Micro Data for Macro Models Topic 5: Trends in Concentration, Competition, and Markups

Micro Data for Macro Models Topic 5: Trends in Concentration, Competition, and Markups Thomas Winberry November 27th, 2017 1

Overview of Topic 5 1. Potentially related trends since 1980 Aggregate factor shares Composition of firms Market power? 2. Possible explanation of those trends Market power and markups (De Locker and Eeckhout 2017) Market power and investment (Gutierrez and Phillipon 2017) Superstar firms (Autor et al. 2017) 2

Global Decline in the Labor Share Source: Karabarbounis and Neiman (2014) 3

Capital Share vs. Profit Share? 1 = w tl t Y t + r tk t Y t + Π t Y t Split between capital and profit share depends on user cost of capital Barakai (2017) argues that aggregate user cost has been falling 4

Capital Share Falling 1 = w tl t Y t + r tk t Y t + Π t Y t 5

Profit Share Rising 1 = w tl t Y t + r tk t Y t + Π t Y t 6

Entry Rate Falling Source: Pugsley and Sahin (2015) 7

Firms are Aging Firm dynamics constant = average age increasing 8

Dynamism Falling Source: Decker, Haltiwanger, Jarmin, and Miranda (2014) 9

Concentration Rising Source: Autor et al. (2017) 10

Profit Share Rising 1 = w tl t Y t + r tk t Y t + Π t Y t 11

Markups Rising? Source: De Loecker and Eeckhout (2017) 12

Summary of Aggregate Trends 1. Aggregate factor shares Labor share falling Capital share potentially falling Profit share potentially rising 2. Composition of firms Entry rates falling Average age increasing Heterogeneity in growth rates falling Concentration rising 3. Market power? Some evidence that markups rising 13

Definitions of Costs Consider the production function y it = z it k θ it vν it, where θ + ν 1 v it = variable input (like labor), price p v it k it = fixed input in period t Variable cost of producing y it units is ( VC it = p v it Marginal cost of producing y it is MC it = VC it y it y it z it k θ it ) 1 ν = λ it 1 ν pv it ( ) 1 y 1 ν ν it z it kit θ 15

Markups The markup is the ratio of price of marginal cost µ it p it λ it, where p it = price of firm i s output Perfect competition = µ it = 1 A measure of market power NB: markup is defined relative to variable costs Two approaches to estimating markups 1. Demand side: specify demand system; estimate elasticity of price w.r.t. total demand = infer markups 2. Supply side: specify production function; estimate marginal cost = infer markups 16

Supply-Side Approach To Estimating Markups De Locker and Warzynski (2012) Consider the value-added production function of firm i y it = z it f(k it, v it ) Lagrangian for cost minimization w.r.t. variable input L = r it k it + p v it v it λ it (z it f(k it, v it ) y it ) First order condition w.r.t. v it p v it = λ itz it f(k it, v it ) v it 17

Supply-Side Approach To Estimating Markups De Locker and Warzynski (2012) Consider the value-added production function of firm i y it = z it f(k it, v it ) Lagrangian for cost minimization w.r.t. variable input L = r it k it + p v it v it λ it (z it f(k it, v it ) y it ) Plug in markup µ it = p it /λ it and rearrange µ it = θ v it p ity it p v it v it 17

Estimating Output Elasticity: Some IO Approximate production function with translog form ỹ it = β v ṽ it + β k kit + z it + ε it ỹ it = log y it, etc. ε it = measurement error Identification problem: choice of v it correlated with z it Solution: two-step procedure 1. Proxy for z it with a control function 2. Use that to consistently estimate β v 18

Step 1: Control Function Assume that ṽ it = f t ( z it, k it ) Timing assumptions: capital predetermined + variable inputs chosen after observing productivity Dependence on aggregate factor prices implicit in t subscript Invert to get z it = f 1 t (ṽ it, k it ) Monotonicity assumption: function is invertible Plug into translog production function ỹ it = β v ṽ it + β k kit + ft 1 (ṽ it, k it ) }{{} ˆfit +ε it 19

Step 2: Production Function Assume that productivity is AR(1): z it = ρ z it 1 + ω it Impose constant returns = β k = 1 β v Estimate β v using GMM: 1. Construct z it (β v ) = ˆf it β v ṽ it (1 β v ) k it 2. Estimate ρ from regressing z it (β v ) on z it 1 (β v ) 3. Construct ω it (β v ) as innovation to productivity 4. Moment condition: E [ω it (β v )ṽ it 1 ] = 0 Timing assumption: v it 1 chosen before period t 20

Back to the Markup Data: annual Compustat, 1955-2014 Infer the markup µ it = θit v p ity it p v it v using: it θ v it = β v from production function estimation p it y it is measured as sales p v it v it is measured as cost of goods sold (COGS): the cost of producing or acquiring goods sold by firm i in period t Key limitations: 1. COGS potentially a poor measure of variable cost 2. Assumption that θit v constant over time is strong 3. Sensitivity of results to production function estimation? 21

Average Markup Over Time (Weighted by Sales) θ v it constant = just plotting trend in sales-to-cogs ratio 22

Average Markup Over Time (Unweighted) Larger firms charge lower markup 23

Decomposition of Changes Over Time Change in markup mainly occuring within sector Not driven by increasing output elasticity 24

Student Presentations 1. Potentially related trends since 1980 Aggregate factor shares Composition of firms Market power? 2. Potential explanation of those trends Market power and markups (De Locker and Eeckhout 2017) Market power and investment (Gutierrez and Phillipon 2017) Superstar firms (Autor et al. 2017) 26