Innovation, productivity, and growth

All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
of 37

Please download to get full document.

View again

Innovation, productivity, and growth. Bronwyn H. Hall University of Maastricht and UC Berkeley. Innovation and productivity . Many papers looking at the link using data from the Community Innovation Survey and others like it My goal here: Provide a framework for interpreting results
Innovation, productivity, and growth Bronwyn H. Hall University of Maastricht and UC Berkeley Innovation and productivity
  • Many papers looking at the link using data from the Community Innovation Survey and others like it
  • My goal here:
  • Provide a framework for interpreting results
  • Draw some conclusions about how we might improve the data/analysis
  • Along the way, discuss employment impacts
  • Macro implications
  • KITeS - U. L. Bocconi Innovation and productivity
  • What are the mechanisms connecting innovation and productivity?
  • Improvements within existing firms
  • Creation of new goods & services, leading to increased demand for firm’s products
  • Process and organizational innovation leading to efficiency gains in production
  • Entry of more efficient firms
  • Entry offirms on technology frontier
  • Exitof less efficient firms
  • KITeS - U. L. Bocconi Measuring innovation
  • Large literature using R&D (capitalized) as a proxy for innovation input
  • Hall, Mairesse, Mohnen 2010 survey, inter alia
  • Smaller literature using patents as a proxy for intermediate innovation output
  • Both measures have well-known weaknesses, especially outside the manfacturing sector.
  • Now we have more direct measures – do they help?
  • KITeS - U. L. Bocconi Innovation surveys contain…..
  • Data on innovation:
  • Product or process new to firm/market (yes/no)
  • Share of sales during past 3 years from new products
  • More recent surveys have expenditures on various kinds of innovation investments
  • Data on productivity and employment:
  • Usually sales per worker (labor productivity)
  • Sometimes TFP (adjusted for changes in capital)
  • Issues arising from deflation and level of aggregation
  • of goods, and of enterprises More info: Mairesse and Mohnen (2010)
  • KITeS - U. L. Bocconi Raw data
  • Next slide – share of process and product innovators in selected sectors:
  • Manufacturing, telecommunications, computer services and software publishing, finance, and some technical professional services
  • As close as we can get to matching OECD coverage to US coverage
  • Suggests the difficulty in measuring innovation with a dummy
  • KITeS - U. L. Bocconi KITeS - U. L. Bocconi Interpretive framework
  • Innovation-productivity regression use revenue productivity data
  • Include coarse sectoral dummies
  • Relative within-sector price changes not accounted for
  • Quality change not generally accounted for
  • In the case of innovative activity, omitting price change at the firm level is problematic
  • alternative analysis - derived from Griliches and Mairesse 1984
  • KITeS - U. L. Bocconi Conventional productivity eq q= log value added (sometimes just output) c= log tangible capital l= log labor input ait = TFP (total factor productivity) Coefficients α, βmeasured as shares (growth accounting) or by regression (econometric) KITeS - U. L. Bocconi Revenue productivity If firms have market power and idiosyncratic prices, we observe real revenue r, not output q: r = p+q(all in logs) Add a CES demand equation: qit = ηpit , η<0 Then the revenue productivity relationship is If demand is inelastic (0>η>-1), revenue falls with increased output KITeS - U. L. Bocconi Adding innovation Add two terms involving knowledge stock: process: γkitin the production function, γ>0 product: φkitin the demand function, φ>0 This yields the following revenue function: Product improvement (-φ/η) always positive Process improvement (γ(η+1)/η) could be small or even negative KITeS - U. L. Bocconi Implication for prices Recall that qit= ηpit + φkit Then If demand elasticity is constant, price falls with innovation if γ-φ > 0 (recall η<0) That is, if efficiency enhancement effect outweighs product improvement effect Impact of innovation on price greater the more inelastic is demand, c.p. KITeS - U. L. Bocconi An example of price impact
  • U.S. deflators for the computer hardware industry and the communications equipment industry are hedonic (account for quality change)
  • see next slide
  • Deflate firm sales by these 2-digit deflators instead of one overall deflator
  • Result: true productivity is substantially higher than revenue productivity, because of hedonic price declines in the computer/electronics sector
  • Benefits of “Moore’s Law”
  • Banca d'Italia Hedonic Price Deflator for Computers Banca d'Italia Estimated R&D Elasticity – U.S. Manufacturing Firms GMM-system estimation with lag 3 & 4 instruments. Sample sizes: 7156, 6507, and 6457 observations Conclusion: much of the R&D in computing hardware went to lower prices for consumers (γ-φ> 0) Banca d'Italia What do the data say?
  • Productivity – look at results from a large set of papers that used the CDM model for estimation (Crepon Duguet Mairesse 1998):
  • Innovation survey data reveals that some non-R&D firms innovate and some R&D firms do not innovate
  • Data is usually cross-sectional, so simultaneity between R&D, innovation, and productivity
  • Sequential model: R&Dinnovationproductivity
  • Employment – results from estimation of a labor demand model with product & process innovation included
  • KITeS - U. L. Bocconi CDM model
  • Proposed originally by Crépon, Duguet and Mairesse (CDM, 1998)
  • Relationship among
  • innovation input (mostly, but not limited to, R&D)
  • innovation output (process, product, organizational)
  • productivity levels (sometimes growth rates)
  • Closer look at the black box of the innovation process at the firm level:
  • unpacks the relationship between innovation input and productivity by looking at the innovation output
  • Banca d'Italia The model parts
  • The determinants of R&D choice: whether to do it and how much to do.
  • Knowledge production function with innovation variables as outcomes as a function of predicted R&D intensity.
  • Production function including the predicted innovation outcomes to measure their contribution to the firm’s productivity. Need bootstrap s.e.s if sequentially estimated.
  • Banca d'Italia CDM model applied to CIS data
  • Estimated for 18+ countries
  • Confirmed high rates of return to R&D found in earlier studies
  • Like patents, innovation output statistics are much more variable (“noisier”) than R&D,
  • R&D tends to predict productivity better, when available
  • Next few slides summarize results for regressions of individual firm TFP on innovation
  • Source: Hall (2011), Nordic Economic Policy Review
  • KITeS - U. L. Bocconi Productivity-innovation relationship in TFP levels Innovation and Productivity TFP levels on innov sales share
  • Robustly positive, supports the view that product innovation shifts the firm’s demand curve out
  • Elasticities range from 0.04 to 0.29 with a typical standard error of 0.03
  • K-intensive and hi-tech firms have higher elasticities (=> equalized rates of return)
  • Coefficient of process innovation dummy usually insignificant or negative, suggesting either inelastic demand or (more likely) measurement error in the innovation variables
  • KITeS - U. L. Bocconi Productivity-innovation using dummies Innovation and Productivity TFP level results with dummies
  • Product dummy supports innovation sales share result, although noisier.
  • There is substantial correlation between product and process innovation, especially when they are instrumented by R&D and other firm characteristics.
  • May lead to bias in both coefficients (upward and downward)
  • KITeS - U. L. Bocconi Summary
  • Elasticity wrt innovative sales center on (0.09, 0.13)
  • higher for high tech and knowledge-intensive
  • Lower on average for low tech and developing countries, but also more variable
  • With product innovation included, process innovation often negative or zero
  • Without product innovation, process innovation positive for productivity
  • When not instrumented, little impact of innovation variables in production function (unlike R&D)
  • See Mairesse & Mohnen (2005), Hall et al. (2012)
  • TFP growth rates
  • Similar results, somewhat lower and noisier
  • KITeS - U. L. Bocconi Discussion
  • Innovation dummies at the firm level may be too noisy a measure to be useful.
  • Share of sales due to new products is more informative.
  • What measure would be useful (and reportable) for process innovation?
  • Further exploration with innovation investment (instead of R&D) is warranted
  • KITeS - U. L. Bocconi Aggregation
  • How does individual firm relationship aggregate up to macro-economy?
  • productivity gains in existing firms
  • exit and entry
  • Haltiwanger & co-authors have developed decompositions that are useful
  • Bartelsman et al. (2010): Size-productivity more highly correlated within industry if regulation is “efficient”
  • Evidence on Eastern European convergence
  • Useful approach to the evaluation of regulatory effects without strong assumptions
  • Similar analysis could assess the economy-wide innovation impacts
  • KITeS - U. L. Bocconi Entry and exit
  • Aghion et al (2009); Gorodnichenko et al (2010)
  • Competition and entry encourages innovation unless the sector is very far behind
  • Djankov (2010) survey – cross country
  • stronger entry regulation and/or higher entry costs associated with fewer new firms, greater existing firm size and growth, lower TFP, lower investment, and higher profits
  • Foster, Haltiwanger, and Syverson (2008) – US data
  • Distinguish between revenue and quantity, and include exit & entry
  • Revenue productivity understates contribution of entrants to productivity growth
  • Demand variation is a more important determinant of firm survival than efficiency in production (consistent with productivity impacts)
  • KITeS - U. L. Bocconi But Full set of links between innovation, competition, exit/entry, and productivity growth not yet explored KITeS - U. L. Bocconi BACKUP SLIDES The CDM Model and employment effects A brief overview of the CDM model Three blocks of equations
  • equations explaining the “R&D” decision and the amount of R&D performed
  • Innovation output equations (KPF) with R&D as input
  • Productivity equation, in which innovation output indicators appear as explanatory variables Estimation is recursive using single equation blocks, or simultaneous.
  • KITeS - U. L. Bocconi Econometrics (1) Only some firms report R&D; use standard selection model: Selection eq Conditional on doing R&D, we observe the level: Assume joint normality => generalized tobit or Heckman selection model for estimation. KITeS - U. L. Bocconi Econometrics (2) Output of the KPF are various binary innovation indicators or the share of innovative sales. For example, DI = Dummy for innovation (process, product, organizational) Why include the latent R&D variable RD*?
  • Account for informal R&D effort that is often not reported
  • Instrument for errors in variables and simultaneity Estimation is via multivariate probit
  • KITeS - U. L. Bocconi Econometrics (3) Production function: y= log sales per employee k= log capital stock per employee DIare predicted probabilities of innovation from second step or predicted share of innovative sales (with logit transform) Zincludes size, age, industry, region, year, wave Estimated by OLS KITeS - U. L. Bocconi What about employment? Assume capital C and knowledge stock K are predetermined. Can show optimal labor choice is Similar conclusion for labor as for output (if demand is elastic or not very inelastic):
  • Product improvement (-φ/η) always positive
  • Process improvement (γ(η+1)/η) can be negative
  • KITeS - U. L. Bocconi 2. Employment
  • Uses an equation in growth rates, allowing for price changes: l = employment growth (i = industry) π = growth of sector price deflator g1, g2 = growth in sales of old, new products Dproc= dummy for process innovation β= relative efficiency of producing new vs old products φ1, φ2 = rel. change in price of old, new products If2>0, the quality improvement of the new prod is passed to consumers via higher prices (lower employment impact, c.p.) If2<0, quality improvement leads to lower “effective” prices
  • KITeS - U. L. Bocconi Results for Europe Manufacturing sector firms 1998-2000
  • Labor efficiency of production of old and new products roughly the same (except possibly in Italy)
  • Process innovation has no impact in Italy, France, and Spain, leads to reduced labor in Germany & UK (increased efficiency)
  • => Suggests the importance of labor market regulation, although effects are fairly small.
  • KITeS - U. L. Bocconi Decomposition of e growth Manufacturing sector firms 1998-2000 KITeS - U. L. Bocconi
    Related Search
    We Need Your Support
    Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

    Thanks to everyone for your continued support.

    No, Thanks