Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Infrastructure Investment News and Business Cycles:Evidence from the VAR with External Instruments
Etsuro Shioji (Hitotsubashi)
CFE2018@PisaDecember 15, 2018
Acknowledgement
Research for this work has been funded by • MEXT through the Hitotsubashi Institute for Advanced Study (HIAS)
• Grant‐in‐aid for Scientific Research– A‐17H00985– C‐15K03418– C‐18K01605
• Nomura Foundation.
2
Objective
Propose a new approach to tackle the
3
“Fiscal Foresight” Problem
Structure of presentation
1. Introduction2. VAR with External Instruments (VAR‐IV)3. News Indicator: details4. Results from VAR‐IV with news indicator5. Conclusions
4
1. Introduction
Why Public Investment?
• Always a subject of heated debate in Japan.
• And... suddenly, also in the US! (since late 2016...)
6
Difficulty in estimating the impact=“Fiscal Foresight” Problem
7
Most fiscal policy measures are pre‐announced.
Main idea
Estimate the effects of a “News Shock”
to public investment
= Changes in the public’s perception about the future course of the policy.
8
Step 1: Construction of a news indicator
• Shioji and Morita (2017) constructed a dailyindicator which captures changes in people’s perceptions about future policy. This combines
–News approach (Ramey)
–Stock market approach (Fisher and Peters)
= Look at responses of stock prices of construction companies when major news about policy arrived.
9
Step 2: Incorporate the news indicator into a time series analysis
‐‐‐ How??
• Previous paper: Put it into a regular VAR as another endogenous variable.
• This paper: Use this as the instrument in the VAR with External Instruments (VAR‐IV).
10
2. On VAR‐IV
VAR‐IV
• Stock and Watson (2012), Mertens and Ravn(2013), Gertler and Karadi (2015)
• Survey paper by Stock and Watson (NBER‐WP24216, January 2018)
• Identification without exclusion restrictions.
12
Identifying assumptions
• IV is correlated with the true shock contemporaneously.
• IV is orthogonal to the other types of shocks
13
VAR‐IV: 2 variables, 1 lag example
14
1,
2,
tt
t
yY
y
1t t tY AY
t tB
Reduced form VAR
Structural relationship
1,
2,
tt
t
Endogenous variables
Structural shocks(mutually orthogonal)
11 12
21 22
b bB
b b
VAR‐IV, continued
15
1,
2,
t
t
11 12
21 22
b bb b
Suppose we are just interested in the first shock…
…then we just need to know the first column of B!
Assuming invertibility, ( )t tY C L B 1where ( ) ( )C L I AL
VAR‐IV, continued
16
Assumption 1: “relevance”
11
21
then, t t
bE Z
b
1, 0t tE Z
Suppose we have an instrument Z which satisfiest
Assumption 2: “exogeneity”(wrt the other shocks) 2, 0t tE Z
Normalize to equal 1.We can focus on b21.
VAR‐IV, estimation
17
2, 21 1, 1 1, 1 2 2, 1 22 2,t t t t ty b y d y d y b
Step 1: IV stage
Step 2: VAR stage
Using Zt as the instrument, estimate:
21ˆget b
1t t tY AY
Estimate the reduced form VAR:
-1ˆ ˆget ( ) ( - )C L I AL
VAR‐IV, Impulse responses
18
Compute the h period ahead Impulse Response Function as:
21
1ˆˆh hIRF Cb
Our case: Use the news indicator as an IV
• Our news indicator = Captures only a part of shocks to expectations about future policies.
– But it is correlated with true shocks to expectations.
– And it is uncorrelated with the other types of shocks.
19
3. News Indicators
20
Its construction: a rough sketch
21
Identify the dates on which important news arrived.
Examine the reaction of construction companies’ stock
prices.
continued
22
Companies that are more dependent on public procurements
Companies that are less dependent.
Study the difference in their responses.
Dependence on Public Investment = Share of Public work in Total (as of 2000)
230%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76
Cross‐group heterogeneity?Example from a big “news” event…
240%
5%
10%
15%
20%
25%
30%
35%
HIGH gov dependence group (>34%) LOW
Great East Japan Earthquake (March 11, 2011)sum of excess returns, March 14‐15
Two stock market indices
• Stock Mkt Index 1 = “High – Low”= (Avg of Upper Half) – (Avg of Bottom Half)
• Stock Mkt Index 2 = “G‐factor”– Extract 5 common factors ‐> Rotate them!– Target rotation: Select a rotation which gives the closest factor loadings to… (see next page)
25
26
(1)Industry‐wide Factor
(2) Home Builders Factor
(3) G‐Factor(Gov. Dependence)
(4) Electric Facilities Builders Factor
(5) Plant Builders Factor
Mid‐sizedContractors
1 0 0/1 0 0
Big FourContractors
1 1 0 0 0
Home Builders (all big)
1 1 0 0 0
Electric Facilities Builders
1 0 0/1 1 0
Plant Builders 1 0 0/1 0 1
Target for rotation
Stock Mkt Index 1 & 2(and 0), Cumulative
27
-1.5
-1-.5
0
01jan1990 01jan1995 01jan2000 01jan2005 01jan2010 01jan2015
Mean Excess Returns
-1.5
-1-.5
0.5
01jan1990 01jan1995 01jan2000 01jan2005 01jan2010 01jan2015
High (Blue) vs Low (Red)-.4
-.20
.2.4
01jan1990 01jan1995 01jan2000 01jan2005 01jan2010 01jan2015
High - Low
-60
-40
-20
020
01jan1990 01jan1995 01jan2000 01jan2005 01jan2010 01jan2015
G-Factor1 2
0
News indicator (1 & 2and 0)
Defined as
(News dates)*(Stock mkt index 1 or 2 or 0)
28
News indicators (daily)
29
News 0: based on the Mean Excess Returns-.0
50
.05
.1.1
5.2
01jan1990 01jan1995 01jan2000 01jan2005 01jan2010 01jan2015
News 1: based on "High-Low"
-.10
.1.2
01jan1990 01jan1995 01jan2000 01jan2005 01jan2010 01jan2015
News 2: based on "G-Factor"
-50
510
1520
01jan1990 01jan1995 01jan2000 01jan2005 01jan2010 01jan2015
News indicators (quarterly aggregates)
30
News 0: based on the Mean Excess Returns-.1
0.1
.2
1990q1 1995q1 2000q1 2005q1 2010q1 2015q1
News 1: based on "High-Low"
-.10
.1.2
.3
1990q1 1995q1 2000q1 2005q1 2010q1 2015q1
News 2: basd on "G-Factor"
-10
010
2030
1990q1 1995q1 2000q1 2005q1 2010q1 2015q1
4. VAR‐IV analysis:specification and results
VAR‐IV
32
IV = the news indicator
Endogenous variables = See the list on the next page
List of endogenous variables
• X1 =Stock Mkt Index 1 or 2(or 0)
• Construction orders from the public sector (top 50 companies)
• Nominal Public Investment (SNA)
• Public Investment Deflator (SNA)
• X5 = One of the macro variables (GDP etc.)
33
Specifications
• All in log differences except for the news variables.
• # of lags = 4
• Dummies for the 3 major earthquakes & Consumption tax hike.
34
X1 = “Stock Mkt Index 1”, X5 = Real GDP, IV =News 1
35
X1 = Stock Mkt Index 2, X5 = Real GDP, IV =News 2
36
For comparison:
X1 = Stock Mkt Index 0, X5 = Real GDP, IV =News 0
37
5. Summary
• What we have done:– Proposed a new way to estimate effects of an anticipated shock to public investment.
• Combine stock market info and news.• Use VAR‐IV
• The identified shock has a positive and significant impact on GDP.
Impact elasticity = 0.2‐0.3→ Impact multiplier =2‐6! (too large?)
39
Thank you!Your comments welcome!
40
Appendix 1Details about the news indicator
41
Literature (1) News‐based approach
• Ramey & Shapiro (Carnegie 1997), Ramey (QJE 2011): news about future US military spending.
• For Japan: Fukuda & Yamada (JJIE 2011): News on Emergency Fiscal Stimulus Packages.
• Drawback = No sense of magnitude or surprise
42
Literature (2) Stock based approach
• Fisher & Peters (EJ 2010)–Excess return on four large military contractors in the US.
• Drawbacks = They are Contaminated signals.
• Morita (Ph.D. thesis, 2014)– Excess returns of the Construction Industry for Japan.– “Purified”measure based on SVAR.
43
[1] News Analysis side: List of FP events
1. Extension of the Fukuda‐Yamada list of Emergency Stimulus Measures beyond 2010.
2. Reconstruction Budget after the Great East Japan Earthquake.
3. Important National Elections.
4. Natural Disasters (three earthquakes and a tunnel collapse).
5. Future Sports Events (Nagano, World‐cup, Tokyo)
6. “Negative” Fiscal Events (Hashimoto reform, Koizumi reform, “Shiwake”).
Identified 38 FP events; 159 dates.
44
[2] Stock market side
• Original data: Construction industry’s 177 firms, listed on Tokyo Stock Exchange (1st or 2nd), at some point between 1974 and 2014.
• Returns = log difference of the close price.
• We regress them on the Market (TOPIX) return to obtain excess returns.
• Are they really informative? Let’s see…
45
(a) Great East Japan Earthquake (March 14‐15, 2011)
0 20 40 60 80 100 120 140 160 180-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
Ranking based on the total market value as of 2012 (if present). 46
Excess returns by firm
(b) Sasako Tunnel Failure (December 3‐5, 2012)
0 20 40 60 80 100 120 140 160 180-0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
47
(c) IOC gives the Olympics 2020 to Tokyo (Sept 9‐11, 2013)
0 20 40 60 80 100 120 140 160 180-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
48
(d) FIFA gives World Cup 2002 to Korea/Japan (June 3, 1996)
0 20 40 60 80 100 120 140 160 180-0.2
-0.15
-0.1
-0.05
0
0.05
49
(e) “Shiwake” (Nov 10‐27, 2009)
0 20 40 60 80 100 120 140 160 180-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
50
How do we combine the two sides?
• Take a simple average? • But it may reflect all sorts of things.
• Instead, we take advantage of within‐industry heterogeneity.
• From here, data is limited to 76 firms that existed throughout the period 1990‐2014.
51
52
0%2%4%6%8%
10%12%14%16%18%20%
HIGH gov dependence group (>34%) LOW
IOC announces Tokyo to hold the Olympics Gamessum of excess returns, Sept. 8‐10, 2013
Appendix 2Factor loadings
53
54
‐0.4
‐0.2
0
0.2
0.4
0.6
0.8
1926
1929
1950
1822
1846
1805
1899
1898
1882
1938
1835
1814
1969
1881
1896
1944
1803
1883
1819
1959
1847
1983
1834
1925
1964
6366
Factor1 Factor loadings
More gov dependent Less gov dependent
55
‐0.4
‐0.2
0
0.2
0.4
0.6
0.819
26
1929
1950
1822
1846
1805
1899
1898
1882
1938
1835
1814
1969
1881
1896
1944
1803
1883
1819
1959
1847
1983
1834
1925
1964
6366
Factor2
Taisei Kajima ObayashiShimizu
Daiwa
56
‐0.4
‐0.2
0
0.2
0.4
0.6
0.819
26
1929
1950
1822
1846
1805
1899
1898
1882
1938
1835
1814
1969
1881
1896
1944
1803
1883
1819
1959
1847
1983
1834
1925
1964
6366
Factor3
Sata
Appendix 3More IRFs
(X1= High‐Low)
57
X5 = Real consumption
58
X5 = Real Business Investment
59
X5 = GDP Deflator
60
X5 = Nominal GDP
61