Discussion of The Gravity of Experience by Dutt, Santacreu and Traca - PowerPoint PPT Presentation

About This Presentation
Title:

Discussion of The Gravity of Experience by Dutt, Santacreu and Traca

Description:

First part a model-driven interpretation, so relevant part is: ... Discussion of The Gravity of Experience by Dutt, Santacreu and Traca Author: Microsoft account – PowerPoint PPT presentation

Number of Views:76
Avg rating:3.0/5.0
Slides: 15
Provided by: Micros479
Category:

less

Transcript and Presenter's Notes

Title: Discussion of The Gravity of Experience by Dutt, Santacreu and Traca


1
Discussion ofThe Gravity of Experience byDutt,
Santacreu and Traca
  • Andrew K. Rose
  • ABFER, CEPR, NBER and Berkeley-Haas

2
Ludicrously Provocative
  • Essential Result Empirical highlighted in
    abstract
  • an additional year of experience at the
    country-pair level reduces trade costs by 2.0
    and increases bilateral exports by 8.
  • First part a model-driven interpretation, so
    relevant part is
  • an additional year of experience increases
    bilateral exports by 8.
  • Who can believe that each additional year of
    interaction raises international commerce by such
    an enormous percent?
  • Accordingly, this discussion is a search and
    destroy mission

3
First Line of Attack Empirical Setup
  • Result depend on some implausible empirics?
  • Problematic because plain-vanilla setup
  • Dyadic (country-pair) exports modelled in
    standard gravity model
  • Examine exports (also extensive and intensive
    margins for trade)
  • Conventional regressors
  • Country x year fixed effects for both exports and
    importers included
  • Dyadic fixed effects added for sensitivity
    analysis (!)
  • Large number of observations implies many
    countries, years
  • Two different data sets
  • All apparent from (self-contained) Table 1

4
Second Wave Functional Form
  • Experience in paper measured as number of years
    for which the country-pair had strictly positive
    trade
  • Seems crude!
  • Aggregate do Paramounts exports benefit from
    Universals foreign sales?
  • More importantly hard to believe functional form
    is linear in years
  • Surely years diminish in importance
  • Does Canadas 150th year of exporting to US
    matter as much as South Somalias first?
  • Easy to test in principle, given the data
  • Thanks to Pushan for quickly providing data and
    sample program
  • (Wouldnt be necessary to request if adhere to
    replication standard)

5
What does the Data Set Really Reveal?
  • Stick to Dutt et al setup, using their data set
  • Standard export gravity model, country x year
    dummies
  • Just mess with measure of experience

6
Sensitivity Analysis
  • (Almost) replicate their results (admittedly
    their data, program)

Coefficient (robust std error)
Default (log exports) .060 ( .001)
Regressand Extensive Margin .030 (.001)
Regressand Intensive Margin .030 (.001)
Experience Log time .869 (.018)
Experience decays (slowly) .070 (.001)
Log experience decays (slowly) .887 (.018)
7
Splines for Functional Form Robustness
Coefficient (robust std error)
Experience 1-2 years .056 (.043)
Experience 3-5 years .168 (.048)
Experience 6-10 years .592 (.053)
Experience 11-20 years 1.12 (.057)
Experience 21-30 years 1.66 (.062)
Experience gt30 years 2.46 (.067)
8
More Checks
Coefficient (robust std error)
Add Lag Dependent Variable .0143 (.0004) ? lr effect of .051
Data before 2000 .064 (.001)
Data after 1995 .058 (.001)
Data after 2000 .057 (.001)
9
Sending in the Reserves The Footnote
  • Footnote 14 (p9), highlights added
  • We restrict our sample to 1988-2006 since we
    decompose total trade into an extensive and
    intensive margin based on COMTRADE HS-6 data.
  • Aha!
  • Cant test importance of sample restriction with
    Dutt et al data set
  • But possible with an independently-constructed
    but similar data set (ongoing work with Reuven
    Glick, FRBSF)
  • Slightly different gravity model
  • Much larger span of time (back to 1948, forward
    to 2013)

10
How Sensitive are Results to Sample?
Coefficient (robust std error)
Dutt et al default .060 (.001)
My replication, Dutt et al sample .071 (.001)
My replication, larger sample .078 (.001)
Before 1960 .195 (.007)
Before 1980 .139 (.002)
Before 2000 .094 (.001)
After 1960 .078 (.001)
After 1980 .071 (.001)
After 2000 .063 (.001)
11
A Flanking Maneuver
  • Can test impact of experience without any
    assumptions about functional form
  • Just add experience dummies year by year
    (possible with many df)
  • Can do this with my data set
  • Then check for disintegration with larger sample

12
(No Transcript)
13
(No Transcript)
14
Surrender!
  • This is a very powerful and strong result!
  • Stunning impact and importance.
  • The way forward more verification
  • Others (make data accessible)
  • Micro data
  • A tip of the hat!
Write a Comment
User Comments (0)
About PowerShow.com