Geographic, Horizontal and Vertical Diversification Juan Alcacer, Harvard Business School Milan, October 31st 2015 Agenda n Breaking up value chains n Geographic and vertical disaggregation ¨ ¨ ¨ n Fort (2013) Alcacer, Lecuona, Oxley (2015) Delgado (2015) Why within-firm collocation can be beneficial? ¨ ¨ Alcacer & Delgado (2015) Alcacer & Chauvin (2015) 2 Allocating Activities Allocating Activities Internal links Dynamic links External links Allocating Activities Internal links Dynamic links External links Two geographic expansions n n Wal-Mart Target 6 Agenda n Breaking up global value chains n Geographic and vertical disaggregation ¨ ¨ ¨ n Fort (2013) Alcacer, Lecuona, Oxley (2015) Delgado (2015) Why within-firm collocation can be beneficial? ¨ ¨ Alcacer & Delgado (2015) Alcacer & Chauvin (2015) 7 Geographic fragmentation and outsourcing Breaking up is hard to do: Why firms fragment production across locations Teresa Fort Tuck Business School Dartmouth College 8 Geographic fragmentation n n Motivation ¨ Despite great interest in offshoring and multinational sourcing little is known of its magnitude and drivers Question ¨ How and why firms break up their productions process across space? n n n Whether to fragment production Conditional on fragmentation, whether to source domestically or offshore Empirical setting ¨ Novel micro-level dataset from US Census Bureau’s 2007 Census of Manufacturing., LBD and US custom import transactions. n Contract Manufacturing Services (CMS): fragmentation of inputs that are customized to meet a firm’s specific production criteria. n Analysis at plant and firm level. 9 Geographic fragmentation n Findings No Purchases Domestic Purchases Offshore Purchases Domestic & Offshore ¨ ¨ Plants 0.71 0.27 0.02 na Plant Shares Sales 0.57 0.39 0.04 na Emp 0.61 0.35 0.04 na Firm Shares Firms Sales Emp 0.69 0.31 0.42 0.28 0.42 0.39 0.02 0.03 0.03 0.01 0.24 0.16 The majority of plants do not fragment their production process for customized inputs (even on sales or employment-weighted basis). The majority of U.S> manufacturing sales and employment takes place at firms with at least one plant that purchases CMS 10 Geographic fragmentation n Findings Domestic Purchases 5-­‐10% 10-­‐20% 20-­‐35% 35-­‐50% 50-­‐60% Total ¨ 86 Manufacturing Industries (NAICS 4) (% of e stabs i n i ndustry) Offshore Purchases 0% 0-­‐5% 5-­‐10% 10-­‐20% Total 0 2 0 0 2 1 22 2 0 25 1 31 6 2 40 0 13 4 1 18 0 1 0 0 1 2 69 12 3 86 Domestic fragmentation is more prevalent than offshoring, both across and within industries 11 Geographic fragmentation n Findings No Purchases Domestic Purchases Offshore Purchases All Plants ¨ ¨ ¨ ¨ Raw Means Salesᵅ Emp ln(VAP) 19,487 51.3 4.51 37,077 79.8 4.63 51,457 137 4.74 24,686 60.4 4.55 Relative Ind. Means Sales Emp ln(VAP) 0.87 0.91 -­‐0.03 1.28 1.2 0.07 2.17 1.69 0.2 1 1 0 Plants that fragment production are larger and more productive that non-fragmenters Plants that fragment production offshore are larger and more productive than domestic fragmenters 22% of plants purchase CMS doemstically in low-wage states compared to 25% and 30% in mid-wage and high-wage states, Firms with no CMS purchases and domestic CMS purchase import 28 and 19 % of their manufactured goods from low-icnome countries. 48% of offshores do. 12 Outsourcing and innovation Forgetting by outsourcing: evidence form the wireless handset industry Ramon Lecuona Joanne Oxley Juan Alcacer Duke University Duke University Harvard Business School 13 Outsourcing and innovation n Motivation ¨ Debate about effect of production outsourcing on technical innovation and national competitiveness goes way back: n 1980s: “Hollowing Out” (e.g.. Cohen & Zysman, 1987); n 1990s: Dark side of ‘learning alliances’ (e.g. Hamel, 1991); n 2000s: The debate continues… (Arrunada and Vazquez, 2006; Pisano and Shih, 2009); ¨ Debate has generated copious passionate rhetoric, but limited systematic empirical study at firm level n Dearth of empirical research due to lack of extensive firm-level data on outsourcing n Most prior studies on jobs at country/region level n Firm-level evidence based on cases / small-scale surveys 14 Outsourcing and innovation n Question ¨ ¨ n Empirical setting ¨ Comprehensive data of GSM handests introduced since 1992 Information on whether the handset was outsourced or manufactured ¨ Detailed technical information for each handset ¨ Patent data for each firm in the industry ¨ n Does outsourcing manufacturing affect innovative capabilities of firms? Introduce less advanced products Findings (very preliminary) ¨ ¨ ¨ Firms that lag in technology are more likely to outsource, they start outsourcing as the lag in market share Firms that outsource are more likely to exit industry Conditional on becoming and outsourcer n n Firms are less likely to introduce a new feature, more likely to copy a feature fast Firms loose market share 15 Collocation of innovation and production clusters The collocation of innovation and production in clusters Mercedes Delgado Sloan- MIT 16 Collocation of innovation and production clusters n Question ¨ ¨ n How the co-location of innovation and production varies across different clusters The goal is to inform the debate and offer some implications for future research on the innovation potential of regions and their firms. Empirical setting ¨ ¨ ¨ US Cluster Mapping Project dataset to measure employment/patenting at the region-cluster level. U.S. Benchmark Cluster Definitions developed by Delgado, Porter, and Stern (2015) grouping related industries based on input-output links, shared labor occupations, and co-location patterns. Data n n spans the years 1998-2011, includes 177 mutually exclusive Economic Areas (EAs), and incorporates 778 traded industries (6-digit NAICS), grouped into 51 clusters of related industries for each EA Focus on 35 clusters with meaningful patenting 17 Por$olio of 51 Traded Clusters and their Connec6ons 2 Source: Delgado, Porter, and Stern (2014, 2015) Collocation of innovation and production clusters n Methodology ¨ ¨ ¨ Innovation strength is measured by patent specialization of a region in the cluster (location quotient of patent) Production strength is measured by employment specialization (both mfg. & services) of an economic area (location quotient of employment) Using these measures, the paper develops indicators of the collocation of innovation and production in clusters n n Dual Specialization Correlation (DSCct) captures the dual specialization of regional clusters in employment and patenting: i.e., the correlation btw regional cluster specialization in employment and patenting DSCct = Corr(Cluster SpecializationEmployment crt , Cluster SpecializationPatent crt) 19 Changes in Collocation in Clusters in the US: 1998-2011 n Findings Year Dual Spec Correlation 1998 2011 Change98-11 DSCc,t [1,-1] Mean (N=35) 0.24 0.22 -0.02 ¨ ¨ ¨ Strong Clusters Patenting SCPc,t (%) Mean (N=35) 28 27 -1 SCPAll,t (%) Total 40 44 4 Strong Clusters Employment SCEAll,t (%) SCEc,t (%) Mean (N=35) Total 51 50 50 43 -1 -8 Dual specialization in patenting and employment has remained significant during the whole period, with an annual average DSCct score of 0.20. The average DSCct declined slightly from 1998 to 2011 (from 0.24 to 0.22). This decline was accompanied by a small decrease in the average SCPct and SCEct (-1%) The findings suggest that innovation concentration in strong clusters has increased by 4%, (from 40 to 44%), but employment in strong clusters declined by 8% during the period 20 IT & Analytical Instruments n Findings Year 1998 2011 Change98-11 ¨ ¨ IT & Analytical Instruments DSCct SCPct SCEct Score Rank Pat Rank Emp Rank [1,-1] 1-35 % 1-35 % 1-35 0.74 1 66 1 64 9 0.53 2 70 1 62 11 -0.22 4 -2 While the DSC is high as of 2011, it has declined by 0.22 points during the period (from 0.74 in 1998 to 0.53 in 2011) This large decline is due in part to an increase in the number of clusters with employment strength but low patenting strength. 21 IT & Analytical Instruments, San Francisco EA 1998-2011 n Findings Year Employment 1998 2011 EA Growth98-11 US Growth98-11 in 1k Shr Pctile 234 14% 100 121 11% 100 -48% -35% ¨ ¨ ¨ ¨ SpecializationEmploy LQ 4.0 3.4 LQ-Pctile 98 98 Patents No. Shr Pctile 3258 20% 100 7602 25% 100 133% 90% SpecializationPate nt LQ 1.8 1.5 LQ-Pctile 97 97 Patents per 1k Employ 14 63 351% 197% Wages in 1k $73 $144 97% 84% Top IT and Analytical Instruments regional cluster by employment and patenting High strength in employment and patenting in 1998/2011 It has outperformed other regional clusters in innovation/wages, but lost many jobs during 2008-2011 The loss of employment may be associated with several factors: ¨ increased productivity, ¨ re-allocation of some of the production capacity to related clusters (e.g., Medical Devices and Distribution and eCommerce clusters) in the region/nearby, and/or ¨ loss in production capacity (e.g., lost suppliers ?) 22 Agenda n Breaking up global value chains n Geographic and vertical disaggregation ¨ ¨ ¨ n Fort (2013) Alcacer, Lecuona, Oxley (2015) Delgado (2015) Why within-firm collocation can be beneficial? ¨ ¨ Alcacer & Delgado (2015) Alcacer & Chauvin (2015) 23 Why would same-firm collocation be beneficial? Spatial organization of firms: Location choices through the value chain Mercedes Delgado Juan Alcacer Sloan- MIT Harvard Business School 24 Why would same-firm collocation be beneficial? n Motivation ¨ ¨ n Rich research on how external environment – location endowments and agglomeration economies – impacts firm performance and location choices … …but there are many multiunit firms and tight coordination among activities is often core for a firm’s competitive advantage (Porter 1996, 1985) Question ¨ Is the location of firm activities influenced by where a firm’s existing facilities are (internal agglomerations)? ¨ Do the internal agglomerations effects vary by activity (R&D, manufacturing and sales)? How internal agglomerations compares to external agglomerations? ¨ 25 Conceptual framework n Firm location strategies are a function of two forces: ¨ External agglomerations n Geographically bounded inter-firm linkages that result in better firm performance ¨ ¨ Internal agglomerations n Geographically bounded intra-firm linkages that result in better firm performance ¨ n Collocation with other firms: Bringing firm activities apart? Collocation with same-firm: Bringing firm activities together? These two forces complement or oppose each other depending on firms’ previous locations. ¨ ¨ For firms located in clusters (i.e. geographic concentrations of related economic activity), internal and external agglomerations may work in the same direction Otherwise, external act as centrifugal forces that may drive firms to disperse their activities geographically, and internal agglomerations as centripetal forces that may drive within-firm collocation. 26 Why would same-firm collocation be beneficial? ¨ Collocation is beneficial because of internal agglomerations: • Within-firm Marshallian economies • Knowledge/information flows, • Specialized labor, • Specialized suppliers • Control and monitoring • Coordination 27 Defining the Value Chains n Bio-Value Chain R&D (SIC-8731) n Sales, Distribution (SIC-5120) Other-Value Chains ¨ ¨ Firms could be diversified into multiple business units – medical devices, downstream chemical products, … (e.g., Johnson and Johnson) Excludes Bio-Value Chain activities R&D (SIC-8732-8734) n Mfg (SIC-2830) Mfg (SIC-20-39) Sales, Distribution (SIC-50 to 59) Support Activities for any Value Chain n n E.g., Business services (HQs), medical labs, Financial/Insurance/Real State Any industry codes excluded from Bio-Value Chains and Other-Value Chains. 28 Descriptive statistics: location frequency External Agglomerations t-1: Low High Top-10 EAs EAs with zero firm employ Low : Internal Agglomerationst-1: Medium: EAs with any firm employ High: Firm Base EA for Bio-VC 29 High External Agglomerations: Top-10 EAs by share of US Bio-VC employment (and high specialization Share US Bio-­‐Value Chain employ LQ Employ, Bio-­‐Value Chain Employ Bio-­‐Value Chain 2.12 101972 Year EA name 1992 New York-­‐Newark-­‐Bridgeport, NY-­‐NJ-­‐CT-­‐PA 18.5% 1992 Chi ca go-­‐Na pervi l l e-­‐Mi chi ga n Ci ty, I L-­‐IN-­‐WI 6.7% 1.58 36661 1992 Wa s hi ngton-­‐Ba l ti more-­‐Northern Vi rgi ni a , D C-­‐MD-­‐VA-­‐WV 4.7% 1.52 26010 1992 Bos ton-­‐Worces ter-­‐Ma nches ter, MA-­‐NH 4.7% 1.37 25678 1992 Phi l a del phi a -­‐Ca mden-­‐Vi nel a nd, PA-­‐NJ-­‐DE-­‐MD 4.4% 1.61 24149 1992 Sa n Jos e-­‐Sa n F ra nci s co-­‐Oa kl a nd, CA 4.3% 1.29 23858 1992 Ra l ei gh-­‐Durha m-­‐Ca ry, NC 1.7% 1.95 9362 1992 Indi a na pol i s -­‐Anders on-­‐Col umbus , I N 1.7% 1.36 9107 1992 Knoxvi l l e-­‐Sevi ervi l l e-­‐La F ol l ette, TN 1.5% 4.2 8274 1992 Sa n D i ego-­‐Ca rl s ba d-­‐Sa n Ma rcos , CA 1.5% 1.69 8262 50% 30 Descriptive statistics: location frequency External Agglomerations t-1: Low High Top-10 EAs EAs with zero firm employ 372 (30%) 84 (7%) Low : Internal 419 (34%) 213(17%) Agglomerations t-1: Medium: EAs with any firm employ Firm Base EA for Bio-VC 54 (4%) 84 (7%) High: 69% 31% • • 37% 52% 11% 100% • • Few expansions in locations where both Internal and External are high (7%) Many expansions (63%) in locations with same-firm presence (Internal drivers); and the rest (37%) into new locations (External drivers) Many expansions (69%) outside the top-10 EAs. For Low-External, most expansions (38%) in locations with same-firm presence • Same patterns if we exclude Bio-Sales Internal and External Agglomeration Matters 31 Alcacer & Delgado (2015): Findings n Internal agglomerations have a positive effect on location choices: ¨ ¨ ¨ They vary by activity in the value chain (lower for Sales), They arise within activity (e.g., plants) and across activities (e.g., R&D & Mfg) Even after controlling for factors that could influence same-firm collocation: n n n n ¨ n Sociological factors (social capital), Initial location of a firm, Strategic location behavior, and Different sources of firm heterogeneity. They are positively associated with firm survival External agglomerations have a positive effect on location choices, but failing to consider internal agglomerations leads to an omitted variable problem, that may bias estimates external agglomerations. 32 Same-firm collocation and survival n Explore the survival of the new establishments that were opened during the period of analysis (i.e. the 1,226 new bio activities examined in the baseline location models) n Estimating a cox proportional hazard model of the survival of each new biopharmaceutical activity n Failure = Exit (death or acquisition) n We find a positive relationship between same-firm collocation in Bio Value chain & survival: bio establishments opened in locations where a firm was previously present are less likely to fail 33 Agglomeration and diversified firms Agglomeration economies and the multibusiness firm Jasmina Chauvin Juan Alcacer Harvrad Business School Harvard Business School 34 Agglomeration and diversified firms n Motivation ¨ Most existing studies of agglomeration ignore the unique aspects of location of MB firms and analyze firm location without regards to organizational form n n n E.g. Ellison, Glaeser, Kerr (2010) calculate coagglomeration of U.S. industries using the entire sample of manufacturing firms Alcacer and Delgado (forthcoming) provide an exception by decomposing location drivers into “internal” and “external” agglomeration economies in the biopharmaceutical industry Question ¨ ¨ Do multi-business firms coagglomerate their disparate manufacturing activities? What drives coagglomeration in multi-business firms? Coagglomeration-Stand alone Coagglomeration - multibusiness Coagglomeration - Chevron Agglomeration and diversified firms n Empirical setting ¨ We focus on one value chain activity - manufacturing - at the level of 140 different (SIC3 level) industries n ¨ We calculate measures of coagglomeration separately for establishments belonging to MB firms and SA firms n n We define a MB firm as any firm active in >1 SIC3 industry Our sample includes 220,246 SA establishments and 173,889 establishments belonging to a MB firm from Dun & Bradstreet (2011 vintage) Methodology ¨ We measure coagglomeration with the continuous Duranton-Overman index at various distance thresholds ¨ This tells us the likelihood that two randomly picked establishments in industries i and j are located within d miles of one another n n ¨ d can be any continuous distance, from 1 mile to >1,000 miles n is the number of observations, h is the optimal bandwidth, and f is the Gaussian kernel function We also measure coagglomeration for discrete spatial units as the share of establishments in industries i and j located in the same county, CBSA, state Kernel density: Stand-alone, multi-business, random at 50 miles Industry coagglomeration by firm type 0 Number of observations 50 100 150 Stand-alone Multi-business Random 0 .02 .04 .06 .08 .1 Likelihood that two establishments are within distance of 50 miles kernel = epanechnikov, bandwidth = 0.0005 Agglomeration and diversified firms n We use three main measures of agglomeration economies, all symmetric for industry-pair ij: ¨ Direct input-output links (1992 U.S. IO tables) max{max{InputReqij, InputReqji }, max{OutputShrij, OutputShrji}} ¨ Labor similarity (using occupational shares from NIOEM matrix from the Bureau of Labor Statistics) Vector correlation of occupational shares the 2001 industry-occupation matrix from the U.S. Bureau of Labor Statistics (BLS ¨ Technology links (NBER patents database / EGK (2010)) max{max{PatInij, PatInji}, max{PatOutij, PatOutji}} Preliminary takeaways n Multi-business co-locate their disparate activities more than stand-alone firms in the same industries ¨ n The drivers of agglomeration differ for multi-business and standalone firms are somewhat distinct ¨ ¨ ¨ n Suggests internal agglomeration economies potentially more important than external ones Labor sharing across establishments particularly important for multibusiness firms Proximity due to tech linkages less important for MB firms suggests firm has greater ability to share knowledge at larger distances Proximity due to IO linkages less important at short distances suggests lower holdup concerns For stand-alone firms, study confirms prior findings that labor market pooling, access to inputs, and technology all associated with greater coagglomeration Departing thoughts n As a phenomenon, breaking up value chains is not new and it has been studied by (and matters for) other communities of scholars ¨ IB scholars may be better equipped…but whatever is found (conceptualized) should be shared ¨ Look at international economics (Pol Antras, Laura Alfaro), organizational economics (Bob Gibbons, Eric Van den Steen) ¨ Still many facts that we are not aware of n Geographic disaggregation is a piece of overall strategy, includes environment variation AND firm-variation 43