The Rise of Market Power and the Macroeconomic Implications

The Rise of Market Power and the Macroeconomic Implications Jan De Loecker 1 Jan Eeckhout 2 1 Princeton and University of Leuven 2 University College London and UPF NBER Summer Institute 18 July, 2017

Motivation Several secular trends in last decades: wages, LF participation, labor share, labor mobility, migration rates, capital share, output growth slowdown Propose common cause: the rise in market power since 1980 Little known about evolution & cross-section markup in macro 1. Data needed: long time series of firm-level data 2. Estimation methods: demand approach uses model of consumer behavior and competition This paper: 1. Document time-series and cross section of markup 1950-2014 2. Cost-based method; no inference from demand; mkt structure 3. Illustrate how this can explain 7 secular trends since 1980 We have nothing (little) to say about causes of change markup

Motivation Secular Increase since 1980: markup 1.18 1.67 1.7 1.6 Share weighted Markup 1.5 1.4 1.3 1.2 1.1 1960 1970 1980 1990 2000 2010 year Figure: The Evolution of Average Markups (1950 2014), weighted

Data Compustat data on publicly listed firms: Long time series: 1950 2014 Broad Cross Section: average 5,000 firms per year Selection? Large firms; miss many small firms Small subset of all firms Publicly traded privately held firms But: Covers all sectors and industries (contrast: Cens. of Manuf.) 40-45% GDP; 35% employment (Cens. of Manuf. 8.8%) Allow for markup variation across producers and time; heterogeneity has substantial economic implications

Production technology Producer Behavior Q it (V it, K it, Ω it ) = F it (V it, K it )Ω it, where V it : variable inputs (labor, intermediate inputs) K it : capital stock Ω it : Hicks-neutral productivity term (TFP) Associated Lagrangian function (with one composite input): L(V it, K it, λ it ) = P V it V it + r it K it λ it (Q it ( ) Q it ) Consider FOC wrt the variable input V : L it = P V Q it ( ) it λ it = 0 V it V it Rearranging expression of output elasticity of input V it : θ V it Q it( ) V it V it = 1 Pit V V it Q it λ it Q it

Producer Behavior Lagrangian multiplier λ is a direct measure of marginal cost Define markup µ = P λ or µ it = θit V P it Q it Pit V V. it depending on Sales S it = P it Q it and expenditure share θ V it, which is specific to technology Method: Hall (1988): aggregate data; De Loecker-Warzynski (2012): micro data

Robustness Benchmark Industry-Specific CD 1.7 1.7 Share weighted Markup 1.6 1.5 1.4 1.3 1.2 1.6 1.5 1.4 1.3 1.2 1.1 1.1 1960 1970 1980 1990 2000 2010 year 1960 1970 1980 1990 2000 2010 year Share weighted Markup Markup (ind techn.) Census Manufacturing Translog 1.7 2.5 1.7 1.6 1.5 1.4 1.3 1.2 Markup (ind techn. Translog) 2 1.5 1.6 1.5 1.4 1.3 1.2 Share weighted Markup 1.1 1960 1970 1980 1990 2000 2010 year Share weighted Markup Share weighted markup (Manufacturing) 1960 1970 1980 1990 2000 2010 year Markup (ind techn. Translog) Share weighted Markup

Cross-Sectional Decomposition Higher Markups for Smaller Firms U t = i s it µ it = µ t + i (s it s t )(µ it µ t ) where: U t : weighted average markup µ t : unweighted average markup µ it : firm i s individual markup s it : firm i s market share of sales Cov(s it, µ it ) < 0 Higher Markups for Smaller Firms

Cross-Sectional Decomposition Higher Markups for Smaller Firms 2.5 2 1.5 1 1960 1970 1980 1990 2000 2010 year Share weighted Markup mu_mean Figure: Unweighted vs. Weighted Average Markup

Time-Series Decomposition Predominantly Within Industry, in All Industries U t = s s,t 1 µ st + µ s,t 1 s s,t + µ s,t s s,t. s s s }{{}}{{}}{{} within between reallocation Markup Markup Within Between Realloc. 1964 1.319 0.135 0.067-0.011 0.079 1974 1.231-0.088-0.084 0.042-0.046 1984 1.236 0.004-0.008 0.025-0.012 1994 1.360 0.124 0.126 0.004-0.007 2004 1.519 0.159 0.116 0.031 0.012 2014 1.667 0.151 0.187-0.018-0.020 Table: Decomposition 10 year change in Markup at 4-digit industry

Dispersion of Markup All Action in Upper Half Distribution 2.5 2 1.5 1 1960 1970 1980 1990 2000 2010 year Share weighted Markup p90 (ms) p50 (ms) p75 (ms)

Dispersion of Markup All Action in Upper Half Distribution 2.5 2 1.5 1 1960 1970 1980 1990 2000 2010 year Share weighted Markup p90 (ms) p50 (ms) p75 (ms) Harberger (1954): roughly equally distributed profits; not now

Markup = Market Power? Dividends and Market Value 1.7 2000000 1.7 8.00e+07 Markup 1.6 1.5 1.4 1.3 1.2 1500000 1000000 Share weighted Dividend Markup 1.6 1.5 1.4 1.3 1.2 6.00e+07 4.00e+07 2.00e+07 Share weighted Market Value 1.1 500000 1.1 0 1960 1970 1980 1990 2000 2010 year 1960 1970 1980 1990 2000 2010 year Markup Share weighted Dividend Markup Share weighted Market Value

Markup = Market Power? Aggregate Profit Rate External validation: compare increase in aggregate profit rate with increase in profit rate from micro data Incomplete comparison: 1. aggregate has profits of all firms vs. micro data only publicly traded firms (large; 40% of GDP) 2. total profits Π versus variable profits Π V (levels are different!) Microdata Π V ( GDP = ΠV GNI PQ GDP = 1 1 ) GNI µ GDP Π V 2014 GDP 2014 = 2.34 Π V 1980 GDP 1980 Aggregate Data: Total Profits Π increase by factor 4 Even bigger increase in aggregate profits. Either: 1. Smaller firms have even bigger markups 2. or, capital expenditure has increased less than variable costs

Markup = Market Power? Aggregate Profit Rate.02.04.06.08.1.12 1950q1 1960q1 1970q1 1980q1 1990q1 2000q1 2010q1 2020q1 time Figure: Profit rates. Data FRED, from national accounts. Quarterly.

Summary of Facts 1. Sharp increase in Markup since 1980: 42% 2. High markup firms tend to be smaller 3. Only in the upper half of Markup distribution (espec. at top) 4. Mostly within industry (in all; no particular industries) 5. Markup = Market Power: total profits 4

Macroeconomic Implications

A Simple Model Unskilled Labor is Variable Input Let V = L, demand P(Q) and quantity Q i = Ω i L θ i K 1 θ i where: Ω i : firm productivity L i : quantity of efficiency units of labor S i = P i Q i FOC wl i S i = θ µ i Inverse relation labor share and markup µ i

1. Decline in Labor Share 110 100 90 80 70 1940 1960 1980 2000 2020 Data Year - Fiscal Share weighted (Inverse) Markup (1950==100) Share weighted (Inverse) Markup (labor) (1950==100) Labor Share (Fred) (1950==100) Figure: Labor share of GDP (BLS), and inverse of the markup (Two measures: weighted by Sales, and weighted by Employment)

2. Decline in Capital Share K is capital investment and r is the user cost. FOC (long run): Accounting identity: rk i = θ k S i µ i wl + rk + Π i = S i wl i S i + rk i S i + Π i S i = 1 (θ l + θ k ) 1 µ i = 1 π i If profit share labor and capital share (provided Q i is complementary in L i, K i )

2. Decline in Capital Share.16.85.14.8 k_share_agg.12.1.75.7 MARKUP_INV.08.65.06 1960 1980 2000 2020 year.6 k_share_agg MARKUP_INV Figure: The Evolution of the capital share (own computations), and inverse of the markup (weighted by Sales Share) (1980-2014). Gross Capital adjusted by the input price deflator, the federal funds rate, and an exogenous depreciation rate of 12%.

Market Power Conduct Parameter λ Follow Bresnahan (1982) Pricing equation (given a marginal cost c): P(Q) = c + λh(q) where Q = i Q i and where h(q) = P(Q) Q Q; Linear demand P(Q) = a bq for example, h(q) = bq Our measure for markup µ = P c µ = 1 + λ h(q) c can then be written as:

Market Power Market power can change with technology, externalities, network goods, entry barriers, preferences (Dixit-Stiglitz), market structure, consumer behavior (Burdett-Judd),... Our model: change in λ = Cournot 1. Market power changes exogenously with λ 2. Number of firms per market N = 1 λ 3. In a given market, all N firms have same technology Ω i 4. Linear Demand, to capture incomplete pass-through ( CES)

Market Power P a Market Power λ [0, 1] P = a bq MR = a (1 + λ)bq a 2b a b Q

Market Power P a Monopoly Market Power λ = 1 P = a bq MR = a 2bQ a 2b a b Q

Market Power P a Perfect Competition Market Power λ = 0 P = a bq MR = a bq a b Q

Equilibrium Impact of Market Power Constant returns θ = 1, identical firms within market Ω Firm objective: First-order Condition: max L i P(Q)Q i wl i, where Q i = Ω i L i w P(Q) = c + λh(q) a (1 + λ)bq = Ωθ L = a w Ω (1 + λ)bω, L i = λ a w Ω 1 + λ bω L λ = a w Ω bω 1 (1 + λ) 2 < 0 and L i λ = a w Ω bω 1 (1 + λ) 2 > 0. Lemma For a given wage and a given market Ω, the market labor demand L is decreasing in market power λ; and an individual firm s labor demand L i is increasing in market power λ.

Compustat Data Firm Size and Number of Firms 20000 10000 8000 L_mean 15000 10000 6000 4000 nrfirms 2000 5000 1940 1960 1980 2000 2020 Data Year - Fiscal L_mean nrfirms 0

Equilibrium Impact of Market Power Labor Demand: aggregate over different markets Ω i F (Ω i ) L D = Ω Ω L(Ω i ; w; λ)df (Ω i ) = 1 (1 + λ)b Ω Ω ( a Ω i w Ω 2 i Labor Supply: heterogeneous workers z G(z) supply efficiency units of labor; outside option is U L S = 1 Equilibrium: L D = L S w, L U w zdg(z) ) df (Ω i )

Equilibrium Impact of Market Power Labor Force: L λ < 0; and Nominal Wages: w λ < 0; Ω = 1

Equilibrium Impact of Market Power Labor Force: L λ P, w w < 0; and Nominal Wages: λ < 0; Ω = 1 P λ=0 = w λ=0 L S L D λ=0 L λ=0 L

Equilibrium Impact of Market Power Labor Force: L λ P, w w < 0; and Nominal Wages: λ < 0; Ω = 1 P λ=1 P λ=0 = w λ=0 L S w λ=1 L D λ=1 L D λ=0 L λ=1 L λ=0 L

3. Decline in (Low Skill) Wages Evidence Without Growth: nominal wages decline With Growth: nominal wages relative to GDP decline Double Impact since 1980: 1. Nominal wages 2. P 42%: real wages w P further relative to perf. comp.

3. Decline in (Low Skill) Wages Real Median Wages and Relative to GDP 310 320 330 340 350 1980q1 1990q1 2000q1 2010q1 2020q1.6.8 1 1.2 1.4 1980q1 1990q1 2000q1 2010q1 2020q1

4. Decline in Labor Force Participation Evidence.58.6.62.64.66.68 1950m1 1960m1 1970m1 1980m1 1990m1 2000m1 2010m1 2020m1 time Notes. From CPS.

Labor Reallocation and Pass-through Labor adjustments: in response to productivity shocks (Jovanovic, Hopenhayn-Rogerson) Evidence (Davis-Haltiwanger); Theory/Calibration (Schaal) Then if pass-through is incomplete, a shock leads to a less than proportional increase/decrease in Q and L Linear demand: pass-through decreasing in market power λ Market power labor market adjustment for same shock

Labor Reallocation and Pass-through P P w λ=0 (Ω) = Ω w P λ=0 (Ω) = Ω L D λ=0 L λ=0 ΩL

Labor Reallocation and Pass-through P P λ=0 (Ω) = w Ω w P λ=0 (Ω) = Ω L D λ=1 L D λ=0 L λ=1 L λ=0 ΩL

Labor Reallocation and Pass-through P P λ=1 (Ω) P λ=1 (Ω) P λ=0 (Ω) = w Ω w P λ=0 (Ω) = Ω L D λ=1 L D λ=0 L λ=1 L λ=0 ΩL

5. Decline in Labor Market Flows total.04.06.08 1975q1 1980q1 1985q1 1990q1 1995q1 2000q1 2005q1 2010q1 2015q1 dq.015.025.03 (UE+NE)/( U+N) (left axis) ee (right axis)

6. Decline in Migration Rates.01.015.02.025.03.035 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 Migration Rates Migration Rate, between MSA s (1962-2016). CPS Data.

7. Slowdown in aggregate output growth 7.5 8 8.5 9 9.5 10 1950q1 1960q1 1970q1 1980q1 1990q1 2000q1 2010q1 2020q1

7. Slowdown in aggregate output growth Productivity Growth Firm s FOC for Labor: PΩ i L θ 1 i K 1 θ i θµ 1 w 1 i = w Ω i = µ i P Constant Marginal Product of Labor: θ = 1 Hall (1988) Ω i = µ i w P θ L1 θ i K 1 θ i.

7. Slowdown in aggregate output growth Productivity Growth.1.06.05.04 0 wedge.02 0 -.05 -.02 -.1 1960 1980 2000 2020 omega_gamma_gr TFP_gamma_gr lpoly smooth: omega_gamma_gr lpoly smooth: TFP_gamma_gr -.04 1960 1980 2000 2020 Data Year - Fiscal 90% CI wedge lpoly smooth

Conclusions 1. Sharp rise in Market Power since 1980 2. Significant macroeconomic implications: 7 secular trends

Conclusions Open Questions 1. Causes? Technology, M&A,... 2. Other secular trends? 1. decrease in startup rate of new firms 2. decrease long term interest rate: capital D, S 3. increase in wage inequality: profit sharing managers 4. the great moderation... 3. Two (uncomfortable) Consequences Food for Thought 1. Inflation is too high: 42% increase price level; 1% per year Policy: anti-trust, not Federal Reserve! 2. Stock market over-valued (compared to Perfect Competition) Stock market increase economic growth

Conclusions M&A Share-weighted Markup (DLW) 1.7 1.6 1.5 1.4 1.3 1.2 1960 1980 2000 2020 Year 20000 15000 10000 5000 0 Share-weighted Markup (DLW) Nr M&A Average value M&A (10,000 USD)

Conclusions Great Moderation -5 0 5 10 15 1950q1 1960q1 1970q1 1980q1 1990q1 2000q1 2010q1 2020q1 time

Conclusions Stock Market Valuation 1.7 20000 Share weighted Markup 1.6 1.5 1.4 1.3 1.2 1960 1970 1980 1990 2000 2010 year 15000 10000 5000 0 DOW JONES (Deflated CPI) Share weighted Markup DOW JONES (Deflated CPI)

Conclusions Open Questions 1. Causes of Market Power? Technology, M&A,... 2. Other secular trends? 1. decrease in startup rate of new firms 2. decrease long term interest rate: capital D, S 3. increase in wage inequality: profit sharing managers 4. the great moderation... 3. Two (uncomfortable) Consequences Food for Thought 1. Inflation is too high: 42% increase price level; 1% per year Policy: anti-trust, not Federal Reserve! 2. Stock market over-valued (compared to Perfect Competition) Stock market increase economic growth

Whenever a theory appears to you as the only possible one, take this as a sign that you have neither understood the theory nor the problem which it was intended to solve Karl Popper

Causes? Production Technology: Big Retail vs. manufacturing and services Network goods Finance: Mergers and Acquisitions Cross-ownership of competitors Private Equity under the radar Vertical Integration Preferences: Price Differentiation improved through technology Brand dependence

Translog Production Technology Unusually long sample period: 1950-2014 Industry-specific, time-varying output elasticities Preserves identification results, De Loecker-Warzynski (2012) Moment conditions from static optimization of variable inputs: E ( ξ it (β) [ vit 1 vit 1 2 ]) = 0, With translog production function for each industry: q it = β v1 v it + β k1 k it + β v2 v 2 it + β k2 k 2 it + ω it + ɛ it Variation output elasticity over time (or firms), no longer attributed to markup variation Output elasticity of the composite variable input: θ v it = β v 1 + 2β v2 v it Markup defined as before; level difference, but normalization

Cross-Sectional Decomposition Across All Industries 1.7 1.6 1.5 1.4 1.3 1.2 1940 1960 1980 2000 2020 year Markup Mean 4digit Figure: Industry disaggregation

Markup = Market Power? Dividends Individual firm level: firm-level average markup (all years) and the share of total dividends in total sales 3.5 Local polynomial smooth markupmean_firm 3 2.5 2 1.5 0.2.4.6.8 1 divmargin_av kernel = epanechnikov, degree = 0, bandwidth =.05

Growth Accounting First order approach: no productivity slowdown Solow residual: shows a decrease in productivity since 2000 4.6 4.7 4.8 4.9 5 5.1 1980 1990 2000 2010 2020 0 1 2 3 4 1980 1990 2000 2010 2020

Growth Accounting Solow residual based on aggregate production technology Q = Ω S L θ K 1 θ Ω S = where L = i L i and K = i K i Micro data: Q L θ K 1 θ Ω = i Ω i = i Q i L θ i K 1 θ i Qi i Lθ i i K 1 θ i = Ω S Level difference, but does not affect growth rate of Ω as long as distribution of firm inputs and output remains unchanged

Growth Accounting Standard Deviation of Employment Size 50 40 emp_sd 30 20 1940 1960 1980 2000 2020 Data Year - Fiscal

Growth Accounting Distribution of Q i L i and Q i K i.00001.3 kdensity s_l 8.000e-06 6.000e-06 4.000e-06 2.000e-06 kdensity s_k.2.1 0 0 500000 1000000 1500000 x 1960 1970 1980 1990 2000 2014 0 0 5 10 x 1960 1970 1980 1990 2000 2014

Growth Accounting Distribution of Q i L θ K 1 θ i i.0004.0003 kdensity s_lk.0002.0001 0 0 5000 10000 15000 20000 25000 x 1960 1970 1980 1990 2000 2014

Growth Accounting Estimated Labor Productivity Growth.2.15.1.05 0 -.05 1940 1960 1980 2000 2020 lpoly smoothing grid lpoly smooth: JJ_gr lpoly smooth: SOLOW_gr