It’s Not You, It’s Me: Breakups in U.S.-China Trade Relationships Ryan Monarch Federal Reserve Board
[email protected]
August 7, 2014
Any opinions and conclusions expressed herein are those of the author and do not necessarily represent the views of the Federal Reserve Board of Governors, any other person associated with the Federal Reserve System, or the U.S. Census Bureau. All results have been reviewed to ensure that no confidential information is disclosed.
Introduction
1/ 39
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Mismatch between producers and suppliers can be a source of inefficiency. If buyers do not know about lower-priced alternatives to their current supplier, or are unable / unwilling to pay the cost of changing, this can lead to high prices. Anecdotal evidence suggests that finding the right supplier is difficult:
Introduction
1/ 39
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Mismatch between producers and suppliers can be a source of inefficiency. If buyers do not know about lower-priced alternatives to their current supplier, or are unable / unwilling to pay the cost of changing, this can lead to high prices. Anecdotal evidence suggests that finding the right supplier is difficult:
Introduction
1/ 39
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Mismatch between producers and suppliers can be a source of inefficiency. If buyers do not know about lower-priced alternatives to their current supplier, or are unable / unwilling to pay the cost of changing, this can lead to high prices. Anecdotal evidence suggests that finding the right supplier is difficult:
Little empirical work on why one supplier is chosen and not another.
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
In this paper, I:
2/ 39
Use two-sided trade data from U.S. Customs and Border Protection to describe how U.S. importers match with Chinese exporters. Model an importer’s choice of exporter as a dynamic discrete choice problem with switching costs: Cost of switching partner Additional cost of switching city.
Estimate these switching frictions across industries. Determine how import prices will change based on switching becoming more common and/or more efficent.
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Preview of Empirical Results: Importers are very likely to remain with their partner over time.
3/ 39
45% of U.S. continuing importers from China maintain their partner year-to-year (average of 30 choices in an HS10 product).
A significant share of switching occurs within-city. 1/3 of switching importers remain in the same city (average of 9 cities).
Higher prices strongly affect the probability of switching.
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Preview of Quantitative Results:
4/ 39
Structural estimates of switching costs are large and heterogeneous across industries. The impact of switching frictions on prices is sizable: Halving costs leads to a 12.5% lower U.S.-China Import Price Index. Eliminating partner costs while maintaining geographic costs lowers prices by 15.2%.
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Related Research Use of importer-exporter relationship data:
5/ 39
Kamal and Krizan (2012), Eaton, Eslava, Jinkins, Krizan, Kugler, and Tybout (2012), Kamal and Sundaram (2013), Bernard, Moxnes, and Ulltveit-Moe (2013)
Discrete Choice Modeling: Rust (1987), Berry, Levinsohn, and Pakes (1995), Fox (2010), Su and Judd (2012), Dub´e, Fox, and Su (2012)
Relationships, networking and contract formation among suppliers: Joskow (1985), Doney and Cannon (1997), Kranton and Mineheart (2001), Rauch (2001), Rauch and Watson (2004), Ahn, Khandewal and Wei (2011), Tang and Zhang (2012)
Relationship frictions in importer-exporter choice Alessandria (2009), Kleshchelski and Vincent (2009), Drozd and Nosal (2012)
Data
Introduction
7/ 39
Data
Model
Implementation
Results
Counterfactuals
U.S. import data is from the Linked-Longitudinal Foreign Trade Transaction Database (LFTTD), collected by U.S. Customs and Border Protection and maintained by U.S. Census Bureau. Each transaction contains (a) (b) (c) (d)
Value and quantity (and thus “unit value”) HS 10 industry code. Unique exporter identifier (known as Manuf. ID). Exporter location (city) information.
Conclusion
Introduction
Data
Model
Implementation
Results
Counterfactuals
U.S. import data is from the Linked-Longitudinal Foreign Trade Transaction Database (LFTTD), collected by U.S. Customs and Border Protection and maintained by U.S. Census Bureau. Each transaction contains (a) (b) (c) (d)
Value and quantity (and thus “unit value”) HS 10 industry code. Unique exporter identifier (known as Manuf. ID). Exporter location (city) information.
Example: MONARCH PAPER PRODUCTS 426 THOMPSON ST, BASEMENT ANN ARBOR, MI 48104 USA
7/ 39
Internal Validity
External Validity
Conclusion
Introduction
Data
Model
Implementation
Results
Counterfactuals
U.S. import data is from the Linked-Longitudinal Foreign Trade Transaction Database (LFTTD), collected by U.S. Customs and Border Protection and maintained by U.S. Census Bureau. Each transaction contains (a) (b) (c) (d)
Value and quantity (and thus “unit value”) HS 10 industry code. Unique exporter identifier (known as Manuf. ID). Exporter location (city) information.
Example: MONARCH PAPER PRODUCTS 426 THOMPSON ST, BASEMENT ANN ARBOR, MI 48104 USA
8/ 39
Internal Validity
External Validity
Conclusion
Introduction
Data
Model
Implementation
Results
Counterfactuals
U.S. import data is from the Linked-Longitudinal Foreign Trade Transaction Database (LFTTD), collected by U.S. Customs and Border Protection and maintained by U.S. Census Bureau. Each transaction contains (a) (b) (c) (d)
Value and quantity (and thus “unit value”) HS 10 industry code. Unique exporter identifier (known as Manuf. ID). Exporter location (city) information.
Example: MONARCH PAPER PRODUCTS 426 THOMPSON ST, BASEMENT ANN ARBOR, MI 48104 USA
9/ 39
Internal Validity
External Validity
USMONPAP426ANN
Conclusion
771673 Introduction
354052 Data
478006 Model
539859 Implementation
Results
Counterfactuals
Conclusion
Fact 1: Many Importer-Exporter Relationships Remain Over Time Fact 2: Strong Geographic Component to Switching Behavior Figure: “Staying” Percentages of Continuing Importers from China, 2003-2008 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Same Partner
Same City
Same Province
Staying means “highest value share from same supplier” (Mean = 83.9%). Does not inlcude “related party” imports.
Internal Validity 10/ 39
External Validity
Robustness
2003 39,098,921,196 Introduction Data 2004 62,308,431,843 2005 83827616010 2006 1.12478E+11 2007 1.25419E+11 2008 140,239,928,638
Model
35182994843 Implementation Results 55926603344 74123737186 92421195287 1.10957E+11 1.22529E+11
19367929172 Counterfactuals Conclusion 29198387690 37454826628 47534086862 58207220996 61416071528
Fact 1: Many Importer-Exporter Relationships Remain Over Time Fact 2: Strong Geographic Component to Switching Behavior Figure: U.S. (Non-Subsidiary) Imports from China, 2003-2008 160 140 120
152,436,100,000
Billions USD
100 80 60 40 20 0 2003
2004 Total Imports
2005
2006
2007
2008
Total Imports, Same Partner
Staying means “highest value share from same supplier” (Mean = 83.9%). Does not inlcude “related party” imports.
Internal Validity 11/ 39
External Validity
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Fact 3: The decision to switch is correlated with price and exporter/importer covariates. Dep. Var.: Stayed with Chinese Exporter Year-to-Year, 2002-2008 (1) (2) (3) (4) ∗∗∗ ∗∗∗ ∗∗∗ Log Price -0.0084 -0.0099 -0.0104 -0.0106∗∗∗ (0.001)
(0.001)
(0.001)
Log Supplier Size
(0.001)
0.0399∗∗∗
0.0643∗∗∗
0.0643∗∗∗
(0.001)
(0.001)
(0.001)
Supplier Age
-0.0024∗∗∗
-0.0031∗∗∗
-0.0029∗∗∗
(0.000)
Importer Size Entry Year FE N R2
No 510,485 0.07
No 510,485 0.09
(0.000)
(0.000)
-0.0322∗∗∗
-0.0306∗∗∗
(0.001)
(0.001)
No 510,485 0.10
Yes 510,485 0.10
HS10 product and year fixed effects included. Standard errors clustered at product level.
12/ 39
Introduction
Data
Model
Implementation
Results
Counterfactuals
Fact 3: The decision to switch is correlated with price and exporter/importer covariates.
13/ 39
Conclusion
Model
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Dynamic Discrete Exporter Choice Model
15/ 39
Importer m picks the exporter xtm ∈ Xt giving highest profits, knowing m E px,t : Expected price from each xt βx , βc : Frictions from switching exporter and city. λx,t : Quality for each xt m x,t : Match-specific shock for each xt (i.i.d).
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Dynamic Discrete Exporter Choice Model
15/ 39
Importer m picks the exporter xtm ∈ Xt giving highest profits, knowing m E px,t : Expected price from each xt βx , βc : Frictions from switching exporter and city. λx,t : Quality for each xt m x,t : Match-specific shock for each xt (i.i.d).
Price transition process- parametrized with LFTTD: xtm = xtm−1 :
m m m px,t = px,t−1 + ηc,t + ux,t
xtm = xem 6= xtm−1 :
pxem,t =
m 2 ux,t ∼ N 0, σx,t
1 N
PN
n=1
pxen,t−1 + ηce,t + uxem,t
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Sketch of the Model: which x maximizes profits?
16/ 39
Possible Choices
X1
Importer
X2 (current)
Choice Affected By E[p1], λ1, ε1, must pay βx
E[p2], λ2, ε2
X3 E[p3], λ3, ε3, must pay βx and βc
Introduction
Data
Model
Implementation
Results
Counterfactuals
Importers Final good producer m requires J inputs, indexed j = 1, ..., J. m J . Thus an xj,t for each product j, with whole set Xtm = xj,t j=1
Monopolistically competitive in its final good. Production is Cobb-Douglas in labor and inputs. CES demand for its final good, with elasticity of substitution σ.
17/ 39
Conclusion
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Importers Final good producer m requires J inputs, indexed j = 1, ..., J.
17/ 39
m J . Thus an xj,t for each product j, with whole set Xtm = xj,t j=1
Monopolistically competitive in its final good. Production is Cobb-Douglas in labor and inputs. CES demand for its final good, with elasticity of substitution σ.
Choice of which exporter xj,t to obtain input j. i h m Form expectations about the price from an exporter E pj,x,t . Frictions from changing exporters, measured as an additional component of the price paid: m m m m m m p¯x,j,t = E pj,x,t exp ζx,j 1{xj,t 6= xj,t−1 } + ζc,j 1{cj,t 6= cj.t−1 }
Introduction
18/ 39
Data
Model
Implementation
Results
Counterfactuals
Conclusion
n oJ m : Expected importer profits at t with choice vector Xtm = xj,t j=1
πtm = maxm pm Qm − MC (Xtm ) Qm pm ,Xt
Specific form of MC
Each product j is additively separable in log terms, and thus I focus on the purchase of good j only.
Introduction
19/ 39
Data
Model
Implementation
Results
Counterfactuals
Conclusion
For any importer, the log expected exporter-specific profit term from a match xtm is: m m m πm t (xt , β) + x,t = βp E ln px,t + ξ λx,t − |{z} βx |{z}
1
{xtm
6=
m xt−1 }
PartnerCost Specifics
Preference Heterogeneity
−
βc |{z}
CityCost
1
{ctm
6=
Quality m ct−1 } +
m x,t |{z}
Profit Shock
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Value Function Importer m chooses an exporter x in each period to maximize profits. " V (pt−1 , xt−1 , t−1 ) =
max
{xt ,xt+1 ,...}
E
∞ X
# δ τ −t (π τ (xτ , pτ −1 , xτ −1 , β) + x,τ )
τ =t
which can be rewritten as a Bellman Equation: 0 0 0 + δEV x , p, x, π x , p, x, β + V (p, x, ) = max x 0 x
for
20/ 39
0
EV x , p, x, =
Z Z p0
0
V p 0 , x 0 , 0 h p 0 , 0 |p, x, x 0 , dp 0 d0
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Assumption (1) Conditional Independence h (pt+1 , t+1 ) = g (t+1 ) f (pt+1 |pt , xt , xt+1 ) Assumption (2) Type I Extreme Value (Gumbel) Shocks Pr (t < y ) = G (y ) = exp{− exp{−y − γ}}, γ = 0.577
21/ 39
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Assumption (1) Conditional Independence h (pt+1 , t+1 ) = g (t+1 ) f (pt+1 |pt , xt , xt+1 ) Assumption (2) Type I Extreme Value (Gumbel) Shocks Pr (t < y ) = G (y ) = exp{− exp{−y − γ}}, γ = 0.577 Proposition For s = {p−1 , x−1 }, and any variable a0 as the one-period ahead of a: exp π x C , s, β + δEV x C , s C P x |s, β = P (1) x , s, β) + δEV (b x , s)] xb∈X exp [π (b
21/ 39
for (
Z EV (x, s) =
log s0
X
) exp π x 0 , s 0 , β + δEV x 0 , s 0 f s 0 |s, x
x 0 ∈X
(2)
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Maximum Likelihood The parameters β can then be solved for via maximum likelihood m , pm estimation. For observed {xt−1 x,t−1 }, the likelihood of observing the set of outcomes at time t is
22/ 39
L (β) =
M Y m=1
m m m m m P xtm |xt−1 , px,t−1 , β · f px,t |px,t−1 , xt−1 , xtm
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Maximum Likelihood The parameters β can then be solved for via maximum likelihood m , pm estimation. For observed {xt−1 x,t−1 }, the likelihood of observing the set of outcomes at time t is
22/ 39
L (β) =
M Y
m m m m m P xtm |xt−1 , px,t−1 , β · f px,t |px,t−1 , xt−1 , xtm
m=1
The log-liklihood maximization problem can be written: max ln L (β) = max β
β
M X
P m=1
m m m m exp [π m t (xt , st , β) + δEV (xt , st )] + ... m m xt , st , β) + δEV (b xtm , stm )] xbtm ∈X exp [π (b
(3)
s.t Z EV (xt , st ) =
log st+1
X
xt+1 ∈X
exp [π t+1 (xt+1 , st+1 , β) + δEV (xt+1 , st+1 )]
× f (st+1 |st , xt )
(4)
Implementation
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Bringing the Model to the Data In order to compute EV (xt , st ) in (4), I approximate the integral using the equation: EV (xt , sbt ) = N X b st+1 =1
24/ 39
log
X
xt+1 ∈X
st+1 , xt+1 )] exp [π t+1 (xt+1 , sbt+1 , β) + δEV (b
· Pr (b st+1 |b s t , xt )
Many importers have more than one exporting partner: I make the observed choice “who do I get the majority of my imports from”. The average share of imports from a U.S. importer’s main Chinese exporte is 83.9%, standard deviation of 22%.
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
I use the MPEC methodology of Su and Judd (2012) and Dub´e, Fox and Su (2012) to solve the MLE problem The only values of β and EV that are used as test candidates are those that satisfy the fixed point equation above.
Data from U.S.-China trade between 2005-2006. Use TOMLAB / KNITRO to compute the Jacobian and the gradient numerically, and then solve the above MLE problem. Fixed parameters: δ=0.975, N=5 Quality λ is approximated similar to the control function approach of Kim and Petrin (2008) and Khandewal (2011). Quality Monte Carlo Results
25/ 39
Results
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Summary of Results
27/ 39
Model estimates of switching costs are large βx βc
Partner Cost City Cost
Mean 2.99 1.61
For 2 identical partners, switching requires an unobservable shock ≈ 2.3 standard deviations above the mean.
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Summary of Results
27/ 39
Model estimates of switching costs are large βx βc
Partner Cost City Cost
Mean 2.99 1.61
For 2 identical partners, switching requires an unobservable shock ≈ 2.3 standard deviations above the mean.
Estimates are heterogeneous across industries.
Introduction
Data
Model
Implementation
Results
Counterfactuals
Low out-city switching implies high out-city switching costs...
28/ 39
βx βc
HS6 841320
Table: Parameter Estimates HS6 841320 820310 Partner Cost 1.67 2.74 Out-City Cost 3.95 2.95
HS6 820310
Conclusion
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Low out-city switching implies high out-city switching costs...
28/ 39
βx βc
HS6 841320
Table: Parameter Estimates HS6 841320 820310 Partner Cost 1.67 2.74 Out-City Cost 3.95 2.95
HS6 820310
Very few switching importers (≈ 20%) switch outside of their city.
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
and high out-city switching implies low out-city switching costs.
29/ 39
βx βc
Table: Parameter Estimates HS6 841320 820310 847130 Partner Cost 1.67 2.74 3.00 Out-City Cost 3.95 2.95 0.43
HS6 847130
HS6 871120
871120 3.91 0.19
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
and high out-city switching implies low out-city switching costs.
29/ 39
βx βc
Table: Parameter Estimates HS6 841320 820310 847130 Partner Cost 1.67 2.74 3.00 Out-City Cost 3.95 2.95 0.43
HS6 847130
HS6 871120
871120 3.91 0.19
Most switching importers (≈ 80%) switch outside of their city.
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
High sensitivity to price changes implies a strong effect on switching...
30/ 39
βp
Table: Parameter Estimates HS10 6116106500 Price Elasticity of Switching -0.06
HS10 6116106500
HS10 6403406000
6403406000 -0.05
Import price elasticity is large (9-14)
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
and low sensitivity to price changes implies a non-effect on switching.
31/ 39
βp
Table: Parameter Estimates HS10 611610 640340 6116930800 Elas. of Switching -0.06 -0.05 0.25
HS10 6116930800
HS10 6107110010
6107110010 0.51
Import price elasticity is small (1-2)
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Model Fit
32/ 39
I estimate parameters for 50 randomly chosen industries. Compute BLS Import Price Index: Weight by firm size to make Industry Price Index, then weight by industry size.
Price Index Weighted Average Median
Data
Median over 1000 runs
%
84.6239 66.1725
76.4979 61.7019
90.4 93.2
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Model Fit
32/ 39
I estimate parameters for 50 randomly chosen industries. Compute BLS Import Price Index: Weight by firm size to make Industry Price Index, then weight by industry size.
Price Index Weighted Average Median
Data
Median over 1000 runs
%
84.6239 66.1725
76.4979 61.7019
90.4 93.2
Total Switching Partner Total Switching City
Data 714 416
Industry Median 711 469
% 99.6 112.7
Counterfactuals
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Counterfactual Experiments
34/ 39
I estimate parameters for 50 randomly chosen industries.
Counterfactual I: Adjust the switching cost parameters, reassign importers to exporters, and calculate the BLS Import Price Index for each set of outcomes. Partner Cost βx ↓ 50%, City Cost βc ↓ 50% =⇒ Prices decrease by 12.5% % of Staying Importers ↓ from 57% to 18%
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Counterfactual I: Adjust the switching cost parameters, and reassign importers to different exporters: Figure: Kernel Density Plots, Original β vs. Reduced By Half
Individual Industries
35/ 39
Introduction
36/ 39
Data
Model
Implementation
Results
Counterfactuals
Figure: Kernel Density Plots, Selected Industries
Conclusion
Overall Price Index
HS 401120, Rubber Tires
HS 610432, W/G Cotton Jackets
HS 847130, Laptop Computers
HS 640340, Metal Toe-Cap Ftwr
HS 852520, Cell Phones
HS 850940, Mixers/Blenders
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Counterfactual I: Adjust the switching cost parameters, and reassign importers to different exporters: βx = 0, βc unchanged =⇒ Prices decrease by 15.20% % of Staying Importers ↓ from 57% to 8%
βx unchanged, βc = 0 =⇒ Prices decrease by 7.37% % of Switching Importers ↓ from 57% to 31%
βx triples, βc triples =⇒ Prices increase by 7.62% % of Staying Importers ↑ from 57% to 90% Summary
37/ 39
Introduction
38/ 39
Data
Model
Implementation
Results
Counterfactuals
Counterfactual II: Allow for an additional supplier, that can be switched to with no city-switching friction: Table: Counterfactual Results (II) Reduction from average Offered Price Trade Share U.S. Exporter Price Median 3.86% 56.27% 75th Pct. 3.18% 34.04%
Conclusion
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Conclusion
39/ 39
Demonstrated new empirical regularities for exporter switching. Solved a Dynamic Discrete Choice model to solve for geographically diverse switching costs. Estimated cost measures that connect well to our conceptions of substitutability and geographic switching behavior. Performed counterfactual experiments to demonstrate the importance of the switching channel for import prices.
Thank You!
Introduction
Data
Model
Implementation
Results
Counterfactuals
Figure: Robustness Checks, % of Importers Staying
Conclusion
Go Back
Importer_counts Importer_counts
Original Partner Found in Later Year Original Partner Found in Later Year 1
1
0.9
0.9
0.8
0.8
0.7
0.7 0.6
0.6
0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1 0
2002 2003 2004 2005 2006 2007
Same Partner
1.00 Same 0.44 0.29 Sheet1 0.19 0.13 0.10
Same City
1.00 0.61 0.50 0.42 0.37 0.34
Sheet1 2002 1.00 2003 0.68 2004 0.60 2005 0.53 2006 0.49 2007 0.47
1.00 0.61 0.61 0.61 0.61 0.61
1.00 0.68 0.68 0.68 0.68 0.68
Province
Importer as Firm-HS6 Code 1.0
1.0
0.9
1.00 0.9 0.44 0.8 0.44 0.7 0.44 0.44 0.6 0.44 0.5
0.8 0.7 0.6 0.5 0.4
0.1
0.0
0.0 Same Partner
Same City
0.9 0.8 0.7
Same Partner
0.00 0.80 2006
2007
Same Province
0.53 0.49 0.47
Same City
0.00
Same Province
Same City
1.0
Same Province
Page 4
0.9 0.8 0.7 0.6 0.5 0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0.0
1.00
1.00 0.90
Same Partner
Same City
2002
Same Par
2002
Same Pa
1.00 Same Pa 0.68 0.68 0.68 0.68 0.68
Same Partner
2008
0.00
2002
1.00 0.61 0.61 0.61 0.61 0.61
Importer as Firm
0.5
0.40 0.20
0.0
Same Partner
0.60
Manufacturing Firms Only 0.20
0.1
Same Province
0.6
0.0
Sheet1 0.00 0.80 2005 1.00 2002 0.60 0.68 Same City Same Par 0.60 0.40
2004
0.2
Page 2
1.0
1.00 0.61 0.50 0.42 0.37 0.34
0.3
0.2
0.1
1.00 0.44 0.8 0.44 0.7 0.44 0.44 0.6 0.44 0.5 0.9
2003
0.4
0.3
0.2
41/ 39
1.0
0.4
0.3
0
1.00 0.44 0.29 0.19 0.13 0.10
0.90 Same Province
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Are these identified exporters some other firm, not the manufacturer?
42/ 39
Possibility 1: This is actually the shipping / forwarding firm, not the manufacturer. Possibility 2: This is a trading house or intermediary.
Possibility 1: This is actually the shipping / forwarding firm.
Possibility 2: This is a trading house or intermediary.
Is the exporter code is uniquely capturing one exporting firm?
Is the exporter code is uniquely capturing one exporting firm?
Panel A: Uniqueness of the “MID”, 2005 Industry (CIC) # of Exporters # of “MID”s CIC 3663 39 38 CIC 3689 27 26 CIC 3353 37 37 CIC 3331 35 35 CIC 4154 74 73 Panel B: Uniqueness of the City Code Industry (CIC) # of Cities # of City Codes,2005 CIC 3663 22 21 CIC 3689 15 14 CIC 3353 28 24 CIC 3331 15 13 CIC 4154 19 18
% 97.4 97.3 100 100 98.6
% 95.5 93.3 85.7 86.7 94.7
Panel C: Changes in the “MID” over Time, 2005-2006 Industry (CIC) # of Exporters # of with Identical “MID” CIC 3663 33 33 CIC 3689 26 26 CIC 3353 31 28 CIC 3331 20 17 CIC 4154 63 62
Go back
% 100 100 90.3 85.0 98.4
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Fact 1: Prevalent Breakups in Import-Export Relationships Fact 2: Strong Geographic Component to Switching Behavior
Table 1: Firm‐HS10 Code Year Total Importers Same Partner 2003 99,116 55,272 2004 84,367 35,193 2005 74,506 22,594 2006 64,323 15,683 2007 50,105 9,840
Sheet1 Same City Same Province 65,905 69,310 50,478 54,967 40,755 45,865 34,033 39,317 25,697 30,290
Table: Later Year “Staying” Counts of U.S. Firms Importing from China in 2002
Go Back
49/ 39
China 80,963 67,908 60,143 53,514 42,201
Introduction
50/ 39
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Production is Cobb-Douglas. Unit cost bundle: cm (Xtm ) = w α
J Y
1−α m p¯x,j,t
γj
j=1
Importer productivity depends on both individual firm productivity and the product quality/ specific match from the exporter it uses: φm (Xtm ) = ψm
J Y
λνx,j
j=1
Marginal cost function is: MC (Xtm ) = cm (Xtm ) Go back
1 φm (Xtm )
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Assuming CES demand with e.o.s. σ, profits are: πtm
1 = max B m Xt σ
σ σ−1
1−σ
[φm ((Xtm )]σ−1 cm (Xtm )1−σ
After taking logs, and focusing on the choice of input j, the supplier choice profit maximization problem is: X m m ln πtm = A + ln πj,t + ln πk,t k6=j
where m ln πj,t = max ν (σ − 1) ln λx,j,t m xj,t
m m m m m + (1 − α) (1 − σ) γj E ln px,j,t + ζx,j 1 xj,t 6= xj,t−1 + ζc,j 1 cj,t 6= cj,t−1 Go back
51/ 39
Introduction
52/ 39
Data
Model
Implementation
Results
Counterfactuals
Conclusion
If importer m stays with its current partner x in city c, using (1) and taking logs, we can write the price process as: pm,x,t = pm,x,t−1 + ηc,t + um,x,t
(5)
A high price increase from one exporter could be due to city reasons or idiosyncratic reasons. Key importer decision: far city, lower average price vs. same city, high probability for high average price.
If importer m decides to use a different partner x 0 in the same city c, then the price process is: pm,x 0 ,t =
N 1 X pn,x 0 ,t−1 + ηc,t + vm,x,t N n=1
n = 1, ..., N are the importers from x 0 in time t − 1.
(6)
Introduction
53/ 39
Data
Model
Implementation
Results
Counterfactuals
Conclusion
If the importer decides to go to an exporter x 00 in another city c 0 , then the price process is: 0
pm,x 00 ,t
N 1 X = 0 pn0 ,x 00 ,t−1 + ηc 0 ,t + wm,x 00 ,c 0 ,t N 0 n =1
(7)
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
If the importer decides to go to an exporter x 00 in another city c 0 , then the price process is: 0
pm,x 00 ,t
N 1 X = 0 pn0 ,x 00 ,t−1 + ηc 0 ,t + wm,x 00 ,c 0 ,t N 0
(7)
n =1
I use these to characterize a distribution of expected prices to avoid a Heckman-type selection problem. Go back
53/ 39
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
To estimate unobserved exporter heterogeneity, I assume a simple price process for exporters: m px,t =µ
1 β (λx,t ) zx
Then take logs and regressing export price on a number of exporter-specific covariates: Number of HS 10 products exported Age Size of exports to the U.S. Total number of transactions Total number of import partners. Location fixed effects.
Following the control function estimation technique of Kim and Petrin (2008), use the residual unobserved heterogeneity as a component of the profit function. Go back
54/ 39
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Table: Monte Carlo Replication Results, based on 250 Replications Conv. βp βx βc Rate Pre-Set Values −0.5 −1 −3 Sample A: M = 6, X = 3, C = 2 .456 Mean −0.378 −35.45 −3.633 Median −0.341 −2.293 −3.429 Sample B: M = 30, X = 33, C = 3 1 Mean −0.543 −0.843 −8.209 Median −0.540 −0.837 −5.017 Sample C: M = 30, X = 33, C = 9 1 Mean −0.538 −0.985 −3.427 Median −0.512 −1.055 −3.081 Go Back
55/ 39
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Value Function Importer m chooses an exporter x in each period to maximize profits. " V (pt−1 , xt−1 , t−1 ) =
max
{xt ,xt+1 ,...}
E
∞ X
# δ τ −t (e π (pτ −1 , xτ −1 , xτ , β) + x,τ )
τ =t
which can be rewritten as a Bellman Equation: 0 0 0 + δEV p, x, x , V (p, x, ) = max π e p, x, x , β + x 0 x
for EV p, x, x
0
Z Z = p0
Go Back
56/ 39
0
V p 0 , x 0 , 0 h p 0 , 0 |p, x, x 0 , dp 0 d0
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
Bringing the Model to the Data Many importers have more than one exporting partner: I make the observed choice “who do I get the majority of my imports from”. The average share of imports from a U.S. importer’s main Chinese exporte is 83.9%, standard deviation of 22%.
Data from U.S.-China trade between 2005-2006. Quality λ is approximated similar to the control function approach of Kim and Petrin (2008). Quality Monte Carlo Results
57/ 39
Introduction
58/ 39
Data
Model
Implementation
Results
Counterfactuals
Conclusion
I use the MPEC methodology of Su and Judd (2012) and Dube, Fox and Su (2012) to solve the MLE problem The only values of β and EV that are used as test candidates are those that satisfy the fixed point equation (2) above.
Introduction
Data
Model
Implementation
Results
Counterfactuals
Conclusion
I use the MPEC methodology of Su and Judd (2012) and Dube, Fox and Su (2012) to solve the MLE problem The only values of β and EV that are used as test candidates are those that satisfy the fixed point equation (2) above.
Use TOMLAB / KNITRO to compute the Jacobian and the gradient numerically, and then solve the above MLE problem. Price space discretized into N = 5 intervals. Rate of time preference δ = 0.975. Go Back
58/ 39
Introduction
59/ 39
Data
Model
Implementation
Results
Counterfactuals
Conclusion
The above equation does not account for serial correlation in : If the importer has some characterstic that makes them prefer exporter x in both periods, this will show up as high switching cost.
Reduced form check: The correlation of lagged quality λx,t−1 on exporter choice xt . Not related to current profits, but if correlated with future choice, then can augment the error term for serial correlation.
Introduction
59/ 39
Data
Model
Implementation
Results
Counterfactuals
Conclusion
The above equation does not account for serial correlation in : If the importer has some characterstic that makes them prefer exporter x in both periods, this will show up as high switching cost.
Reduced form check: The correlation of lagged quality λx,t−1 on exporter choice xt . Not related to current profits, but if correlated with future choice, then can augment the error term for serial correlation. m m m πm t (xt , β) + x,t = βp E ln px,t + ξλx,t m m m −βx 1{xtm 6= xt−1 } − βc 1{ctm 6= ct−1 } + αxm + νx,t
Estimating α with an average of 35 partners means a large increase in the state variables. Can estimate a ‘limited memory” version where α is not observed until the match is made (α is importer-exporter-time specific). Go Back
Introduction
Data
Model
Implementation
Results
Counterfactuals
Table: Counterfactual Results (I) βx = Original βx ↓ 50% βx ↓ 100% Sample βc ↓ 50% βc = βc ↓ 100% Price Index Staying City Stay Go Back
60/ 39
Conclusion
βx ↑ 200% βc ↑ 200%
-
-12.50%
-15.20%
-7.37%
+7.62%
57% 75%
18% 43%
8% 47%
31% 46%
90% 93%