Slides - HBS People Space
Transcription
Slides - HBS People Space
Americans do I.T. Better: US Multinationals and the Productivity Miracle Nick Bloom, Stanford & NBER Raffaella Sadun, LSE John Van Reenen, LSE, NBER & CEPR March 2008 European productivity had been catching up with the US for 50 years… …but since 1995 US productivity accelerated away again from Europe. The “productivity miracle” occurred as quality adjusted computer prices began to fall very rapidly In the US the “miracle” appears linked in to the “IT using” sectors… Sources: Stiroh (2002, AER) See also: Oliner and Sichel (2000 JEP, 2002 Fed) & Jorgenson (2001, AER), … but no acceleration of productivity growth in Europe in the same “IT using” sectors. - Change in annual growth in output per hour from 1990 –95 to 1995 –2001 % U.S. EU ICT-using sectors ICT-producing sectors Non-ICT sectors -0.5 3.5 1.9 -0.1 1.6 -1.1 Source: O’Mahony & Van Ark (2003, Gronnigen Data & European Commission) 3 So why did the US achieve a productivity miracle and not Europe? Two types of arguments proposed (not mutually exclusive): (1) Standard: US advantage lies in geographic, business or demographic environment (e.g. more space, younger workers) (2) Alternative: US advantage lies in their firm organizational or management practices Paper uses two micro data sets (one from the UK and one from Europe) that support (2) -Idea is to look within UK and Europe (holds environment constant) and compare US and non-US multinationals Summary of Results (1) Use new data on 11,000 UK establishments, 1995-03, find: • US multinationals use IT more effectively (and invest more in IT) than non-US multinationals • This occurs in same sectors driving the macro story • Even true for takeovers (with a lag) One possible interpretation is • US firms are managed in a way that make them more IT intensive, both in the US and as multinationals abroad • When IT prices fell rapidly in mid-1990s onwards they benefited more than European firms (2) Test with a second new dataset: on 720 firms, 1998-2005, which contains accounts, management and IT data, finding: • US firms & multinationals are indeed differently managed • This explains much of the higher US productivity of IT Macro facts and motivation Evidence from UK establishments Evidence from an EU panel Conclusion Why use UK micro data? • The UK has a lot of multinational activity – In our sample of 11,000 establishments 10% are US multinational and 30% non-US multinational – Frequent M&A generates also lots of ownership change • UK census data is well suited for this research – Data on IT and productivity for manufacturing and services (where much of the “US miracle” occurred) – Data from 1995 to 2003, the productivity miracle period (note: US Census has no annual service sector data) Descriptive statistics already show US multinationals are particularly different in IT use % difference from 4 digit industry mean in 2001 60 50 40 US Multinationals Non-US Multinationals UK domestic 30 20 10 0 -10 -20 -30 Employment Value added per Employee Non-IT Capital per Employee IT Capital per Employee Observations: 576 US; 2228 other MNE; 4770 Domestic UK Conceptually want to see if there are differences between US and European production functions Output (Q) function of TFP (A), Non-IT Capital (K), Labor (L), Materials (M) and IT-Capital (C) Q = A KαLβMγCδ Interested whether there is any difference between the US and Europe in the coefficients α, β, γ and δ Empirically will show: δUS>δEU and βUS<βEU Econometric Methodology (1) Estimate a production function for establishment i at time t: ln(Qit ) = ln( Ait ) + ! K ln( K it ) + ! L ln( Lit ) + ! L ln( M it ) + ! C ln(Cit ) K M C ln(Q / L)it = ln( Ait ) + ! ln( K / L)it + ! ln( M / L)it + ! ln(C / L)it + (1 " ! K " ! L " ! M " ! C ) ln( L)it Allow TFP and factor coefficients to vary by ownership (US, non-US multinational and domestic firms) Where Q = Gross Output K = Non-IT capital M = Materials A = TFP L = Labor C = IT capital Econometric Methodology (2): Other Issues • Include full set of SIC-3 digit industry dummies interacted with year dummies to control for output price differences • Main specifications also include establishment fixed effects • Standard errors clustered by establishment TABLE 2: PRODUCTION FUNCTIONS Depend Var Sectors Ln(Q/L) Ln(Q/L) Ln(Q/L) Ln(Q/L) Ln(Q/L) All All All IT Using Others USA×ln(C/L) 0.020*** 0.038*** 0.012 MNE×ln(C/L) 0.004 -0.001 0.006 0.046*** 0.043*** 0.037*** 0.046*** Ln(C/L) Ln(M/L) 0.558*** 0.547*** 0.548*** 0.622*** 0.507*** Ln(K/L) 0.139*** 0.127*** 0.127*** 0.111*** 0.146*** Ln(L) -0.005* -0.011*** -0.011*** -0.009** -0.012*** USA 0.071*** 0.064*** 0.073*** 0.044** 0.089*** MNE 0.039*** 0.034*** 0.037*** 0.015 0.044*** Obs 21746 21746 2175 7784 13962 USA×ln(C/L)=MNE×ln(C/L) 0.032 0.004 0.527 USA=MNE 0.011 0.176 0.015 0.021 0.023 Notes: Log (output/employees) is the dependent variable. C=‘IT Capital’, M=‘Materials’, K=‘Non-IT Capital’, L=‘Employees’, USA=‘USA Multinational’ and MNE=‘Non-US multinational’ (domestically owned is baseline). Stiroh (2002) “IT Intensive / Non-Intensive” and Services / Manufacturing split IT Intensive # obs IT non-intensive # obs Wholesale trade 2620 Food, drink and tobacco 1116 Retail trade 1399 Hotels & catering 1012 Machinery and equipment 736 Construction 993 Printing and publishing 639 Supporting transport 740 services (travel agencies) Professional business 489 services Real estate Industries (SIC-2) in blue are services and in black are manufacturing 700 Table 2, Production Functions with Fixed Effects Sectors IT Using Others Fixed effects YES YES USA×ln(C/L) 0.037*** -0.006 MNE×ln(C/L) -0.003 0.001 Ln(C/L) 0.012** 0.016*** Ln(M/L) 0.502*** 0.361*** Ln(K/L) 0.106*** 0.067*** Ln(L) -0.128*** -0.247*** USA 0.045 -0.007 MNE 0.017 -0.001 Observations 7,784 13,962 USA×ln(C/L)=MNE×ln(C/L) 0.009 0.521 Test USA=MNE 0.815 0.430 Note: C=‘IT Capital’, M=‘Materials’, K=‘Non-IT Capital’, L=‘Employees’, USA=‘USA Multinational’, MNE=‘Non-US multinational’ (domestic owned the baseline) Quantification suggests UK micro data can account for about half of US macro productivity surge • US firms have a 0.037 larger coefficient on IT (in IT sectors) • IT grew at around 22% per year 1995-2005 in (US and EU) • This implies a faster Q/L growth rate of 0.81% in the US (calculated as: 0.81%=0.037×22%) • IT sectors about ½ of all employment – so if applied to US economy would imply faster Q/L growth in US of about 0.4% • Since US productivity growth about 0.8% faster over 19952005 this suggests UK results can account for half of the gap • Even this probably an underestimate as IT grew faster in IT sectors than non-IT sectors Robustness Tests (1/2) - Endogeneity • Results due to reverse causation – e.g. – IT in US firms correlated with productivity shocks, but • Only in IT intensive industries (IT/non-IT > median, including retail, wholesale & high-tech manufacturing) • Only for US firms (not other multinationals) • Only for IT in US firms (not labor, capital or materials) • Unfortunately no clean natural experiment • As a partial check use Blundell-Bond GMM and Olley-Pakes and find results robust (Table A4) Table 3, Runs Some Robustness Tests Experiment All inputs interact Another IT measure Trans log USA×ln(C/L) 0.033** 0.065** 0.033** MNE×ln(C/L) 0.000 0.003 -0.001 -0.005 Ln(C/L) 0.013** 0.029*** 0.033 -0.025 Skills (wages) 0.028** Ln(Wage) 0.280*** Ln(Wage)×Ln(C/L) 0.012* Split out EU MNEs 0.038** 0.012** EU×ln(C/L) 0.002 Non-EU×ln(C/L) -0.014 USA×ln(C)= MNE×ln(C) 0.022 0.012 0.024 0.058 0.046 Obs 7,784 2,196 7,784 7,780 7,784 ‘All inputs interacted’ allows labor, capital and materials to interact with ownership – these are individually and joint insignificant. ‘Another IT measure’ is “% of employees using a computer” Robustness Tests (2/2) • Could this all be due to transfer pricing? – Higher US coefficient not observed for any other factor inputs (e.g. materials) – Takes time to arise (see takeover table 5) • Software – US multinationals have more/better software? – US multinationals global size the same as non-US multinationals (i.e. not a simple HQ fixed cost story) – Within US multinationals global size plays no role (the interaction global size with IT negative & insignificant) TABLE 4, IT INTENSITY EQUATION Dependent var: Sectors (1) (2) (3) (4) (5) (6) ln(C/L) ln(C/L) ln(C/L) ln(C/L) ln(C/L) ln(C/L) All IT Using Others All IT Using Other USA 0.263*** 0.339*** 0.209*** 0.241*** 0.313*** 0.193*** MNE 0.163*** 0.212*** 0.133*** 0.151*** 0.194*** 0.123*** Extra controls NO NO NO YES YES YES Observations 21,746 7,784 13,962 21,746 7,784 13,962 0.076 0.211 0.053 0.097 0.251 Test USA=MNE 0.031 Notes: All columns include SIC3 * time dummies & ln(Q). Additional controls = age, region & multi-plant. SE clustered by establishment. What About Unobserved Heterogeneity? • Maybe US firms “cherry pick” plants with high IT productivity? • Look at production functions before & after establishment is taken-over by US and non-US multinationals (domestic baseline) • No difference before takeover. After takeover results look very similar to table 3 (and interesting dynamics) Table 5, Before and After Takeovers Takeover timing: Before Before After USA×ln(C) -0.067 0.054*** MNE×ln(C) -0.043 0.007 USA -0.066 -0.106 0.062 MNE 0.032 -0.001 0.021 Ln(C) 0.074*** 0.094** 0.029*** After 0.029*** USA×ln(C), 1 year after 0.019 USA×ln(C), 2+years 0.066** MNE×ln(C), 1 year after -0.009 MNE×ln(C), 2+ years 0.012 Obs 261 USA×ln(C)=MNE*ln(C) 261 1,066 0.704 0.097 1,066 USA×ln(C)=MNE*ln(C), 1 year after 0.495 USA×ln(C)=MNE*ln(C), 2+ years 0.073 Macro facts and motivation Evidence from UK establishments Evidence from an EU panel Conclusion Why Do US firms have Higher IT productivity? Macro and micro estimates consistent with the idea of an unobserved factor which is • Complementary with IT • Abundant in US firms relative to others Range of possible explanations – one we think may explain part of this is the different management practices of US firms • Briefly sketch out the idea (model in the paper) • Provide a test using a new cross-country firm-level management, IT and performance dataset The Management Story Based on Prior Literature Literature suggests tough “people” management (hiring, firing, promotions & rewards) associated with higher IT productivity: • Econometric evidence in Caroli and Van Reenen (2001) and Bresnahan et al. (2001) • Case study evidence surveyed in Blanchard et al. (2004) Argument is IT changes informational flow, changing the optimal firm structure (Arrow, 1974). Good “people” management enables: • reorganization more quickly to exploit this • decentralization more effectively to allow experimentation Test Using New Firm-Level Management Practices Data Across Countries Developed questions on managerial & organizational practices • ~45 minute phone interview of manufacturing plant managers • Randomized from medium sized firms (100 to 5000 employees) Used “Double-blind” interviews to try to reduce survey bias • Interviewers do not know the company performance in advance • Managers are not informed (in advance) they are scored Getting firms to participate in the interview • Introduced as “Lean-manufacturing” interview, no financials • Official Endorsements (e.g. Bundesbank, PBC, RBI) • Run by 51 MBA types (loud, persistent & business experience) Example Management Question on Promotions See Appendix and Bloom and Van Reenen (2007) for details People Management by Country of Location Note: Uses 4,003 firms. Z-score of 4 people management questions (hiring, firing, promotion and rewards). People Management by Country of Origin Note: Uses 631 multinational subsidiaries in Europe. Z-score of 4 people management questions (hiring, firing, promotion and rewards) Aside: This is part of a set of results suggesting multinationals take domestic organizational and management practices abroad • Growing literature on multinationals often assumes they take firm-level ‘attributes’ across countries • Productivity – Helpman, Melitz and Yeapple (2004) • Communication/organization – Antras, Garicano & RossiHansberg (2008) • Management - Burstein and Monge (2008) • These results, and those in Bloom, Sadun and Van Reenen (2008) are completely consistent with this • Multinationals appear to have management and organizational characteristics partly based on their country of origin and partly based on their country of location We Matched the Firm-Level Management Data to Panel Company Accounts and IT Data • Obtained accounts for all European firms (public and private) • Purchased firm-level IT panel data from Harte-Hanks (an IT survey firm) for the European firms • HH runs annual surveys on all firms with 100+ employees • HH achieves about a 50% coverage ratio of this group • High quality data as sold for marketing purposes • Join cross-sectional management data with panel accounts and IT data, yields dataset on 719 firms with 2,555 obs TABLE 6: EU PANEL PRODUCTION FUNCTIONS Dependent Var: Ln(Q/L) Ln(Q/L) Ln(Q/L) Ln(Q/L) Ln(Q/L) USA×Log(C/L) 0.179** 0.078 0.052 MNE×Log(C/L) -0.026 -0.024 0.022 Management 0.019 0.019 Manag.×Log(C/L) 0.145*** 0.140*** 0.128* 0.184*** 0.178*** 0.179*** 0.235** 0.126*** 0.143*** 0.146*** -0.049 Log (K/L) 0.236*** Log(C/L) USA 0.270*** 0.078 0.111** 0.084* MNE 0.193*** 0.160*** 0.160*** 0.162*** Log(Degree) 0.043** 0.037** 0.037** (USA=MNE)×ln(C/L) 0.019 Firms 1633 719 Observations 7420 Fixed Effects NO 0.235 0.631 719 719 719 2555 2555 2555 2555 NO NO NO YES TABLE 6 CONTINUED: EU PANEL PRODUCTION FUNCTIONS AND IT INTENSITY Dependent Variable Ln(Q/L) USA×Log(PC/L) 0.019 MNE×Log(PC/L) 0.023 Ln(PC/L) People Management Ln(PC/L) 0.088*** Management×Log(PC/L) 0.099* Log (K/L) 0.232*** Log(PC/L) -0.228 USA 0.260*** 0.215*** MNE 0.049 0.037 Log(Degree)×Log(PC/L) 0.070 (USA=MNE)×ln(C/L) 0.955 0.001 0.027 Firms 719 719 719 Observations 2555 2555 2555 Fixed Effects YES NO NO Macro facts and motivation Evidence from UK establishments Evidence from an EU panel Conclusion Currently looking at why US firms have better people management • Bloom and Van Reenen (2007) suggest two factors important in improving overall US management practices – Greater product market competition – Fewer primo geniture family firms • Currently investigating two other factors that may play a role: – Lower labor market regulation in US – Higher skill levels in the US Both factors correlated with people management in our data • These two factors are also correlated with cross-country IT investment and productivity experience Labor market regulation and IT investment Source: GGDC Labor market regulation and productivity growth Source: GGDC Source: John Fernald, EF&G discussion Fall 2007 Flexible labor markets are correlated with IT use and productivity growth —but so is higher education IT Contribution to output growth, 1990-03 IT Contribution to output growth, 1990-93 1 1 0.8 0.8 US UK 0.6 Italy German Francey 0.4 0.2 0.6 UK German y France 0.4 0.2 US Italy 0 0 4 3 2 1 Employment Protection Index (Increasing flexibility →) 0 10 20 30 40 50 Share with tertiary education Sources: IT contribution to output growth (annual average, percentage points) and share with tertiary education from OECD. Employment Protection Index from Nicoletti et al (2000). 40 Conclusions 1) New UK census micro data: – US MNEs higher intensity of IT than non-US MNEs – Driven by sectors responsible for US “productivity miracle” – Magnitudes can account for ≈ ½ US productivity miracle 2) New international firm IT and management data: – Suggests US firms differently managed at home & abroad – This can explain much of the higher US intensity of IT use Currently working on trying to understand why US and other firms are differently managed and organized across countries Back Up Econometric Methodology (2) • TFP can depend on ownership (UK domestic is omitted base) ait = " USA h USA it D +" MNE h MNE it D ~ + ! h ' zit Non-US MNE US MNE • Coefficient on factor J depends on ownership (and sector, h) J it ! =! J ,0 h +! J ,USA h USA it D US MNE +! J , MNE h MNE it D Non-US MNE Empirically, only IT coefficient varies coefficient in US higher than non-US MNEs) significantly (IT Table A1 BREAKDOWN OF INDUSTRIES (1 of 3) IT Intensive (Using Sectors) IT-using manufacturing 18 Wearing apparel, dressing and dying of fur 22 Printing and publishing 29 Machinery and equipment 31, excl. 313 Electrical machinery and apparatus, excluding insulated wire 33, excl. 331 Precision and optical instruments, excluding IT instruments 351 Building and repairing of ships and boats 353 Aircraft and spacecraft 352+359 Railroad equipment and transport equipment 36-37 miscellaneous manufacturing and recycling IT-using services 51 Wholesale trades 52 Retail trade 71 Renting of machinery and equipment 73 Research and development 741-743 Professional business services BREAKDOWN OF INDUSTRIES (2 of 3) IT Producing Sectors (Other Sectors) IT Producing manufacturing 30 Office Machinery 313 Insulated wire 321 Electronic valves and tubes 322 Telecom equipment 323 radio and TV receivers 331 scientific instruments IT producing services 64 Communications 72 Computer services and related activity BREAKDOWN OF INDUSTRIES (3 of 3) Non- IT Intensive (Other sectors – cont.) Non-IT intensive manufacturing 15-16 Food drink and tobacco 17 Textiles 19 Leather and footwear 20 wood 21pulp and paper 23 mineral oil refining, coke and nuclear 24 chemicals 25 rubber and plastics 26 non-metallic mineral products 27 basic metals 28 fabricated metal products 34 motor vehicles Non-IT Services 50 sale, maintenance and repair of motor vehicles 55 hotels and catering 60 Inland transport 61 Water transport 62 Air transport 63 Supporting transport services, and travel agencies 70 Real estate 749 Other business activities n.e.c. 90-93 Other community, social and personal services 95 Private Household 99 Extra-territorial organizations Non-IT intensive other sectors 01 Agriculture 02 Forestry 05 Fishing 10-14 Mining and quarrying 50-41 Utilities 45 Construction