1 Sources of Inflation Dynamics in Japan: AS-AD and Monetary Policy Effectiveness Seiji Komine, University of Wisconsin-Madison Department of Economics Abstract Using a simple macroeconomic model, this paper inspects the driving factors of deflation long observed in Japan, and accordingly, it assesses the effectiveness of monetary policy, including the very recent policy instruments the BOJ has employed, such as Quantitative Easing (QE), Yield-Curve- Control, and negative interest rates. More specifically, I decompose the historical inflation dynamics into four components by sources, using identified structural residuals from a SVAR model of AS-AD framework with two monetary policy benchmarks: Taylor rule and Fisher’s equation of exchange. Then, I discuss lessons for monetary policy, looking at the coordination of the impulse responses and structural shocks folded inside the historical decomposition. The empirical results suggest the following inferences. 1) Long-run inflation dynamics is pushed by AS shocks, while short-run fluctuations are motivated by other structural deviations. 2) QE is now possibly getting excessive and losing its efficiency, while monetary easing by interest rates performs relatively well. This claim is reasoned from those observations: 2-a) It is likely there exist effective channels (significant impulse responses), respectively for both of the policy instruments. When sizes of impulses in each type of policy are sufficient to stimulate the system or enough large according to the market surroundings, the inflation rates response significantly; 2-b) However, that does not necessarily mean all the recent BOJ’s trials are effective. The variances in structural money supply shocks are decreasing in these years, reflecting that the BOJ is losing control on money supply in the saturated monetary environment. * This paper is written for the semester work for Econ706, Fall 2017 at UW-Madison. I would like to thank Professor Bruce Hansen at UW-Madison for helpful comments. All analysis is made with the software EViews 10.
28
Embed
Sources of Inflation Dynamics in Japan: AS-AD and Monetary ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Sources of Inflation Dynamics in Japan: AS-AD and Monetary Policy Effectiveness
Seiji Komine, University of Wisconsin-Madison
Department of Economics
Abstract
Using a simple macroeconomic model, this paper inspects the driving factors of deflation long
observed in Japan, and accordingly, it assesses the effectiveness of monetary policy, including the
very recent policy instruments the BOJ has employed, such as Quantitative Easing (QE), Yield-Curve-
Control, and negative interest rates.
More specifically, I decompose the historical inflation dynamics into four components by sources,
using identified structural residuals from a SVAR model of AS-AD framework with two monetary
policy benchmarks: Taylor rule and Fisher’s equation of exchange. Then, I discuss lessons for
monetary policy, looking at the coordination of the impulse responses and structural shocks folded
inside the historical decomposition.
The empirical results suggest the following inferences. 1) Long-run inflation dynamics is pushed by
AS shocks, while short-run fluctuations are motivated by other structural deviations. 2) QE is now
possibly getting excessive and losing its efficiency, while monetary easing by interest rates performs
relatively well. This claim is reasoned from those observations: 2-a) It is likely there exist effective
channels (significant impulse responses), respectively for both of the policy instruments. When
sizes of impulses in each type of policy are sufficient to stimulate the system or enough large
according to the market surroundings, the inflation rates response significantly; 2-b) However, that
does not necessarily mean all the recent BOJ’s trials are effective. The variances in structural money
supply shocks are decreasing in these years, reflecting that the BOJ is losing control on money
supply in the saturated monetary environment.
* This paper is written for the semester work for Econ706, Fall 2017 at UW-Madison.
I would like to thank Professor Bruce Hansen at UW-Madison for helpful comments.
All analysis is made with the software EViews 10.
2
1. Introduction
1.1. Background story
“Japanese economy is kind of mystery.”
Deflation, for a long time beyond the last couple of decades (“Lost Decades”), Japanese economy
has experienced. I do not know how many discussions on this topic were made by outstanding
economists in and outside the country. Still it does not seem that we have achieved consensus on
several related questions.
Why the deflation survives for such a long while, nevertheless economy keeps the moderate output
growth (Fig. 1 and 2)? What is the driver of deflationary development (AS vs AD, or others)? Is
deflation depressing? Can we say monetary policy is still effective, even though we are not quite
sure about the sources of deflation?
Besides the discussions, the policy menu of the Bank of Japan (BOJ) is like “taking everything,”
among untraditional ones: Quantitative Easing (QE), Commitment, Yield-Curve-Control, negative
interest rate. Off course, here are controversies.
Before estimating a system, to certify I have a covariance stationary vector process, I investigate the
time-series properties of each variables. As a result, I will take first differential formation
𝑥𝑥 ≡ [∆𝑦𝑦,∆p,∆r,∆m]′ in the SVAR estimation.
4.1. Unit root and cointegration tests
Table 1 summarizes the outcomes of unit root tests for level variables. At first, I examine unit root
of each level variable by Augmented Dickey-Fuller test (1979). In the table, there are shown both
results from tests with trend and without trend. Shortly, we can say m is nonstationary. On the
other hand, hypothesis of r’s unit root is rejected when a time trend is included. I interpret r is
trend-stationary process (TSP), supported by that the coefficient on the trend term was significant.
Anyway, I use Δr in SVAR analysis to eliminate the linear trend. Series y and p need an additional
step. ADF tests with trend did not reject unit roots, while those with trend did, suggesting
possibilities of non-stationarity. The powers of those tests, however, are presumably low, because
estimators on lagged terms were very close to 1 (0.99 for y, 0.96 for p). Thereat, I employ
Kwiatkowski-Phillips-Schmidt-Shin (1992) (KPSS) tests for y and p in a supplemental way, resulting in
rejections null hypothesis that they are stationary3. Thus, a plausible view is that the two series are
family of random walk. Also, it is confirmed there are no unit roots in first differences of all variables
by ADF tests (Table 2).
After checking unit roots, possibilities of cointegrations are tested, with Johansen and Juselius’s
(1990) maximum eigenvalue statistics. As reported in Table 3, the system is not likely a cointegrated
process. In this test, four lags are chosen by AIC, also in the forthcoming SVAR estimation (Table 4).
So, again, the system to estimate below is of 𝑥𝑥 ≡ [∆𝑦𝑦,∆p,∆r,∆m]′ with four lags.
3 Note that the design of null and counter hypothesis in KPSS test is reversed to Dickey-Fuller.
7
Table 1
Table 2
Level variables: Unit Root Test (lags selected by SIC, max 4)ADF test KPSSTrend No Trend Selected Lags
m -2.843 0.446 0 -
r -5.003 * -0.883 0 -
y -1.190 -4.842 * 0 0.412 *
p -3.085 -3.484 * 4 0.368 *
* denotes null rejection by 1% level.ADF null: the series has a unit root. Critical values by MacKinnon (1996) one-sided.KPSS null: the series is stationary. Critical values by Kwiatkowski-Phillips-Schmidt-Shin (1992) .
First differences: Unit Root Test (lags selected by SIC, max 4)Trend No Trend Selected Lags
⊿m -7.770 * -7.783 * 0
⊿y -12.668 * -11.365 * 0
⊿r -10.852 * -10.846 * 0
⊿p -4.084 * -4.002 * 3
* denotes null rejection by 1% level. Critical values by MacKinnon (1996) one-sided.
8
Table 3
Table 4
Johansen Cointegratin Rank Test (Max-Eigenvalue Stat, lag=4 by AIC)# Coint 1) No Trend in Cointegration Equation 2) Trend in Cointegration Equation
Max-Eigen Stat 5% Critical Value Max-Eigen Stat 5% Critical ValueNone 18.540 27.584 23.064 32.118At most 1 15.777 21.132 18.200 25.823At most 2 12.284 14.265 13.623 19.387At most 3 1.362 3.841 8.687 12.518Critical values by MacKinnon-Haug-Michelis (1999) .
The very key part of SVAR method is identification strategy. Basically, I follow Gali (1992), whose
approach captures well the nature of AS-AD mechanism4, i.e., I combine the long-run and the short-
run restrictions, as summarized in Table 5.
For the system of four variables, I am estimating 4×4 coefficients matrices (conditionally on a
matrix of normalized orthogonal shocks), thus there are six arbitrary constraints, or six linear
independences, required to just-identify the structure.
Table 5
The long-run restrictions R1-R3 are what distinguish AS variances, associated with the theoretical
guidepost, a vertical long-run Philips curve.
For three additional constraints, I employ the short-run restrictions R4-R6. They are reasoned with
lags for monetary policy to affect other parts of economy. Blown by changes in interest rate or
money supply, GDP within the quarter hardly responses, since there would be idle time before
recognitions, judgements and adjustments of production schedules and investment plans. As well,
price stickiness is almost a stylized fact. I pick MS shock rather than MD shock on p’s slowness, since
velocity is supposed to be much flexible and to absorb by itself the short-run fluctuations on
Fisher’s exchange.
Off course, from the technical viewpoint, there are other possible combinations of restrictions.
4 Blanchard and Quah (1989) are ones of originators of AS-AD identification in VAR with the long-run restrictions. Besides Lippi and Reichlin (1993) made a criticism for this approach, Faust and Leeper (1997) summarized the conditions where it may/may-not be plausible. Stock and Watson (2001) noted their general appetite to VAR inferences. In this paper, I take the first differences of endogenous, which enables to avoid some criticisms. And I believe the prior that technology is the anchoring factor of the long-run production is the common among many economists.
Identifying RestrictionsLong-run restrictionsR1: no long-run effects of AD shocks on y.R2: no long-run effects of MD shocks on y.R3: no long-run effects of MS shocks on y.Short-run restrictionsR4: no contemporaneous effects of MD shocks on y.R5: no contemporaneous effects of MS shocks on y.R6: no contemporaneous effects of MS shocks on p.
10
However, from qualitative priors, it is not so plausible to have constraints on reactions of r and m,
since the financial market adapts very quickly to surprises.
4.3. Econometrical expressions
The VAR system has several formularizations. As noted, our system is of 𝑥𝑥 ≡ [∆𝑦𝑦,∆p,∆r,∆m]′,
whose dynamics are driven by structural disturbances 𝜀𝜀 ≡ [𝜀𝜀𝐴𝐴𝐴𝐴, 𝜀𝜀𝐴𝐴𝐴𝐴, 𝜀𝜀𝑀𝑀𝐴𝐴, 𝜀𝜀𝑀𝑀𝐴𝐴]′, which are mutually
The SVAR with the noted specifications reports the structural shocks as in Fig. 4-Panel A (left
column). It is seen that this stochastics is zero-mean, stable variance process, as tested by Ljung-
Box’s Q-stats as in Table 6. Fig. 4-Panel B (right column) plots the smoothed version by ETS
exponential smoothing (Hyndman, et al., 2002)5, which is for a supplemental reference to grab their
historical characteristics (Note that their appearances are relying upon transitions of the mid-term
mean, so just focus the big picture of momentum6). AS deviation dropped sharply at two oil-shocks
in the 70’s and the credit crisis in 1991-1993. AD shock grew rapidly in the 70’s, but it slowed down
in the late 80’s, corresponding to the end of the financial boom, and stagnated during the Lost
Decades. We can also find the collapses in 2001 and 2009. MS showed similar movement as AD, but
it became flatter after declining by the 90’s credit crunch. Meanwhile, MS behaved somehow
oppositely to MS, losing its initial height through the bubble economy.
Table 6
5 Spec: additive, no trend and no seasonality. 6 It is understandable if it is thought the appearances of Panel B invokes they would have a sort of
mid-term trends. Technologically speaking, however, it does not necessarily imply that (by the nature of ETS smoothing method). Even if it does, the normality of residuals holds at least in the long horizon, as tested in Table 6. Anyhow, I use this here for a brief characterization, as mentioned.
14
Normality Tests for Identified Structural Shocks (Ljung-Box's Q-statistics )Shocks AS AD MD MS
Numbers in brackets are p-values (significant when non-normal).
15
Fig. 4 Identified structural shocks
Panel A: Identified version Panel B: Smoothed version
Panel A (left column) series are identified structural shocks, which are raw without any manupulations. Panel B (right coumn) series are smoothed by ETS exponential method (additive, no trend and no seasonality).
Since AS stagnation is usually untouchable for the central bank, there would be a limit for monetary
policy to overwhelm deflation. However, I do not mean by this that monetary policy is not totally
helpless. As we saw, the MD component in the decomposition is large. Noticing that the MD
component includes both of policy effects and market behaviors7, anyway, I am not hesitating to
say that monetary policy via interest rate could be effective. Nevertheless, the quantitative easing is
not a big deal. As mentioned before, there would be an active channel for QE, but structural
deviations are getting smaller these days, because BOJ is losing control of market liquidity.
Therefore, MS factor is not playing an important role these days in the historical decomposition,
despite the immense expansion of recent monetary base.
Fig. 13
7 MD shocks are composed not only by arbitrary policy disturbances, but also by structural deviations from the MD curve, made by any economic agents (The SVAR does not eliminate them, by its nature). Thus, not all of the movements with MD factor in the historical decomposition derives from monetary policy.