switch to internet procurement or not. The model also allows us to derive the part o oreign input demand traceable back to the internet switch. Due to the dierent properties o materials and services input, the discussion distinguishes between them. We provide a simple actor demand model that may be applied to both materials and services. Ater presenting the model setup, we derive the optimal allocation between oreign and domestically outsourced inputs (Section 2.2). The model predictions will later be interpreted in the light the dierential eects o internet usage or outsourcing services and materials inputs respectively, given their dierent properties (Section 2.3). A graphical illustration in Figure 1 highlights the argumentation or the speciic case o oreign outsourced services, thereby assigning the eects to switching and nonswitching irms. 2.1 Formal representation o optimal input demand To ormalize the problem, we rely on the seminal work o Sandmo (1971) and Leland (1972). We assume a risk averse representative irm that produces a homogenous good thereby utilizing oreign and domestically outsourced actor inputs. The irm is well inormed about the price and quality o domestically outsourced inputs whereas there is uncertainty w.r.t. the price/quality relationship or oreign outsourced inputs. Formally spoken, the price a irm has to pay or the domestically outsourced input is deterministic whereas the price or the oreign input is stochastic. The 6 One might argue that the make-or-buy decision was already inalized as a irst part o a two step- optimization problem. 5

8 objective o the optimizing irm is to maximize expected utility * [ ] E Z out o proits as given by 7 * * E Z{ } = E Z p X ( v, v ) v q v ( q t ) (1) d d d c Outsourced inputs comprise domestic (d) and oreign () shares, the amounts thereo denoted by v d and v or domestic and oreign inputs respectively. The corresponding input prices are q d and The parameter t c denotes transportation costs. The irm's production conditions are given by a generalized production unction X v d, v ) and we assume perectly competitive input and product ( markets. For the sake o simplicity, we normalize the price o the inal product, p, to unity. The irst order conditions then relect the equalization o the value o the marginal product with that o the considered actor price. For both the domestic and oreign outsourced inputs, the irst order conditions can be shown to be as ollows; 8 q. v * d X v d = q d (2) X v = E [ q ] t (3) * * c v As argued beore, there is no uncertainty concerning the domestically outsourced input and q d represents the corresponding deterministic actor price (compare (2)). In contrast to this, uncertainty * about oreign inputs requires us to assume an expected input price, E [ ]. In this vein, increasing the transparency on (national and international) actor markets, induces a convergence o domestic and international actor prices E q qd. Moreover risk aversion is captured in a (input speciic) risk premium,. Physical transportation, t c, may represent a signiicant part o the overall actor price a irm has to pay (especially when outsourcing materials inputs). All three components on the right hand side in (3) together with the actor demand unction co-determine equilibrium demand or the oreign outsourced input. It is reasonable to assume that these components dier or services and materials. q 7 The act that the irm maximizes expected utility out o proits and not just expected proits is due to the assumption o risk aversion. This is a characteristic o irms which can later be connected to some o the drivers o outsourcing (See empirical part in Section 5). 8 For a ormal derivation o (3) and especially the composition o the risk premium,, see Appendix 3. 6

9 2.2 Optimal allocation o oreign and domestically outsourced inputs Equations (2) and (3) also capture the optimal ratio o oreign and domestic outsourced inputs, v / v. 9 We irst assume a constant output level or the inal good. Other arguments come into d play i output increases, thereby inducing higher overall input demand (as veriied in Section 2.3). For a given output level, the optimal ratio between oreign and domestically outsourced inputs is then determined by the marginal rate o substitution * v dv E * [ q ] t d = vd dv qd c (4) This implies a reallocation e.g. in avor o oreign at the expense o domestically outsourced inputs, * the lower the stochastic price o the oreign input ( E [ q ] t ) compared to the deterministic domestic input price ( q d ). The resulting ratio v / vd is thus mainly driven by the aorementioned components that result rom the input s properties. 10 For example, it is reasonable to assume that the level o the risk premium is aected by the degree o input standardization (which e.g. is easier to gauge or materials than or services with the consequence that is smaller or materials than or services). Conversely, while transportation costs are negligible or services, these remain a relevant cost component or materials. 2.3 The role o the internet: an application to demand o oreign services The outsourcing o oreign services is still a growing ield and the accompanying eiciency gains or irms are not yet ully exhausted. Amiti and Wei (2006) documented an average annual rate o 6.3 percent or US manuacturing, albeit rom low starting levels. In this sense, the internet allows irms to select inputs rom an ever widening menu o options ( access eect ). Since there are stronger limits to standardization, the outsourcing o services is requently tied to uncertainty over value or money. Another moderating inluence in the purchasing o services is the magnitude o a irm's risk aversion. Why should the irm s degree o risk aversion matter? Purchasing irms can experience higher ex post costs with oreign IT suppliers (e.g. supplying sotware programming), where purchasers ind that service inputs may need reworking. Providers o services inputs (e.g. R&D) in locations such as China may be less eicient in producing a product comparable with that o a domestic supplier. c v 9 / v Notice that v / vd = with v = v d vd / v v. Our measure o these ratios v / v and v d / v is discussed in the empirical part o the paper and the notes to Table 5. v / v may also change i the two inputs exhibit dierent growth. 10 Notice that the optimal ratio d 7

10 In what ollows, we ocus on the impact o the internet on demand or oreign outsourced services. We restrict ourselves to considering the quantity eect (change in oreign outsourced inputs) instead o the aorementioned compositional eects (ratio o oreign over domestically outsourced inputs) and discuss the incentives or switching vs. non-switching irms. Our starting point is the stylized act that oreign outsourced services are expected to become more tradable with the advent o internet technologies and more easily purchased at arm s length. The basic argumentation or materials is analogous despite our expectation o a dierent overall eect due to the speciic properties o materials inputs, as discussed later on in this section. With respect to the actor demand unctions, two eects are to be distinguished: The competition eect increases transparency across international actor markets thereby inducing a convergence o domestic and oreign input prices E q qd. This eect basically holds independent o internet usage and it is stronger, the more standardized the inputs and the more transparent the actor markets are. Moreover, it is reasonable to assume that the competition eect is more pronounced or irms using the internet as they are better inormed. This allows them to realize even higher savings on purchased inputs. Altogether, the overall demand or oreign outsourced inputs increases and the impact on quantity demanded may be assigned to switching and nonswitching irms. In contrast to this, the access eect is exclusively due to internet usage and does not apply or non-switchers. It provides (exclusively) switching irms access to so ar unknown actor markets, e.g. to state-o-the-art processes rom specialized suppliers. Concerning the overall input demand, 'internet switch', may then be interpreted as an exogenous shock. 11 (Figure 1 about here) In line with the uncertain quality o oreign services inputs above, Figure 1 illustrates the underlying eects o the internet on demand or oreign outsourced services. The initial situation (no internet, no switch) is given by point A with amount v A and price or services q A. 12 The competition eect leads to an overall actor price decline rom q > q, thereby inducing an increase in demand as represented by a move along the actor demand unction rom point A to point B in Figure 1 (green arrowed line; quantity v AvB ). As argued beore, this eect is due to the mere presence o the internet and arises independent o internet usage. Additionally, switching irms get an even clearer picture o the potential input suppliers, allowing them to beneit rom still stronger price reductions, e.g. up to q C with v C, respectively. This eect is probably the more pronounced, the less A B 11 It is reasonable to assume that the access eect impacts small irms more strongly vis-à-vis large irms. The ormer are not integrated into more complex organizations and thereore the marginal utility o switching to online outsourcing should be higher or smaller irms. 12 Notice that the index,, is suppressed since the entire discussion now reers to the demand or oreign outsourced services. Besides, as aorementioned, the stochastic input price as relected by the horizontal q lines covers all three components,. 8

11 standardized the inputs are. The overall quantity eect o increased competition is thus v AvC which may be assigned to non-switchers ( v AvB ) and switchers ( v BvC ). The access eect is illustrated by a shit o the actor demand unction in Figure 1 to the right, thus relecting increased access to speciic inputs ( widening menu o options as argued beore), a beneit enjoyed by online outsourcing irms only. Given the 'initial' actor price q C, demand or oreign outsourced services would increase even more, up to v C. The entire access eect thus covers the quantity switching irms. vcv C. It may only be realized by So ar we did not address the impact o the stochastic actor price. It is quite reasonable to assume that or service inputs rom as yet unknown suppliers, actual demand crucially depends upon the purchasing irm's characteristics, among them its degree o risk aversion. Another component o the stochastic actor price is transportation costs; however these are probably low or services. Both risk aversion as well as transportation costs result in an upward shit o the stochastic actor price line as indicated by q D. The extent o the shit depends upon the actually prevailing levels o risk aversion,, and transportation costs, t c. For services, we assume that the ormer may play a signiicant role whereas the latter may be almost neglected. 13 actor price line reduces the actual demand, e.g. to v D. 14 Accordingly, a positive shit o the The inal equilibrium outcome is codetermined by the stochastic actor price and the input demand unction. In Figure 1 this equilibrium is denoted by the quantity vv A D. To summarize: The net increase in quantity is based on the competition and the access eect, on the one hand, and risk aversion and transportation costs, on the other hand. The competition eect induces a movement along the actor demand unction which as a consequence o the access eect shits to the right. As a consequence, overall demand increases. Dierent levels o risk aversion and transportation costs are associated with an upward shit o the stochastic actor price line up to q D.which basically dampens the overall eect. Concerning materials, similar eects should be observed or irms which outsource materials inputs as those witnessed or services inputs. However, it is reasonable to assume that the magnitude o the eects diers. Speciically, in contrast to services, the market or material inputs is likely to be more mature and hence more saturated (Amiti and Wei, 2006). Higher transparency due to enhanced scope or standardization o materials inputs also helps underpin the competition eect.15 Thus the 13 In case o materials the opposite argumentation applies (see below). 14 Notice that q D and accordingly v D have been chosen randomly. Depending upon the actual level o ψ or t c, any inal amount smaller than would be easible. 15 Put dierently: dye is just dye, independent o the internet and its usage. 9

12 scope or both the competition and access eect to aect the demand or inputs is likely to be rather small or materials. Even i one negates the impact o risk aversion (which does not aect materials signiicantly), the potential gains or oreign outsourcing are modest due to remaining transportation costs shiting the line o the stochastic actor price, q, signiicantly upwards. Compared to services, the overall eect o the internet and its usage on outsourced materials are likely to be more modest or insigniicant. 16 The arguments discussed so ar are picked up in the empirical part o the paper. We show that the overall the ratio o oreign to domestic inputs increases as a consequence o the mere existence o the internet. We also derive which share o the increase arises rom the internet switch and which increase would have been registered even by non-switching irms. As argued beore, it is reasonable to assume that the access eect is quite important in case o services. This assumption would urthermore support our later empirical indings whereby oreign vs. domestical outsourced services respond more to the internet switch than materials. 3 Methodology When we examine the web switching/outsourcing relationship, we ace a standard selection problem. I irms which commence outsourcing services online ( internet switchers ) happen to be already highly involved in outsourcing services rom supplier irms (compared with other irms which avoid using the internet or outsourcing), we would expect to witness above-average outsourcing quantities or internet switcher irms even i they had not started to use the online outsourcing. Internet switchers could be a sel-selecting group, in which case, the impact on outsourcing growth has less to do with the move to online outsourcing and more to do with inherent characteristics o this group. The lack o statistically relevant and intuitively compelling instruments or online switching (actors prompting irms to switch to online usage but which do not aect outsourcing growth) makes it diicult to deal with sel-selection unless we isolate rom our sample the group o newly switching and non-switching irms. Following Heckman et al. (1997) we can calculate the average eect o switching to online outsourcing on expected outsourcing growth. Speciically, and represent outsourcing quantities pre- and post the switching period respectively or irms switching to online outsourcing (SWITCH = 1) as: Note that, although made explicit in igure 1, our argumentation can be extended to explain a decrease in oreign input demand as a consequence o internet procurement. 10

13 where the last expression term is needed in order to iner the oreign outsourcing intensities or the group o irms that did not switch to online outsourcing. To get this term, we match each irm that switched to online outsourcing with a derived counteractual, constructed over the distribution o nonswitching irms. We apply the STATA propensity score routine, psscore, based on Rosenbaum and Rubin (1983). Speciically, the irst-stage Probit captures the likelihood that irms switch to online outsourcing based on observable pre-switching attributes o the irm (irm size and ownership status, R&D status, technology status and growth). 17 Both control (never using online outsourcing) and treatment (newly online outsourcing) irm groups are then assigned to strata according to the propensity score and the balancing property checked or each stratum. 18 A note on the covariates used to explain outsourcing growth. Amiti and Wei (2006) argue that there might be a concern that oreign services outsourcing is correlated with omitted variables such as total outsourced inputs (levels). This would make it diicult to interpret the overall impact o a switch on quantities o oreign services/materials outsourced. To address this, we include lagged outsourcing (levels) in some speciications o the empirical model. The transactions cost literature suggests a strong connection between a irm s technological ability and its outsourcing intensity (e.g. see Bartel et al., 2005). Thereore, we control or irm-level R&D to prevent overstating the impact o online outsourcing on oreign outsourcing growth. To derive our outsourcing intensity measure, we scale the share o inputs rom oreign suppliers by the total amount o inputs supplied (oreign and domestic). 19 In our empirical analysis, size is measured as number o employees, R&D is a irm s total R&D divided by sales and the ownership dummy indicates whether a irm is oreign owned. Finally, our outcome variable oreign outsourcing is deined as oreign services (materials) used, divided by total services (materials). 20 In estimating oreign outsourcing, we opt or the STATA attk procedure proposed by Heckman et al. (1997) which builds on traditional pairwise matching by using the ull distribution o irms alling under common support in the pre-exporting Probit. 21 The nonparametric matching estimator constructs a match or each irm which has just begun outsourcing online using a kernel-weighted average over multiple irms which do not carry out online outsourcing. Assuming that the common 17 Labour productivity and irm age are also likely drivers o oreign outsourcing growth. They were used in earlier regressions though their relative inability to explain variations in outsourcing growth led to them being subsequently dropped. 18 We assume that the assumption o conditional independence holds: i.e. that irms in the control and treatment group largely select into online outsourcing based on these observable pre-switching attributes. Speciically, their diering willingness/ability to manuacture in-house vs. to outsource. 19 See ootnote In this we depart rom Abramovsky and Griith (2006) or Görg and Hanley (2011) who both scale oreign outsourcing by sales. Our rationale or scaling by total outsourced inputs is to discern the changing bias between oreign and domestic inputs with online outsourcing. 21 We use the STATA deault Gaussian kernel with bandwidth Smith and Todd (2005) give an excellent summary o this and other matching techniques. An advantage o this matching technique is that it reduces the asymptotic mean squared error ound in traditional pairwise matching. 11

14 support conditions hold, we now have a consistent estimator o the growth in oreign outsourcing or irms newly outsourcing online, had they not decided to outsource online: 1 Our empirical analysis outlines how oreign outsourcing growth is expected in both groups (treatment and control) in line with the competition eect as the internet becomes widespread with implications or the supply o services and materials. Accordingly, it is possible (though improbable) or irms in both groups to draw equally on this wider palette o suppliers, even i they do not transact online. Hence the need or Propensity Score Matching with Dierence-in-Dierences (DID) to net out these growth dierences due to the competition eect (a common shock) and isolate the impact o online outsourcing on oreign outsourcing growth. In combining Propensity Score Matching with DID, we ollow Girma and Görg (2007) allowing us to tackle both endogenous selection into online outsourcing where irms select into this group on the basis o covariates ound to be signiicant in the literature whilst neutralizing the bias o common macroeconomic shocks (DID). 4 Data We use plant level inormation rom data collected by Forás, the Irish policy and advisory board with responsibility or Enterprise, Trade, Science, and Technology in Ireland. Speciically, our data source is the Annual Business Survey o Economic Impact (ABSEI), covering the period rom 2000 until The ABSEI survey is an annual survey o Irish plants with at least 10 employees, although a plant, once it is included, is generally still surveyed even i its employment level alls below the 10 employee cut-o point. The response rate is estimated by Forás at circa 55 to 60 percent o the targeted population per year. This data contains inormation on oreign and domestic services outsourced. Further data available rom this source relevant to the current paper are total ull time employment, sales, R&D expenditure, irm age, nationality o ownership and the our digit sector o production. Importantly or this study, the survey also included questions on ICT usage or the period 2002 to Some recent work is now emerging or other EU member states where the same ICT usage survey was carried out (See Abramovsky and Griith, 2006 or the UK). 22 Data cleaning and variable generation Since the Internet usage variables were covered only in the period rom 2002 when the E-commerce survey was piloted and the last year where inormation is recorded or outsourcing is 2004, this means that our sample rame contains 3 years o data. A main novelty o our paper is estimating the impact 22 See Inormation society: ICT impact assessment by linking data rom dierent sources, Eurostat Report, August

15 o having switched to online outsourcing on oreign outsourcing growth. Accordingly, we identiy 2003 as the year in which a switch to online outsourcing is made and divide the 3-year time window into pre-treatment (2002), treatment (2003) and post-treatment (2004) periods respectively. The raw data contained 2,022 irms in the initial ABSEI sample (Table 1). 22 percent o the sample was discarded because it comprised irms which were already outsourcing online rom beore irms were assigned to the treatment sample (7.9 percent o the raw data), i they used online outsourcing or the irst time in 2003 but had not already done so the previous year. Analogously, 691 irms were assigned to the control sample (70.1 percent o the raw data) i they had never at any point during the 3-year time window done any o their outsourcing online. All covariates such as sales were delated by the Consumer Price Index. 23 The key variable o interest in our analysis is the proportion o oreign outsourced inputs (services vs. materials) or irms moving to online outsourcing (treatment group) compared to the proportion or irms which avoid online outsourcing (control group). Accordingly, Table 2 gives the breakdown o oreign outsourcing across these two groups or the pre-treatment period (2002) and thereater. Looking at the composition o outsourced oreign services, what is clear is that irms in the treatment group start out with low levels o oreign outsourcing (13 percent vs. 20 percent). Once treatment irms introduce online outsourcing in 2003, the dierences between both groups are reduced (18 percent vs. 22 percent). The greatest convergence in oreign services outsourcing between the treatment and control groups is contemporaneous with the switch to online outsourcing ( ) and not thereater ( ). This same pattern o convergence seen in oreign services outsourcing is not replicated or materials where the composition o oreign materials outsourcing (pre- and post- treatment) remains stable at percent. Figure 2 illustrates the same inormation graphically. What is clear rom Figure 2 is the relatively minor role played by oreign services compared to oreign materials inputs (e.g. high transportation costs would shit q B upwards thereby inducing a decrease in the demand or oreign outsourced inputs). However, over the period , there is some growth in oreign services inputs. Although overall, irms increased their intake o oreign services rom 19 to 21 percent over the 3-year period, compared to an increase o 50 to 51 percent in materials inputs, much o the oreign services growth seems to have come rom the 99 irms in the treatment group who started online outsourcing between 2002 and 2003 where the increase in oreign services amounts to 8 percent (rom 13 to 21 percent). Figure 2 makes two points clear: 1) the market or oreign outsourced services is small compared to that or oreign outsourced materials and 2) this market in oreign services appears to be growing. These two points are consistent with Amiti and Wei (2006) who used data or the US. (Figure 2 about here) 23 See Appendix 1 or a ull list o variables used in our analysis 13

16 How similar are irms in both treatment and control groups? Table 3 reports the breakdown o key covariates or both groups o irms in the pre-treatment period. However, irms switching to online outsourcing (treatment) appear to be, at irst glance, smaller, Irish-owned, higher-growth and more R&D intensive than their non-switching counterparts. One possibility is that larger, oreign irms that do not make the move to online outsourcing are perhaps better vertically integrated into global production networks. Moreover, the propensity to move online seems to be, to some extent, conditioned on high-growth (e.g. see Falk and Koebel, 2002). These are arguments that support the aorementioned access eect where a reduction in transacting costs may coner relatively higher utility to irms with little ability to procure within a global production network. However, we need to examine this growth-outsourcing nexus in a multivariate ramework. The patterns in Table 3 showing the bias towards small, Irish, higher-growth irms hint that a matching ramework might be in order. This approach would help correct the pre-treatment biases and allow us to observe the eect o a switch to online outsourcing or observationally equivalent treatment and control irms in Analysis The next step is to regress the online outsourcing dummy on oreign outsourced services growth while controlling or lagged values o key variables ound signiicant in the literature. Table 4 shows the results o this exercise. In column 1, we merely control or irm size and ownership while in columns 2 and 3 lagged growth and R&D intensity are added sequentially. In the latter two estimations, the online outsourcing dummy is ound to be signiicant at the 5 percent level where a switch to online outsourcing is associated with a 45 percent increase in the average ratio o oreign to total outsourced services. Consistent with our earlier observation that larger irms are likely to be more vertically integrated, an inverse relationship is reported on the irm size coeicient. When we repeat the procedure or outsourced oreign materials (columns 4 6), the eect is signiicant but smaller than or that o services. Much o the explained variation in oreign outsourcing growth can be explained by the irm s current outsourcing proile. I current oreign outsourcing levels are high, outsourcing growth will be lower the next year. 24 Earlier, in the descriptive statistics (Table 3) we observed dierences in the average irm size, ownership, R&D intensity, growth and sector in the pre-treatment period. This hinted at the possibility that the switch to online outsourcing may be endogenous to these covariates. To check or endogeneity o the internet switch indicator, we ollow Eichenbaum et al. (1988), by calculating the Dierence-in-Sargan statistic (C-statistic) with p-values o and or the switch to online outsourcing or services and materials inputs respectively. This indicates that we are correct in rejecting the null that the switching dummy is exogenous. 24 Amiti and Wei (2006) highlighted the importance o including this measure o total imports (levels) in regressions in order to reduce the bias in the elasticity o oreign inputs growth to online outsourcing. 14

17 Having veriied that the switch to online outsourcing is endogenous to the lagged regressors, we proceed to the Propensity Score Matching procedure which will allow us to condition the switching dummy in a irst step on the covariates and calculate a propensity score. In a second step, subject to the STATA procedure (STATA attk ) having perormed the balancing tests satisactorily (See Appendix 2 or balancing tests), we can move to the Kernel matching to see whether there are any systematic dierences in oreign outsourcing growth rates or treatment and control irms. In the selection equation, we see that larger and oreign-owned irms are signiicantly less likely to start outsourcing online (Table 5a). Arguably, they are better vertically integrated and have thereby less use o this outsourcing mode. The high and positive coeicient or R&D intensity may have to do with the complementarity between these activities e.g. a presence o systems savvy sta involved in R&D is perhaps helpul when a irm is working out ways o outsourcing online. Table 5b reports the indings or the Propensity Score Matching procedure or the eects o the switch to online outsourcing on growth (reduction) in the oreign outsourcing share. Our estimations report Dierence-in-Dierence estimates, thereby purging any common time-variant shocks which would aect would groups o irms (see Girma and Görg, 2007). The oreign services ratio grew over the treatment period by 70 and 22 percent or irms in the treatment and control groups respectively. This corresponds to a 48 percent increase or irms in the treatment sample vis-à-vis irms in the control sample. The dierence is marginally signiicant (at the 10 percent level) or the bootstrapped standard errors. However, the corresponding growth increase o 22 percent or oreign materials outsourced is insigniicant. The estimates o 48 percent rom Propensity Score Matching compare reasonably well with those obtained under standard OLS (Table 4) where increases o percent were registered. 6 Conclusions and Policy Implications Consistent with the competition eect outlined in our model, irms grew their share o oreign outsourced services inputs over the treatment and post-treatment periods ( ) by 22 percent. 25 Consistent with the access eect, irms switching to online outsourcing registered increases in the ratio o oreign outsourced service inputs o 48 percent, a result broadly in line with the OLS estimates (42 46 percent). Our estimates or the eects o switching to online procurement represent a lower bound or the overall potential eect o the internet on oreign outsourcing. When both competition and access eects are added together, the overall eect is 70 percent (22 percent + 48 percent). The Propensity Score Matching estimates reveal online outsourcing to be the choice o irms not integrated into global production networks (e.g. not oreign-owned or large). Such irms lack the percent corresponds to the exogenous growth in the oreign services share o control irms during the period and thereore relects the background growth in the sector as more services become tradable and their visibility rises. 15

18 critical mass and global reach to get their inputs in other ways (through overseas vertically integrated sister irms or in-house domestic production). Our study dates rom the time when online outsourcing was a relatively new phenomenon. Would the impacts or oreign services outsourcing be the same i the growth in oreign services outsourcing had already peaked? Our estimates or the period suggest promising oreign services growth (compared to the market or materials inputs). Estimates using more recent data might reveal ewer dierences in the growth rates or service vis-à-vis materials inputs as the market or traded services matures. 16

19 Reerences Abramovsky, L. and R. Griith, 2006, Outsourcing and Oshoring o Business Services: How Important is ICT? Journal o the European Economic Association, April-May 2006, v4, n2-3, pp Acemoglu, D., P. Aghion, R. Griith and F. Ziliboti, 2010, `Vertical Integration and Technology: Theory and Evidence', Journal o the European Economic Association, v8, pp Amiti, M. and S.J. Wei, 2006, `Service Oshoring, Productivity and Employment: Evidence rom the US', CEPR Discussion Paper Bartel, A., S. Lach and N. Sicherman, 2005, `Outsourcing and Technological Change`, Centre or Economic Policy Research, Discussion Paper Series No Bhagwati, J., A. Panagariya and T. Srinivasan, 2004, `The Muddles over Outsourcing`, Journal o Economic Perspectives, v18, n4, pp Eichenbaum, M., P. Hansen and K. Singleton, 1988, `A Time Series Analysis o Representative Agent Models o Consumption and Leisure`, Quarterly Journal o Economics, v103, n1, pp Eurostat Report (as yet unpublished), `Inormation society: ICT Impact Assessment by Linking Data rom Dierent Sources`, August 2008 drat Falk, M. and B. Koebel, 2002, 'Outsourcing, Imports and Labour Demand', Scandinavian Journal o Economics, v104, n4, pp Girma, S. and H. Görg, 2007, `Evaluating the Foreign Ownership Wage Premium using a Dierencein-Dierences Matching Approach`, Journal o International Economics, v72, n1, pp Görg, H. and A. Hanley, 2011, `Services Outsourcing and Innovation: An Empirical Investigation`, Economic Inquiry, v49, n2, pp Grossman, S. and O. Hart, 1986, The Costs and Beneits o Ownership: A Theory o Vertical and Lateral Integration, Journal o Political Economy, v94, n, pp Haller/ESRI, S., 2008, `ICT Adoption in Irish Manuacturing: An Example o merging two CSO Business Microdata Sets`, CSO Business Statistics seminar, Economic and Social Research Institute, (ESRI) Heckman, J., H. Ichimura, J. Smith, and P. Todd, 1997, `Matching as an Econometric Evaluation Estimator: Evidence rom Evaluating a Job Training Program,` Review o Economic Studies, v64, pp Houseman, S., 2007, `Outsourcing, Oshoring, and Productivity Measurement in U.S. Manuacturing`, Upjohn Institute Sta Working Paper No , W. E. Upjohn Institute or Employment Research Heckman, J., H. Ichimura, J. Smith and P. Todd, 1997, `Matching as an Econometric Evaluation Estimator: Evidence rom Evaluating a Job Training Programme`, Review o Economic Studies, 64, pp Hijzen, A, T. Inui and Y. Todo, 2010, `Does Oshoring Pay? Firm-Level Evidence rom Japan`, Economic Inquiry, v48, n4, pp Leland, H. E. 1972, Theory o the Firm Facing Uncertain Demand, American Economic Review, v62, pp

20 Lileeva, A. and J. van Biesebroeck, 2008, `Outsourcing When Investments are Speciic and Complementary`, NBER Working Paper No. w14477 Milgrom, P. and J. Roberts, 1990, `The Economics o Modern Manuacturing: Technology, Strategy, and Organization`, American Economic Review, v80, n3, pp Rosenbaum P. and D. Rubin, 1983, `The Central Role o the Propensity Score in Observational Studies or Causal Eects`, Biometrika, v79, n1, pp Sandmo, O., 1971, On the Theory o the Competitive Firm Under Price Uncertainty, American Economic Review, v61, pp Smith, J. and Todd, P., 2005, 'Rejoinder', Journal o Econometrics, v125, pp UNCTAD, 2004, World Investment Report: Shit Towards Services 18