• No results found

The effect of quantitative easing on expected inflation in Japan

N/A
N/A
Protected

Academic year: 2021

Share "The effect of quantitative easing on expected inflation in Japan"

Copied!
23
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

The effect of quantitative easing on expected

inflation in Japan

ABSTRACT

Japan was the first country in the world to apply the unconventional monetary policy, Quantitative Easing (QE) to stimulate the stagnating economy and fight the deflationary pressure. This research focuses on the current more extensive QE policy that is in place since April 2013. Former research does not suggest a strong effect of the first QE policy on economic activity and the price level. But the literature does suggest that a monetary expansion can be effective in altering inflation expectations when a credible commitment is made by the central bank. Some literature suggests that the effectiveness of QE can be improved by financial reforms, since the unsolved banking problems could have been an explanation for the insignificant effect of first QE policy. The empirical analysis performed does not suggest a strong relationship between QE and expected inflation. Although it should be taken into account that there might be a reliability issue arising concerning the estimated model, caused by an endogeneity problem and the use of a weak instrument.

Name: Adriaan Tax

Student number: 5690870

University: University of Amsterdam

Faculty: Economics and Business

Field: Macro-economics

Supervisor: Mr C.G.F. van der Kwaak MSc

(2)

Table of content

1 INTRODUCTION ... 3 2 LITERATURE REVIEW ... 5 2.1 QE ... 6 3 EMPIRICAL STUDY ... 10 3.1 DATA ... 10 3.2 METHOD ... 11 3.3 RESULTS ... 12 3.3.1 ENDOGENEITY ... 14 3.4 DISCUSSION ... 18 4 CONCLUSION ... 20 5 REFERENCES ... 22 6 APPENDIX ... 23

(3)

1 Introduction

In response to the economic downturn triggered by the bursting of the housing bubble early 90’s, Japan was the first country in the world to apply the unconventional monetary policy tool quantitative easing (QE) on 19 March 2001 (Shiratsuka, 2010). By this time Japan had suffered from a decade of declining prices and a stagnating economy. The bursting of the housing bubble and the unsolved banking problems in the early 90’s were the events that caused the economic turmoil leading up to this, so-called, lost decade (Schenkelberg & Watzka, 2013). They stated that in 1995 the Bank of Japan (BOJ) decreased the interest rate to 0 % to stimulate the economy, but this was insufficient to let the economy recover. In the following period the policy rate stayed close to zero, but the economy kept struggling. Japan was caught in a, so-called, liquidity trap where monetary policy is ineffective from the traditional point of view (Krugman, Dominquez, & Rogoff, 1998). But, Krugman et al. (1998) believe that the traditional view must be qualified: monetary policy will in fact be effective if the central bank can credibly promise to be irresponsible to seek a higher future price level. Japan reaching the zero lower bound (ZLB) eventually led to implementing the QE policy in 2001 (Schenkelberg & Watzka, 2013). The BOJ changed their main operating target from the uncollateralized overnight call rate to the current account balances (CABs). The overnight call rate is the base interest rate banks charge when lending out short maturity loans, with no collateral needed, to each other. This base interest rate has a great impact on the interest rates banks charge on commercial products. The CABs the BOJ targeted were the outstanding reserves balances the commercial banks held by the BOJ. So, the BOJ changed their main operating target from the conventional interest rate target to the unconventional reserves-target. In March 2001 the target of the CAB was first set at ¥5 trillion and was throughout the years progressively increased to ¥30 trillion to ¥35 trillion in January 2004 up and until the policy was exited in 2006 (Ugai, 2006). In this way the BOJ was increasing the reserves far in excess of the required ratios. After the core consumer price index (CPI) growth turned positive in November 2005 and was expected to remain positive, the BOJ exited the QE policy and changed their main operating target back to the uncollateralized overnight call rate (Ugai, 2006). Schenkelberg and Watzka (2013) say that while the Japanese quantitative easing policy was effective in temporarely stimulating the real economic activity, it did not lead to a persistent increase in inflation.

(4)

At the end of 2012 Shinzō Abe took office as prime minister. Shortly after taking office he executed the, so-called, “Abenomics” by introducing the Quantitative and Qualitative Monetary Easing (QQE) in April 2013. In this way he wanted to make an end to the stagnating economy and deflationary pressure Japan was still suffering from. This QQE was a much more extensive and aggressive QE policy with respect to the earlier QE policies applied in Japan (Bank of Japan, 2013). The BOJ set a price stability target of 2% in terms of the year-on-year rate of change in the CPI at the earliest possible time with a time horizon of about 2 years. In the outline of this policy the following was stated: “The Bank will continue with the QQE, aiming to achieve the price stability target of 2 percent, as long as it is necessary for maintaining that target in a stable manner.”. In October 2014 they even accelerated the pace of increasing the monetary base. The amount of asset purchases was increased from ¥60 trillion to ¥70 trillion per year up to ¥80 trillion per year (Bank of Japan, 2013). This indicated that the Japanese were serious. The research question of this thesis therefore will be:

Does the aggressive QE policy of Japan have a significant effect on expected inflation?

This thesis will be set up in the following way; the second section consists of a literature review addressing the most important literature regarding quantitative easing and the Japanese economic background. In the third section a quantitative research will be conducted. First, all the relevant variables will be discussed. Second, a model will be estimated by use of an OLS regression. Furthermore, several statistical tests will be performed to evaluate the model and to outline the results. This research will finish with a conclusion.

(5)

2 Literature review

During the 1960s the Japanese economy grew fast. With an average GDP growth rate of 10% a year until the 1970s and an average GDP growth rate of 5% a year between 1970 and 1980, the Japanese became the second largest world economy based on Gross Domestic Product (GDP) (World Development Indicators, 2015). A ‘bubble economy’ was emerging from the later part of the 1980s when the asset prices were rapidly rising. This accompanied with overheating economic activity and a sizable increase in money supply and credit caused an economic bubble to emerge (Okina, Shirakawa, & Shiratsuka, 2001). This economic bubble burst at the beginning of 1992 and caused an economic slowdown that lasted almost 3 years (Okina, Shirakawa, & Shiratsuka, 2001). This bursting of the economic bubble was the event leading Japan into the lost decade during the 1990s. In 1995 Japan lowered the policy interest rate practically to zero to stimulate the economy, but it became clear to be insufficient to let the economy recover.

Japan got stuck at the ZLB, where the interest rate is (close to) zero, and ended up in a liquidity trap were monetary policy is ineffective. The interest rate being this low from 1995 on, made it impossible to use the interest rate as a macroecnomic stimulus because it simply could not be lowered any further. Being caught at the ZLB, makes the interest rate instrument unavailable therefore (Eggertson & Woodford, 2003). Krugman et al. (1998) state that the Japanese were dealing with a credibility problem. Contrary to the usual credibility problem, where the central bankers have problems convincing the economic agents of their commitment to price stability, the economic agents will expect the central bank to target the price stability in a liquidity trap. Therefore a monetary expansion is expected to be merely transitory and is ineffective from the traditional point of view (Krugman, Dominquez, & Rogoff, 1998). They state that the monetary policy will in fact be effective if the central bank can credibly promise to be irresponsible to seek a higher future price level. If monetary policy is ineffective, it must be because the public does not believe the current monetary expansion will be sustained. If the central bank can convince the public that it will allow prices to rise more than sufficient, monetary policy can get the economy out of the liquidity trap (Krugman, Dominquez, & Rogoff, 1998). So, the credibility of the central bank while applying monetary policy under these circumstances is really important. Otherwise only a temporary monetary expansion is expected and this is found to be ineffective.

(6)

After it became clear that the lowering of the policy rate was not sufficient to let the economy recover, the BOJ introduced their Zero Interest Rate Policy (ZIRP) in 1999. The zero interest rate would be kept in place until deflationary concerns were dispelled (Schenkelberg & Watzka, 2013). After signs of recovery the ZIRP was dispelled in August 2000, but the following of the bursting of the global IT bubble caused a setback for the Japanese economy.

2.1 QE

Due to the setback caused by bursting of the global IT bubble, the QE policy was implemented in March 2001. The way how QE is expected to affect inflation expectations is based on the credibility theory as described by Krugman et al. (1998). He uses a monetary neutrality argument, being: an increase in the money supply in the current and all future periods will raise prices in the same proportion. So, if a monetary expansion is expected to be sustained it will affect expected inflation. One of the difficulty for the BOJ lies in the management of expectations of economic agents, being that the monetary expansion will be sustained. By implementing the QE policy in March 2001 the BOJ seemed to be giving a credible sign and was therefore strengthening the commitment of getting Japan out of the stagnating economy and fight the declining prices.

Ugai (2006) summarized the following three pillars which the QE policy consisted of; (i) shifting the BOJ’s main operating target from the uncollateralized overnight call rate to the current account balances (CABs) at the BOJ, and supplying ample liquidity in an amount substantially in excess of the required reserves; (ii) being committed to maintain the policy until the CPI registers stably zero percent or an increase year-on-year; and (iii) increasing the purchase of long-term JGBs if deemed necessary to facilitate meeting the targeted CABs. By explicitly targeting the CABs, the BOJ was increasing the reserves held by the central bank (R), which, combined with the currency in circulation (C), make the monetary base (MB). Whether the increase in the MB causes an increase in the money supply, being currency in circulation (C) and total deposits (D), depends on the creation of deposits by the banks. Because an increase of a monetary base solely will not cause an increase in prices (Mishkin, 2010). Whether an increase in the money supply will cause inflationary pressure depends, as described above, on whether the monetary expansion is believed to be permanent.

An often used concept to identify to what extent the money supply will be increased by an increase of the MB is through the money multiplier. The relationship between the money

(7)

supply (M), the money multiplier (m) and the monetary base (MB) is described as follows by Mishkin (2010):

𝑀𝑀 = 𝑚𝑚 ∗ 𝑀𝑀𝑀𝑀

So, a raise in the monetary base will raise the money supply by a multiple of this monetary base expansion, the money multiplier. The money multiplier is a function of the currency ratio c, the excess reserves ratio e, and the required reserve ratio r:

𝑚𝑚 =r + e + c1 + 𝑐𝑐

The required reserve ratio is set by the central bank. This is the percentage of the deposits of the commercial bank that must be held as reserves at the central bank. The percentage of their deposits that the commercial banks hold at the central bank in excess of the required ratio is the excess reserve ratio. The currency ratio is the fraction of deposits held as currency by the people (Mishkin, 2010).

Ugai (2006) concludes in his empirical research that QE had an accommodative monetary environment preventing the worsening of the Japanese economy from the macroeconomic point of view. But, he found that the effect of the QE policy under the ZLB constraint on aggregated demand and prices was small or negligible. Schenkelberg and Watzka (2013) found similar results in their empirical research. They said that the QE shock was successful in temporary stimulating real economy, but was not able to stimulate prices.

Ueda (2012) concludes that the negligible effect of QE stimulating the economy was caused by a corresponding decline of the money multiplier during periods of expansion of the monetary base. The money supplied through the QE operations was held by the financial institutions instead of being used to increase deposits. The Japanese banks did not trust each other to make interbank loans (Ueda, 2012). This is in line with what Schenkelberg and Watzka (2013) found in their research, being that the monetary base expansion did not lead to sufficient increase in the money supply and could therefore not impact inflation. Ueda (2012) also states that the stagnating of the Japanese economy was partly due to mismanaged macroeconomic policy. The Japanese policymakers were found not to be aggressive enough during the bursting of the economic bubble early 1990s. This made the Japanese economy vulnerable. Japan reaching the ZLB made the interest rate instrument

(8)

unavailable, since it could not be lowered any further. On top of that there was deflationary pressure, which combined with strong expectations of low-growth, made the conventional and unconventional monetary policy highly constraint (Ueda, 2012). This may be an explanation why the effect of the first QE policy on stimulating prices and economic activity was small or even negligible. On top of that, the exiting of the policy in 2006 with a low inflation rate was questioned by the financial markets whether the Japanese Central Bank was really committed to resolving the deflationary pressure and stagnating economy. So, the credibility of the BOJ concerning it was committed to achieving stimulating the economy and prices seemed to have played a role in the effectiveness of the QE policy (Ueda, 2012). Hetzel (2003) thinks that the BOJ could end deflation by raising the money growth more significant than during the first QE period. He argues that the monetary base expansions under the first QE period, up and until he wrote his research in 2003, were not sufficient to have a substantial effect. He thinks that the BOJ should explicitly target the price level and use a more substantial monetary base expansion as a policy instrument he says. In this way the announcement of a commitment to stabilize prices accompanied by a more pronounced monetary base expansion could fight deflationary expectations (Hetzel, 2003). This is in line with what the BOJ did when they announced the QQE in April 2013, since they explicitly set an inflation target of 2 percent and announced a more extensive monetary base expansion. This might be seen as a credible commitment by the BOJ.

Girardin and Moussa (2011) conclude in their empirical research that QE can have a positive effect on the economy when combined with financial reforms. The policy needs to be maintained longer than the first QE policy was applied. Moreover they conclude that to deal with such a serious crisis it should not be expected to solve the problems in the short term since it should be applied long enough to have effect on the economy and the price level. Summarizing, the literature described above does not suggest a really strong relationship between QE and the expected inflation, although there is no literature available on the current QQE policy. The credibility of the commitment of the BOJ to stimulate economic activity and fight deflation seemed to have played a role in the effectiveness of the QE policy. The literature suggests that the problems in the financial sector should be solved as well to make the QE policy be more effective. The in latest October announced acceleration of the QQE policy really strengthens the commitment of fighting deflation. In this way the Japanese seem to be credibly committed to doing whatever it takes to stimulate the

(9)

economy and end deflation and could in fact be successful in doing so. In this way the inflation expectations could be altered. The inflation rate actually picking up during the first months of the current aggressive policy could be a cautious sign of a positive effect on inflation (see figure 2.1 of the appendix).

In the next section the quantitative analysis will be outlined to investigate whether there was a statistically significant effect on expected inflation.

(10)

3 Empirical study

In this section the performed empirical study will be discussed. First the dataset used is explained, followed by an outline of the method applied. At last the results are discussed, including a section about endogeneity, followed by a conclusion.

3.1 Data

Because it is hard to exactly identify inflation expectations, data will be taken from the World Economic Survey (WES) the CESifo Group conducts every quarter. The participants of this survey are economic experts of multinational firms and institutions, so a more rational, forward-looking formation of expected inflation could be expected. The annual inflation expectation in percentage points is taken from the WES. Because QE is a relatively new monetary policy tool that has only been applied since 2001, quarterly data is used to be able to capture enough observations. The available quarterly data on inflation expectations ranges from the first quarter of 1991 until the last quarter of 2014, so 96 observations are captured.

The independent variable QE represents the purchase of Japan Government Bonds (JGBs) with respect to the nominal GDP. In this way a distinction can be made between the magnitudes of the different asset-purchase programs.

Autocorrelation is a feature of time series data and means that what happens in one period tends to be correlated with what happens next period. To control for autocorrelation, a one-quarter lagged inflation expectations is taken into account in the regression.

The quarterly real GDP growth in percentage points will be taken into account to control GDP growth. When GDP growth is high, this may cause inflationary pressure in two ways based on the aggregate supply and aggregate demand model (AS-AD). First, if aggregate demand increases faster than aggregate supply firms can push up prices when they cannot meet demand. Second, high GDP growth causes unemployment to fall which causes upward pressure on wages. Eventually this will lead to rising prices.

The Philips-curve suggests a relationship between the unemployment rate and inflation. To control for this effect, the level of the unemployment rate in percentage points for all people between 15 and 64 years is taken into account.

(11)

The realized annual inflation may alter inflation expectations so, the level of the inflation is taken into account as well. The level of inflation is calculated as the percentage change of the Consumer Price Index (CPI).

The short-term interest rate is taken into account to control for inflationary pressure, caused by an investment stimulating and therefore a GDP stimulating effect of a decrease of interest rate.

The final control variable is the percentage change in the real effective exchange rate. A decrease in the real effective exchange rate causes exports to go up, but become cheaper and have less, but more expensive, imports. This should increase the net export and therefore be stimulating the GDP growth, which will create inflationary pressure.

The quarterly data on the above mentioned variables can be obtained from the Datastream database and are all stated in percentage points. From all the data used, the data on inflation expectations is the least available, so when all the data is matched a dataset containing 96 observations is available for the regression analysis.

3.2 Method

To investigate whether a quantitative easing (QE) policy does affect inflation expectations significantly, a regression analysis will be performed. The dependent variable will be the expected inflation (InfExp). The independent variable, QE. Furthermore the following control variables are added to the regression; the one-quarter lagged inflation expectations (InfExp t-1), the real GDP growth (RealGDP), the inflation rate (Inf), interest rate (Int), depreciation of real effective exchange rate (RealDepr), the unemployment rate (UE). The t subscript denotes the respective quarter. This results in the following equation:

InfExpt, year = β0 + β1 * QEt + β2 * InfExpt-1 + β3 * Inft + β4 * RealGDPt + β5 * Intt

+ β6 * RealDeprt + β7 * UEt + ut (3.1)

An OLS regression will be performed on the model described at 3.1. The robustness of the model will be tested by discarding insignificant control variables from the first OLS regression and check for changes of the coefficients and the significance of the relevant variables. A five-percent significance level is used in the estimated model.

(12)

3.3 Results

In the first column of table 3.1 the results of the OLS regression performed on equation 3.1 are summarized. The measure of fit represented by the R2 is 0.9126. So the regressors in this

model explain 91.26 percent of the variation in expected inflation. The variable of most interest, QE, is not significantly different from zero given the p-value being 0.127. So, from these results it can be concluded that QE is not having a significant effect on inflation expectations in this model.

Furthermore, a one-percent increase of the lagged inflation expectations increases the current inflation expectations with 0.454 percent. This effect is statistically significant. A one-percent increase of the inflation rate has a positive effect of 0.237 one-percent on inflation expectations, and an increase of the unemployment rate with 1 percent will decrease the inflation expectations with 0.202 percent. Both variables are significantly different from zero. These results are in line with what you would expect from economic theory described above. The short-term interest, real GDP growth and real depreciation rate are not statistically different from zero and therefore can be ignored.

Given that the coefficients on real depreciation, the short-term interest rate and the real GDP growth are not statistically significant, another OLS regression is performed to test the robustness of the model. First, real depreciation is left out of the equation. This results in the following equation:

InfExpt,year = β0 + β1 * QEt + β2 * InfExpt-1 + β3 * Inft + β4 * RealGDPt + β5 * Intt

+ β6 * UEt + ut (3.2)

The results of the second OLS regression performed on equation 3.2 are presented in the second column of table 3.1. Leaving real depreciation out of the model does not change the R2, still 91.26 percent of the variation in inflation expectations is explained by the included regressors. Furthermore, with respect to the significance of the variables everything stays the same compared to the first regression. Still, QE is not statistically significant. The effect of inflation, the unemployment rate, the interest rate and real GDP growth stay just about the same. The real GDP growth and the interest rate variables are still not statistically significant.

(13)

To complete the robustness test a third regression will be performed with real GDP growth left out of the equation instead of real depreciation, resulting in the following model:

InfExpt,year = β0 + β1 * QEt + β2 * InfExpt-1 + β3 * Inft + β4 * Intt + β5 * RealDeprt

+ β6 * UEt + ut (3.3)

The results of the third OLS regression are summarized in the third column of table 3.1. The R2 decreases a little compared to the previous regression. Still 91.20 percent of expected inflation is explained by the regressors in this model. With respect to the significance of the variables nothing changes. The coefficient on QE is still statistically insignificant. The coefficients on lagged inflation expectations, inflation and the unemployment rate have changed from 0.454 to 0.434, from 0.237 to 0.233 and from -0.201 to -0.214 respectively. The interest rate and real depreciation are still insignificant. Given all these minor changes and no changes in the significance of the regressors the model seems to be robust.

Table 3.1 OLS regression expected inflation

Regression 3.1 Regression 3.2 Regression 3.3

QE 0.0566086 0.0564189 0.0603225 (0.0367819) (0.0362822) (0.0363891) InfExpt-1 0.4532207** 0.4536587** 0.4337737** (0.0840125) (0.0828566) (0.0800295) Inf 0.2373113** 0.2374518** 0.2325145** (0.0478737) (0.0474793) (0.0473687) RealGDP 0.0250447 0.0248077 (0.0321842) (0.0314808) Int 0.0341615 0.0343007 0.0406137 (0.0353975) (0.0350339) (0.0343354) RealDepr 0.0002759 -0.0006646 (0.0067295) (0.0066052) UE -0.2018374** -0.2012717** -0.2138862** (0.0696161) (0.067847) (0.0677201) Constant 0.9822705 0.9796005 1.041068 (0.3286907) (0.3203429) (0.3191705) R2 0.9126 0.9126 0.9120 # observations 95 95 95

(14)

3.3.1 Endogeneity

The BOJ was committed to let the QE policy stay in place until the inflation was stable at zero or increasing. The QE policy was exited in 2006, because prices started to rise and it was expected the prices would stabilize. In other words, the expected inflation had an effect on QE as well. Besides this, the relationship between the nominal interest rate and the expected inflation described by Fisher suggests a simultaneous causality as well. This causality running both ways could cause a simultaneous causality bias, which could make the OLS estimator biased and inconsistent and could make the estimated model therefore unreliable. In other words, the regressors (QE and the interest rate) might be correlated with the error term and are called endogenous. To account for this problem a two-stage least squares (TSLS) regression in an Instrumental Variable model is applied. The TSLS estimates are calculated in two stages. First the endogenous variable is regressed on the instruments and the exogenous variables (all the control variable in this case). The first stage regression is as follows:

QEt,year = π0 + π1 * Inft-1 + π2 * InfExpt-1 + π3 * RealGDP%t + π4 * Inft + π5 * Intt

+ π6 * RealDeprt + π7 * UEt + vt (3.4)

In the second stage of the TSLS regression the first regression model (3.1) is estimated with the predicted value, 𝑄𝑄𝑄𝑄� , obtained from the first-stage regression, instead of the QE𝑡𝑡 t

variable used in equation 3.1:

InfExpt, year = β0 + β1 * 𝑄𝑄𝑄𝑄� + β𝑡𝑡 2 * InfExpt-1 + β3 * Inft + β4 * RealGDP%t + β5 * Intt

+ β6 * RealDeprt + β7 * UEt + ut (3.5)

With multiple endogenous variables each variable has its own first-stage regression, only including the endogenous regressor, the instrument(s) used and the exogenous variables (Stock & Watson, 2012, p. 475).

In this model the QE variable and interest rate variable are indicated as endogenous. Furthermore there are two instruments indicated. A valid instrument must satisfy two conditions. First, it must be correlated with the endogenous variable (instrument relevance) and second, it must not be correlated with the error term (instrument exogeneity) (Stock & Watson, 2012, p. 476).

(15)

The instrument used for QE is the one-period lagged inflation rate, because it is expected that the lagged inflation would have an effect on QE since in the second pillar of the QE policy is stated that the policy is maintained until the CPI becomes stable at zero percent or is increasing. The one quarter lagged inflation might not affect the inflation expectations in the estimated model, because the inflation expectations come from economic experts which, I think, tend to be more rational and mostly forward looking when forming expectations. For the interest rate a one-period lagged interest rate is used, because the lagged interest rate is expected to affect the current interest. Policy makers decide whether to let the interest rate rise or fall, among other things, based on the interest rate level that was set in a prior period. So, it can be expected that the lagged interest rate will be correlated with the current interest rate. When inflation expectations are formed the change in the interest might be more of interest than the level of the interest one period before, if the interest rate alters the inflation expectations at all. Again, the data on inflation expectations comes from economic experts which, I think, tend to be more rational and mostly forward looking when forming expectations. The tests described in the next section have to point out whether both instruments chosen satisfy both conditions for a valid instrument.

Through a Durbin-Wu-Hausman test endogeneity will be tested. Furthermore a first-stage F-statistic will be calculated to evaluate the chosen instruments relevance. Finally to test the exogeneity of the instruments an overidentifying test is performed by calculating a Sargan Χ2

score and Basmann Χ2 score.

The results of this IV (2SLS) regression are summarized in the first column of table 3.2. All variables are not statistically significant different from zero. The Durbin test is significant at a one-percent significance level, while the Wu-Hausman test is only significant at a five-percent significance level. Given these results it can be concluded that there is an endogeneity problem in this model. When there are multiple endogenous variables indicated, the test for weak instruments goes through Shea’s partial R2. This partial R2 measure takes the intercorrelation between the instruments into account (Baum, 2006, p. 208). The partial R2 values for the QE variable and the interest rate are extremely low, 0.0005 and 0.0042 respectively. On top of that the minimumeigenvaluestatistic is extremely low (0.0238584). The H0 hypotheses, that the instruments are weak, cannot be rejected. The

(16)

weak instruments could explain why all variables are not statistically different from zero and some of the variables having extremely high coefficients. Therefore, the model estimated may not be reliable (Stock & Watson, 2012, p. 480).

To look more detailed into the instruments used, more regressions are performed. In this way it can be concluded which of the two instruments used in the first regression is a weak instrument. First, a regression is performed with only the interest rate variable indicated as endogenous variable, using the one-quarter lagged interest rate as an instrument. In the second column of table 3.2 the results of the second regression are summarized. It can be concluded that the interest rate is not an endogenous variable in this model, since the endogeneity test fails. The H0 hypotheses, all variables beingexogenous, cannot be rejected,

the p-values of the Durbin and Wu-Hausman test being 0.3955 and 0.4193 respectively. The first-stage F-statistic is 2271.82 which indicates that the one quarter lagged interest rate is a reliable instrument since the first-stage F-statistic exceeds the rule of thumb value of 10. Next, a regression is performed with the QE variable indicated as an endogenous variable, using the one-quarter lagged inflation rate as an instrument. The results of this regression are summarized in the third column of table 3.2. Again, all the variables are not statistically significant different from zero just like the first 2SLS regression. The endogeneity test does not fail this time, which indicates that QE is indeed an endogenous variable. But, the first stage F-statistic is extremely low (0.045878) which means the instrument used is weak. The very low partial R2 (0.0005), which indicates the correlation between QE and the lagged

inflation after partialling out the effect of the other regressors in this model, indicates that the instrument relevance condition has not been satisfied.

To be able to test for instrument exogeneity a final regression is performed with QE indicated as an endogenous variable and with the lagged interest rate and lagged inflation rate as instruments. In this way an indirect test for instrument exogeneity can be performed by calculating a Sargan Χ2 score and Basmann Χ2 score. Two things are simultaneously tested by a test of overidentifying restrictions of which one is the instrument exogeneity for the indicated instruments. The other is whether the equation is misspecified. The results of the performed regression are summarized in the fourth column of table 3.12. It can been seen that the test of overidentifying is not significant (p-values are large) which implies that the instruments are indeed exogeneous and the model is not misspecified. So, the reason why

(17)

the lagged inflation rate is a weak instrument is solely because of an almost negligible correlation between the instrument and the endogenous variable. The instrument being weak could explain why the coefficients on all the variables in this model are not statistically significant, because using a weak instrument could make the estimated model unreliable and cause a size distortion on the coefficient as mentioned earlier in this section (Stock & Watson, 2012, pp. 461-497).

Table 3.2 IV (2SLS) regression expected inflation

IV regression 1 IV regression 2 IV regression 3 IV regression 4

QE -4.406642 0.0544899 -4.466527 -2.886422 (19.1597) (0.0352924) (20.272) (9.190894) Int -1.512489 0.0285413 1.384797 -0.9933568 (6.78669) (0.0345189) (4.30731) (3.220433) InfExp_1 1.361961 0.4580636** 2.966229 1.059361 (4.108102) (0.0806107) (12.24236) (2.015294) Inf 2.932162 0.2377236** -1.545029 2.012913 (11.85992) (0.0458229) (7.089983) (5.556161) RealGDP 0.5295068 0.0262417 0.5385854 0.3591859 (2.248098) (0.030836) (2.336761) (1.076317) RealDepr -0.102791 0.0003784 -0.1038087 -0.0674479 (0.4613152) (0.006442) (0.4740775) (0.218576) UE 1.567085 -0.204122** 1.583282 0.9596708 (7.836002) (0.0666844) (8.046876) (3.670703) _cons -3.025625 0.9959139 -3.041896 -1.6361 (18.15711) (0.3150019) (18.50436) (8.612192)

Instruments Inft-1, Intt-1 Intt-1 Inft-1 Inft-1, Intt-1

Endogenous variable(s) QE, Int Int QE QE

Durbin Χ2 score (p-value) (p = 0.0078) 9.70568** 0.722111 (p = 0.3955) (p = 0.0032) 8.70761** (p = 0.0057) 7.65621** Wu-Hausman F-statistic (p-value) (p = 0.0103) 4.8361* (p = 0.4193) .658708 (p = 0.0041) 8.6781** (p = 0.0074) 7.53842** (Shea's) Partial R2 QE = 0.0005 0.9633 0.0005 0.0011 Int = 0.0042 (minimumeigenvalue statistic = 0.0238584) First-stage F-statistic 2282.71 0.045878 0.047093 Overidentifying test: Sargan Χ2 (p-value) Basmann Χ2 (p-value) 0.027478 (p = 0.8683) 0.024882 (p = 0.8747)

(18)

Summarizing the results from the regressions described in this section gives: the interest rate is not an endogenous variable in this model, but QE is. The instrument used for QE, the lagged inflation rate, happened to have a very low correlation with QE. So, the instrument relevance condition is not satisfied, causing the 2SLS estimates to be unreliable and inconsistent.

3.4 Discussion

The variable of interest, QE, does not have a significant effect on expected inflation in the OLS regression described at 3.1. Furthermore, the model described seems to be robust given the tests performed. But, because the QE variable seems to be endogenous, there is also an endogeneity problem which might cause the OLS estimates to be unreliable and inconsistent. Therefore this outcome is not reliable. This problem is caused by a simultaneous causality bias, a causality running both ways from expected inflation to QE and the other way around. To control for this bias an IV (2SLS) regression is performed. Because of a very low correlation between the lagged inflation rate, used as an instrument in this regression, and the QE variable, the instrument relevance condition for a valid instrument has not been met. This makes the lagged inflation a weak instrument, causing the model estimated to be unreliable and inconsistent. The unreliability of the estimated models makes it hard to conclude something about the effect of QE on expected inflation. Any future research will have to point out whether there is a significant effect of QE on expected inflation. With the availability of more data on the current, aggressive QE policy and the use of a valid instrument, it should be possible to get more reliable results.

Since the QE policy makers might probably not be looking at a single lagged inflation rate to make any decisions on the continuation of the policy, a suggestion could be to add multiple instruments to the regression. On top of that, the QE policy makers might be more interest in the increase or decrease of the inflation rate between two preceding periods than the level of the lagged inflation itself. The difference between a three-period lagged inflation rate and a four-period lagged inflation rate could be added as an instrument. Furthermore, the difference between the two-period lagged inflation rate and the three-period lagged inflation rate could be added as a second instrument. These instruments might be more correlated with the QE variable than the one-period lagged inflation rate used in this research, so the instrument relevance condition is met. The instrument exogeneity condition

(19)

might also be met if the more rational and forward looking forecast of the economic experts are taken into account so no correlation between expected inflation and the instruments described above would be expected. But, again, future research has to point out whether this is indeed the case.

(20)

4 Conclusion

In this research the impact of QE on expected inflation is examined. This research was centered on the following research question:

Does the aggressive QE policy of Japan have a significant effect on expected inflation?

First, a literature review outlined the economic background of Japan and the most important literature about the first QE policy applied between 2001 and 2006. Furthermore an empirical analyses was conducted to investigate the effect of QE on expected inflation. An OLS regression was performed, and because of an expected endogeneity problem, an IV (2SLS) regression was performed as well. Both regressions were subject to several tests to test the reliability of the results.

From the literature review and the empirical study the following conclusions can be made. The effect of the first QE policy between 2000 and 2006 on prices and economic activity was found to be small or negligible (Ugai, 2006) (Schenkelberg & Watzka, 2013). Ueda (2012) states that the declining money multiplier, combined with the underlying interbank trust issues, could be an explanation for the negligible effect of QE on stimulating economic activity and price level. Schenkelberg and Watzka (2013) found little increase in the money supply after a monetary base expansion during the first QE period between 2001 and 2006, which could explain the negligible effect on prices. The constraint on the effectiveness of monetary policy could have been caused by macroeconomic mismanagement by the BOJ during the 1990s (Ueda, 2012). By exiting of the policy in 2006 with a low inflation rate it was questioned whether the BOJ was really committed in resolving the deflationary pressure and stagnating economy (Ueda, 2012).

Despite the first QE policy not being successful in reversing the deflationary pressure, the BOJ seem to be really committed to end deflation with the current aggressive QQE policy in place. Since they explicitly set an inflation target and use a more extensive monetary base expansion, as suggested by Hetzel (2003). Girardin and Moussa (2011) suggest in their empirical research that the QE policy can be even more effective if the policy is combined with financial reforms, moreover the policy should be applied longer than the first QE policy between 2001 and 2006 was applied.

The most important issue concerning the effectiveness of the QE policy can be found in the theory described by Krugman et al. (1998) concerning the credibility of a central bank. They

(21)

think monetary policy at the ZLB can be effective if the central bank can credibly commit to seek a higher price level. If it is believed that a permanent monetary expansion is applied, inflation expectations could be altered. The credibility of the BOJ is really important because if a temporary money expansion is expected this is ineffective in stimulating the economy and the price level (Krugman, Dominquez, & Rogoff, 1998). Future research has to point out whether the current QQE policy was a credible commitment of the BOJ to end the stagnating economy and fight the deflationary pressure.

From the empirical study there is no evidence of a statistically significant effect of QE on expected inflation. The QE variable is insignificant in both the OLS and IV (2SLS) regressions. Because it is expected that the causality runs both ways from expected inflation on QE as well as from QE on expected inflation an endogeneity problem might be arising from a simultaneous causality bias, making the results unreliable. The instrument used in the IV (2SLS) regression happened to be weak, causing the results of the IV (2SLS) regression to be unreliable too. The availability of more data on the current QE policy and the use of a valid instrument should make it possible to obtain a more reliable estimate in any future research. A suggested improvement to the estimated model would be to use multiple instruments. And instead of using the one-period lagged inflation rate level, the difference between a three-period lagged inflation rate and a four-period lagged inflation rate could be added as an instrument and the difference between the three-period lagged inflation rate and the two-period lagged inflation rate could be added as a second instrument. Future research has to make clear whether this improvement makes the results of the IV (2SLS) estimated model more reliable.

(22)

5 References

Bank of Japan. (2013, April 4). Price Stability Target" of 2 Percent and "Quantitative and Qualitative Monetary Easing. Retrieved from website Bank of Japan: https://www.boj.or.jp/en/mopo/outline/qqe.htm/

Baum, C. F. (2006). An introduction to Modern Econometrics Using Stata. Stata Press.

Eggertson, G. B., & Woodford, M. (2003). Zero Bound on Interest Rates and Optimal Monetary Policy. Brookings Papers on Economic Activity, 139-233.

Girardin, E., & Zakaria, M. (2011). Quantitative easing works: Lessons from the unique experience in Japan 2001-2006. Journal of International Financial Markets, Institutions & Money, 461-495.

Hetzel, R. L. (2003). Japanese Monetary Policy and Deflation. FRB Richmond Economic Quarterly, 21-52.

Krugman, P. R., Dominquez, K. M., & Rogoff, K. (1998). It's Baaack: Japan's Slump and the Return of the Liquidity Trap. Brookings Papers on Economic Activity, 1998(No. 2), 137-205.

Mishkin, F. S. (2010). The Economics of Money, Banking and Financial Markets. Pearson. Okina, K., Shirakawa, M., & Shiratsuka, S. (2001). The Asset Price Bubble and Monetary

Policy: Japan's Experience in the Late 1980s and the Lessons. Monetary and Economic Studies, 395-450.

Schenkelberg, H., & Watzka, S. (2013). Real effects of quantitative easing at the zero bound: Structural VAR-based evidence from Japanq. Journal of International Money 33, 327-357.

Shiratsuka, S. (2010). Size and Composition of the Central Bank Balance Sheet: Revisiting Japan’s Experience of the Quantitative Easing Policy. Monetary and Economic Studies. Stock, J. H., & Watson, M. W. (2012). Introduction to Econometrics. Pearson.

Ueda, K. (2012). Deleveraging and Monetary Policy: Japan since the 1990s and the United States since 2007. Journal of Economic Perspectives, 26(3), 177-202.

Ugai, H. (2006). Effects of the Quantitative Easing Policy: A Survey of Empirical Analyses. Bank of Japan Working Paper Series.

World Development Indicators. (2015, March). Retrieved from The World Bank: http://databank.worldbank.org/data/

(23)

6 Appendix

Figure 2.1 Inflation rate Japan

Referenties

GERELATEERDE DOCUMENTEN

When we considered the impact of obesity on life expectancy in the 26 European countries, we found that without obesity, the increase in e 0 between 1975 and 2012 would have

ABSTRACT: Though 1,4-disubstituted 1,2,3-triazole rings have been utilized as electronic bridges in the solution phase, the use of a triazole ring to serve as an electronic bridge

Sergio Espagna, Politecnical University of Valencia, Spain Robert Feldt, Blekinge Institute of Technology, Sweden Vincenzo Gervasi, University of Pisa, Italy. Smita Ghaisas,

Unfortunately, to the best of our knowledge, whether or not sales promotions (i.e., free product offers or price reductions) through large-scale promotional events influence customer

Comparable to modulating allosteric regulation with light, self-assembly of proteins can be activated/deactivated upon illumination by introducing photo-responsive moieties

We compared the functional traits of invasive plant species to those of indigenous and non-invasive alien plant species on Sub-Antarctic Marion Island. Invasive species

The two-continua model, the buffering effects of well-being in general, and the adaptive impact of specific positive characteristics such as positive emotions (Garland et

This paper presents the research protocol and design rationale for a personalised coaching system and the evaluation of persuasive coaching strategies for emotional eaters