This document summarizes a study examining the relationship between knowledge sharing and individual performance using data from a Japanese bank. The data includes records of every document access and question/answer exchange for over 2,800 loan officers over two years. Descriptive statistics show variation in individual productivity both within and between groups. Transition probabilities between productivity percentiles are also sticky, indicating persistent performance differences. The study aims to identify the causal effect of knowledge sharing behaviors like asking and answering questions on individual productivity.

1. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Knowledge Sharing & Individual
Performance:
Evidence from a Japanese Bank
Marco Di Maggio and Marshall Van Alstyne
MIT Economics Dept. & Boston University and MIT
NSF Grant #0925004
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

2. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
The Motivation: Productivity
In U.S. manufacturing, the average difference in productivity
between the highest and lowest percentiles is roughly 1.92 and
the autoregressive coefficients are on the order of 0.6 to 0.8.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

3. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
The Motivation: Productivity
In U.S. manufacturing, the average difference in productivity
between the highest and lowest percentiles is roughly 1.92 and
the autoregressive coefficients are on the order of 0.6 to 0.8.
Existing research has linked productivity to a number of
factors e.g. technology, demand, human capital and market
structure.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

4. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
The Motivation: Productivity
In U.S. manufacturing, the average difference in productivity
between the highest and lowest percentiles is roughly 1.92 and
the autoregressive coefficients are on the order of 0.6 to 0.8.
Existing research has linked productivity to a number of
factors e.g. technology, demand, human capital and market
structure.
However, to create persistent performance differences the
advantageous inner workings must be difficult to imitate.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

5. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
The Motivation: Productivity
This suggests that performance variation might be due to
management practices, communication, and individual talent
which are softer and more informal than factors identified in
the literature.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

6. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
The Motivation: Productivity
This suggests that performance variation might be due to
management practices, communication, and individual talent
which are softer and more informal than factors identified in
the literature.
Main empirical challenge for economics: the absence of
high-quality data on performance and individual practices.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

7. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
The Motivation: Productivity
This suggests that performance variation might be due to
management practices, communication, and individual talent
which are softer and more informal than factors identified in
the literature.
Main empirical challenge for economics: the absence of
high-quality data on performance and individual practices.
We have very fine-grained data including every database
access and all individual productivity of bank loan officers for
two years.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

8. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
The Question
Does access to shared information increase
the productivity of knowledge workers?
Consider a knowledge management platform for shared
documents and Q&A
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

9. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Outline
1. Data
2. Existence of Persistent Performance Differences (PPDs)
3. Identification Strategy
4. Main Results
5. Robestness
6. Concluding Remarks
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

10. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
The Data
Scope: We collected terabytes of data containing every access
to every document and Q&A exchange at a major Japanese
bank.
Observations: Our data include all loan officers, roughly 2800
people, located across 290 branches. We also have individual
loan performance data, and tenure at the bank.
Multi-Level:We also have branch data. Branches vary in size
(7-110) and primary type of business, mainly due to location.
Work Context: Six main groups to which each loan officer
might belong: large existing account, small existing account,
restructuring group, public sector, new strategic account, and
new account.
Time period: October 2006 - September 2008.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

11. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Individual Output Measures
Quantitative Qualitative
bank gross profit customer service
individual loan profit contribution to branch
liquid deposit profit operations
loan volume contribution to
reduced disclosed debt organization operations
reduced estimated losses loan reinforcement
Bank headquarters set branch Branch manager evaluation of
targets, passed to loan officers. individual officers.
Loan officer performance is reviewed semi-annually.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

12. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Q&A Example
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

13. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
A Document
Corporate
Banking
Community
Library
/
View
Document
Return
to
Documents
list
/
Bookmark
[Links]
"Se?lement
products
(corporate)"
list
and
"Personnel
in
charge"
list
2006/11/14
10:50:56
(posted
by)
Corporate
se?lement
products
division,
product
development
group
[Last
update]
2010/10/18
[Body]
This
is
the
link
list
of
the
se?lement
products
(services)
pages.
Check
this
list
first
if
you
want
to
know
about
"processes",
"forms",
and
"proposal
templates"
of
the
se?lement
products!
If
you
want
to
inquire
about
those
products,
see
"personnel
in
charge"
list.
[TIPS]
• There
are
"product
links
in
alphabeVcal
order"
and
"product
links
by
category".
You
can
find
products
from
category
in
case
you
don't
know
the
exact
name.
• The
proposal
templates
are
renewed
for
the
8
key
products
and
4
new
products.
The
templates
present
soluVons
to
customers'
issues,
using
pictures
and
real
screens/forms/papers
for
easy
understanding.
• The
direct
links
to
those
proposal
are
available
at
the
top
of
the
list.
[How
to
get
there]
Portal
top
page
>
Divisions
>
Corporate
business
navigaVon
>
Products
and
Services
>
Se?lement
[Inquiry]
[NoVce]
See
"Personnel
in
charge"
list.
"Personnel
in
charge"
list
is
added.
[Update
history]
2010/10/18
.......
2010/10/1
.......
2010/9/8
.......
2010/9/7
.......
.......
A?achment
file
Se?lement
products
and
personnel
in
charge
list
20101019.xls
(429KB)
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

14. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Alternate Explanations
Table: Competing Hypotheses
Questions Answers
H1:People who ask questions get H3:People who answer questions
useful advice, reduce search costs signal their greater expertise
and are more productive. and are more productive.
H2:People who ask questions are H4:People who answer questions
inexperienced, less able and are distracted from their work
are less productive. and are less productive.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

15. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Alternate Explanations
Table: Causal Hypotheses
Questions Answers
H1:People who ask questions get H3:People who answer questions
useful advice, reduce search costs signal their greater expertise
and are more productive. and are more productive.
H2:People who ask questions are H4:People who answer questions
inexperienced, less able and are distracted from their work
are less productive. and are less productive.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

16. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Alternate Explanations
Table: Selection Hypotheses
Questions Answers
H1:People who ask questions get H3:People who answer questions
useful advice, reduce search costs signal their greater expertise
and are more productive. and are more productive.
H2:People who ask questions are H4:People who answer questions
inexperienced, less able and are distracted from their work
are less productive. and are less productive.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

17. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Summary of the Descriptive Statistics
Individual loan officers (2800)
On average: 500 (400) documents consulted every term, 78
(116) questions posted and 300 (150) answers provided.
On average: a score of 50 (22) for total performance and 31
(17) for quantitative productivity; 10 years of previous
experience in the bank.
Branches (290)
From 7 to 110 loan officers in each branch. Corr(Y , I d )=.03;
Corr(Y , I q )=−.1;Corr(Y , I a )=.3.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

18. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Facts: Individual Productivity Differences
Kernel Density Estimation
Te rm 1 Te rm 2
.03
.02
.01
Density
0
Te rm 3 Te rm 4
.03
.02
.01
0
0 50 100 0 50 100
Total Productivity Density
Kernel Density
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

19. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Facts: Heterogeneity within and between Groups
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

20. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Individual Transition Probabilities Are Sticky
Table: Transition Probabilities for Total Performance
<5% 5-25 25-50 50-75 75-95 > 95% Total
<5% 14.78 29.29 29.55 16.89 7.12 2.37 100
5-25 6.42 25.61 28.62 22.95 13.25 3.14 100
25-50 2.41 19.94 28.06 27.12 18.24 4.24 100
50-75 1.59 13.81 26.44 29.33 21.6 7.22 100
75-95 1.64 9.83 20.61 25.73 30.58 11.6 100
> 95% 1.13 4.51 17.46 21.69 35.77 19.44 100
For a person in bottom 5%, there is a ∼ 15% chance he remains
there, a ∼ 44% chance he remains in the lowest quartile, but only
a ∼ 2% chance he jumps to the top 5%.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

21. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Facts: Lower performers have fewer answers but more
questions.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

22. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Empirical Model
d q a
Pijkt = α1 Iijt + α2 Iijt + α3 Iijt + βXijt + ε ijt (1)
Pijkt : Total performance of officer i at branch j in group k at
time t.
d
Iijt : Number of documents accessed by officer i in branch j in
term t.
I q : Number of questions.
I a : Number of answers provided.
Xijt : tenure and indicator for previous experience of officer i;
time, branch, group and individual fixed effects.
We also allow for non-linear effects and interactions.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

23. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Threats to Identification
Correlation with unobserved shocks over time;
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

24. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Threats to Identification
Correlation with unobserved shocks over time;
Solution: we include Time Fixed Effects.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

25. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Threats to Identification
Correlation with unobserved shocks over time;
Solution: we include Time Fixed Effects.
Information usage is bundled with many other branch-level
management practices.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

26. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Threats to Identification
Correlation with unobserved shocks over time;
Solution: we include Time Fixed Effects.
Information usage is bundled with many other branch-level
management practices.
Solution: we include Branch Fixed Effects.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

27. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Threats to Identification
Correlation with unobserved shocks over time;
Solution: we include Time Fixed Effects.
Information usage is bundled with many other branch-level
management practices.
Solution: we include Branch Fixed Effects.
Correlation with unobserved individual characteristics.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

28. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Threats to Identification
Correlation with unobserved shocks over time;
Solution: we include Time Fixed Effects.
Information usage is bundled with many other branch-level
management practices.
Solution: we include Branch Fixed Effects.
Correlation with unobserved individual characteristics.
Solution: we shall show Individual Fixed Effects.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

29. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Threats to Identification
Correlation with unobserved shocks over time;
Solution: we include Time Fixed Effects.
Information usage is bundled with many other branch-level
management practices.
Solution: we include Branch Fixed Effects.
Correlation with unobserved individual characteristics.
Solution: we shall show Individual Fixed Effects.
Our measure of information access may be noisy.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

30. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Threats to Identification
Correlation with unobserved shocks over time;
Solution: we include Time Fixed Effects.
Information usage is bundled with many other branch-level
management practices.
Solution: we include Branch Fixed Effects.
Correlation with unobserved individual characteristics.
Solution: we shall show Individual Fixed Effects.
Our measure of information access may be noisy.
Solution: our estimates identify a lower bound.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

31. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Threats to Identification
Correlation with unobserved shocks over time;
Solution: we include Time Fixed Effects.
Information usage is bundled with many other branch-level
management practices.
Solution: we include Branch Fixed Effects.
Correlation with unobserved individual characteristics.
Solution: we shall show Individual Fixed Effects.
Our measure of information access may be noisy.
Solution: our estimates identify a lower bound.
Endogeneity
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

32. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Threats to Identification
Correlation with unobserved shocks over time;
Solution: we include Time Fixed Effects.
Information usage is bundled with many other branch-level
management practices.
Solution: we include Branch Fixed Effects.
Correlation with unobserved individual characteristics.
Solution: we shall show Individual Fixed Effects.
Our measure of information access may be noisy.
Solution: our estimates identify a lower bound.
Endogeneity
Solution: we propose an Instrumental Variable approach.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

33. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Dependent Variable: Log of Performance
Table: Panel Model I – Fixed Effects Estimates
Log(Tot. Performance) (1) (2) (3) (4) (5)
Log(Number Documents) 0.0493*** 0.0562*** 0.0226** 0.0300*** 0.0216**
(0.0103) (0.0106) (0.0104) (0.0107) (0.0109)
Log(Number Questions) -0.0505*** -0.0335*** -0.0246*** -0.0217*** -0.0195***
(0.00452) (0.00506) (0.00498) (0.00504) (0.00519)
Log(Number Answers) 0.0508*** 0.00848 0.00533 0.00337 0.00475
(0.00301) (0.0065) (0.0064) (0.00643) (0.00692)
Log(Tenure) 0.0632*** 0.0674*** 0.105*** 0.100*** 0.1000***
(0.0086) (0.00862) (0.00848) (0.00827) (0.00812)
College 0.108*** 0.113*** 0.105*** 0.0924*** 0.0891***
(0.0276) (0.0276) (0.0263) (0.0258) (0.0255)
Time Fixed Effects YES YES YES YES
Group Fixed Effects YES YES YES
Branch Fixed Effects YES YES
Time * Branch Fixed Effects YES
Observations 9,805 9,805 9,805 9,805 9,805
Number of Officers 2,916 2,916 2,916 2,916 2,916
R-squared 0.0629 0.0582 0.1567 0.3049 0.467
Tenure and college freshness both significant.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

34. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Dependent Variable: Log of Performance
Table: Panel Model I – Fixed Effects Estimates
Log(Tot. Performance) (1) (2) (3) (4) (5)
Log(Number Documents) 0.0493*** 0.0562*** 0.0226** 0.0300*** 0.0216**
(0.0103) (0.0106) (0.0104) (0.0107) (0.0109)
Log(Number Questions) -0.0505*** -0.0335*** -0.0246*** -0.0217*** -0.0195***
(0.00452) (0.00506) (0.00498) (0.00504) (0.00519)
Log(Number Answers) 0.0508*** 0.00848 0.00533 0.00337 0.00475
(0.00301) (0.0065) (0.0064) (0.00643) (0.00692)
Log(Tenure) 0.0632*** 0.0674*** 0.105*** 0.100*** 0.1000***
(0.0086) (0.00862) (0.00848) (0.00827) (0.00812)
College 0.108*** 0.113*** 0.105*** 0.0924*** 0.0891***
(0.0276) (0.0276) (0.0263) (0.0258) (0.0255)
Time Fixed Effects YES YES YES YES
Group Fixed Effects YES YES YES
Branch Fixed Effects YES YES
Time * Branch Fixed Effects YES
Observations 9,805 9,805 9,805 9,805 9,805
Number of Officers 2,916 2,916 2,916 2,916 2,916
R-squared 0.0629 0.0582 0.1567 0.3049 0.467
Tenure and college freshness both significant.
One standard deviation rise in document access predicts 11% rise in output.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

35. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Dependent Variable: Log of Performance
Table: Panel Model I – Fixed Effects Estimates
Log(Tot. Performance) (1) (2) (3) (4) (5)
Log(Number Documents) 0.0493*** 0.0562*** 0.0226** 0.0300*** 0.0216**
(0.0103) (0.0106) (0.0104) (0.0107) (0.0109)
Log(Number Questions) -0.0505*** -0.0335*** -0.0246*** -0.0217*** -0.0195***
(0.00452) (0.00506) (0.00498) (0.00504) (0.00519)
Log(Number Answers) 0.0508*** 0.00848 0.00533 0.00337 0.00475
(0.00301) (0.0065) (0.0064) (0.00643) (0.00692)
Log(Tenure) 0.0632*** 0.0674*** 0.105*** 0.100*** 0.1000***
(0.0086) (0.00862) (0.00848) (0.00827) (0.00812)
College 0.108*** 0.113*** 0.105*** 0.0924*** 0.0891***
(0.0276) (0.0276) (0.0263) (0.0258) (0.0255)
Time Fixed Effects YES YES YES YES
Group Fixed Effects YES YES YES
Branch Fixed Effects YES YES
Time * Branch Fixed Effects YES
Observations 9,805 9,805 9,805 9,805 9,805
Number of Officers 2,916 2,916 2,916 2,916 2,916
R-squared 0.0629 0.0582 0.1567 0.3049 0.467
Tenure and college freshness both significant.
One standard deviation rise in document access predicts 11% rise in output.
One standard deviation rise in questions predicts 5% fall in output.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

36. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Loan Officer Fixed Effects
Table: Panel Model II - Individual Fixed Effects
Log(Tot. Performance) (1) (2) (3) (4) (5)
Log(Number Documents) 0.0480*** 0.0430*** 0.0467*** 0.0292* 0.0374**
(0.0104) (0.0161) (0.0172) (0.0171) (0.0178)
Log(Number Questions) -0.0543*** -0.0408*** -0.0176** -0.0153** -0.0104
(0.00452) (0.00664) (0.00708) (0.00702) (0.00715)
Log(Number Answers) 0.0511*** 0.0519*** -0.00604 -0.00668 -0.0062
(0.00301) (0.00326) (0.00741) (0.00733) (0.00748)
Individual Fixed Effects YES YES YES YES
Time Fixed Effects YES YES YES
Group Fixed Effects YES YES
Branch Fixed Effects YES
Observations 9,806 9,806 9,806 9,806 9,806
R-squared 0.032 0.048 0.059 0.081 0.157
Number of questions is no longer significant.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

37. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Loan Officer Fixed Effects
Table: Panel Model II - Individual Fixed Effects
Log(Tot. Performance) (1) (2) (3) (4) (5)
Log(Number Documents) 0.0480*** 0.0430*** 0.0467*** 0.0292* 0.0374**
(0.0104) (0.0161) (0.0172) (0.0171) (0.0178)
Log(Number Questions) -0.0543*** -0.0408*** -0.0176** -0.0153** -0.0104
(0.00452) (0.00664) (0.00708) (0.00702) (0.00715)
Log(Number Answers) 0.0511*** 0.0519*** -0.00604 -0.00668 -0.0062
(0.00301) (0.00326) (0.00741) (0.00733) (0.00748)
Individual Fixed Effects YES YES YES YES
Time Fixed Effects YES YES YES
Group Fixed Effects YES YES
Branch Fixed Effects YES
Observations 9,806 9,806 9,806 9,806 9,806
R-squared 0.032 0.048 0.059 0.081 0.157
Number of questions is no longer significant.
But, effect of document access is even larger.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

38. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Quantile Regression Estimates
d q a
Quant θ (yijt |·) = α1θ Iijt + α2θ Iijt + α3θ Iijt + β θ Xijt + ε ijt (2)
All variables are as previously defined.
Bootstrapped standard errors based on 1000 replications.
Effect of document, question, and answer access on loan
officer performance at the θ th conditional quantile of log
performance is measured by vector αθ .
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

39. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Table: Quantile Regression Estimates
Log(Total Performance) 10th 25th 50th 75th 90th
Log(Number Documents) 0.0391** 0.0409*** 0.0250* 0.002 -0.012
(0.017) (0.012) (0.014) (0.013) (0.010)
Log(Number Questions) -0.0207*** -0.0251*** -0.0372*** -0.0361*** -0.0164***
(0.007) (0.008) (0.007) (0.006) (0.006)
Log(Number Answers) 0.004 0.008 0.013 0.0209** 0.006
(0.010) (0.012) (0.012) (0.009) (0.007)
Observations 9,805 9,805 9,805 9,805 9,805
Effect of document access is 0 for top two quantiles, but positive and significant
for bottom three quantiles. A 10% increase is associated with a rise in
performance of at least 20%.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

40. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Table: Quantile Regression Estimates
Log(Total Performance) 10th 25th 50th 75th 90th
Log(Number Documents) 0.0391** 0.0409*** 0.0250* 0.002 -0.012
(0.017) (0.012) (0.014) (0.013) (0.010)
Log(Number Questions) -0.0207*** -0.0251*** -0.0372*** -0.0361*** -0.0164***
(0.007) (0.008) (0.007) (0.006) (0.006)
Log(Number Answers) 0.004 0.008 0.013 0.0209** 0.006
(0.010) (0.012) (0.012) (0.009) (0.007)
Observations 9,805 9,805 9,805 9,805 9,805
Effect of document access is 0 for top two quantiles, but positive and significant
for bottom three quantiles. A 10% increase is associated with a rise in
performance of at least 20%.
Effect of questions is negative and significant across all quantiles.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

41. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Table: Quantile Regression Estimates
Log(Total Performance) 10th 25th 50th 75th 90th
Log(Number Documents) 0.0391** 0.0409*** 0.0250* 0.002 -0.012
(0.017) (0.012) (0.014) (0.013) (0.010)
Log(Number Questions) -0.0207*** -0.0251*** -0.0372*** -0.0361*** -0.0164***
(0.007) (0.008) (0.007) (0.006) (0.006)
Log(Number Answers) 0.004 0.008 0.013 0.0209** 0.006
(0.010) (0.012) (0.012) (0.009) (0.007)
Observations 9,805 9,805 9,805 9,805 9,805
Effect of document access is 0 for top two quantiles, but positive and significant
for bottom three quantiles. A 10% increase is associated with a rise in
performance of at least 20%.
Effect of questions is negative and significant across all quantiles.
Information access appears to help loan officers in the left tail of the distribution
but is not significant for officers in the right tail.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

42. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Dependent Variable: Qualitative Measure
Table: Panel Model III - Qualitative Performance Measure
Log(Qualitative Performance) (1) (2) (3) (4) (5)
Log(Number Documents) 0.0587*** 0.0616*** 0.0416*** 0.0382*** 0.0459***
(0.00872) (0.00866) (0.00866) (0.00894) (0.0157)
Log(Number Questions) -0.0812*** -0.0251*** -0.0171*** -0.0188*** -0.00715
(0.00383) (0.00421) (0.00419) (0.00427) (0.0063)
Log(Number Answers) 0.139*** 0.00157 -0.001 4.76E-05 -0.00969
(0.00271) (0.00554) (0.00551) (0.00554) (0.00659)
Log(Tenure) 0.0648*** 0.0791*** 0.100*** 0.104***
(0.0069) (0.00683) (0.00687) (0.00676)
College 0.0654*** 0.0752*** 0.0747*** 0.0641***
(0.0224) (0.022) (0.0214) (0.0212)
Time Fixed Effects YES YES YES YES
Group Fixed Effects YES YES YES
Branch Fixed Effects YES YES
Individual Fixed Effects YES
Observations 9,801 9,801 9,801 9,801 9,801
R-squared 0.3023 0.3692 0.3716 0.3965 0.421
Restricting the model to qualitative estimates of performance leaves the key results
unchanged.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

43. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Promotion & Advancement
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

44. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Promotion
Table: Promotion Probability
Promotion (Probit) (1) (2) (3) (4)
Number Documents 3.63E-06 3.40E-06 4.61E-06 9.43E-05
(3.77E-06) (4.00E-06) (3.56E-06) (0.00021)
Number Questions -0.00172** -0.00174** -0.00154** -0.00245**
(0.000719) (0.000718) (0.00069) (0.00099)
Number Answers 0.00106*** 0.00107*** 0.00106*** 0.00153***
(2.88E-04) (2.87E-04) (0.00029) (0.00041)
Lag Productivity 0.00194 0.00177 0.00152 0.00146
(0.00205) (0.00203) (0.00207) (0.00252)
Tenure 0.107*** 0.107*** 0.108*** 0.148***
(0.00963) (0.00967) (0.00965) (0.0164)
College -0.227 -0.229 -0.22 -0.294
(0.254) (0.254) (0.258) (0.267)
Time Fixed Effects YES YES YES
Group Fixed Effects YES YES
Branch Fixed Effects YES
Observations 6,971 6,971 6,971 6,971
We observe just over 200 promotions. Tenure is by far the largest predictor.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

45. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Promotion
Table: Promotion Probability
Promotion (Probit) (1) (2) (3) (4)
Number Documents 3.63E-06 3.40E-06 4.61E-06 9.43E-05
(3.77E-06) (4.00E-06) (3.56E-06) (0.00021)
Number Questions -0.00172** -0.00174** -0.00154** -0.00245**
(0.000719) (0.000718) (0.00069) (0.00099)
Number Answers 0.00106*** 0.00107*** 0.00106*** 0.00153***
(2.88E-04) (2.87E-04) (0.00029) (0.00041)
Lag Productivity 0.00194 0.00177 0.00152 0.00146
(0.00205) (0.00203) (0.00207) (0.00252)
Tenure 0.107*** 0.107*** 0.108*** 0.148***
(0.00963) (0.00967) (0.00965) (0.0164)
College -0.227 -0.229 -0.22 -0.294
(0.254) (0.254) (0.258) (0.267)
Time Fixed Effects YES YES YES
Group Fixed Effects YES YES
Branch Fixed Effects YES
Observations 6,971 6,971 6,971 6,971
We observe just over 200 promotions. Tenure is by far the largest predictor.
Lagged productivity is not significant.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

46. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Promotion
Table: Promotion Probability
Promotion (Probit) (1) (2) (3) (4)
Number Documents 3.63E-06 3.40E-06 4.61E-06 9.43E-05
(3.77E-06) (4.00E-06) (3.56E-06) (0.00021)
Number Questions -0.00172** -0.00174** -0.00154** -0.00245**
(0.000719) (0.000718) (0.00069) (0.00099)
Number Answers 0.00106*** 0.00107*** 0.00106*** 0.00153***
(2.88E-04) (2.87E-04) (0.00029) (0.00041)
Lag Productivity 0.00194 0.00177 0.00152 0.00146
(0.00205) (0.00203) (0.00207) (0.00252)
Tenure 0.107*** 0.107*** 0.108*** 0.148***
(0.00963) (0.00967) (0.00965) (0.0164)
College -0.227 -0.229 -0.22 -0.294
(0.254) (0.254) (0.258) (0.267)
Time Fixed Effects YES YES YES
Group Fixed Effects YES YES
Branch Fixed Effects YES
Observations 6,971 6,971 6,971 6,971
We observe just over 200 promotions. Tenure is by far the largest predictor.
Lagged productivity is not significant.
Loan officers might be signaling via answers in the data.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

47. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Summary of Panel Results
A standard deviation increase in information access is
associated with an increase in total and qualitative
performance of 6–10%.
A standard deviation increase in the number of questions is
associated with a reduction in performance of 4%, while an
increase in answers predicts a performance rise of 2%.
The probability of being promoted is positively correlated with
information production and negatively correlated with
information gathering.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

48. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Identification Strategy: Exogenous Variation
We observe 650 loan officers switching branch. This is an
involuntary transfer due to a Japanese term limit law to
prevent corruption.
This allows us to disentangle the individual fixed effects from
the branch fixed effects.
The identification assumption: the rotation and the allocation
to a new branch are not correlated with individual productivity.
”Job rotation is quite regular thing for RMs, just like the solar
system rotate around the sun.” It is not performance based.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

49. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Causal Model: The Term Limit
We can use the average information access at the previous
branch (excluding officer i) at t − 1 as an instrument for
information access of officer i at time t.
Motivation: branches differ in their cultural inclinations to
use the knowledge platform. Branch 1 instrument correlated
with information consumption and productivity at branch 1
but not branch 2.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

50. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Causal Model: The Term Limit
We can use the average information access at the previous
branch (excluding officer i) at t − 1 as an instrument for
information access of officer i at time t.
Motivation: branches differ in their cultural inclinations to
use the knowledge platform. Branch 1 instrument correlated
with information consumption and productivity at branch 1
but not branch 2.
It is invalid if there exists correlation between unobserved
ability and original assignment to branches.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

51. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Causal Model: The Term Limit
We can use the average information access at the previous
branch (excluding officer i) at t − 1 as an instrument for
information access of officer i at time t.
Motivation: branches differ in their cultural inclinations to
use the knowledge platform. Branch 1 instrument correlated
with information consumption and productivity at branch 1
but not branch 2.
It is invalid if there exists correlation between unobserved
ability and original assignment to branches.
It is invalid if the effect of the loan officer on the branch
practices is too strong.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

52. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Causal Model: The Term Limit
We can use the average information access at the previous
branch (excluding officer i) at t − 1 as an instrument for
information access of officer i at time t.
Motivation: branches differ in their cultural inclinations to
use the knowledge platform. Branch 1 instrument correlated
with information consumption and productivity at branch 1
but not branch 2.
It is invalid if there exists correlation between unobserved
ability and original assignment to branches.
It is invalid if the effect of the loan officer on the branch
practices is too strong.
It is invalid if they learned some other practices from the
previous branch.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

53. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Job Rotation
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

54. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Switching Increases Demand for Information
Table: Effect of Switching on Document Access
Log(Number Documents) (1) (2) (3) (4)
Switch 0.0199** 0.0218**
(0.009) (0.009)
After Switch 0.0125 0.0150*
(0.008) (0.008)
Log(Tenure) -0.0218 -0.0217
(0.014) (0.014)
College 0.278*** 0.279***
(0.042) (0.042)
Time Fixed Effects YES YES YES YES
Group Fixed Effects YES YES YES YES
Branch Fixed Effects YES YES YES YES
Individual Fixed Effects YES YES
R-squared 0.3342 0.347 0.334 0.346
Observations 10,055 10,055 10,055 10,055
Increase in demand for documents on changing jobs is consistent with a
learning hypothesis that diminishes over time.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

55. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Table: Effect of Switching on Performance
Log(Total Performance) (1) (2) (3) (4) (5) (6) (7)
Switch -0.111*** -0.0317 -0.118*** -0.0461* -0.108***
(0.012) (0.027) (0.012) (0.027) (0.013)
Switch*Tenure -0.00778*** -0.00726***
(0.002) (0.002)
After Switch -0.0889*** -0.0915***
(0.010) (0.012)
Log(Number Documents) 0.0470*** 0.0469*** 0.0287*** 0.0283*** 0.0274** 0.0387** 0.0348**
(0.010) (0.010) (0.010) (0.010) (0.010) (0.018) (0.018)
Log(Number Questions) -0.0515*** -0.0516*** -0.0210*** -0.0210*** -0.0205*** -0.00851 -0.0078
(0.005) (0.005) (0.005) (0.005) (0.005) (0.007) (0.007)
Log(Number Answers) 0.0569*** 0.0571*** 0.0059 0.00618 0.00476 -0.00394 -0.0052
(0.003) (0.003) (0.006) (0.006) (0.006) (0.007) (0.007)
Log(Tenure) 0.0636*** 0.0700*** 0.0994*** 0.105*** 0.0994***
(0.009) (0.009) (0.008) (0.008) (0.008)
College 0.106*** 0.104*** 0.0915*** 0.0907*** 0.0897***
(0.027) (0.028) (0.026) (0.026) (0.026)
Time Fixed Effects YES YES YES YES YES
Group Fixed Effects YES YES YES YES YES
Branch Fixed Effects YES YES YES YES YES
Individual Fixed Effects YES YES
Observations 9,805 9,805 9,805 9,805 9,805 9,805 9,805
R-squared 0.0698 0.0701 0.3078 0.308 0.3054 0.166 0.165
Switching jobs appears to significantly reduce performance.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

56. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Instrument: Variation in branch attitudes toward information access
Table: First Stages
Documents Answers Questions
Documents Prev Branch 7.398*** -0.654* -0.214
(0.744) (0.362) (0.248)
Answers Prev Branch -14.439*** 4.190*** -1.637**
(2.282) (1.112) (0.757)
Questions Prev Branch 8.420*** 2.271** 7.437***
(2.157) (1.052) (0.718)
Tenure -10.419*** -4.132*** -2.688***
(2.622) (1.278) (0.873)
College 54.021 19.242 -2.409
(69.580) (33.915) (23.161)
F-Test 50.289 16.774 49.343
Time Fixed effects YES YES YES
Group Fixed Effects YES YES YES
Observations 618 618 618
Document IV specification F-test >> 20 ⇒ instrument is not weak.
Tenure reduces demand for information.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

57. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
IV: Variation in branch attitudes
Table: IV Estimates
(1) (2) (3) (4) (5) (6)
Total Performance OLS IV OLS IV OLS IV
Number Documents -0.00123 0.0197** -0.00138 0.0225*** -0.00347 0.0194**
(0.003) (0.008) (0.003) (0.008) (0.003) (0.008)
Number Questions -0.0306*** -0.133** -0.0287*** -0.104*** -0.0169** -0.0908***
(0.008) (0.061) (0.008) (0.034) (0.008) (0.033)
Number Answers 0.0175*** 0.112 0.0148** 0.0727** 0.0114* 0.0674**
(0.007) (0.080) (0.007) (0.036) (0.007) (0.034)
Tenure 0.387** 0.871** 0.385** 0.689*** 0.603*** 0.819***
(0.172) (0.382) (0.174) (0.242) (0.183) (0.223)
College -3.736 -7.538 -3.714 -6.748 -3.414 -6.015
(4.809) (6.436) (4.806) (5.74) (4.839) (5.544)
Time Fixed Effects YES YES YES YES
Group Fixed Effects YES YES
Observations 618 618 618 618 618 618
OLS estimates of document access n.s. but IV estimates are significant.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

58. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
IV: Variation in branch attitudes
Table: IV Estimates
(1) (2) (3) (4) (5) (6)
Total Performance OLS IV OLS IV OLS IV
Number Documents -0.00123 0.0197** -0.00138 0.0225*** -0.00347 0.0194**
(0.003) (0.008) (0.003) (0.008) (0.003) (0.008)
Number Questions -0.0306*** -0.133** -0.0287*** -0.104*** -0.0169** -0.0908***
(0.008) (0.061) (0.008) (0.034) (0.008) (0.033)
Number Answers 0.0175*** 0.112 0.0148** 0.0727** 0.0114* 0.0674**
(0.007) (0.080) (0.007) (0.036) (0.007) (0.034)
Tenure 0.387** 0.871** 0.385** 0.689*** 0.603*** 0.819***
(0.172) (0.382) (0.174) (0.242) (0.183) (0.223)
College -3.736 -7.538 -3.714 -6.748 -3.414 -6.015
(4.809) (6.436) (4.806) (5.74) (4.839) (5.544)
Time Fixed Effects YES YES YES YES
Group Fixed Effects YES YES
Observations 618 618 618 618 618 618
OLS estimates of document access n.s. but IV estimates are significant.
Most conservative Model (6) implies causal performance effect of ≥ 5%.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

59. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Large Branches
Table: IV Estimates: Large Branches (> 50)
(1) (2) (3) (4) (5) (6)
Total Performance OLS IV OLS IV OLS IV
Number Documents 0.007 0.026 0.006 0.0129* 0.004 0.0156**
(0.005) (0.042) (0.005) (0.007) (0.005) (0.007)
Number Questions -0.0455*** 0.224 -0.0384*** -0.0422** -0.0199** -0.0377*
(0.009) (0.683) (0.009) (0.019) (0.010) (0.021)
Number Answers 0.0316*** -0.387 0.019 0.020 0.011 0.0392*
(0.011) (1.049) (0.013) (0.025) (0.013) (0.021)
Tenure 0.243 -0.039 0.211 0.269 0.603** 0.695**
(0.272) (1.100) (0.278) (0.269) (0.282) (0.286)
College -25.25*** -13.650 -22.65*** -24.11*** -19.94*** -23.44***
(7.129) (33.150) (6.421) (6.669) (7.420) (8.608)
Time Fixed Effects YES YES YES YES
Group Fixed Effects YES YES
Observations 240 240 240 240 240 240
A standard deviation increase in
Documents ⇒ increases output by 10%;
Questions ⇒ decreases output by 21%;
Answers ⇒ increases output by 20%.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

60. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Interpretation
There exists substitution between individual ability and access
to others’ knowledge.
Local Average Treatment Effect: loan officers very susceptible
to the branch environment.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

61. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Results
1. We document the existence of PPDs within and across branches of
a major Japanese bank. Shared information might explain a
significant portion of this heterogeneity.
2. Access to information appears to substitute for officers’ ability. Low
skilled workers benefit the most from access to others’ information.
3. The likelihood of being promoted is negatively correlated with the
number of questions asked (screening effect) and positively
correlated with the number of answers provided (signaling effect).
4. Anti-corruption law job rotation reduces loan officers’ performance,
explained by destruction of specialized human capital.
5. Controlling for unobserved heterogeneity over time, branches, and
officers, a standard deviation increase in information access
increases performance by ≈ 10%. Output falls in questions and rises
in answers. Findings are robust to exogenous cultural variation.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

62. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Thank You
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

63. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Appendices: Further Results
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

64. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Officer Assignment is Weakly Correlated with Branch
Performance
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

65. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Relative Measure of Information Access
We constructed the following relative measures
rankij − 1
scoreij = 100 − × 100
number of officersj − 1
For information access, gathering and production, where
higher in the rank means more information usage.
Knowledge and Performance MIT Economics Dept. & Boston University and MIT

66. The Introduction The Data Existence of PPDs Identification Strategy IV IV Conclusion
Relative Measure
Table: Panel Model IV - Relative Information Access
Total Performance (1) (2) (3) (4)
Information Access 0.0158 0.0376***
(0.011) (0.0125)
Information Gathering -0.0264** -0.0446
(0.0108) (0.0275)
Information Production -0.0246** 0.000149
(0.0113) (0.0284)
Tenure 0.483*** 0.474*** 0.475*** 0.460***
(0.0514) (0.0515) (0.0516) (0.0517)
College 4.635*** 4.840*** 4.827*** 4.788***
(1.261) (1.26) (1.26) (1.26)
Time Fixed Effects YES YES YES YES
Group Fixed Effects YES YES YES YES
Branch Fixed Effects YES YES YES YES
Time * Branch F. E. YES YES YES YES
Observations 10,045 10,045 10,045 10,045
Knowledge and Performance MIT Economics Dept. & Boston University and MIT