Measuring he reurns o innovaion (1) Prof. Bronwyn H. Hall Globelics Academy May 26/27 2005 Ouline This morning 1. Overview measuring he reurns o innovaion 2. Measuring he reurns o R&D using produciviy regressions 3. Measuring he privae reurns o R&D using marke value equaions This afernoon 1. Paen daa 2. Measuring innovaion using paen daa May 2005 Globelics Academy - Lisbon 2 Why is his an ineresing problem? Economiss Models of innovaion and growh Managers Allocaion of resources for invenion Measure resuls of innovaion Accounans Accurae reporing of inangible value in company accouns Policy makers How much o spend on innovaion? wha policy insrumen o use? How o choose he level of subsidy? Evaluaion of resuls May 2005 Globelics Academy - Lisbon 3 Framework for analysis Invesmen in innovaion (R&D, raining, licenses) creaes an asse ha pays off in he fuure Enerprise level: asse ends o become less producive over ime (i depreciaes) Indusry/counry/world level: invesmens in innovaion by many agens creae aggregae knowledge asse depreciaes more slowly - when privae firms no longer earn reurns from an innovaion, he knowledge hey have creaed remains useful May 2005 Globelics Academy - Lisbon 4 Framework for analysis Innovaion invesmen R a ime = R Knowledge asse K = f(r, R -1, R -2, ) PDV ( π( K, X )) f ( R, R 1, R 2,...) Gross rae of reurn ρ K R Ne rae of reurn ρ δ where PDV = presen discouned value X = oher inpus δ = depreciaion of innovaion asses π(k) = profis or welfare produced by K May 2005 Globelics Academy - Lisbon 5 Map of innovaion inpus and oupus Demand pull (ase, marke size) Tech. opporuniy (science base) Economy-wide knowledge Paens, oher IPR Marke srucure and indusry; appropriabiliy environmen; Governmen insiuions R&D and oher innovaion invesmens Knowledge firm-level capial creaed by innovaion invesmen Innovaion oupu Firm size and marke share, diversificaion, ech ransfer and experience Innovaive sales Oucomes: Produciviy, Diffusion Physical capial Profiabiliy, Value, process Worker skills and Economic Growh May 2005 Globelics Academy - Lisbon 6 1
Inpu measures R&D spending wihin firm alliance and join venure paricipaion Purchase of new capial equipmen Technology purchases/licensing Markeing relaed o new producs Training and educaion of workers Spillover variables Based on geography or echnology Innovaion survey variables Wheher a firm is innovaive Sources of knowledge suppliers, parners, consumers, inernal May 2005 Globelics Academy - Lisbon 7 Inermediae inpus/oupus Paen couns Raw Weighed by ciaions received Innovaion/new produc couns From news journals From surveys Innovaion surveys shares of sales ha is New o marke (radical?) New o firm (incremenal?) May 2005 Globelics Academy - Lisbon 8 Oupu measures Individual innovaions Licensing fees Paen renewals as a funcion of fee schedule (Schankerman-Pakes) Surveys (Harhoff, Scherer, PATVAL) Firm level Profis or revenue produciviy Sock or financial marke value - covers a broad range of echnology & indusry, bu requires acive sock marke (Griliches, Hall, ec.) Economy level (social reurns) Consumer willingness-o-pay (Trajenberg) Relaing inpus and oupus 1. Producion funcion approach privae and/or social reurns 2. Marke value approach privae reurns 3. Paens as indicaors of innovaion aciviy 4. Using innovaion surveys Aggregae produciviy growh May 2005 Globelics Academy - Lisbon 10 May 2005 Globelics Academy - Lisbon 9 1. Producion funcion framework Cobb-Douglas producion firs order log approximaion o producion funcion general ool o relae quan measures of oupu o inpu Line of business, firm, indusry, or counry level Variey of esimaing equaions: Convenional producion funcion Parial or oal facor produciviy funcion R&D inensiy formulaion Semi-reduced form (add variable facor demand equaions) May 2005 Globelics Academy - Lisbon 11 Concepual issues Wha is oupu? usual measures exclude benefi of governmen spending on R&D defense, environmen, healh Unmeasured qualiy change and new goods Revenue or oupu? Wha is knowledge capial? Varying lags in producing knowledge Depreciaion is endogenous a he firm level Own capial depends on he effors of ohers as well as he firm iself (spillovers) May 2005 Globelics Academy - Lisbon 12 2
Produciviy approach (1) α β γ u Y = AL C K e L = labor C = capial u = random shock K = research or knowledge capial, consruced from invesmens R: K = (1 δ) K + R 1 May 2005 Globelics Academy - Lisbon 13 Produciviy approach (2) Take logarihms and model he inercep wih year and firm (or indusry) effecs: y = η + λ + αl + βc + γk + u i i i i i i i = 1,..., N = 1,..., T Economerics: The error u may possibly be correlaed wih he curren (and fuure) inpu levels. The firm effec η may also be correlaed wih inpu levels. May 2005 Globelics Academy - Lisbon 14 Alernaive formulaions Differencing o remove firm effec: y = λ + α l + β c + γ k u i i i i i R&D inensiy version: R δk i i, 1 Ri ki = K K i, 1 i, 1 if depreciaion δ is near zero Y i Ri Ri γ k i γ = ρ K i Yi Yi Yi where ρ =, he gross rae of reurn o R&D capial K i May 2005 Globelics Academy - Lisbon 15 Alernaive formulaions Parial or oal facor produciviy version: Parial: y ˆ α l = λ + β c + γ k u i i i i i Toal: y ˆ α l ˆ β c = λ + γ k u i i i i i α and β may be esimaed using facor shares a he firm level (when available). Ofen combined wih he R&D inensiy approach. Noe change in he assumpions on u required for consisen parameer esimaes. May 2005 Globelics Academy - Lisbon 16 Some measuremen issues Daa is sales, no value added Assume maerials share consan across ime for each firm Resul is ha coefficiens are inflaed by (one-share of maerials) confirmed in pracice Double couning of R&D (Schankerman 1981) R&D expendiure is also in labor and capial Under simple assumpions, elasiciy is downward biased by share of R&D in growh of labor/capial Effecs of choice of deflaors (inpu and oupu) May 2005 Globelics Academy - Lisbon 17 Simulaneiy Sources of endogeneiy: Inpus and oupu chosen simulaneously - favorable produciviy/profis experience leads o increased R&D effor in he fuure Firm knows is efficiency level (fixed effec) when choosing inpus Inpus measured wih error Soluions Difference o remove fixed effec, exacerbaes measuremen error bias Toal or parial produciviy moves some inpus o lef hand side Insrumenal variables, GMM for panel daa May 2005 Globelics Academy - Lisbon 18 3
French Firms 1981-1989 Sales vs Value added Dep var Log S Log VA Log C.043 (.002).193 (.008) Log K.024 (.001).092 (.004) Log L.193 (.005).699 (.012) (1-.74)*VA Coeff.050.024.183 Log M.735 (.004) -- Sum 0.995 0.984 0.257 R 2.993.926 s.e..115.349 Source: Mairesse and Hall 1999 May 2005 Globelics Academy - Lisbon 19 197 French firms 1980-1987 Log(C/L) Log(K/L) s.e. Double couning Unadjused.21 (.01).18 (.01) logl -.03 (.01) -.04 (.01) -.00 R 2.996.996.998.336 Adjused.20 (.01).25 (.01).344 Source: Hall and Mairesse 1995 Pooled OLS esimaes Parial Produciviy Labor share = 0.67.11 (.01).22 (.01).347 Labor share esimaed -.05 (.02).49 (.02).10.974 1.234 May 2005 Globelics Academy - Lisbon 20 French Firms 1981-1989 Toal Log C/L.20 (.01) Log K/L.25 (.01) Dep Var = log(y/l) Wihin Long diff. Firs diff..17 (.06).20 (.13).23 (.09).07 (.03).13 (.03).05 (.07) Log L -.04 (.01) -.06 (.05).17 (.12) -.60 (.10) R 2.996.103.030.183 s.e..344.186.051.193 Source: Hall and Mairesse 1995 Noe: all esimaes are inconsisen if RHS vars no sricly exogenous; wihin are probably leas biased. May 2005 Globelics Academy - Lisbon 21 Approximae gross rae of reurn Y Y ρ = = γ R R Large R&D-doing manufacuring firms Counry Y R/Y γ dy/dr France (1981-1989) VA UK (1988-1996) Sales Germany (1988-96) Sales 4%.069 1.72 2.42%.065 3.30 5.84%.079 1.35 US (1990-1998) Sales 8.00%.118 1.48 Chile (1998) VA 1.5%.131 8.7 May 2005 Globelics Academy - Lisbon 22 Oupu deflaion a he firm level Inerpreing produciviy growh regressions a he firm level: versus y = λ + α l + β c + γ k u (1) i i i Y i i s y p λ α l β c γ k u (2) i = i i = i i + S i i If (2) is esimaed insead of (1), we obain an esimae of γ = γ + γ S Y P The revenue produciviy of R&D is he sum of rue produciviy of R&D he effec R&D has on he prices a which goods are sold (due o qualiy improvemens, produc differeniaion, and cos reducion) May 2005 Globelics Academy - Lisbon 23 Inerpreaion Revenue produciviy is a deerminan of privae reurns True produciviy (more consan qualiy oupu for a given se of inpus) is relevan for social reurns The difference represens pecuniary exernaliies benefis received by downsream producers and consumers in he form of lower prices in some cases, hese can be large May 2005 Globelics Academy - Lisbon 24 4
Illusraion Some deflaors a he indusry level are hedonic (in he US and some OECD daa) e.g., for he compuer indusry and he communicaions equipmen indusry Deflae firm sales by 2-digi deflaors insead of one overall deflaor rue produciviy is subsanially higher han revenue produciviy, because of price declines in hese R&D-inensive indusries innovaion invesmens largely direced a produc improvemen (~2/3 of R&D) May 2005 Globelics Academy - Lisbon 25 Index number Hedonic Price Deflaor for Compuers 3.0 2.5 2.0 1.5 1.0 0.5 0.0 Shipmens Deflaors for U.S. Manufacuring NBER Barlesman-Gray Produciviy Daabase 1961 1965 1969 1973 1977 1981 1985 1989 1993 1997 May 2005 Globelics Academy - Lisbon 26 Year Compuers & elecronics Insrumens & Comm. Equip. Oher manufacuring Esimaed R&D Elasiciy U.S. Manufacuring Firms Dep. Var = Log Sales, 2-digi deflaors (Y) Difference ("price effec") (P) Dep. Var = Log Period Sales (S) 1974-1980 -.003 (.025).102 (.035) -0.099 1983-1989.035 (.030).131 (.049) -0.096 1992-1998.118 (.031).283 (.041) -0.165 Firm sock marke value Measuremen of privae reurns o invesmen in innovaion Mehod of esimaion is GMM-sysem wih lag 3 and 4 insrumens. Sample sizes for he hree subperiods are 7156, 6507, and 6457. γ S = γ Y + γ P May 2005 Globelics Academy - Lisbon 27 May 2005 Globelics Academy - Lisbon 28 Why marke value? Reurns o innovaion are he profis earned in he fuure from invesmens made oday Forward looking measure, allows ineremporal producion of innovaions Under an efficien markes assumpion, equal o he expeced value of he discouned cash flows ha will be received in he fuure from he asses of he firm Observable for a wide range of firms and counries (alhough no as wide as we would like) Measuring inangible asses a presen-day problem for economiss and accounans Exploring his mehodology helps our undersanding of how o measure innovaion asses May 2005 Globelics Academy - Lisbon 29 Theoreical framework Measured marke value = value funcion associaed wih firm s profi-maximizing dynamic program References Hayashi (Economerica 1982) condiions under which marginal = average Q (including axes) Wildasin (AER 1982) same hing for muliple capials Hayashi & Inoue (Economerica 1991) same model wih capial aggregaor funcion May 2005 Globelics Academy - Lisbon 30 5
Theoreical Q model Tobin s original Q = raio of he marke value V of a (unique) asse o is replacemen cos A Q>1 => inves o creae more of he asse Q<1 => disinves o reduce asse Q=1 in equilibrium Hayashi (1982) - he asse is a firm derived Q from he firm s dynamic program gave condiions under which marginal Q (dv/da) equal o average (V/A) Hayashi-Inoue (1991) and Wildasin (1984) developed he heory wih more han one capial May 2005 Globelics Academy - Lisbon 31 Pracice: hedonic regression Linear approx: V i (A i,k i ) = b [A i +γk i ] log V i - log A i = log Q i = log b + γ K i /A i Non linear: log Q i = log b + log(1+γ K i /A i ) Q i =V i /A i is Tobin s q b = overall marke level (approximaely one) K i /A i = raio of inangible innovaion asses o angible γ = relaive shadow value of K asses (γ = 1 if depreciaion correc, invesmen sraegy opimal, and no adjusmen coss). May 2005 Globelics Academy - Lisbon 32 Typical firm s balance shee Asses (denominaor) Propery, plan, & equipmen Invenories Invesmens in oher firms Shor erm financial asses; cash; receivables Good will; booked invesmen in inangibles Inangibles no on balance shee Liabiliies (numeraor) Common sock Preferred sock Long erm deb; bonds Shor erm deb; bank loans; payables Subordinaed deb; oher financial claims Owner s equiy (residual) May 2005 Globelics Academy - Lisbon 33 Wha belongs in he value eq? Only he asses (resource base) of he firm Physical capial (A) Knowledge capial (K), including IT capial such as sofware Purchased inangibles (I) Repuaional capial, brand name value (sock of adverising) Human capial, o he exen ha i is no capured in wages Oher infrasrucural capial, such as he exisence of a disribuion nework No such hings as growh in sales or profiabiliy unless hey are used as proxies for lef-ou ypes of capials (similarly for fixed effecs) May 2005 Globelics Academy - Lisbon 34 Consrucing innovaion socks K = (1-δ)K -1 + R where K = knowledge sock a end of period R = flow of innovaion invesmen during δ = depreciaion rae of K, usually = 15% If R grows a a consan rae g over ime, hen K R /(δ+g) Used Acual g, δ 5%, 15% 5%, 45% γk γ 5R 2.5γ 2R Empirical evidence Indusry aggregaes - indusries wih high Q have high R&D inensiy Firm-level Funcional form? Changes over ime Low coefficien on K or R may imply δ >>0.15 May 2005 Globelics Academy - Lisbon 35 May 2005 Globelics Academy - Lisbon 36 6
Log of Tobin's Q Median Q versus Median K/A by Indusry 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 Pharmaceuicals Compuer sofware -1.5-1.0-0.5 0.0 0.5 1.0 Log of R&D sock o asses raio LogQ = 0.58 + 0.40 log(k/a) May 2005 Globelics Academy - Lisbon 37 Median Q and K/A for seleced indusries in US Indusry K/A Q=V/A Pharmaceuicals 3.39 8.92 Compuer sofware 2.92 8.61 Compuing equipmen 1.44 3.68 Medical insrumens 0.96 3.81 Auos 0.18 1.65 Prining and publishing 0.15 2.08 Rubber & plasics 0.15 1.61 Telecommunicaion services 0.12 2.27 Food & obacco 0.09 2.16 Primary meals 0.06 1.28 Lumber & wood 0.04 1.14 May 2005 Globelics Academy - Lisbon 38 K/A Coefficien Relaive Marke Value of R&D Sock - U. S. Manufacuring Secor 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00 1972 1976 1980 1984 1988 1992 1996 2000 Year Loose rim (.1<q<10;KA<5) Tigh rim (.2<q<5; KA<1) May 2005 Globelics Academy - Lisbon 39 A Puzzle? Compare changes 1972-1999 1. Marke value of R&D capial using hedonic model 2. Revenue produciviy of R&D capial 3. Average R&D o sales raio Resuls 1. Marke value declines during 1980s from 1 o around.2 2. R&D produciviy increases seadily from.02 o.10 3. Firms invesmen rae jumps during 1980s from.02 o.04. Why? May 2005 Globelics Academy - Lisbon 40 Coefficien or raio U. S. Trends in R&D Produciviy R&D in U.S. Manufacuring 0.12 1.60 1.40 0.10 1.20 0.08 1.00 0.06 0.80 0.60 0.04 0.40 0.02 0.20 0.00 0.00 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 Year R&D Produciviy Average R&D Inensiy Marke Value of R&D May 2005 Globelics Academy - Lisbon 41 Some open quesions Relaionship beween firm-level (revenue) produciviy and aggregae produciviy Puzzles Has he produciviy of R&D declined? How do we reconcile R&D inensiy and R&D growh versions of producion funcion? Marke value and produciviy versions of rae of reurn compuaion? Firm and indusry resuls? R&D Sock compuaion R&D is cumulaive, creaes knowledge Decay of useful knowledge no he same as decay of privae reurns from ha knowledge How o measure and accoun for his fac in our models? May 2005 Globelics Academy - Lisbon 42 7