International Shocks and Domestic Prices: How Large Are Strategic Complementarities? Preliminary and Incomplete Mary Amiti Oleg Itskhoki Jozef Konings NY FRB Princeton Leuven and BNB ERWIT Sciences Po, Paris June 2015 1 / 31 Motivation • How strong are strategic complementarities across firms in price setting? — do firms mostly respond to own cost shocks or put a large weight on the prices of their competitors? — a fundamental mechanism in both micro- and macroeconomics 1 / 31 Motivation • How strong are strategic complementarities across firms in price setting? — do firms mostly respond to own cost shocks or put a large weight on the prices of their competitors? — a fundamental mechanism in both micro- and macroeconomics • A particular application: — How do international shocks affect domestic marginal costs, markups and prices? 1 / 31 Motivation • How strong are strategic complementarities across firms in price setting? — do firms mostly respond to own cost shocks or put a large weight on the prices of their competitors? — a fundamental mechanism in both micro- and macroeconomics • A particular application: — How do international shocks affect domestic marginal costs, markups and prices? • Other applications: slope of the Phillips Curve in Monetary Economics 1 / 31 Our Approach 1 Directly estimate extent of strategic complementarities across a broad set of manufacturing industries (a) Develop a general accounting framework — decompose firm price changes into response to own cost shocks and to changes in competitor prices (b) Develop an identification strategy — two major challenges: simultaneity of price setting and measurement error in firm marginal costs (c) Construct a new micro-level dataset for Belgium manufacturing to carry out this estimation — intensive data requirements on firm prices, marginal costs, and firm’s competitor prices within sectors 2 Develop a quantitative framework to evaluate counterfactual scenarios — a flexible model tightly calibrated to the Belgian microdata — explore response of firm markups and prices to various shocks 2 / 31 Main Findings 1 Strong evidence of strategic complementarities: (i) response to own marginal cost = 0.65–0.70 (or idiosyncratic cost pass-through elasticity) (ii) response to competitor prices = 0.35–0.40 (or strategic complementarities elasticity) 2 These are average (sales-weighted) responses. Yet, a lot of heterogeneity in responses across firms: — small firms: no strategic complementarities (as in standard models) — large firms: strong strategic complementarities and variable markups 3 Implications for aggregate pass-through — decomposition across firms — variation across sectors 3 / 31 Related Literature 1 IO-style studies: — Industry studies: Feenstra et al. (1996, cars), Nakamura and Zerom (2010, coffee), Goldberg and Hellerstein (2013, beer) — Alternative: De Loecker and Warzynski (2012), De Loecker and Goldberg (2014) 2 Domestic prices and exchange rates: — industry-level: Goldberg and Campa (2010) — product level: Auer and Schoenle (2013), Cao, Dong and Tomlin (2012), Pennings (2012) 3 International prices and exchange rates: — Gopinath and Itskhoki (2011) — Berman, Martin and Mayer (2012), Amiti, Itskhoki and Konings (2014) 4 Domestic prices and trade shocks: — De Loecker, Goldberg, Khandelwal and Pavcnik (2012) — Edmond, Midrigan and Xu (2012) 4 / 31 THEORETICAL FRAMEWORK 5 / 31 Accounting Framework • Log markup: µit = pit − mcit • Markup elasticities: Γit = −∂µit /∂pit and Γ−i,t = ∂µit /∂p−i,t • Change in markup: ∆µit = −Γit ∆pit + Γ−i,t ∆p−i,t + ε̃it • Decomposition of price change: ∆pit = Non-CRS case PT and Heterogeneity Γ−i,t 1 ∆mcit + ∆p−i,t + εit 1 + Γit 1 + Γit 5 / 31 A Model of Strategic Complementarities • Modeling variable markups requires relaxing either CES or monopolistic competition assumptions • We follow Atkeson and Burstein (2008) by adopting a CES-model with oligopolistic competition — nested-CES demand — finite number of firms within sector and Cournot competition — flexible price setting — focus on industry equilibrium 6 / 31 Demand • Nested-CES demand: Qit = ξit Pit−ρ Ptρ−η Dt , ρ > η ≥ 1, where s industry (omitted) and i firm-product • Sectoral price index: Pt ≡ hP 1−ρ N i=1 ξit Pit i 1 1−ρ h i 1 1−ρ 1−ρ = ξit Pit1−ρ + (1 − ξit )P−i,t where N is a finite number of firms in the industry • Market share: Pit Qit Sit ≡ PN j=1 Pjt Qjt = ξit Pit Pt 1−ρ ∈ [0, 1] 7 / 31 Market Structure • Cournot (quantity) competition across firms results in optimal markup pricing rule: Pit = Mit · MCit , Mit = σit σit − 1 Fixed Point Why Cournot? where and σit ≡ −1 1 1 Sit + (1 − Sit ) η ρ • Markup elasticity: Γit = Γ−i,t = (ρ − η)(ρ − 1)σit Sit (1 − Sit ) ρη(σit − 1) increases in market share (over relevant range) • Decomposition of price change in the model: ∆pit = 1 Γit Γit ∆ξit ∆mcit + ∆p−i,t + 1 + Γit 1 + Γit 1 + Γit (ρ − 1)(1 − Sit ) 8 / 31 Cost Structure • Total Cost: TCit = AVCit · Yit + Fit • Average Total Cost: MCit = AVCit = TVCit Yit • Marginal cost: MCit = Wit1−φit (Vit∗ Et )φit e it Ω • Marginal cost in log changes: ∆mcit = φit ∆vit + (1 − φit )∆wit + (vi,t−1 − wi,t−1 )∆φit − ∆e ωit , ∆mcit∗ = φit ∆vit 9 / 31 DATA 10 / 31 Dataset 1 Belgium firm-product level data 1995-2008 on values and quantities — PC 8-digit (1,700 products) — All manufacturing firms with minimum of 10 employees 2 Import and export data on values and quantities at HS 8-digit (over 10,000 products) 3 Firm-level data on firm characteristics — includes material costs, wagebill and employment 10 / 31 Variables • Domestic Prices: ∆pijt = ∆ log Domestic Valueijt Domestic Quantityijt • Marginal Cost Variable: ∆mcit = ∆ log TVCit Yit • Instrument: ∆mcit∗ = φit P m ωimt ∆vimt where ∆vimt is the log change in firm input prices (m is at source-country–HS-8-digit level) 11 / 31 Variables • Competition Variables: D F ∆p−i,t = (1 − θst )∆p−i,t + θst ∆pst X X D F ∆p−i,t = ωjt−i ∆pjt , ∆pst = ωjtF ∆pjt j∈Ds ,j6=i j∈Fs • Instruments: ωjt−i ∆mcjt∗ P 2 Trade Weighted Industry Exchange Rate: ∆est = ωmt ∆emt 1 ∗ Competitor marginal cost: ∆mc−i,t = 3 eu Change in export prices to other EU countries: ∆pst P j6=i 12 / 31 EMPIRICAL FINDINGS 13 / 31 Strategic Comlementarities • Our main empirical specification: ∆pijt = Γ−i,t 1 ∆mcit + ∆p−i,t + εit 1 + Γit 1 + Γit OLS Dep. var: ∆pit (1) IV (2) (3) (4) ∆mcit 0.348∗∗∗ (0.040) 0.348∗∗∗ (0.041) 0.667∗∗∗ (0.117) 0.757∗∗∗ (0.150) ∆p−i,t 0.400∗∗∗ (0.079) no 0.321∗∗∗ (0.095) yes 0.467∗∗∗ (0.143) no 0.315∗∗ (0.151) yes Industry FE Warmup ERPT regressions • IV regressions: First Stage 1 pass weak instrument test 2 pass over-id test 3 cannot reject the equality to 1 of the sum of the coef. 13 / 31 Strategic Complementarities Robustness Two-period Dep. var.: differences ∆pit (1) Alternative input definition (2) Major product (3) 5-digit industry (4) ∆mcit 0.642∗∗∗ (0.245) 0.684∗∗∗ (0.148) 0.658∗∗∗ (0.173) 0.750∗∗∗ (0.167) ∆p−i,t 0.434∗ (0.234) 0.388∗ (0.200) 0.374∗ (0.200) 0.410∗∗∗ (0.147) # obs. 50,600 64,320 27,027 63,511 14 / 31 Strategic Complementarities Firm Heterogeneity Dep. var: ∆pijt ∆mcit Employment ≥ 100 (1) (2) (3) 0.929∗∗∗ (0.152) ∆mcit × Largeit ∆P−i,t 49,462 0.947∗∗∗ (0.226) 0.599∗∗ −0.315 (0.237) (0.351) −0.270 (0.356) 0.142 (0.225) 0.063 (0.180) 0.469∗∗ (0.202) 0.279 (0.319) 0.355 (0.325) 15,353 64,815 64,815 0.078 (0.189) ∆P−i,t ×Largeit # observations 0.949∗∗∗ (0.201) Top 20% (4) 15 / 31 QUANTITATIVE MODEL 16 / 31 Quantitative Model • Atkeson-Burstein demand and market structure: — Nested CES demand with elasticities ρ > η ≥ 1 — Oligopolistic quantity (Cournot) competition • Firms differ in size (productivity) and import intensity, with the joint distribution approximating the data • Industry equilibrium model: firm costs follow exogenous processes. We use the model to calculate the evolution of Price setting markups and prices • We simulate multiple industries calibrated to approximate typical Belgian 4-digit sectors 16 / 31 Cost Structure • Marginal cost: Wt1−ϕi Vt Et MCit = Ωit ϕi • Trade-weighted exchange rate (in logs: et ≡ log Et ): et = et−1 + σe ut , ut ∼ iidN (0, 1) • Idiosyncratic productivity (in logs: ωit ≡ log Ωit ) ωit = µ + ωi,t−1 + σω vit , vit ∼ iidN (0, 1), with a reflecting barrier at ω to ensure an ergodic Pareto distribution of productivities in the cross-section 17 / 31 Three Types of Firms 1 2 3 Belgian (domestic) firms European (non-Belgian) firms Non-European firms • The groups of firms differ in two respects: (i) Exchange rate exposure, ϕi (ii) Mass of entrants • Conditional on entry, the firms in each group have the same market share distribution (Eaton, Kortum and Sotelo 2013) • Exchange rate exposure of domestic firms: X E X B ϕi = φEi ψ E + φX i ψ + (1 − φi − φi )ψ 18 / 31 Calibration Parameter Expected number firms: — Belgian — European union — Non-EU Elasticity of substitution: — across sectors — within sectors Productivity distribution: — Pareto shape parameter — St.dev. of innovation — Drift — Reflecting barrier St.dev. of ∆et Exchange rate exposure: — European firms — Non-EU firms — Belgian firms — Pass-through — Import intensity Value Moment in the data N̄B = 48 N̄E = 21 N̄X = 9 Number of Belgian firms Sales share Sales share η=1 ρ=8 Pass-through heterogeneity k = 6.6 σω = 0.034 2 /2 µ = −kσω ω=0 σe = 0.06 Size distribution of firms std(∆sit ) = 0.0042 Distribution stationarity Normalization Trade-weighted ER ϕE = 0.8 ϕX = 1 φB ψB + φE ψE + φX ψX ψB = 0.15, ψE = 0.6, ψX = 1 φE , φX ∼ Beta Aggregate pass-through Aggregate pass-through Pass-through into input prices Import intensity 19 / 31 Moments Moment Data Number of firms: — Belgian 41 (48) [22,87] — EU — Model std(∆sit ) 0.0042 48 [40,57] 21 [16,27] — Non-EU — 9 [5,13] Top Belgian 10% (12%) 11% market share [5%,21%] [6%,23%] 0.0042 corr(sit , si,t+12 ) 0.90 (0.85) [0.69, 0.98] 0.88 Moment Sales share: — Belgian Data corr(sit , φB i ) 0.26 (0.24) [0,0.44] 0.05 (0.08) [-0.03, 0.37] Model 0.64 (0.62) 0.62 [0.39,0.86] [0.46,0.77] — EU 0.26 (0.27) 0.26 [0.12,0.42] [0.14,0.41] — Non-EU 0.08 (0.11) 0.09 [0.01,0.25] [0.04, 0.22] Inverse HHI 16.4 (20.8) 13.7 for Belgian firms [7.1,38.4] [6.5,24.3] B corr sit , φX i /φi 0.28 0.16 20 / 31 Market Share Distribution Log r ank of fir m, logR a n k ( k ) 3 Model Median Data Median Data 10% Data 90% 2.5 2 1.5 1 0.5 0 −3 −2.5 −2 −1.5 −1 −0.5 0 Log r e lat ive mar ke t s har e , log[s ( k ) /s ( 1) ] 21 / 31 Import Intensity Belgian Firms (b) Kernel Density (model) (a) Averages by firm rank 15 3 Outside Euro Zone Outside Belgium Model Data Log rank of firm, log Rank(k) 2.5 10 2 Outside Belgium 1.5 1 5 Outside Euro Zone 0.5 0 0 0.05 0.1 0.15 0.2 0.25 Average import intensity 0.3 0.35 0.4 0 0 0.1 0.2 0.3 0.4 0.5 0.6 Import intensity 22 / 31 SIMULATION RESULTS 23 / 31 Aggregate ERPT • Sector-level and pooled firm-level regressions reproduce the aggregate pass-through patterns in the data See data Sector-level All firms MC Price 0.494 0.488 Domestic firms MC Price 0.286 0.321 Foreign firms MC Price 0.866 0.792 Firm-level pooled: — unweighted — sales-weighted 0.460 0.473 0.233 0.268 0.828 0.836 0.460 0.464 0.247 0.308 0.804 0.751 23 / 31 Strategic complementarities Dep. var.: ∆ log Pit ∆ log MCit ∆ log MCit ×Largeit ∆ log P−i,t ∆ log P−i,t ×Largeit OLS W/out size W/size interaction interaction 0.778 0.952 — −0.258 0.183 0.041 — 0.223 IV W/out size interaction 0.779 — 0.221 — W/size interaction 0.952 −0.258 0.047 0.262 • The model slightly underpredicts the extent of strategic complementarities among the large firms • IV for competitor prices is important for large firms 24 / 31 Strategic Complementarities Across firm-size deciles 1.2 Strategic Complementarities Marginal Cost Pass-through elasticity 1 0.8 0.6 0.4 0.2 0 0 0.5 1 2 3 5 7.5 10 15 25 40 Market share bins (%) Γ 1 −i • Estimates of 1+Γ and 1+Γ by firm size deciles i i Without IV 25 / 31 ERPT heterogeneity Across firm-size deciles 0.45 Markup Marginal Cost Exchange rate pass-through 0.4 0.35 0.3 0.25 0.2 0.15 0 0.5 1 2 3 5 7.5 10 15 25 40 Market share bins (%) • ERPT into MC (red) and prices (red+blue) by firm-size deciles 26 / 31 COUNTERFACTUALS 27 / 31 ERPT Decomposition • Aggregate ER pass-through of 0.35 is split as follows: Marginal Cost Markup Small Firms 39.3% 2.2% 41.5% Large Firms 51.1% 7.4% 58.5% 90.4% 9.6% • 10% of the largest firms account for 50% of sales and almost 60% of pass-through • Small firms’ markups are stable, while for large firms they account for about 15% of price adjustment 27 / 31 Response to a 10% Depreciation Prices 0.5 Price ERPT 0.4 0.3 0.2 0.1 0 0.1 0.08 0.5 0.45 0.06 0.4 0.35 0.04 0.3 0.25 0.02 Market share 0 0.2 ER exposure 28 / 31 Response to a 10% Depreciation Marginal costs 0.5 Marginal Cost ERPT 0.4 0.3 0.2 0.1 0 0.1 0.08 0.5 0.45 0.06 0.4 0.35 0.04 0.3 0.25 0.02 Market share 0 0.2 ER exposure 28 / 31 Response to a 10% Depreciation Markups Markup ERPT 0.15 0.1 0.05 0 0.1 0.08 0.5 0.45 0.06 0.4 0.35 0.04 0.3 0.25 0.02 Market share 0 0.2 ER exposure 28 / 31 Model distributions Firm percentiles 25 50 75 90 95 97.5 99 99.5 99.75 Sales percentile 5.3 14.2 29.6 49.3 62.7 74.0 85.1 90.7 94.4 Market share (%) 0.36 0.57 1.13 2.62 4.60 7.51 12.45 16.62 21.67 Exchange rate exposure 0.199 0.244 0.289 0.333 0.363 0.392 0.425 0.450 0.472 Markup 1.147 1.149 1.156 1.174 1.198 1.236 1.305 1.371 1.460 29 / 31 EXPLORING INDUSTRY HETEROGENEITY 30 / 31 Heterogeneity across sectors I. By domestic sales share 0.4 Markup Marginal cost 0.35 0.3 0.25 39.1 54.4 56.9 58.7 60.2 61.6 63 64.5 66.2 68.6 82.5 Domestic share bins (by number of firms, %) • ERPT into MC (red) and prices (red+blue) across sectors 30 / 31 Heterogeneity across sectors II. By sector import intensity 0.4 Markup Marginal cost 0.35 0.3 0.25 21.9 26.1 26.8 27.4 27.9 28.4 28.9 29.5 30.2 31.4 44.6 Sector import intensity • ERPT into MC (red) and prices (red+blue) across sectors 30 / 31 Heterogeneity across sectors III. By market share of the largest domestic firm 0.4 Markup Marginal cost 0.35 0.3 0.25 0.8 5.6 7 8.3 9.6 11 12.7 14.9 17.8 22.7 80 Market share bins (%) • ERPT into MC (red) and prices (red+blue) across sectors 30 / 31 Heterogeneity across sectors IV. By correlation b/w size and import intensity 0.4 Markup Marginal cost 0.35 0.3 0.25 -24.1 18.9 25.1 29.5 33.4 36.9 40.5 44.3 48.6 54.2 82.1 Correlation (Sit , ϕi ) bins • ERPT into MC (red) and prices (red+blue) across sectors 30 / 31 Conclusions • We provide direct evidence on the strength of strategic complementarities in firm price setting — on average, responsiveness to own cost is 65–70% and to competitor prices is 35–40% • Uncover substantial heterogeneity in the extent of strategic complementarities across firms: — small firms do not respond to competitor prices and have complete pass-through of own cost shocks — large firms exhibit substantial strategic complementarities and variable markups • The interplay of these forces with heterogeneous exposure to foreign inputs is crucial for understanding aggregate ERPT 31 / 31 APPENDIX 32 / 31 Generalization to non-CRS case • Assume marginal cost equals: MCit = Cit Yitα where α is the DRS parameter (α = 0 under CRS) • The decomposition in this case is: ∆pit = Γ−i,t + ασ̃−i,t 1 ∆cit + ∆p−i,t + εit 1 + Γit + ασ̃it 1 + Γit + ασ̃it • Under Atkeson-Burstein demand, Γ−i,t = Γit and σ̃it = σ̃−i,t + η, so that the sum of the coefficient equals: ψit + γit = 1 − αηψit < 1 Back to slides 33 / 31 Heterogeneity Matters • ERPT of firm i: Back to slides Ψit = Γ−i,t ϕit + Ψ−i,t 1 + Γit 1 + Γit 34 / 31 Heterogeneity Matters • ERPT of firm i: Back to slides Ψit = Γ−i,t ϕit + Ψ−i,t 1 + Γit 1 + Γit • Aggregate PT into marginal costs, prices and markups: X ΨMC =Φ≡ Sit ϕit , t X 1 Sit ϕit ΨPt = P Sit Γ−i,t 1+Γit , 1− 1+Γit XS Γ i Xh 1 Γit jt −j,t ΨM Sit ϕit P Sit Γ−i,t t =− 1+Γit − 1+Γjt 1− 1+Γit • Cross-sectional heterogeneity: ϕit , Sit , Γit , Γ−i,t • Aggregate pass-through into domestic prices: ΨD t = 1 1− P D SitD (1−StF )Γ−i,t 1+Γit X SitD h ϕit 1+Γit + StF Γ−i,t F 1+Γit Ψt i D 34 / 31 Price Setting Fixed Point • In each industry, given a vector of firm marginal costs {MCit }, we find the equilibrium vector of prices {Pit } • This is a fixed point problem: Pit = Mit · MCit , Mit = σit /(σit − 1), −1 σit = η −1 Sit + ρ−1 (1 − Sit ) , 1−ρ Sit = ξit Pit /Pt , PN 1−ρ 1/(1−ρ) Pt = i=1 ξit Pit All prices are determined simultaneously • The solution to this problem can be found numerically by iteration Market Structure QModel 35 / 31 Sensitivity Demand and Market Structure (a) Markup, Mit = σit σit −1 (b) Pass-through, Ψit = 1 1+Γit 1 1.6 0.9 ρ =5 Be r t r and Pas s -t hr ough, Ψ i t Mar k up, M i t 1.5 1.4 1.3 0.8 η=2 0.7 0.6 ρ =5 0.5 η=2 1.2 0.4 Be r t r and 1.1 0 0.05 0.1 0.15 0.2 0.25 0.3 0 0.05 Mar ke t s har e , s i t 0.1 0.15 0.2 0.25 Mar ke t s har e , s i t Figure : Markups and pass-through in a calibrated model Note: σit = h 1 S η it + ρ1 (1 − Sit ) i−1 under Cournot and σit = [ηSit + ρ(1 − Sit )] under Bertrand. Market Structure Calibration 36 / 31 Exchange Rate Pass-Through Firm-level regressions Table : Aggregate ERPT Dep. var.: ∆pst 0.489∗∗∗ (0.061) ∆est D ∆pst F ∆pst 0.311∗∗∗ (0.085) 0.642∗∗∗ (0.059) Table : Firm-level ERPT Dep. var.: ∆est ∆pijt 0.279 (0.187) ∆mcit ∆mcit∗ 0.395∗∗ 0.246∗∗∗ (0.187) (0.040) ∆vit 0.651∗∗∗ (0.122) Recall that ∆mcit = ∆mcit∗ + o.t. and ∆mcit∗ = φit ∆vit Back to slides Back to slides 37 / 31 Warmup II Non-importers • Preliminary evidence on strategic complementarities: Dep. var.: ∆est ∆est × Largeis ∆mcit ∆pijt 0.056 (0.129) 0.350∗∗ (0.153) ∆pijt 0.029 (0.138) 0.643∗∗ (0.325) Back to slides 38 / 31 Strategic Complementarities First Stage Regressions Column (3) ∆mcit ∆p−i,t Column (4) ∆mcit ∆p−i,t ∆mcit∗ 0.614∗∗∗ (0.111) 0.173∗∗∗ (0.039) 0.597∗∗∗ (0.014) 0.174∗∗∗ (0.033) ∆est -0.222 (0.230) 0.270∗∗ (0.120) -0.169 (0.258) 0.343∗∗ (0.144) ∗ ∆mc−i,t 0.392∗∗∗ (0.098) 0.468∗∗ (0.148) 0.379∗∗∗ (0.124) 0.580∗∗∗ (0.106) EU ∆pst 0.194∗∗∗ (0.054) 0.304∗∗∗ (0.053) 0.215∗∗∗ (0.049) 0.274∗∗∗ (0.053) Dep. var.: Ind. FE no no yes yes Back to slides 39 / 31 Strategic Complementarities Across firm-size deciles—Without IV 1.2 Strategic Complementarities Marginal Cost Pass-through elasticity 1 0.8 0.6 0.4 0.2 0 0 0.5 1 2 3 5 7.5 10 15 25 40 Market share bins (%) Γ 1 −i • Estimates of 1+Γ and 1+Γ by firm size deciles i i back to slies 40 / 31