Banque nationale de Belgique SA · EURONEXT:BNB
Banque nationale de Belgique SA — Technical Presentation (February 2013)
50 slides from Banque nationale de Belgique SA's technical presentation, Feb 26, 2013
Slide 1: Prices, Markups and Trade Reform
Prices, Markups and Trade Reform J. De Loecker, P. Goldberg, A. Khandelwal and N. Pavcnik Princeton, Yale, Columbia and Dartmouth NBER, CEPR and BREAD NBB February 26, 2013
Slide 2: The goal of this paper
The goal of this paper: 1. Estimate Effects of India’s Trade Liberalization on I Costs, I Markups, I Prices. 2. In order to answer this we develop a method to jointly estimate markups and marginal cost of production using production and price data.
Slide 3: Why do we care?
Why do we care? I Gains from trade depend on markups (currently few models allowing for this channel) I Help us identify how gains from trade are distributed across producers and consumers. Leads to political economy of trade and explain often cited resistance to trade liberalization. I Relationship
Slide 4: Motivation
Motivation I Conventional wisdom: 1. Trade liberalization increases productivity (reallocation effects and within-firm productivity improvements) 2. Trade liberalization decreases markups (because of the pro-competitive effects of the reforms) I Conventional wisdom subject to caveats related to the
Slide 5: Motivation ctd: why one more study?
Motivation ctd: why one more study? I Better Data: Firm/Plant Level Data but with information on physical output and price by product I Methodology delivers full distribution of markups and marginal costs, while addressing the caveats associated with earlier studies (multi-product, prices, etc.), al
Slide 6: General Approach
General Approach I Use Firm/Plant-level data to estimate production functions I Use optimal input choices and production to infer productivity and markups I Marginal Costs are backed out in standard IO way: use Price/Markup I Relate estimates to trade reform I Advantage of approaching the questions
Slide 7: Novelties on Methodological Side
Novelties on Methodological Side 1. Product-level output-based production functions on single product firms. 2. Explicit Treatment of multi-product firms. 3. Use price and input tariff data to control for unobserved variation in firm input prices. 4. Obtain estimates of productivity for multi-produc
Slide 8: Road map for this talk
Road map for this talk I Brief discussion of trade liberalization and data. For details, see GKPT (previous papers). I Methodology: 1. Deriving Markups and Costs from Production Functions: 2. Estimation and Identification of Single-Product Production Function with Output Data 3. Inference of Product
Slide 9: India's Trade Liberalization: Average Tariffs
Trade liberalization in India India’s Trade Liberalization 1.2 0.6 We exclude 1 0.5 data after 1997 Std. Dev Tariffs Average Tariffs 0.8 0.4 0.6 0.3 0.4 0.2 0.2 0.1 0 0 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 Average Tariffs Std Deviation Tariffs Source: Topalova (
Slide 10: Data on firms, products and tariffs
Data on firms, products and tariffs I Center for Monitoring of Indian Economy (CMIE) Covers 60-70% of Indias economic activity in organized industrial sector I Firm-level panel data 1989-2003 and spans the period of trade reforms I Medium and large Indian firms: Well suited to study changes among bi
Slide 11: Definition of a product
Definition of a product I Based on CMIE’s internal product classification and products reported in companies annual reports I 1,886 products: Comparable to BRS (2006a) product definition (ca 1500 5-digit SIC), Linked to 108 4-digit NIC industries I Coverage across firms: Product info available for v
Slide 12: Example of product space
Example of product space džĂŵƉůĞƐŽĨ/ŶĚƵƐƚƌŝĞƐ͕^ĞĐƚŽƌƐĂŶĚWƌŽĚƵĐƚƐ E/ŽĚĞ ĞƐĐƌŝƉƚŝŽŶ Ϯϳ ĂƐŝĐDĞƚĂů/ŶĚƵƐƚƌŝĞƐ;^ĞĐƚŽƌƐͿ ϮϳϭϬ DĂŶƵĨĂĐƚƵƌĞŽĨĂƐŝĐ/ƌŽŶΘ^ƚĞĞů;/ŶĚƵƐƚƌLJŝͿ ϭϯϬϭϬϭϬϭϬϬϬϬ WŝŐŝƌŽŶ ϭϯϬϭϬϭϬϮϬϬϬϬ ^ƉŽŶŐĞŝƌŽŶ ϭϯϬϭϬϭϬϯϬϬϬϬ Ϳ ;Ɛ &ĞƌƌŽĂůůŽLJƐ ũ ϭϯϬϭϬϲϬϰϬϴϬϬ ƚĐtĞůĚĞĚƐƚĞĞůƚƵďƵůĂƌƉŽůĞƐ Ƶ ϭϯϬϭϬϲϬϰϬϵϬϬ Ě^ƚĞĞůƚƵďƵůĂ
Slide 13: Empirical Framework
Empirical Framework I A firm faces a production function of the general form: Qft = Ft (Xft ) exp(ωft ) (1) I where X captures: 1. variables inputs (V 1 , ..., V V ): such as materials, electricity, etc. 2. dynamic inputs of production K, such as capital and potentially labor. I For now we supress t
Slide 14: Firm behavior: cost minimization
Firm behavior: cost minimization I We follow DLW and rely on cost minimization of variable input(s), conditioning on dynamic inputs: V v X L(Vft1 , ..., VftV , Kft , λft ) = Pft V vft +rft Kft +λft (Qft −Qft (.)) v =1 (2) ∂Lft v ∂Qft (.) v = Pft − λft =0 (3) ∂Vft ∂Vftv ∂Qft (.) Vftv 1 Pftv Vftv = (4
Slide 15: Cost minimization and markups
I Cost minimization implies that optimal input demand is realized when a firm equalizes the output elasticity of any Pftv Vft variable input Vftv to λ1ft Q ft . I Define markup µft as µft ≡ Pft . DLW show that the λft cost-minimization condition can be rearranged to write markup as: v −1 µft = θftv
Slide 16: Markups at the firm-product level
Markups at the firm-product level I In our data we have product-level information on output and prices. I Using the framework and moving to multi-product firms does not pose problems for theory, in fact we get: v v −1 µfjt = θfjt (αfjt ) (6) I The problem arises in getting an estimate of both RHS va
Slide 17: Multi-Production Functions: challenges
Multi-Production Functions: challenges I We consider the following production function: qfjt = f (xfjt ; β) + ωft + fjt (7) I However, it is hard (if not impossible) to measure xfjt across a wide set of products, producers, industries and time. I We do observe, for a given input xft and xfjt ≡ ρfjt
Slide 18: Identification I
Identification I I Identification strategy: use the (unbalanced) set of single product firms in a given industry to recover production function using extended proxy approach. I In addition, include selection correction for non-random event of becoming multi-product based on underlying productivity o
Slide 19: Implications multi-product production
Implications multi-product production I Economies of scope in production are allowed for, or in our case: X Cft (Q) ≤ Cfjt (Qfjt ) (10) j I We simply restrict product j’s production to follow Qfjt = Ft (Xfjt )Ωft , regardless of whether the firm produces any other product. Economies of scope can sti
Slide 20: Identification II
Identification II I In addition, we directly observe quantity produced and can therefore immediately eliminate demand and price effects that plague production function identification (De Loecker 2011). I Remember: (qfjt + pfjt ) = xfjt β + ωft + pfjt + fjt (11) I With revenue data unobserved demand
Slide 21: Identification III
Identification III I Key identification assumption in this class of models is productivity process. We move away from the standard assumption of exogenous Markov process, and consider: ωft = gt−1 (ωft−1 , τt−b ) + ξft (12) I Also see Doraszelski and Jaumandrei (2008) and De Loecker (Forthcoming) for
Slide 22: Identification IV: selection
Identification IV: selection I Using the set of single product producers potentially introduces a selection bias if adding one or more products depends on the firm’s productivity and input use. I If adding products only depends on the firm’s productivity, the use of an unbalanced panel of single pro
Slide 23: Selection control: more details
Selection control: more details I Define an indicator of remaining single product at time t given the firm’s information set at t − 1 as χft = 1; and the probability to remain single product Pr (χft = 1) is: Pr [ωft ≤ ω̄ft (lft , kft , zft )|ω̄ft (lft , kft , zft ), ωft−1 ] = κt−1 (ω̄ft (lft , kft ,
Slide 24: Estimation: single product firms
Estimation: single product firms I We consider a flexible production function, and note that although single product we need f , j and t: qft = f (xft ; β) + ωft + ft (14) I Standard simultaneity bias through unobserved productivity shocks ωfjt where we rely on extension of proxy estimator approach.
Slide 25: Input demand and productivity
Input demand and productivity I We rely on a static input demand equation to proxy for unobserved productivity, simply restates cost min. conditions: mft = mt (kft , zft , ωft ) (15) ωft = ht (mft , kft , zft ) (16) I where zft affecting choice of mft , including both demand and cost factors. zft =
Slide 26: Estimation algorithm
Estimation algorithm 1. Predicting output: Using input demand to control for ωft : qft = φt (xft , zft ) + ft (18) 2. GMM: We form moments constructing shock to productivity: ξft (β) = ωft (β) − E (ωft (β)|Ift−1 ) (19) E (ξft (β)Yft ) = 0 (20) with instruments Yft containing lfjt−1 (lfjt ), mfjt−1 a
Slide 27: Input price/quality bias
Input price/quality bias I Once we take the physical production function seriously we run into similar problem of omitted price bias (KG, DL). I In the data we only observe deflated (by sector and not firm) input expenditures, x̃ft . Therefore: qft = f (x̃ft ; β) + ωft + Bft (x̃ft , νft ) + ft (21)
Slide 28: Second stage: creating moments
Second stage: creating moments I No restriction on production function f (.) so far, only Hicks neutrality, and we now know productivity up to a vector of parameters and taking stand on f (.): ωft (β, δ) = φft − xft β − dt (τiti , τito , pft , msft ; δ) (22) I where inclusion of d(.) is important to
Slide 29: Full model
Full model: I We form moments: E (ξft (β)Yft ) = 0 (23) I with instruments Y ft containing lfjt−1 (lfjt ), mfjt−1 and kfjt , and their appropriate higher order terms, and lagged prices, market shares, and current tariffs. I We exploit fact that current shocks to productivity immediately impact a fir
Slide 30: Recovering productivity, markups and marginal costs: single-product firms
Recovering productivity, markups and marginal costs: single-product firms I We recover markups and marginal costs as follows: Rft µft = θftv (β) (24) Cftv Pft mcft = (25) µft I where C v denotes product j’s expenditure on input v . ft
Slide 31: Recovering markups and marginal costs: multi-product firms
Recovering markups and marginal costs: multi-product firms I Our identification strategy eliminates the problem of dealing with multi-product production, and we back out bfjt = qfjt − x̃ft β ω (26) = ωft + ρ(ρfjt , xft , β) (27) n o X̃fjt I where ρfjt = ln , are cost shares of an input j. X̃ft I The
Slide 32: Multi-product firms ctd.
Multi-product firms ctd. I Under translog we solve, by firm and time, for (ρj , ω) where j = {1, ..., J} using: bj = ω + aρj + bρ2j + cρ3j ω (28) X exp (ρj ) = 1 (29) j bj = E (qj |.) and a, b are data related to translog term I where ω I System of (J + 1) equations to solve for (J + 1) unknowns, wh
Slide 33: Markups and marginal costs for multi-product firms
Markups and marginal costs for multi-product firms I We compute the markup and marginal costs using price data according to v Rfjt µfjt = θfjt (β) (30) exp (ρfjt )Cftv Pfjt mcfjt = (31) µfjt
Slide 34: Results: Overview
Results: Overview 1. Production Function coefficients, 2. Firm performance correlation matrix, 3. Change in distributions, 4. Regression framework to identify trade reform’s effect.
Slide 35: Results: Output elasticities and RTS
Results: Output elasticities and RTS KďƐ͘ŝŶ^ŝŶŐůĞͲWƌŽĚƵĐƚ ZĞƚƵƌŶƐƚŽ &ŝƌŵƐƚŝŵĂƚŝŽŶ >ĂďŽƌ DĂƚĞƌŝĂůƐ ĂƉŝƚĂů ^ĐĂůĞ ^ĞĐƚŽƌ ;ϭͿ ;ϮͿ ;ϯͿ ;ϰͿ ;ϱͿ ϭϱ&ŽŽĚƉƌŽĚƵĐƚƐĂŶĚďĞǀĞƌĂŐĞƐ ϭ͕ϲϴϭ Ϭ͘Ϯϵ Ϭ͘ϲϭ Ϭ͘Ϭϴ Ϭ͘ϵϴ Ϭ͘ϬϮ Ϭ͘Ϭϯ Ϭ͘Ϭϯ Ϭ͘Ϭϰ ϭϳdĞdžƚŝůĞƐ Ϯ͕Ϭϱϰ Ϭ͘ϯϭ Ϭ͘ϱϳ Ϭ͘ϭϭ Ϭ͘ϵϴ Ϭ͘Ϭϭ Ϭ͘ϬϮ Ϭ͘ϬϮ Ϭ͘ϬϮ ϭϴtĞĂƌŝŶŐĂƉƉĂƌĞů
Slide 36: Results: The importance of corrections
Results: The importance of corrections ƐƚŝŵĂƚĞƐǁŝƚŚŽƵƚŽƌƌĞĐƚŝŶŐĨŽƌ/ŶƉƵƚWƌŝĐĞƐƚŝŵĂƚĞƐŽŶĂĂůĂŶĐĞĚWĂŶĞůŽĨ^ŝŶŐůĞͲ sĂƌŝĂƚŝŽŶ WƌŽĚƵĐƚ&ŝƌŵƐ ZĞƚƵƌŶƐƚŽ ZĞƚƵƌŶƐƚŽ >ĂďŽƌ DĂƚĞƌŝĂůƐ ĂƉŝƚĂů ^ĐĂůĞ >ĂďŽƌ DĂƚĞƌŝĂůƐ ĂƉŝƚĂů ^ĐĂůĞ ^ĞĐƚŽƌ ;ϭͿ ;ϮͿ ;ϯͿ ;ϰͿ ;ϱͿ ;ϲͿ ;ϳͿ ;ϴͿ ϭϱ&ŽŽĚƉƌŽĚƵĐƚƐĂŶĚďĞǀĞƌĂŐĞƐ ϭ͘ϮϮ Ϭ͘ϱϬ Ϭ͘ϯϲ Ϯ͘Ϭϵ Ϭ͘Ϯϱ
Slide 37: Results: Markups
Results: Markups DĂƌŬƵƉƐ ^ĞĐƚŽƌ DĞĂŶ DĞĚŝĂŶ ϭϱ&ŽŽĚƉƌŽĚƵĐƚƐĂŶĚďĞǀĞƌĂŐĞƐ Ϯ͘ϬϮ Ϭ͘ϵϳ ϭϳdĞdžƚŝůĞƐ ϭ͘ϰϱ ϭ͘ϭϭ ϭϴtĞĂƌŝŶŐĂƉƉĂƌĞů ϭ͘ϭϱ ϭ͘ϭϬ ϮϭWĂƉĞƌĂŶĚƉĂƉĞƌƉƌŽĚƵĐƚƐ ϭ͘Ϭϱ Ϭ͘ϵϴ ϮϯŽŬĞ͕ƌĞĨŝŶĞĚƉĞƚƌŽůĞƵŵƉƌŽĚƵĐƚƐ ϰ͘ϭϳ Ϭ͘ϵϯ ϮϰŚĞŵŝĐĂůƐ Ϯ͘ϭϭ ϭ͘ϬϬ ϮϱZƵďďĞƌĂŶĚWůĂƐƚŝĐ ϭ͘ϵϲ ϭ͘Ϯϴ ϮϲEŽŶͲŵĞƚĂůůŝĐŵŝŶĞƌĂůƉƌŽĚƵĐƚƐ ϰ͘ϰϬ Ϯ͘Ϭϱ ϮϳĂƐŝĐ
Slide 38: Firm Performance measures
Firm Performance measures WƌŝĐĞƐ DĂƌŬƵƉƐ DĂƌŐŝŶĂůŽƐƚƐ WƌŽĚƵĐƚŝǀŝƚLJ WƌŝĐĞƐ ϭ͘ϬϬ DĂƌŬƵƉƐ Ϭ͘ϭϲ ϭ͘ϬϬ DĂƌŐŝŶĂůŽƐƚƐ Ϭ͘ϵϰ ͲϬ͘ϭϵ ϭ͘ϬϬ WƌŽĚƵĐƚŝǀŝƚLJ ͲϬ͘ϯϵ Ϭ͘Ϭϰ ͲϬ͘ϰϬ ϭ͘ϬϬ EŽƚĞƐ͗dĂďůĞƌĞƉŽƌƚƐƚŚĞƉĂŝƌǁŝƐĞ ĐŽƌƌĞůĂƚŝŽŶŵĂƚƌŝdžĂĐƌŽƐƐĨŽƵƌƉĞƌĨŽƌŵĂŶĐĞ ǀĂƌŝĂďůĞƐ͗ƉƌŝĐĞƐ͕ŵĂƌŬƵƉƐ͕ŵĂƌŐŝŶĂůĐŽƐƚƐĂŶĚƉƌŽĚƵĐƚŝǀŝƚLJ͘ůůǀĂƌŝĂďůĞƐĂƌĞ Ğ
Slide 39: Back to our product listing: Ferro Alloys
Back to our product listing: Ferro Alloys (130101030000) Year Price Markup Mc Output elasticity 1992 2.1 1.3446 1.5 0.72 1993 2.0 1.3336 1.4 0.70 1994 1.6 1.5390 1.0 0.73 1996 1.8 1.2304 1.4 0.72 1997 1.7 1.3994 1.2 0.61 1998 1.7 1.5062 1.1 0.62 1999 1.5 1.5439 1.0 0.60 2000 1.7 1.5578 1.1 0.58 2001
Slide 40: Pass-through
Pass-through I We consider standard pass-through regression exploiting time series variation in prices and marginal costs. I We perform this at the product-firm level, while controling for output tariffs’ effect on residual demand. I We find incomplete pass-through of around 0.6 indicating strong de
Slide 41: A look at pass-through
A look at pass-through >ŽŐWƌŝĐĞĨũƚ ;ϭͿ ;ϮͿ >ŽŐDĂƌŐŝŶĂůŽƐƚĨũƚ Ϭ͘ϮϵϯΎΎΎ Ϭ͘ϲϬϯΎΎΎ Ϭ͘Ϭϯϳ Ϭ͘ϬϱϮ KďƐĞƌǀĂƚŝŽŶƐ ϮϬ͕ϲϯϭ ϮϬ͕ϲϯϭ &ŝƌŵͲWƌŽĚƵĐƚ&Ɛ LJĞƐ LJĞƐ EŽƚĞƐ͗dŚĞĚĞƉĞŶĚĞŶƚǀĂƌŝĂďůĞŝƐ;ůŽŐͿƉƌŝĐĞ͘dŚĞƌĞŐƌĞƐƐŝŽŶƐĞdžĐůƵĚĞ ŽƵƚůŝĞƌƐŝŶƚŚĞƚŽƉĂŶĚďŽƚƚŽŵϯƌĚƉĞƌĐĞŶƚŽĨƚŚĞŵĂƌŬƵƉĚŝƐƚƌŝďƵƚŝŽŶ͘ ůůƌĞŐƌĞƐƐŝŽŶƐŝŶĐůƵĚĞĨŝƌŵͲƉƌŽĚƵĐƚĨŝdžĞĚĞ
Slide 42: The role of trade liberalization: distributions
The role of trade liberalization: distributions I We exploit the exogenous trade liberalization episode and plot for 1989 and 1997 1. Markups 2. Marginal costs 3. Prices
Slide 43: Distribution of Prices
Change in prices Distribution of Prices 2.5 21.5 Density 1 .5 0 -.5 0 .5 Log Prices 1989 1997 Sample only includes firm-product pairs present in 1989 and 1997. Observations are de-meaned by their time average, and outliers above and below the 3rd and 97th percentiles are trimmed.
Slide 44: Change in markup and mc distribution
Change in markup and mc distribution Distribution of Marginal Costs 1 Density .5 0 -2 -1 0 1 2 Log Marginal Costs 1989 1997 Sample only includes firm-product pairs present in 1989 and 1997. Observations are de-meaned by their time average, and outliers above and below the 3rd and 97th percentiles ar
Slide 45: The impact of trade liberalization
The impact of trade liberalization I We now consider the following specifications: afjt = αfj + αst + β1 τitoutput + β2 τitinput + ηfjt (32) I where a: prices, markups and marginal costs. I Cluster s.e. on industry everywhere.
Slide 46: Firm performance and trade liberalization: a decomposition
Firm performance and trade liberalization: a decomposition Log Pricesfjt Log Marginal Costfjt Log Markupfjt (4) (5) (6) Output Tariffit 0.135 *** 0.083 0.052 0.053 0.116 0.083 Input Tariffit 0.343 1.405 * -‐1.062 ** 0.450 0.876 0.544 marginal cost R-‐squared 0.02 0.03 0.03 Observations 20,654 20,6
Slide 47: Identifying the mechanism
Identifying the mechanism >ŽŐDĂƌŬƵƉĨũƚ ;ϭͿ ;ϮͿ >ŽŐDĂƌŐŝŶĂůŽƐƚĨũƚ ͲϬ͘ϳϮϲΎΎΎ ͲϬ͘ϳϮϲΎΎΎ Ϭ͘Ϭϲϯ Ϭ͘Ϭϲϯ KƵƚƉƵƚdĂƌŝĨĨŝƚ Ϭ͘ϬϵϯΎΎ Ϭ͘ϬϵϯΎΎ Ϭ͘Ϭϯϴ Ϭ͘ϬϰϮ /ŶƉƵƚdĂƌŝĨĨŝƚ Ϭ͘ϬϮϱ Ϭ͘ϯϬϵ ZͲƐƋƵĂƌĞĚ Ϭ͘ϲϮ Ϭ͘ϲϮ KďƐĞƌǀĂƚŝŽŶƐ ϭϵ͕ϳϲϴ ϭϵ͕ϳϲϴ &ŝƌŵͲWƌŽĚƵĐƚ&Ɛ LJĞƐ LJĞƐ ^ĞĐƚŽƌͲzĞĂƌ&Ɛ LJĞƐ LJĞƐ EŽƚĞƐ͗dŚĞĚĞƉĞŶĚĞŶƚǀĂƌŝĂďůĞŝƐ;ůŽŐͿŵĂƌŬƵƉ͘dŚ
Slide 48: A Stylized Illustration of our findings
A Stylized Illustration of our findings Price P1 μ1 P2 mc1 μ2 mc2 MR D Quan+ty Q1 Q2
Slide 49: Conclusion
Conclusion I Provide a new method to deal with multi-production function estimation using product-level quantity data. We think there is a lot of room for applying our framework. I Recover full distribution of markups, marginal costs and productivity using minimal assumptions on firm behavior, i.e.
Slide 50: Implications of findings
Implications of findings I Implications for demand systems we use in pass-through and welfare analysis of policy changes. In particular it will help shed light on dynamic competition: how do firms adjust prices in light of changes in market structure and facing a demand curve. We are not restricting
Banque Nationale De Belgique — Technical Presentation (February 2013) on 6ix