Embed Size (px)
Citation preview
Discussion onMiyamoto, Nguyen and Sergeyev
“Government Spending Multipliers under the Zero Lower Bound: Evidence from Japan”
Etsuro Shioji(Hitotsubashi)
JES 2017, February 10
Major comment:This is a great paper with many good elements, but most importantly,
it shows…
… that the Japanese economy, in many ways, is a pioneer!
The world should try to learn more from the Japanese experiences.
More concretely: what this paper does
• Big debate over the size of the G multiplier at ZLB among macroeconomists.
• This paper takes advantage of the fact that only Japan has spent a long time at ZLB.
• This allows the authors to compare the multipliers between the ZLB and the noZLBperiods.
4
Other major strengths of the paper
• Use of the advanced time series technique (projection method).
• Careful studies on the robustness.
• Close linkage between theory and empirics (quantitative assessment)
• But the rest of my discussion will focus mostly on one aspect of the paper.
5
“Fiscal foresight” = Common challenge in the estimation
of the effects of G
Most of G is announced in advance.= Using the actual amount of G can be
misleading.
“News” approach Stock market based approach
6
Reactions in the literature
Shioji and Morita (JES2015) proposed a way to combine the two.
This paper’s response to the problem is different from either of the two.
• Use the published forecasts as people’s expectations.
• Hence, knowing the nature of this forecast holds a key to evaluating this paper.
7
Use JCER’s short‐term forecasts Great idea!
JCER forecasts
• In what follows, I will focus on GI (Public Investment).– I did not have access to GC (Government Consumption) data.
• I will look at how forecasts and the “actual” estimates, which keep getting updated, for a particular point in time (i.e., a quarter) evolved over time.
8
How does the JCER forecast look like?Case 1: In the aftermath of
the Great East Japan Earthquake
9
Forecasts and Estimates for Q3 2011change in GI from previous quarter, real
‐4.0
‐3.0
‐2.0
‐1.0
0.0
1.0
2.0
3.0
4.0
5.0
date the estimate was published
forecast actual actual, end of 2016
10
… then Kan disappoints!Big buildup of expectations after the
earthquake.
Prime minister Kan’srecovery spending
• Earthquake: March 11, 2011
• Supplementary budget –Mark 1 : May 2 = 4 trillion JPY
–Mark 2: July 25 = Expected 10 trillion; Got 2 trillion instead.
11
12
‐0.1‐0.05
00.050.1
0.150.2
0.250.3
0.350.4
high, March‐mid May
low, March‐MidMay
high, mid May‐July
low, mid May‐July
Excess stock returns, construction firms"High"=High dependence on Gov
"Low" = Low dependence
Lesson
• This kind of event is perceived as a negative unexpected shock to GI in this paper.
• Correctly, I think.
• This example demonstrates usefulness of the paper’s approach.
13
14
How does the JCER forecast look like?Case 2: The rise of the Abenomics
Forecasts and Estimates for Q2 2013change in GI from previous quarter, real
‐6.0
‐4.0
‐2.0
0.0
2.0
4.0
6.0
8.0
date the estimate was published
forecast, nowcast actual actual, end of 2016
15
Huge expectations after the election!
But Abe’s bold approach beats
even this optimism!Or.. did he?
Lesson
• Data on GI is updated very frequently.
• Even after a few years.
16
question
• What is JCER trying to forecast?– Volume of GI that is perceived to be the reality at that time?
– Or is it going for the “true” reality (= final estimate?)??
• …and, which one is more relevant for this analysis??– Is it people’s perception about the amount of money that move people? or is it the true amount itself?
17
One suggestion
• Nominal G is easier to forecast and to estimate (in real time).
• Deflator for G is more difficult.
• If data availability allows, you might want to forecast them separately and then combine the two.
18
Minor Comments
• Why focus on the effects of unexpected G?–Most G is expected, and we want to know its impact.
• This paper uses G =GC+GI. Should we treat GC and GI separately?– Fig 13 of the paper shows that the IRFs of GI and GC to G shock are very different.
19
continued
• Unfortunately, there is not much going in and out of ZLB. – noZLB = before 1995– ZLB = after 1995– Many things could be different between those two. – How do we know it’s noZLB vs ZLB that’s making the difference?
This leads me to the next comment…
20
continued
• Related literature, mostly by Japanese researchersMostly about decreasing impact of G.
– Productivity effect of KG.
– Structural changes in the impact of G• Ihori, Nakazato & Kawade (2003): VAR• Morita (2015?): regime switching VAR
21
continued
• Is all the ZLB periods alike? For example, perceived reactions of the “shadow rate” could be different across, say, different BOJ Governors.
• FP‐MP interactions: When debt is huge, does it change MP’s reaction to FP? (debt monetization?)
22
AppendixJCER forecasts of future Public
Investment
Next plots show how forecasts and estimates
for a particular Quarter have evolved over time.
• I focus exclusively on Public Investment, as I did not have access to data on Government Consumption.
• Data is real, seasonally adjusted, rate of change from previous quarter (in%).
24
Reaction to the Great East Japan Earthquake, March 2011.
Forecasts and Estimates for Q2 2011
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
9.0
date the estimate was published
forecast actual actual, end of 2016
26
Forecasts and Estimates for Q3 2011
‐4.0
‐3.0
‐2.0
‐1.0
0.0
1.0
2.0
3.0
4.0
5.0
date the estimate was published
forecast actual actual, end of 2016
27
Forecasts and Estimates for Q4 2011
‐6.0
‐4.0
‐2.0
0.0
2.0
4.0
6.0
8.0
10.0
date the estimate was published
forecast actual actual, end of 2016
28
Forecasts and Estimates for Q1 2012
‐2.0
0.0
2.0
4.0
6.0
8.0
10.0
date the estimate was published
forecast actual actual, end of 2016
29
Forecasts and Estimates for Q2 2012
‐3.0‐2.0‐1.00.01.02.03.04.05.06.07.08.0
date the estimate was published
forecast actual actual, end of 2016
30
Forecasts and Estimates, cumulative, Q2 2011 – Q4 2012
‐10.0
‐5.0
0.0
5.0
10.0
15.0
20.0
date the estimate was published
forecast, nowcast actual actual,, end of 2016
31
Reaction to Abe’s victory, December 2012.
Forecasts and Estimates for Q1 2013
‐4.0
‐3.0
‐2.0
‐1.0
0.0
1.0
2.0
3.0
4.0
5.0
date the estimate was published
forecast actual actual, end of 2016
33
Forecasts and Estimates for Q2 2013
‐6.0
‐4.0
‐2.0
0.0
2.0
4.0
6.0
8.0
date the estimate was published
forecast, nowcast actual actual, end of 2016
34
Forecasts and Estimates for Q3 2013
‐4.0
‐2.0
0.0
2.0
4.0
6.0
8.0
date the estimate was published
forecast, nowcast actual actual, end of 2016
35
Forecasts and Estimates for Q4 2013
‐5.0
‐4.0
‐3.0
‐2.0
‐1.0
0.0
1.0
2.0
3.0
date the estimate was published
forecast, nowcast actual actual, end of 2016
36
