Presentation - Freeman Spogli Institute for International Studies

By a Silken Thread
regional banking integration & pathways to financial development in Japan’s Great
Recession
Mathias Hoffmann
Toshihiro Okubo
University of Zurich, URPP FinReg, CESifo & CAMA
Keio University
Stanford APARC, 1 Dec 2015
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
1 / 25
Background
What is the role of banking integration in a financial crisis?
Theoretically, integration
I
I
... increases an economy’s exposure to foreign bank liquidity
shocks
... insulates from idiosyncratic shocks to domestic banking
system
How do historical factors determine the trade-off between
integration and segmentation of banking markets?
We look at the regional spread of Japan’s Great Recession after
1990, exploiting prefecture-level variation in banking integration
and local bank dependence
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
2 / 25
Preview of results
Cross-regional variation in financial (banking) integration was a
key determinant of the transmission of the crisis post-1990
I
I
more integrated prefectures more exposed to the property price
downturn in the major cities
But: FI was good for areas with many small manufacturing
firms (SME ):
F
F
internal capital markets mattered: nationwide banks withdrew
less strongly from areas with many small manufacturing firms
We suggest persistent banking relationships as a driver of de
facto segementation in regional banking markets.
The Silken thread: de facto segmentation in Japan’s banking
market can be traced back in history: comparative advantage in
silk in the late 19th-century led to a regionally-tiered banking
system with close SME -bank relationships.
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
4 / 25
The silken thread
Figure: City and Regional Bank Lending Shares (pre-1990 (1980-1990) averages) vs. number of silk filatures per
head in 1895
0.9
0.45
Tokyo
Yamanashi
0.85
0.4
0.8
0.35
lending share of Shinkins pre−1990
lending share of City Banks pre−1990
Osaka
0.75
0.7
Nara Fukuoka
0.65
Kanagawa
Saitama
Miyagi
Kagawa
0.6
0.5
Hokkaido
Kumamoto
Chiba
Tochigi
Tokushima
Aomori
Hiroshima
Fukui
Hyogo
Yamaguchi
Akita
Okayama
Shizuoka
Mie
Ehime Niigata
Shiga
Oita
Wakayama
Miyazaki
Fukushima
Kagoshima
Iwate
Kochi
0.45
0.4
1.5
Aichi
Ishikawa
Nagasaki
0.55
2
2.5
Kyoto
3
3.5
4
Toyama
Ibaraki
Gunma
Kyoto
Gifu
0.3
Gunma
0.25
0.2
Hiroshima
4.5
Gifu
Nagano
Yamanashi
5
5.5
6
6.5
log # silk filatures per capita in 1895
Aichi
Ishikawa
Toyama
Shimane
Yamaguchi
Miyazaki
Kumamoto
Saga
Yamagata
Ibaraki
Saitama Fukui
Wakayama
Kanagawa
Oita
Fukushima
Niigata
Kagoshima
Okayama
Shiga
Chiba
Aomori
0.15
Tottori
Shimane
Nagano
Shizuoka
Hyogo
Hokkaido
0.1
Tokyo
Nagasaki
0.05
1.5
2
Mie
Osaka
Tottori
Tochigi
Iwate
Nara
Kagawa
Kochi
Ehime
Miyagi
Fukuoka
Saga
Yamagata
Akita
Tokushima
2.5
3
3.5
4
4.5
5
5.5
6
6.5
log # silk filatures per capita in 1895
The silken thread: Regional financial integration in 1990 low in areas in with
a high share of silk-exporting (reeling) firms in 1895
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
5 / 25
Policy implications
Relevant in the context of the Eurozone today
I
Was there too little or too much financial integration in Europe
during the banking & sovereign debt crisis?
F
F
I
’optimal’ degree of banking integration of an economy depends
on the structure of loan demand by local firms: regions with
many SMEs that depend on the local provision of bank loans
have more to gain from banking integration.
local finance may be a poor substitute for finance from
integrated banks in a crisis.
A cautionary lesson: differences in financial integration can
persist in a de iure integrated banking market. For a VERY
long time ....
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
6 / 25
Our story (I): intra-national barriers to capital flows
Remark
Japan is a centralized country, no major regional differences in banking or
financial regulation etc.
Broadly similar levels of financial development (e.g. in terms of credit over
GDP, bank branches p.c and area)
But: a regionally strongly tiered banking system
’city’ banks & 1st tier regional banks → operate nationwide or at least
in severl prefecture
2nd tier regional banks (Sogo (mutual)), industrial cooperative banks
(Shinkin) → regional lenders to SMEs, regional deposit base, limited or
no access to Interbank market
Exploit variation in (pre-1990) share of nationwide banks in
prefecture-level lending as measure of financial (banking)
integration
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
7 / 25
Figure: Geographical distribution of Pre-1990 SME importance and financial
integration and post-1990 p.c. GDP growth rates
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
8 / 25
Theoretical considerations
Figure: A stylized interregional banking model
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
9 / 25
Empirical framework.
Main specification:
i
h
∆gdptk = AggShockt × α0 FI k × SME k + α1 FI k + α2 SME k + ...
+µk + τt + kt
(1)
where
AggShockt = Post1991t
and our theory predicts α0 > 0.
Note: this does not imply that FI is unambigously good. E.g., if
integrated banks are hit harder than local banks overall, then α1 < 0
and the marginal effect α1 + α0 SME k can be negative.
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
10 / 25
A first look: high/low FI sample split
Table: Small business importance, financial integration and the Great Recession
Small manufacturing firms and the effect of the Great Recession on prefecture-level output growth rates
Panel A: Based on value added SME-measure
All
prefectures
k
Post1991t × SMEVA
Regional Banks
high
low
Sample split by importance of ...
City Banks
Regional Banks: Shinkins only
high
low
high
low
-0.07
(-2.04)
-0.13
(-4.04)
-0.01
(-0.19)
-0.0140
(-0.25)
-0.12
(-3.82)
-0.12
(-3.03)
-0.03
(-0.70)
0.55
0.56
0.58
0.6042
0.53
0.57
0.552
R2
Panel B: Based on employment based SME-measure
k
Post1991t × SMEEMP
All prefs.
high
low
high
low
high
low
-0.08
(-1.96)
-0.15
(-3.73)
0.01
(0.01)
-0.002
(-0.02)
-0.15
(-3.78)
-0.15
(-4.06)
-0.04
(-0.64)
R2
0.55
0.55
0.58
0.60
0.53
0.55
0.5866
The table shows the coefficient α in panel regressions of the form ∆gdptk = α × Post1991t × SME k + µk + τt +
kt + constant where Post1991t is a dummy indicating the period from 1991, SME k is small-business importance
andµk and τt are prefecture-fixed and time effects respectively. Sample period is 1980-2005. Cooperative banks
include Shinkin banks and industrial credit cooperatives. OLS estimates, t-statistics in parentheses. Standard errors
are clustered by prefecture.
Example
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
11 / 25
Figure: Cumulative Growth Differential between high and low SME group
in two-way sample split (High/Low City Bank Share and High/Low SME
share by value added )
0
−0.01
−0.02
−0.03
−0.04
−0.05
−0.06
−0.07
−0.08
−0.09
1992
1994
1996
1998
2000
2002
2004
Relative cumulative output loss of credit-dependent prefectures worse
with low financial integration (red, dashed line).
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
12 / 25
Baseline specification: interaction term regs
Table: Baseline results — interaction terms and robustness
I
II
III
IV
V
VI
FI =
VII
High Regional
High City
Regional
City
Regional
City
Regional
FI =
Interactions of Post1990t
with ...
Controls:
Xk :
FI =
VIII
IX
X
FI =
High City
(dummy)
(dummy)
(share)
(share)
(dummy)
(share)
(share)
...SME k × FI k
-0.09
(-2.15)
0.08
(1.93)
-1.42
(-3.24)
0.74
(3.78)
0.09
(-2.15)
0.51
(1.75)
-1.07
(-2.14)
0.012
(2.53)
0.009
(2.24)
0.007
(1.73)
...FI k
0.01
(1.93)
-0.01
(-1.97)
0.24
(3.87)
-0.13
(-5.03)
0.00
(-0.49)
-0.01
(-0.65)
0.01
(0.21)
-0.006
(-1.96)
-0.005
(-1.66)
-0.004
(-1.35)
k
...SMEVA
-0.03
(-1.12)
-0.12
(-3.84)
0.32
(2.72)
-0.48
(-4.06)
-0.08
(-3.64)
-0.08
(-3.51)
-0.08
(-3.80)
-0.013
(-3.97)
-0.013
(-3.97)
-0.012
(-3.56)
...CoreArea
-0.01
(-3.25)
-0.01
(-3.30)
-0.01
(-4.00)
-0.008
(-2.63)
-0.01
(-3.25)
-0.01
(-2.46)
-0.01
(-3.37)
-0.012
(-3.05)
-0.008
(-3.33)
-0.009
(-2.75)
0.55
0.55
0.57
0.57
0.55
0.56
0.56
0.56
0.55
R2
(dummy)
Prefectures
Tokio dropped
All
Tokio dropped
All
Tokio dropped
potential outliers dropped
Remarks
SME, FI demeaned
SME dummy
h
i
k
k
The Table shows results from the regression ∆gdptk = Post1990t × α0 SMEVA
× FI k + α1 FI k + α2 SMEVA
+ α03 Xtk + µk + τt + kt where Post1990t
k
is a dummy indicating the period after 1990 (1991-2005), SMEVA
is small-business importance based on value added, FI k is the measure of financial
integration (regional or city bank share in total lending in prefecture k), as indicated in the column heading. µk and τt are prefecture-fixed and time
k
effects respectively. The vector X captures prefecture characteristics. In the regressions it is interacted with our crisis dummy Post1990t and contains
CoreAreak , a dummy for the core economic areas (Tokyo, Osaka, Aichi, Kanagawa, Chiba, Saitama, Hyogo and Kyoto prefectures). The sample period
is 1980-2005. OLS estimates, t-statistics in parentheses. Standard errors are clustered by prefecture.
In the regressions in column X , we identify a prefecture as a potential outlier if SME or FI are more than 1.64 standard deviations away from the
cross-prefectural mean of the respective variable. This leads us to exclude the following six prefectures: Saitama, Tokio, Gifu, Shiga, Osaka, Nagasaki .
Robustness
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
13 / 25
Geographical profile
Figure: Geographical profile of the interaction between bank dependence
and financial integration
0.01
0.005
0
−0.005
30
31
32
33
34
35
36
37
38
39
40
Kagoshima
29
Miyazaki
28
Oita
Fukuoka
27
Kumamoto
Kochi
26
Nagasaki
Ehime
25
By a Silken Thread
Saga
Kagawa
23
24
Prefecture
Tokushima
22
Yamaguchi
21
Hiroshima
20
Okayama
19
Shimane
18
Tottori
17
Wakayama
16
Nara
15
Hyogo
14
Osaka
13
Kyoto
Shizuoka
12
Shiga
Gifu
11
Mie
Nagano
10
()
Aichi
Yamanashi
9
Fukui
Tochigi
8
Ishikawa
Ibaraki
7
Toyama
Fukushima
6
Niigata
Yamagata
5
Kanagawa
Akita
4
Tokyo
Miyagi
3
Chiba
Iwate
2
Hoffmann & Okubo
Saitama
Aomori
1
Gunma
Hokkaido
−0.01
41
42
43
44
45
46
Stanford APARC, 1 Dec 2015
14 / 25
Our story (I): financial integration and the spread of the
Great Recession
Post-1990 growth lower in prefectures with
I
I
I
many small (credit-dependent) firms (α2 < 0)
high levels of financial integration (α1 < 0).
But: SME and FI interact —negative effect of FI (or SME ) is
mitigated in high SME -regions (α0 > 0)
Interpretation in line with model
I
I
I
FI increase vulnerability to aggregate shocks but can also
provide insulation agains local shocks.
SMEs depend on local access to finance and therefore are
particularly exposed to local bank shocks.
Conditioning on the size of the shock to city and local banks, FI
attenuates the effect in high SME prefectures.
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
15 / 25
Transmission channel
Theoretical model assumes that markets for SME lending are
segmented: SMEs borrowing from local banks face higher
interest rates
I
In integrated markets, SMEs could switch to city banks and
total lending (and GDP) growth should be independent of
pre-crisis city bank lending share – contrary to our finding
We argue that it is the strong relationships between local banks
and local SMEs that segment the market. Regional banks have
long-standing informational advantage w.r.t. their customer base
of small firms. But during a crisis these relationships may be a
fetter for SME’s:
I
I
Nationwide banks may be unwilling to lend to unknown, risky
customers
Regional banks with their locally concentrated portfolio lend to
SME’s only at less favorable terms.
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
16 / 25
Transmission channel: evidence I
City banks keep on lending in areas where they have strong ties to SME
Table: Prefecture-level lending after 1990
Lending growth
total
City Banks
Local Banks
total
City Banks
Local Banks
total
City Banks
Local Banks
I
II
III
IV
V
VI
VII
VIII
IX
Interactions of Post1990t with pre-1991 variables
k
...SMEEMP
× FI l
FI k
SME k
...CoreArea
R2
FI=CityBankShare
Tokyo&Osaka excluded
AggShockt =∆log(LandPricet )
FI = CityBankShare
Tokyo&Osaka excluded
FI =CityBankShare
0.67
(2.67)
-0.18
(-3.91)
-0.40
(-2.65)
-0.02
(-4.06)
1.46
(2.61)
-0.37
(-4.60)
-0.88
(-2.47)
-0.02
(-3.06)
-0.91
(-1.27)
-0.12
(-2.21)
0.11
(1.00)
0.01
(1.04)
0.61
0.80
0.73
Memorandum item:
0.25
(0.61)
-0.05
(-2.39)
-0.04
(-1.16)
-0.02
(-3.76)
1.80
(1.90)
-0.14
(-2.56)
-0.10
(-1.36)
-0.02
(-3.20)
-1.01
(-0.78)
-0.10
(-1.44)
0.12
(1.04)
0.01
(1.21)
0.60
0.81
0.73
-1.77
(-1.44)
0.25
(3.70)
0.40
(4.34)
0.07
(8.18)
-3.00
(-1.76)
0.20
(1.98)
0.36
(2.78)
0.07
(6.80)
0.65
0.81
k
Fraction of SME with City Bank as main bank2002 = 0.6230 × CityBankShare1980−90
− 0.25
0.78
(0.57)
0.08
(1.08)
0.38
(3.65)
0.05
(3.69)
0.74
R 2 = 0.49
(tstat=6.49)
h
i
k
k
The Table shows results from the regression ∆log(Xtk ) = Post1990t × α0 SMEEMP
× FI k + α1 FI k + α2 SMEEMP
+ α03 Xtk + µk + τt + kt where Xtk stands in turn for total
lending (columns I, IV and VII), city bank lending (columns II, V and VIII) and city bank lending relative to regional bank lending (columns III, VI and IX ) in prefecture k.
Post1990t is a dummy indicating the period after 1990 (i.e. 1991-2005), SME k is our measure of bank dependence (small-business importance) , FI k is a measure of financial
integration, the pre-1991 (1980-90) average city bank share in total lending in prefecture k. In the third panel (columns VII − IX ), the aggreagte shock is given by the land
price decline in the core prefectures from ?.µk and τt are prefecture-fixed and time effects respectively. CoreArea is a dummy for the core economic areas (Tokyo, Osaka, Aichi,
Kanagawa, Chiba, Saitama, Hyogo and Kyoto prefectures). The sample period is 1980-1996. The memorandum item at the bottom of the table reports the regression of the
fraction of small firms reporting a city bank as main bank on our pre-1990 measure of financial integration, the average lending share of city banks in a prefecture in 1980-1990.
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
17 / 25
Transmission channel: evidence II
SME-local bank relationships extremely persistent in the silk regions
SME share w/o change in bank relationship 1990−2000
Figure: Silken thread and silken fetters
Hoffmann & Okubo
0.95
Shimane
Akita
Niigata
Oita
0.9
Ishikawa
Yamagata
Tottori
Toyama
Gunma
Saga
Tokushima
Nagasaki
Nara
0.85
Kumamoto
0.8
Aomori
Kochi
Ehime
Shiga
Fukui
Mie
Tochigi
Yamaguchi Fukushima
Miyazaki
Iwate
Kagawa
Hiroshima
Saitama
Shizuoka
Miyagi
Okayama
Chiba
Kagoshima
Hyogo
Tokyo
Osaka
Fukuoka
Nagano
Gifu
Yamanashi
Aichi
Ibaraki
Kyoto
Kanagawa
Wakayama
0.75
Hokkaido
0.7
1.5
2
2.5
3
3.5
4
4.5
5
5.5
6
6.5
log # silk filatures per capita in 1895
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
18 / 25
Transmission channel: evidence III
Table: Transmission mechanism: ancillary implications of model
A: Hold-up
B: Exposure to local shocks
Tier 2 banks
All
High FI
I
II
Real estate exposure
k
SMEVA
...CoreArea
Low FI
All
High FI
Low FI
III
IV
V
VI
Dependent variable is average loan interest rate
FI = CityBankShare
Interactions of Post1990t with pre-1991 variables
k
...SMEVA
× Real estate exposure
Tier 1 banks
0.21
(2.13)
0.01
(0.86)
-0.01
(-1.52)
-0.002
(-3.57)
0.15
(1.00)
0.01
(1.64)
-0.01
(-0.80)
-0.002
(-2.8)
0.27
(2.02)
0.01
(0.42)
-0.0045
(-0.67)
-0.12
(-0.59)
0.002
(0.16)
0.01
(1.46)
-0.002
(-1.21)
Interactions of ∆LocalLandPricetk with pre-1991 variables
0.40
(0.59)
-0.02
(-1.03)
0.01
(1.32)
-0.003
(-1.35)
-0.31
(-2.09)
0.02
(1.54)
0.01
(0.97)
k
...SMEVA
× Tier 2 Real estate exposure
...Tier 2 Real estate exposure
k
...SMEVA
High FI
Low FI
VII
VIII
Dependent variable is GDPgrowth
FI = CityBankShare
-9.03
(-2.22)
0.10
(0.89)
-0.08
(-0.64)
4.46
(4.65)
-0.12
(-3.70)
0.19
(2.21)
0.01
(1.44)
0.13
(1.03)
0.01
(0.93)
0.13
(1.47)
Add’l controls
∆LocalLandPricetk
SME k × ∆Citylandpricet
R2
0.99
0.99
0.98
0.99
0.98
0.99
0.62
0.55
number of prefectures
35
16
19
38
17
21
21
21
The Table shows regressions illustrating the ancillary implications of the stylized
banking model discussed in the main text: hold-up (panel iA) and differential exposure to local shocks
h
k
k
(panel B). Panel A presents regressions of the form Rtk (Tier ) = Post1990t × α0 SMEVA
× REE (Tier )k + α1 REE (Tier )k + α2 SMEVA
+ α03 Xtk + µk + τt + kt where Tier = 1, 2 stands for
either tier 1 (supraregional) or tier 2 (local) banks and R(Tier )kt is the average interest rate charged by banks of the respective tier in prefecture k and REE (Tier )k denotes these banks’
pre-1990 real estate exposure. Regressions are reported for all (colums I, IV), and for high (low) financial integration prefectures (columns II and III for tier 1 and columns V and VI for tier
k
2). As before, Post1990t is a dummy indicating the period after 1990 (i.e. 1991-2005), SME is our measure of small-business importance (based on value added) , FI k is our measure of
financial integration, the pre-1990 lending share of city banks in prefecture k. CoreArea is a dummy for the core economic areas (Tokyo, Osaka, Aichi, Kanagawa, Chiba, Saitama, Hyogo
and Kyoto prefectures). The sample period is 1980-2005.
h
i
k
k
Panel B shows the regressions of the form ∆gdptk = ∆LocalLandpricetk × α0 SMEVA
× REE (2)k + α1 REE (2)k + α2 SMEVA
+ α3 + α4 ∆CityLandpricet × SME k + µk + τt + kt where
∆LocalLandpricetk is the log change in land prices in prefecture k and ∆CityLandpricet is the log change in land prices in the core areas and REE (2)k is the pre-1990 real estate
exposure of local (Tier = 2) banks in prefecture k. The variables SME and FI are as before. The sample period is 1980-2003. In both panels, µk and τt are prefecture-fixed and time
effects respectively.
The memorandum item at the bottom of the table reports the regression of the fraction of small firms reporting a city bank as main bank on our pre-1990 measure of financial integration,
the average lending share of city banks in a prefecture in 1980-1990.
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
19 / 25
Our story (part II): silken threads and silken fetters
What are the deep sources of prefecture-level differences in
financial integration?
Cross-prefecture differences in the historical pathways to
financial development lead to huge differences in financial
integration in 1990.
We argue that the specific financing needs of the silk reeling
industry led to the emergence of a cooperative, regionallly tiered
model of banking.
[...]
We use # of reeling plants (filatures) p.c. at prefecture-level in
1895 as an instrument for the market share of regional banks in
the 1980s
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
20 / 25
[...] Silk and finance
Silk filatures (i.e. reeling plants) were heavily dependent on working
capital:
I
I
cocoons had to be purchased in the spring, reeled silk could only be
shipped to the Yokohama market in the late summer.
purchase of cocoons amounted to 80 percent of the operating costs of a
silk filature.
Silk reelers were located in remote mountain areas and could not usually
borrow from (mainly Yokohama-based) city banks. However, export
market for silk was concentrated in Yokohama.
Local banks in Japan institutional response to this dilemma:
I
I
local silk reelers’ associations and Yokohama silk broker had first-hand
knowledge of market conditions and of the quality produced by
individual silk reelers. This gave them a huge comparative advantage
(relative to city banks) in lending to these local SMEs.
Shinkin (cooperatives) and Sougo (mutuals) were founded by Silk
merchants and by reelers’ cooperatives.
As the silk industry was superseded by other export industries , these
local banks preserved their comparative advantage in lending to SMEs.
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
22 / 25
Figure: City and Regional Bank Lending Shares (pre-1990 (1980-1990) averages) vs. number of silk filatures per
head in 1895
0.9
0.45
Tokyo
Yamanashi
0.85
0.4
0.8
lending share of Shinkins pre−1990
lending share of City Banks pre−1990
0.7
Nara Fukuoka
0.65
Kanagawa
Saitama
Miyagi
Kagawa
0.6
0.5
Hokkaido
Kumamoto
Chiba
Tochigi
Tokushima
Aomori
Hiroshima
Fukui
Hyogo
Yamaguchi
Akita
Okayama
Shizuoka
Mie
Ehime Niigata
Shiga
Oita
Wakayama
Miyazaki
Fukushima
Kagoshima
Iwate
Kochi
0.45
0.4
1.5
Aichi
Ishikawa
Nagasaki
0.55
2
2.5
Kyoto
0.35
Osaka
0.75
3
3.5
4
Toyama
Ibaraki
Gunma
Kyoto
Gifu
0.3
Gunma
0.25
0.2
Hiroshima
Yamaguchi
Miyazaki
Kumamoto
Yamagata
4.5
5
5.5
6
6.5
log # silk filatures per capita in 1895
Aichi
Ishikawa
Toyama
Shimane
Saga
Shimane
Ibaraki
Saitama Fukui
Wakayama
Kanagawa
Oita
Fukushima
Niigata
Kagoshima
Okayama
Shiga
Chiba
Aomori
0.15
Tottori
Gifu
Nagano
Yamanashi
Nagano
Shizuoka
Hyogo
Hokkaido
0.1
Tokyo
Nagasaki
0.05
1.5
2
Mie
Nara
Osaka
Tottori
Tochigi
Iwate
Kagawa
Kochi
Ehime
Fukuoka
Miyagi
Saga
Yamagata
Akita
Tokushima
2.5
3
3.5
4
4.5
5
5.5
6
6.5
log # silk filatures per capita in 1895
The silken thread: Regional financial integration in 1990 low in areas in with
a high share of silk-exporting (reeling) firms in 1895
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
23 / 25
Panel IV regressions
Table: Panel IV Regressions with filatures / head in 1895 as instrument
City
Banks
k
× FI k
SMEVA
0.89
(2.15)
Regional
Banks
All
Shinkin
-1.57
-1.94
(-2.18) (-2.08)
FI k
-0.18
(-2.21)
0.43
(2.00)
0.40
(1.96)
-0.20
(-1.58)
0.28
(1.28)
0.27
(1.28)
-0.16
(-1.86)
0.31
(1.64)
0.33
(1.65)
k
SMEVA
-0.57
(-2.44)
0.32
(1.80)
0.21
(1.61)
-0.65
(-1.81)
0.30
(1.39)
0.17
(1.20)
-0.53
(-1.92)
0.32
(1.73)
0.22
(1.63)
no
no
no
yes
0.01
(0.33)
-0.01
(-1.72)
yes
-0.01
(-0.60)
-0.01
(-2.38)
yes
-0.01
(-2.02)
-0.01
(-2.51)
yes
yes
yes
-0.00
(-0.78)
0.00
(0.93)
-0.01
(-1.58)
0.00
(1.03)
-0.01
(-1.85)
0.00
(2.71)
Interactions terms
of Post1991t with ...
Controls
relative GDP
Core
City
Banks
1.04
(1.69)
Regional
Banks
All
Shinkin
-1.41
-1.42
(-1.50) (-1.42)
Distance to Yokohama
R2
City
Banks
0.86
(1.84)
All
-1.46
(-1.81)
Regional
Banks
Shinkin
-1.65
-1.76
0.69
0.69
0.69
0.70
0.70
0.70
0.70
0.70
0.70
1st-Stage F-stat for SME k × FI k × Post1991t
303.29
288.56
407.01
420.48
279.43
479.21
383.56
297.11
439.05
Kleibergen-Paap rank test
77.26
0.00
37.53
0.00
41.56
0.00
66.78
0.00
25.76
0.00
38.98
0.00
94.57
0.00
37.86
0.00
44.68
0.00
p-value
h
i
ck + α3 SME k + α0 Xt + µk + τt + k where where Post1990t is a dummy indicating the period starting in
\
k × FI k + α FI
The Table shows results from the IV regression ∆gdptk = Post1990t × α1 SME
2
t
4
ck are the first-stage fitted values of SME k × FI k and FI k using
\
k × FI k and FI
1991, SME k is small manufacturing firm importance (value-added or employment based) and Xt is a vector of controls. SME
SME k × Silk k and Silk k as instruments, where Silk k is the log number of silk filatures per head of population in a prefecture in 1895. CoreArea is a dummy for the core economic areas (Tokyo, Osaka,
Aichi, Kanagawa, Chiba, Saitama, Hyogo and Kyoto prefectures). The sample period is 1980-2005, t-statistics appear in parentheses. The bottom of the Table reports information on instrument relevance:
the F-statistics associated with the first stage regression of the interaction term on all instruments and the Kleibergen-Paap (2006) (KP) rank statistics and its associated p-value for the hypothesis of
under-identification. The KP-statistics appears in boldface (italics) if it exceeds the Stock-Yogo(2005) weak-instrument critical values of 7.03 (4.58) (see Table 5.2. in Stock and Yogo (2005) , for the
case of n = 2 endogenous variables and K = 2 excluded instruments), This suggests that the instruments can be taken to be sufficiently strong to ensure a maximal size of no more than 10% (15%) for a
nominal 5% size Wald Test on the IV-estimates.
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
24 / 25
Summary and outlook
Paper studies the regional dimension of Japan’s Great Recession:
local bank dependence and interregional banking integration
interacted in the regional spread of the crisis after 1990.
The silken thread: regional differences in financial integration
are extremely persistent: Comparative advantage in trade (silk)
affected the particular pathway to financial development and de
facto integration at the onset of the crisis of 1990.
’Silken thread’ is reflected in extremely persistent banking
relationships which may have made it hard for SME to switch to
nationwide banks in the trough of the crisis.
Relevance for today: state-level segmentation of the US banking
market till the 1980s, Regulatory and regional segmentation of
banking in Europe today. Even a banking union will not make
this segmentation go away!
Hoffmann & Okubo
()
By a Silken Thread
Stanford APARC, 1 Dec 2015
25 / 25
Part I
Appendix
Bonus slides
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (27)
Data set
Panel data set for 46 prefectures (ex Okinawa).
I
I
I
I
GDP p.c.
Lending by type of bank by prefecture, 1964-96
Data on small manufacturing firms by prefecture, employment
and value added from the ’Manufacturing Census’. Here focus
on SMEs with <300 employees. (This is also the cut-off value
for Shinkin membership).
Data on number of silk filatures, population etc. in 1895 from
various sources
....
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (28)
Our story (I): related findings
Low financial integration prohibits cross-prefectural pooling of
funds (e.g. through nationwide banks’ internal capital markets
(Cetorelli and Goldberg; JoF forthcoming)).
The pattern of nationwide banks withdrawing from areas where
they have low market share is reminiscent of Japanese banks’
behavior overseas after 1990 ((e.g. Peek and Rosengreen
(1997)).
Consistent with ’Evergreening’ and the ’Zombie hypothesis’
(Caballero, Hoshi and Kashyap (2008), Peek and Rosengreen
(2005)): big banks could have withdrawn from their non-core
areas to prop up Zombie firms in their core areas of activity.
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (29)
An example
prefecture
City lending share
SME share
post-1990 average growth
7
19
Fukushima
Yamanashi
45.81
42.29
17.06
20.09
0.58
-0.14
29
40
Nara
Fukuoka
66.14
65.54
19.67
10.49
0.08
0.26
Back
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (30)
’Proper’ interaction term specification
Table: Interaction terms and additional controls
I
Regional
Interactions of Post1990t
with ...
II
City
III
Regional
...SME k × RegionalBankShare k
-1.50
(-2.72)
...SME k × CityBankShare k
...RegionalBankShare k
k
...SMEVA
V
Regional
0.03
(0.82)
0.27
(3.04)
-0.05
(-2.38)
-0.09
(-3.87)
0.33
(2.19)
VII
Regional
0.74
(3.78)
0.24
(3.87)
0.23
(3.23)
-0.45
(-3.55)
...Lending/GDP
0.56
∆gdptk
0.57
0.57
h
0.29
(2.35)
-0.47
(-3.66)
-0.0006
(-1.31)
0.0003
(0.60)
0.57
By a Silken Thread
k
0.57
k
0.32
(2.72)
-0.48
(-4.06)
-0.01
(-4.00)
-0.008
(-2.63)
0.56
0.56
0.57
k
α2 SMEVA
α03 Xtk
i
The Table shows results from the regression
= Post1990t ×
× FI + α1 FI +
+
+µ +τt +kt
k
where Post1990t is a dummy indicating the period after 1990 (1991-2005), SMEVA
is small-business importance based on value
added, FI k is the measure of financial integration (regional and city bank share in total lending in prefecture k), as indicated
in the column heading. µk and τt are prefecture-fixed and time effects respectively. The vector X k captures various prefecture
characteristics. In the regressions it is interacted with our crisis dummy Post1990t and contains prefecture-level Lending k /GDP k
(1980-90 average) and CoreAreak , a dummy for the core economic areas (Tokyo, Osaka, Aichi, Kanagawa, Chiba, Saitama,
Hyogo and Kyoto prefectures). The sample period is 1980-2005. OLS estimates, t-statistics in parentheses. Standard errors are
clustered by prefecture.
Stanford APARC, 1 Dec 2015
k
α0 SMEVA
-0.13
(-5.03)
-0.16
(-4.15)
...CoreArea
R2
VIII
City
-1.42
(-3.24)
0.72
(3.20)
-0.15
(-4.56)
-0.07
(-2.85)
VI
City
-1.35
(-2.89)
0.68
(3.12)
...CityBankShare k
Controls:
Xk :
IV
City
k
Appendix (31)
Robustness
Back to Baseline spec
Table: Robustness – interaction terms and additional controls
Interactions of Post1990t
with ...
...SME k × FI k
...FI k
k
...SMEVA
FI k
k
... SMEVA
Controls:
Xk :
I
Regional
II
City
III
Regional
IV
City
V
Regional
VI
City
-1.28
(-3.04)
0.23
(3.59)
0.34
(2.88)
0.66
(3.33)
-0.11
(-3.72)
-0.39
(-3.17)
-0.66
(-1.23)
0.44
(3.61)
0.17
(0.80)
0.47
(2.51)
0.02
(0.22)
-0.14
(-0.79)
-1.09
(-2.06)
0.25
(1.50)
0.32
(1.48)
0.55
(2.78)
-0.06
(-0.61)
-0.11
(-0.59)
-0.55
(-1.96)
-0.12
(-1.61)
-0.12
(-0.30)
-0.03
(-0.44)
-0.22
(-0.38)
-0.62
(-1.33)
-0.12
(-0.28)
-0.69
(-1.49)
-0.003
(-0.53)
0.01
(0.62)
0.003
(0.58)
0.002
(1.05)
0.02
(2.28)
-0.004
(-0.79)
0.01
(0.94)
0.003
(0.45)
0.002
(0.76)
0.02
(1.44)
Yes
Yes
2
2
...CoreArea
...Share Lowland Areas
...Share of steep areas
...Min. distance to core
...Sectoral Specialization
-0.003
(-0.57)
0.01
(0.62)
0.003
(0.44)
0.002
(1.04)
0.02
(1.99)
-0.003
(-0.44)
0.01
(1.01)
0.004
(0.61)
0.002
(0.91)
0.02
(1.59)
Yes
Yes
Ztk :
Region Fixed Effect
Stanford APARC, 1 Dec 2015 By a Silken Thread
2
Yes
Yes
Appendix (32)
Is it financial integration or local financial development?
Table: Alternative measures of financial development and financial
integration
I
FI =
Interactions of Post1990t
with pre1990 variables:
k
...SMEVA
FD =
...FI k
...SME k × FI k
FD
SME k × FD
k
k
...CoreArea
R2
II
City Bank Lending
Total Lending
#Branches
Population×Area
-0.48
(-3.82)
-0.14
(-3.89)
0.78
(3.00)
FD = Lending
GDP
-0.45
(-3.73)
-0.09
(-2.28)
0.46
(1.73)
0.07
(0.54)
-0.32
(-0.43)
III
FI =
FD =
CityBankLending
GDP
Regional Bank Lending
GDP
IV
FI =
FD =
City Bank Lending
Total Lending
Regional BankLending
GDP
-0.07
(-0.81)
-0.004
(-6.76)
0.03
(4.07)
-0.55
(-4.42)
-0.14
(-5.55)
0.81
(4.52)
-0.002
(-2.09)
0.02
(2.61)
0.01
(1.79)
-0.07
(-1.31)
0.00
0.12
0.02
(0.42)
-0.01
(-2.14)
-0.01
(-3.43)
-0.01
-4.85
-0.01
(-4.01)
0.56
0.56
0.56
0.56
The Table shows results from the regression
h
i
k
k
k
∆gdptk = Post1991t × α1 SMEVA
+ α2 FI k + α3 SMEVA
× FI k + α5 FD k + α6 SMEVA
× FD k + α70 CoreAreak +µk +τt +kt
k
where where Post1991t is a dummy indicating the period from 1991, SMEVA
is small-business importance based
on value added, and FI k and FD k are the measures of financial integration and financial development respectively as
indicated in the column heading. µk and τt are prefecture-fixed and time effects respectively. CoreArea is a dummy
for the core economic areas (Tokyo, Osaka, Aichi, Kanagawa, Chiba, Saitama, Hyogo and Kyoto prefectures). The
sample period is 1980-2005. OLS estimates, t-statistics in parentheses. Standard errors are clustered by prefecture.
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (33)
A first set of results
Great Recession is deeper and more prolonged in areas with
many SME’s – provided financial integration is low.
The effect is big: a prefecture with a 20 percent SME share
would have ˜0.4 percent lower annual growth than the country
as a whole if its banking sector is weakly integrated with the rest
of the country.
Financial frictions seem stronger in areas with low (pre-1990)
levels of banking integration (low share of universal / high share
of regional banks)
Lending channel & low financial integration? City banks
withdraw lending from areas to which they traditionally have
weak ties
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (34)
Endogeneity issues
Lending shares of regional / city banks might be endogenous as
might be small firm importance.
But: using pre-1990 data would counter most endogeneity issues.
However, using pre-1990 data does not entirely preclude
expectational feedbacks:
I
If investment and growth prospects were poor in some areas, the big
city banks might have started to withdraw from such regions even
before the 1990s. This would lead to a high market share of
regional banks.
→ need some instrument for regional bank lending share.
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (35)
Silk and finance
Silk filatures (i.e. reeling plants) were heavily dependent on working
capital:
I
I
cocoons had to be purchased in the spring, reeled silk could only be
shipped to the Yokohama market in the late summer.
purchase of cocoons amounted to 80 percent of the operating costs of a
silk filature.
Silk reelers were located in remote mountain areas and could not usually
borrow from (mainly Yokohama-based) city banks. However, export
market for silk was concentrated in Yokohama.
Instead of banks, Yokohama silk export merchants would issue a letter of
credit to small reelers who would discount it with his local silk
cooperative.
These local coperatives had first-hand insihjt into the quality of the
output of their members, making them ideal intermediaries of credit.
Local cooperatives often were at the origin of regional banks. These
banks were purely regional – and stayed it for more than a century.
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (36)
Silk export finance
Yokohama merchant would advance credit to a reeler in the form
of a ’documentary bill’ issued by a Yokohama bank.
Reeler would obtain trade credit in the form of an advance on
the bill from a local bank.
I
These banks were often cooperative or mutual banks, founded
by silk industry associations (Shinkins) or by Yokohama silk
merchants.
After reeling and shipping of the silk to Yokohama, Yokohama
bank would issue a bill of acceptance to the reeler who would
use this to discount the documentary bill with his local bank.
Regional bank settled payment of the bill with the issuing bank
in Yokohama.
This system of credit is very much like the system of modern
trade finance:
I
(’advising’) bank of the exporter borrows and lends locally.
’International’ transactions occur only with the big Yokohama
banks, which in turn have links to regional banks around the
country.
Regional tiering of Japan’s banking system goes back to the
institutions of silk export finance.
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (37)
Silk finance: the role of mechanization
Huge relative price increase of mechanically vs. hand-reeled silk in the
1890s.
Mechanization central in the quality improvement. But increased
dependence on working capital — it reinforced the separation of
cocoon-growing and reeling.
Early stages of mechanization: cooperatively organized and centralized
second (mechanical) reeling process of (possibly manually reeled) silk.
Centralized re-reeling allowed the implemenation of quality control
system and the development of internationallly recognized brands.
Quality was central in the monitoring of the credit relationship between
silk producers and the Yokohama silk merchants: regional banks would
provide credit (’advances’) against a documentary bill issued by
Yokohama silk merchants.
Ultimately, only those producers could continue to export who
mechanized early. These also had access to the trade credit and export
finance by the Yokohama silk merchants. The others ended up
producing mainly for the domestic market.
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (38)
Mechanized silk filatures in 1895
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (39)
Silk and modern-day lending shares
Table: Modern day (pre-1990) lending and silk filatures
Financial Integration
pre-1990 share in prefecture-level lending by
City Banks
Regional Banks
All (Shinkin+Sogo)
Shinkins only
filatures / population
(log #)
-0.03
(-3.14)
-0.04
(-4.70)
0.03
(4.22)
0.03
(4.11)
0.04
(4.96)
0.04
(4.53)
Financial Development
bank branches
population×area
Lending/GDP
(pre-1990)
(pre-1990)
0.01
(0.87)
0.01
(0.87)
-0.61
(-1.78)
-0.55
(-1.95)
-0.10
(-0.29)
Relative GDP (pre-90)
0.19
(3.32)
-0.01
(-0.18)
-0.01
(-0.24)
0.09
(1.68)
8.56
(4.21)
6.27
(2.88)
Core Dummy
0.07
(2.46)
-0.001
(-0.02)
0.02
(0.71)
-0.02
(-0.57)
1.92
(1.88)
1.06
(1.02)
Distance to Yokohama
(log)
-0.02
(-1.33)
0.01
(0.66)
-0.01
(-0.93)
0.01
(0.74)
0.55
(1.25)
0.74
(1.75)
12.20
(2.28)
City Bank Lending
R2
0.18
0.60
0.29
0.30
0.36
0.40
0.02
0.08
0.07
0.46
The table shows regressions of modern-day (pre-1990) average prefectural lending shares by bank type on our silk
instrument – the number of filatures per head of population in a prefecture in 1895. The control variables are relative
(pre-199) per capita GDP, the (log) distance to Yokohama and a dummy for the core areas (Tokyo, Osaka, Aichi,
Kanagawa, Chiba, Saitama, Hyogo and Kyoto prefectures).
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (41)
0.53
The role of mechanization
Table: Impact of mechanization on pre-1990 financial integration measures and founding year of first
industrial cooperative bank in a prefecture.
(log) share in prefecture-level lending by
City Banks
Regional Banks
All (Shinkin+Sogo)
Shinkins only
Founding year
of first
Shinkin in prefecture
(log)
hand filatures
(log #)
-0.01
(-1.38)
0.01
(0.94)
-0.00
(-0.08)
-0.00
-0.09
mechanized filatures
(log #)
-0.02
(-3.35)
0.02
(2.85)
0.03
(4.22)
-0.00
(-1.89)
output: hand reeled
(log tons)
-0.00
(-0.61)
-0.00
(-0.42)
-0.01
(-0.62)
0.00
(0.66)
output: machine reeled
(log tons)
-0.03
(-3.76)
0.02
(2.65)
0.02
(2.30)
-0.00
(-0.76)
R2
0.62
0.62
0.22
0.17
0.38
0.21
0.21
0.16
Controls
yes
yes
yes
yes
yes
yes
yes
yes
The table shows results from regression of pre-1991 average prefectural lending shares by bank type (left panel) and of founding year of the first
industrial cooperative bank (Shinkin) in a prefecture (right panel) on our alternative silk industry instruments: the number of hand-powered and
machine filatures at prefecture-level, and the output of hand-powered and machine filatures respectively. Controls are: relative GDP pre-1990, a
core area dummy and log distance to Yokohama. Core areas are as described in previous tables. The founding year of the first Shinkin is normalized
by 1900 (the year of the enactment of the first industrial cooperative act). We take the logarithm of this normalized measure as our dependent
variable.
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (42)
Panel IV regressions
Table: Panel IV Regressions with filatures / head in 1895 as instrument
City
Banks
k
× FI k
SMEVA
0.89
(2.15)
Regional
Banks
All
Shinkin
-1.57
-1.94
(-2.18) (-2.08)
FI k
-0.18
(-2.21)
0.43
(2.00)
0.40
(1.96)
-0.20
(-1.58)
0.28
(1.28)
0.27
(1.28)
-0.16
(-1.86)
0.31
(1.64)
0.33
(1.65)
k
SMEVA
-0.57
(-2.44)
0.32
(1.80)
0.21
(1.61)
-0.65
(-1.81)
0.30
(1.39)
0.17
(1.20)
-0.53
(-1.92)
0.32
(1.73)
0.22
(1.63)
no
no
no
yes
0.01
(0.33)
-0.01
(-1.72)
yes
-0.01
(-0.60)
-0.01
(-2.38)
yes
-0.01
(-2.02)
-0.01
(-2.51)
yes
yes
yes
-0.00
(-0.78)
0.00
(0.93)
-0.01
(-1.58)
0.00
(1.03)
-0.01
(-1.85)
0.00
(2.71)
Interactions terms
of Post1991t with ...
Controls
relative GDP
Core
City
Banks
1.04
(1.69)
Regional
Banks
All
Shinkin
-1.41
-1.42
(-1.50) (-1.42)
Distance to Yokohama
R2
City
Banks
0.86
(1.84)
All
-1.46
(-1.81)
Regional
Banks
Shinkin
-1.65
-1.76
0.69
0.69
0.69
0.70
0.70
0.70
0.70
0.70
0.70
1st-Stage F-stat for SME k × FI k × Post1991t
303.29
288.56
407.01
420.48
279.43
479.21
383.56
297.11
439.05
Kleibergen-Paap rank test
77.26
0.00
37.53
0.00
41.56
0.00
66.78
0.00
25.76
0.00
38.98
0.00
94.57
0.00
37.86
0.00
44.68
0.00
p-value
h
i
ck + α3 SME k + α0 Xt + µk + τt + k where where Post1990t is a dummy indicating the period starting in
\
k × FI k + α FI
The Table shows results from the IV regression ∆gdptk = Post1990t × α1 SME
2
t
4
ck are the first-stage fitted values of SME k × FI k and FI k using
\
k × FI k and FI
1991, SME k is small manufacturing firm importance (value-added or employment based) and Xt is a vector of controls. SME
SME k × Silk k and Silk k as instruments, where Silk k is the log number of silk filatures per head of population in a prefecture in 1895. CoreArea is a dummy for the core economic areas (Tokyo, Osaka,
Aichi, Kanagawa, Chiba, Saitama, Hyogo and Kyoto prefectures). The sample period is 1980-2005, t-statistics appear in parentheses. The bottom of the Table reports information on instrument relevance:
the F-statistics associated with the first stage regression of the interaction term on all instruments and the Kleibergen-Paap (2006) (KP) rank statistics and its associated p-value for the hypothesis of
under-identification. The KP-statistics appears in boldface (italics) if it exceeds the Stock-Yogo(2005) weak-instrument critical values of 7.03 (4.58) (see Table 5.2. in Stock and Yogo (2005) , for the
case of n = 2 endogenous variables and K = 2 excluded instruments), This suggests that the instruments can be taken to be sufficiently strong to ensure a maximal size of no more than 10% (15%) for a
nominal 5% size Wald Test on the IV-estimates.
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (43)
Robustness: X-sectional OLS and IV regressions
Table: Cross-sectional Regressions
SMEVA (output based)
City Banks
SMEEMP (employment based)
OLS
0.14
(1.33)
IV
0.36
(1.71)
Regional Banks
All
Shinkin
OLS
IV
OLS
IV
-0.35
-0.77
-0.29
-0.98
(-2.12) (-1.52)
(-1.68) (-1.55)
FI k
-0.04
(-2.36)
-0.08
(-2.01)
0.06
(2.15)
0.18
(1.50)
0.05
(1.59)
0.22
(1.52)
-0.04
(-1.97)
-0.10
(-2.01)
0.07
(2.18)
0.16
(1.92)
0.06
(1.79)
0.18
(1.90)
SME k
-0.10
(-1.79)
-0.23
(-1.94)
0.07
(1.48)
0.15
(1.25)
0.03
(0.79)
0.11
(1.25)
-0.12
(-1.51)
-0.34
(-1.88)
0.12
(1.72)
0.19
(1.43)
0.05
(1.16)
0.14
(1.45)
-0.00
(-2.73)
-0.00
(-1.06)
-0.01
(-4.58)
-0.00
(-1.99)
-0.01
-0.01
(-4.79) (-3.73)
-0.00
(-2.89)
-0.00
(-1.32)
-0.01
(-4.87)
-0.01
(-3.36)
-0.01
(-5.03)
-0.01
(-4.42)
0.46
0.46
0.46
0.45
0.46
0.44
SME k × FI k
City Banks
OLS
0.16
(1.12)
IV
0.56
(1.70)
Regional Banks
All
Shinkin
OLS
IV
OLS
IV
-0.52
-0.85
-0.44
-1.08
(-2.22) (-1.78)
(-1.94) (-1.87)
Controls
Core
R2
0.50
0.46
0.44
0.46
0.48
0.46
First-Stage F-stat
for SME k × FI k
14.21
10.56
17.07
13.13
6.94
12.40
Kleibergen-Paap rank test
p-value
3.50
0.06
1.32
0.25
1.71
0.19
4.19
0.04
3.04
0.08
3.75
0.05
k
k
The Table shows results from the cross-sectional OLS and IV regressions ∆gdppost1990
= α1 SME k × FI k + α2 FI k + α3 SME k + α04 CoreDummy k + const + k where ∆gdppost1990
is average
post-1990 (1991-2005) GDPgrowth in prefecture k, SME k is small manufacturing firm importance (value-added or employment based) and FI k our measure of regional banking integration
(city bank share, regional bank share, Shinkin share). CoreArea is a dummy for the core economic areas (Tokyo, Osaka, Aichi, Kanagawa, Chiba, Saitama, Hyogo and Kyoto prefectures). In
the IV-regressions, SME k × FI k and FI k are instrumented using SME k × Silk k and Silk k , where Silk k is the log number of silk filatures per head of population in a prefecture in 1895. The
F-statistics below the IV-regression pertain the the test for the significance of instruments in the first-stage regression for SME k × FI k . t-statistics in parentheses. The last two rows of the
table reports F-statistics associated with the first stage regression of the interaction term on all instruments and the Kleibergen-Paap (2006) rank statistics and the associated p-value for the
hypothesis of under-identification.
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (44)
Is industry structure endogenous?
RZ-approach has been criticized on the grounds that financially
constrained regions have a comparative advantage in less
finance-dependent industries.
Hence, they should specialize in these industries. If this is the
case, then we would probably tend to overestimate the
aggregate effects of low financial integration.
What we need is an exogenous measure of growth expectations
/ industry structure that is not affected by low financial
integration.
We turn to the literature on agglomeration externalities (Glaeser
et al 1992) to proxy for the importance of knowledge
externalities in manufacturing: distance to the main silk regions
(as opposed to share of silk industry in local economy).
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (45)
Table: Disentangling financial integration & industrial structure
Industrial structure
Small manufacturing
Manufacturing Share
firm share
in GDP
in EMP
in GDP
in EMP
Financial Integration
pre-1990 lending share by
City Banks
Regional Banks
All
Shinkin
distance to most highly mechanized
silk regions (log)
-0.03
(-6.28)
-0.02
(-5.41)
-0.06
(-5.05)
-0.03
(-5.26)
-0.02
(-1.35)
-0.01
(-1.46)
-0.01
(-1.07)
filatures / population
(log #)
0.01
(2.04)
0.01
(2.87)
0.00
(0.31)
0.01
(1.87)
-0.04
(-4.41)
0.02
(3.09)
0.03
(3.60)
Core Dummy
Distance to Yokohama
(log)
-0.01
(-1.68)
-0.03
(-2.30)
-0.01
(-1.61)
-0.03
(-2.77)
-0.03
(-2.03)
-0.05
(-1.39)
-0.02
(-2.32)
-0.03
(-1.77)
-0.03
(-1.96)
0.08
(2.53)
0.01
(1.01)
-0.01
(-0.46)
-0.01
(-0.70)
0.01
(0.37)
R2
0.69
0.68
0.57
0.65
0.56
0.34
0.42
The table shows regressions of modern-day (pre-1990) industrial structure (left) and average prefectural lending shares
by bank type (right) on our two alternative silk-related variables: the (log) distance to the three prefectures with
the most highly mechanized silk industry in 1895 (Kyoto, Nagano, Gifu and Shizuoka) and the (log) number of
filatures per head in 1895 and a set of controls. The control variables are the (log) distance to Yokohama (the
main silk market), a dummy for the Core areas (Tokyo, Osaka, Aichi, Kanagawa, Chiba, Saitama, Hyogo and Kyoto
prefectures).
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (46)
Table: Panel IV Regressions (both credit dependence and financial integration endogenous)
CD = SMEVA
Regional
Banks
All
Shinkin
1.30
-3.25
-3.98
(1.79)
(-1.94) (-1.88)
CD = SMEEMP
Regional
Banks
All
Shinkin
2.68
-5.35
-5.78
(1.98)
(-2.06) (-2.03)
CD =Manufacturing Share in GDP
City
Regional
Banks
Banks
All
Shinkin
0.77
-1.70
-3.28
(1.54)
(-1.67)
(-1.57)
FI k
-0.24
(-1.93)
0.65
(1.90)
0.80
(1.86)
-0.40
(-2.08)
0.86
(2.00)
0.93
(2.00)
-0.20
(-1.64)
0.50
(1.65)
0.98
(1.56)
CD
-0.78
(-1.93)
0.76
(1.83)
0.53
(1.72)
-1.56
(-2.06)
1.30
(2.00)
0.80
(1.92)
-0.44
(-1.68)
0.44
(1.55)
0.49
(1.46)
yes
yes
yes
yes
yes
yes
yes
yes
yes
Interactions terms
of Post1990t with ...
CD × FI k
City
Banks
Controls
R2
City
Banks
0.70
0.70
0.70
0.70
0.70
0.70
0.70
0.70
0.70
1st-Stage F-stat for
CD k × FI k
384.83
723.66
726.13
335.05
757.38
776.77
240.39
396.09
534.91
Kleibergen-Paap rank test
p-value
33.93
0
10.87
0.01
8.15
0.01
19.13
0.00
9.08
0.01
8.14
0.01
23.89
0.00
12.71
0.00
4.62
0.03
i
h
ck + α3 SME k + α0 X k + µk + τt + k where where Post1990t is a dummy
k × FI k + α FI
The Table shows results from the IV regression ∆gdptk = Post1990t × α1 CD\
2
t
4
k × FI k and
indicating the period after 1990, CD k is our measure of credit dependence as indicated in the respective column headings and X k is a vector of controls. CD\
ck are the first-stage fitted values of CD k × FI k and FI k using the log numbers of filatures per head (filatures k ), the (log) distance to one of the three most mechanized
FI
silk regions and the interaction between these two as instruments. Control variates are (log) distance to Yokohama and a dummy for the core economic areas (Tokyo,
Osaka, Aichi, Kanagawa, Chiba, Saitama, Hyogo and Kyoto prefectures). The sample period is 1980-2005. t-statistics in parentheses. The bottom of the Table reports the
F-statistics associated with the first stage regression of the interaction term on all instruments and the Kleibergen-Paap (2006) rank statistics and the associated p-value
for the hypothesis of under-identification. Values of the KP-statistics in boldface or italics indicate that the hypothesis of weak identification is rejected. We reject if the
asymptotic bias of the TSLS estimator is less than 5% (KP in bold) or 10% (KP in italics) based on the critical values tabulated in Table 5.1. of Stock and Yogo (2005).
Since values for our case n = 3 endogenous variables and K = 3 instruments are not directly tabulated, we use the more conservative values for n = 3 and K = 5 which
are 9.53 and 6.61 respectively.
Stanford APARC, 1 Dec 2015
By a Silken Thread
Appendix (47)