ECON 357
Lecture 6: estimating …rm level productivity
Thomas Chaney
University of Chicago
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 1 / 15
Estimating …rm-level productivity
Qit = F (Lit , Mit , Kit ) exp (ω it + uitq )
Lit , Mit variable factors of production (labor and intermediates).
Kit capital stock.
ω it …rm level productivity (unobserved).
uit unobserved shocks.
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 2 / 15
Empirical challenges
1 Simultaneity bias:
ω it a¤ects both productivity and input choices.
β and βk biased upwards.
2 Selection bias:
if low productivity realization, …rms with more kit more likely to survive.
) sprurious negative correlation between capital and productivity.
βk biased downward.
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 3 / 15
Additional challenges
Qit = F (Lit , Mit , Kit ) exp (ω it + uitq )
1 Unobserved prices and mark-ups.
Traditional method: de‡ate sales (rit ) by industry prices (pIt ).
Problem: (pit pIt ) 6= 0 induces bias.
2 Problem: if trade a¤ects prices and mark-ups, impact of trade on
productivity misidenti…ed.
Note 1: we will assume Cobb-Douglas for F ( ).
Note 2: we will assume CES demand system.
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 4 / 15
Tentative solutions
1 Olley-Pakes: use investment decision to control for simultaneity and
selection bias.
2 Levinsohn-Petrin: use intermediate inputs instead.
3 de Loecker: use variation in …rm-level exposure to trade shocks to
identify mark-ups.
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 5 / 15
CES demand system
ηs
Pjt
Qjt = Qst exp ξ jt + ujtq
Pst
Good j in segment s.
ξ jt : unobserved demand shock (may be correlated with price).
ujtq : i.i.d. demand shock.
ηs
Note: constant mark-up ηs 1 .
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 6 / 15
Empirical challenges: summary
(
r̃it = β0 + βl lit + βm mit + βk kit + βs qst + ω it + ξ it + εit
ηs 1
with βh = ηs 1 αh , h = fl, m, k g , βs = ηs .
ω it unobserved and correlated with input choices.
non random exit of …rms (biases estimate of βk ).
demand shock ξ it correlated with input choice.
aggregate demand shifter correlated with unobserved demand
conditions.
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 7 / 15
Multi-product dimension
Qijt = (cijt )γ Qit
with ∑j 2J (i ) cijt h =c
= 1 and cijt ijt for h = fl, m, k g .
Firms split up their production between di¤erent products.
Assumption: same factor intensities for all inputs.
γ measures returns to scale.
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 8 / 15
Multi-product production function
r̃it = β0 + βl lit + βm mit + βk kit +
!
5
+ ∑ βs (sis qst ) + ln ∑ (cijt )γ/η
s =1 j 2J (i )
+ ω it + ξ it + εit
Demand shocks di¤er across segments (qst ), and …rms di¤er in their
exposure to those shocks (sis ).
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 9 / 15
Estimation strategy
Investment function can be inverted:
ω it = ht (iit , kit , qrit , npi )
qrit represents …rm i’s exposure to quota reduction, npi number of
products.
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 10 / 15
Estimation
First stage: (use a polynomial approximation for h̃)
r̃it = β0 + βl lit + βm mit + βk kit
5
+ ∑ βs (sis qst ) + ∑ δj Dij + ∑ δg Dig + φ̃t (iit , kit , qrit , npi ) + εit
s =1 j 2J (i ) g 2G (i )
φ̃t (iit , kit , qrit , npi ) = βk kit + h̃t (iit , kit , qrit , npi )
b̃ . Note: η̂ = 1/ β̂ .
) get β̂l , β̂m , β̂s , δ̂j , δ̂g , φ it s s
Second stage: (use a polynomial approximation for g )
5
r̃it β̂l lit β̂m mit ∑ β̂s (sis qst ) ∑ δ̂j Dij ∑ δ̂g Dig
s =1 j 2J (i ) g 2G (i )
b̃
= β0 + βk kit +1 + gt φ βk kit + ξ̃ it +1 + ζ it +1 + εit +1
it
) get β̂k .
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 11 / 15
Key steps in the estimation
Control for endogenous choice of k and endogenous selection
(unbiased β’s).
Control for mark-ups (η ).
Control for simultanuous impact of demand shocks on input choices,
prices, quantities and selection.
Method:
1 use …rm level variation in demand induced by di¤erent trade shocks.
2 invert investment equation to control for choice of capital.
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 12 / 15
Backing out productivity
η̂ s
ω̂ it = ∑ sis r̃it β̂l lit β̂m mit β̂k kit β̂s (sis qst )
η̂ s 1
s
Key estimates to uncover ω are β’s and η.
Note that segment and sub-segment …xed e¤ects can a¤ect
productivity.
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 13 / 15
Impact of trade on productivity
ω̂ it = λ0 + λ1 qrit + controls + εit
Di¤erent estimates of ω̂ will give di¤erent answers.
If quotas a¤ect demand, not controlling for the impact of quotas on
demand will overestimate λ1 .
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 14 / 15
Impact of trade on productivity
Figure: Jan de Loecker (2009)
Thomas Chaney (Chicago) ECON 357: Lecture 6 (de Loecker) 15 / 15
