Missing Growth from Creative Destruction Philippe Aghion (College de France & LSE) Antonin Bergeaud (LSE) Timo Boppart (IIES) Pete Klenow (Stanford) Huiyu Li (FRB SF) January 2017 Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 1 / 37
Introduction Introduction Products that are newly introduced and cease to exist, constitute a main challenge to estimate time variations in Price Index (PPI, CPI) about 40% of items in the PPI and 30% in the CPI exit in a typical year Standard procedure is to resort to imputation, i.e. to use price changes of surviving products to infer the overall price change In this paper: we argue that imputation leads to an underestimation of the productivity gains from innovation we quantify the resulting missing growth for the overall US economy over the past three decades Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 2 / 37
Introduction Basic model Time is discrete and in each period the economy is populated by mass L of identical one-period lived individuals who consume a final good. Final good production: ( N Y = 0 [q ω y ω ] σ 1 σ ) σ σ 1 dω, where y ω and q ω are the quantity and quality of intermediate input ω, N denotes the number of intermediate varieties currently available, and σ > 0 denotes the elasticity of substitution between intermediate inputs. Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 3 / 37
Introduction Innovation (1) Creative destruction by new entrant (d) : arrival rate λ d [0, 1) in any sector ω, step size γ d = q ω,t+1 /q ω,t. Own improvement by incumbent (i) : arrival rate λ i [0, 1) in any sector ω, step size γ i = q ω,t+1 /q ω,t. Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 4 / 37
Introduction Innovation (2) New varieties (n) at rate λ n new varieties may come at an above average quality by factor γ n Product VS process innovation Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 5 / 37
Introduction Source of missing growth True real output growth: P t Y t+1 = M t+1., Y t M t P t+1 where M = YP is aggregate nominal output (or aggregate expenditure on the final good), and P denotes the aggregate price index Measured real output growth: Ŷ t+1 Y t = M t+1 M t. P t P t+1 Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 6 / 37
Introduction Missing Growth Thus missing growth is entirely due to overstated (quality-adjusted) inflation: MG t+1 = ln( Y t+1 ) Y ln(ŷt+1 ) = ln( P t+1 ) ln( P t+1 ) t Y t P t P t Our assumption is that the Statistical Agency imputes inflation rate from disappearing products based on inflation rate of surviving products. CPI and PPI We obtain: MG = 1 σ 1 log ( 1 + λ d [ γ σ 1 d 1 λ i ( γ σ 1 i 1 + λ i ( γ σ 1 i 1 ) 1 )] + λ n γ σ 1 n ). Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 7 / 37
Market Share Approach Estimating missing growth using market shares Here we estimate missing growth using the market shares of entrant plants/establishments, of incumbent plants that stay in the market, and of exiters Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 8 / 37
Market Share Approach Estimating missing growth using market shares Let s It 1,t denote the aggregate market share in period t of the set of plants operating in both periods t and t 1 (we refer to those firms as continuers from date t 1). Let s It 1,t 1denote the aggregate market share in period t 1 of the same continuers. Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 9 / 37
Market Share Approach Estimating missing growth using market shares Missing growth can be expressed as: ( ) ( ) Pt P t MG t = ln + ln = 1 P t 1 P t 1 σ 1 ln ( sit 1,t 1 s It 1,t ) (1) Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 10 / 37
Market Share Approach Estimating missing growth using market shares Thus true growth will exceed measured growth whenever the market share of continuing incumbents shrinks over time. Intuitively: the difference between true growth and measured growth is equal to the difference between true growth and incumbent average productivity growth......and the market share of incumbents shrinks whenever the average productivity of continuing incumbents grows more slowly than average productivity of the economy. Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 11 / 37
Market Share Approach Estimating missing growth using market shares We do the analysis at the plant level and use the Longitudinal Business Database (LBD) which covers all plants with at least 1 employee from 1976-2013. We focus on period 1983-2013. We then infer s It 1,t 1 and s It 1,t from the LBD information on employment or payroll as measures of relative market shares. Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 12 / 37
Market Share Approach Missing Growth implied by Survivors Market Shares % points per year Missing Measured True 1983 2013 0.56 1.93 2.49 1983 1995 0.60 2.01 2.61 1996 2005 0.41 2.65 3.06 2006 2013 0.69 0.90 1.59 Employment vs Payroll Plant vs Firms Choice of k Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 13 / 37
Indirect inference (1) Rely on algorithm in Garcia-Macia, Klenow and Hsieh (2016) (GHK) GHK uses data from the LBD for two time periods: 1976-1986 and 2003-2013. Over those two time intervals, they compute average (measured) TFP growth, exit rate of firms by age, employment, employment by age, job destruction and creation... to infer arrival rates and step size of innovations Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 14 / 37
Indirect inference (2) We run GHK codes for different initial values of aggregate total productivity growth g u. For each g u we derive the corresponding (λ s, γ s) u by running the GHK algorithm, which in turn yields a value G u for measured growth. We stop at u such that the corresponding computed value of measured growth ĝ u is equal to the actual measured TFP growth rate Missing growth is then taken to be equal to MG = g u ĝ u. Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 15 / 37
Indirect inference (3) Key advantages need not assume that CD and NV only come from new plants need not assume a constant number of products per plant Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 16 / 37
Result from indirect inference σ = 4 1 plant = 1 variety (market share) 1976-1986 0.91 0.46 from CD 0.85 2003-2013 1.29 0.68 from CD 1.19 average over 1983-1986. Akcigit Kerr (2010) Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 17 / 37
Conclusion In this paper we developed a Schumpeterian growth model which suggested two ways of assessing the magnitude of missing growth from imputation. Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 18 / 37
Conclusion The first approach is simple and intuitive and already suggests that MG is large ( 0.6 ppt per year on average) but it relies on simplifying restrictions and in particular it abstracts from CD by incumbents The second approach does not rely on such restrictions, and allows for CD by incumbents, thereby yielding higher MG ( 1.1 ppt per year on average) it allows us to decompose MG into its different sources, and we find that most of it comes from CD Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 19 / 37
Conclusion Some implications 1 Ideas may no get as hard to find as official statistics suggest. 2 US FED may wish to raise its inflation target to achieve price stability. Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 20 / 37
Result from Akcigit Kerr (2010) Back Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 21 / 37
Declining Dynamism 1 Establishments vs. firms 2 Net entry vs. gross entry 3 5-year lag vs. year entered Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 22 / 37
Declining Dynamism Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 23 / 37
Manufacturing All Mfg. Non-Mfg. 1983 2013 0.56 0.03 0.67 1983 1995 0.60 0.23 0.71 1996 2005 0.41-0.13 0.51 2006 2013 0.69-0.07 0.79 Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 24 / 37
Product versus Process innovation If process innovation: Unit prices fall with innovation Might be easy to measure growth from CD Data: elasticity of unit prices wrt revenue 0. e.g. Hottman, Redding and Weinstein (2015) Consistent with product innovation. Back Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 25 / 37
Rate of imputation on CD CPI: 97.4% PPI: If no price report from a participating company has been received in a particular month, the change in the price of the associated item will, in general, be estimated by averaging the price changes for the other items within the same cell (i.e., for the same kind of products) for which price reports have been received. Back Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 26 / 37
Methods to deal with missing prices in the CPI (1) When the BLS cannot track a same item over time (see General Accounting Office (GAO) Report, 1999)) No adjustment when new item and old item are deemed comparable Otherwise Direct quality adjustment Class-mean imputation Linking imputation Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 27 / 37
Methods to deal with missing prices in the CPI (2) Class-mean imputation is based on the rate of price changes experienced by other substitutions (those that the commodity analyst had considered comparable or had directly adjusted) Linking method is based on the rate of price changes of all products (mostly surviving) Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 28 / 37
Methods to deal with missing prices in the CPI (3) GAO report (1999) 58% of substitutions were deemed comparable (these prices enter the CPI without adjustment) The adjustments to non-comparable substitute price broke down as: 31.1% direct quality adjustments 33.4% class-mean imputation 35.5% linking imputation Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 29 / 37
Methods to deal with missing prices in the CPI (4) We make three assumptions: 1 Comparable substitutions do not involve any innovation. 2 The direct adjustments occur only with some of the incumbent improvements to their own products (OI). 3 CD results in imputation by class-mean or linking in the proportions above. Under these assumptions, CD was treated with the equivalent of all-items-in-a-category imputation effectively 97.4% of the time in 1997. Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 30 / 37
Evidence for imputation in other countries PPI Evidence from Eurostat s Handbook on industrial producer price indices (PPI) Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 31 / 37
Alternative measurements (1) Employment Payroll 1989 2013 0.60 0.69 1989 1995 0.77 0.97 1996 2005 0.41 0.38 2006 2013 0.69 0.83 Back Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 32 / 37
Revenue vs. Employment (1) Back CMF (Census of Manufacturing Firms) is every five years so we only know the *cumulative* market share growth of incumbent plants that survive for at least 5 years For these survivors, market share by revenue grew less fast than market share by employment hence MG is higher under revenue share than under employment share CMF and LBD yield similar MG when using employment share Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 33 / 37
Revenue vs. Employment (2) Evidence using French firms Back Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 34 / 37
Alternative measurements (2) New New Plants Firms 1983 2013 0.56 0.08 1983 1995 0.60 0.29 1996 2005 0.41-0.03 2006 2013 0.69-0.14 Back Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 35 / 37
Alternative measurements (3) 5 year old 3 year old 0 year old plants plants plants 1983 2013 0.56 0.47 0.20 1983 1995 0.60 0.54 0.28 1996 2005 0.41 0.38 0.20 2006 2013 0.69 0.46 0.07 Back Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 36 / 37
Motivation for using k=5 Back Aghion, Bergeaud, Boppart, Klenow and Li Missing Growth from Creative Destruction January 2017 37 / 37