Did the expectations channel work? Evidence from quantitative easing in Japan, 2001–06

The Japanese economy experienced a prolonged period of quantitative easing (QE) over the five years from March 2001 to March 2006. The purpose of this paper is to evaluate the direct and exclusive effects of this rather unconventional monetary policy on financial markets, economic activity, and labor markets in Japan empirically by employing exactly the same testing period with the QE period in most of our examinations. Using a range of variables, we first estimate vector error correction models (VECMs) that consider the cointegrating relations between the Japanese monetary base and other variables in our data-set. We also use Markovswitching dynamic regression (MSDR) models, Bayesian vector autoregressive (VAR) models, and causality analyses to test for robustness. Together, all the above analyses consistently provide a number of interesting findings. First, QE lowered shortand medium-term credit spreads and improved Japanese credit market conditions. Second, QE increased stock prices in Japan and improved market expectations. Third, the QE policy recovered labor market conditions and economic productivity in Japan. Finally, additional analyses of fund flows and economic survey data suggest that the primary transmission channel of this period of Japanese QE policy was the expectations channel. *Corresponding author: Chikashi Tsuji, Faculty of Economics, Chuo University, 742-1 Higashinakano, Hachioji-shi, Tokyo 192-0393, Japan E-mail: mail_sec_low@minos.ocn.ne.jp Reviewing editor: David McMillan, University of Stirling, UK Additional information is available at the end of the article ABOUT THE AUTHOR Chikashi Tsuji is a professor of Economics and Finance at Chuo University in Tokyo, Japan. He commenced his academic career as an assistant professor at Ritsumeikan University in 2003, followed by an associate professorship at Tsukuba University in 2007 and a professorship at Chuo University in 2013. His research interests include asset pricing, corporate finance, applied econometrics, financial accounting, and monetary economics. He is also interested in engineering issues related to economics and finance. His published research includes forecasting asset returns and volatility, testing asset-pricing models, investigating financial risk, examining linkages between asset prices and macroeconomic variables, and testing the relations between corporate payout policy, capital structure, financial and accounting ratios, and stock returns. He has published widely in a range of international peer-reviewed journals including the Journal of International Money and Finance, Quantitative Finance, and Applied Financial Economics. PUBLIC INTEREST STATEMENT The Bank of Japan executed a quantitative easing (QE) monetary policy over the five-year period from March 2001 to March 2006. This work empirically evaluates the direct and exclusive effects of this rather unconventional monetary policy on financial markets, economic activity, and labor markets in Japan. Using a wide range of variables, the work yields a number of interesting findings. First, QE lowered shortand mediumterm credit spreads and improved credit market conditions. Second, QE increased stock prices and improved market expectations. Third, the QE policy facilitated the recovery of labor market conditions and economic productivity. Finally, analysis of fund flows and economic survey data suggests that the primary transmission channel of this QE policy was the expectations channel. This study is noteworthy in that it is the first to reveal the working mechanisms of the expectations channel in Japanese QE from 2001 to 2006. Received: 25 April 2016 Accepted: 04 July 2016 Published: 17 August 2016 © 2016 The Author(s). This open access article is distributed under a Creative Commons Attribution (CC-BY) 4.0 license.


ABOUT THE AUTHOR
Chikashi Tsuji is a professor of Economics and Finance at Chuo University in Tokyo, Japan. He commenced his academic career as an assistant professor at Ritsumeikan University in 2003, followed by an associate professorship at Tsukuba University in 2007 and a professorship at Chuo University in 2013. His research interests include asset pricing, corporate finance, applied econometrics, financial accounting, and monetary economics. He is also interested in engineering issues related to economics and finance. His published research includes forecasting asset returns and volatility, testing asset-pricing models, investigating financial risk, examining linkages between asset prices and macroeconomic variables, and testing the relations between corporate payout policy, capital structure, financial and accounting ratios, and stock returns. He has published widely in a range of international peer-reviewed journals including the Journal of International Money and Finance, Quantitative Finance, and Applied Financial Economics.

PUBLIC INTEREST STATEMENT
The Bank of Japan executed a quantitative easing (QE) monetary policy over the five-year period from March 2001 to March 2006. This work empirically evaluates the direct and exclusive effects of this rather unconventional monetary policy on financial markets, economic activity, and labor markets in Japan. Using a wide range of variables, the work yields a number of interesting findings. First, QE lowered short-and mediumterm credit spreads and improved credit market conditions. Second, QE increased stock prices and improved market expectations. Third, the QE policy facilitated the recovery of labor market conditions and economic productivity. Finally, analysis of fund flows and economic survey data suggests that the primary transmission channel of this QE policy was the expectations channel. This study is noteworthy in that it is the first to reveal the working mechanisms of the expectations channel in Japanese QE from 2001 to 2006.

Introduction
In recent years, a number of major central banks around the world have conducted quantitative easing (QE) programs to counter the strong negative effects of the most recent financial crisis on economic activity. For instance, in an attempt to end Japan's current prolonged deflation and economic slump, known as the "Lost Two Decades," the Bank of Japan (BOJ) is currently executing an unconventional monetary easing policy referred to as quantitative-qualitative easing (QQE). However, this policy is ongoing, and it is difficult to fully assess its effectiveness at present. Similarly, the US Federal Reserve has also implemented successive QE monetary policies, known as QE1, QE2, and QE3, and sometimes referred to as large-scale asset purchase (LSAP) programs. 1 Evaluation of the effectiveness of QE policies like these is extremely important for policy-makers and other monetary policy stakeholders. Fortunately, aside from the current QQE policy in Japan, the BOJ also conducted a QE policy from March 2001 to March 2006. Because this program has now concluded, we can now evaluate its effects fully.
Regarding the effectiveness of QE, previous studies present two conflicting views. Concerning the first, traditional macroeconomic theory as Hicks (1937) suggested that when interest rates are at their lower bound, monetary injections into the economy are ineffective. In contrast, Eggertsson and Woodford (2003), Clouse, Henderson, Orphanides, Small, and Tinsley (2003), , and Bernanke, Reinhart, and Sack (2004) argued that central bankers could stimulate the economy, even when interest rates are at their lower bound, because QE monetary policy can stimulate the economy by changing market expectations or the composition and/or size of the central bank's balance sheet.
More recently, Glick and Leduc (2012), D' Amico and King (2013), and Neely (2015) examined the effects of QE in the US. Fujiwara (2006) used Markov-switching vector autoregressive (MSVAR) models to examine Japanese monetary policies empirically, including the partial period of QE from 2001 to 2006, and found that it had little effect. In contrast, Girardin and Moussa (2011), using an extended MSVAR model and factor analysis, found that it had in fact worked as expected. Using a structural VAR (SVAR) approach, Schenkelberg and Watzka (2013) also found evidence of the effectiveness of Japanese QE policy, as did  with simple VAR analysis. Given this mixed evidence, more comprehensive study is required on the effectiveness of QE in Japan from 2001 to 2006.
In this paper, we analyze the effects of Japan's QE program using vector error correction models (VECMs), Markov-switching dynamic regression (MSDR) models, Bayesian VAR models, and causality analyses to obtain robust evidence. Our paper has several noteworthy features. (i) First, we employ a comprehensive data-set to examine the effects of a wide range of economic and financial market variables on productivity, labor market conditions, bond markets, and stock markets in Japan. (ii) Second, we investigate the effects of QE by taking into account the bivariate cointegration between the monetary base and the other variables used in our analysis. The results are also carefully checked by additional analyses by MSDR models, Bayesian VAR models, and causality tests. (iii) Third, our study mainly focuses on the direct effects of the increases in the monetary base on the Japanese economy and financial markets. Nevertheless, we also attempt to identify the transmission channel that was effective during the QE period by examining important related data. We note that in identifying the effective channel, we focus on two important theoretical channels discussed later: the

Literature review
Although the focus of this paper is QE in Japan, we now briefly review the existing empirical literature, starting with QE in the US, followed by QE again in Japan from 2001 to 2006. To start with, Hancock and Passmore (2011) analyzed the purchase program for mortgage-backed securities (MBSs) conducted by the US Federal Reserve in 2008 by focusing on the effects of the purchase program on mortgage rates. 4 More recently, Glick and Leduc (2012) examined the effects of the Federal Reserve's LSAPs since 2008 on financial markets in the US. 5 Moreover, Kandrac and Schlusche (2013) confirmed the existence of "flow effects" in the prices of US Treasury securities during the initial stages of the LSAP program. 6 Kandrac (2013) examined the impact of the MBS purchases by the Federal Reserve on market functions in the US and found that these purchases indeed disrupted markets, although the magnitude of the effects was small. Elsewhere, D'Amico and King (2013) considered the impact of the US Treasury securities purchase program by the Federal Reserve on yields. Lastly, using international data, Neely (2015) examined the effects of the Federal Reserve's QE announcements from 2008 to 2009 and found that these substantially lowered the yields of international long-term bonds and depreciated the spot price of the US dollar.
We next review extant studies of QE in Japan from 2001 to 2006, which is the focus period of our study. To start, Fujiwara (2006) used MSVAR models to analyze QE in Japan from January 1985 to December 2003, corresponding to only part of the QE sample period. However, the focus of that study was to confirm the existence of an economic structural break in Japan resulting from the BOJ's ZIRP. 7 Furthermore, Fujiwara (2006) suggested that during the 1990s, monetary policy became less effective in Japan; consequently, QE after March 2001 had very little effect.
Conversely, Girardin and Moussa (2011) conducted empirical analysis of QE in Japan from 2001 to 2006 and found that QE was indeed effective. More specifically, using an extended MSVAR model and factor analysis, they showed that in Japan a significant economic regime change to a favorable state took place because of QE. On this basis, they concluded that QE was able to both prevent further recession and deflation and to stimulate output and prices substantially. Their analysis using factor analysis was also economy wide because it employed the extracted factors for (i) economic activity, (ii) prices, and (iii) interest rates from among several related series. However, their approach differs from the present analysis in that we examine the effects of QE on a variety of economic and financial variables, whereas their focus was only on output, prices, and interest rates. As we describe later in our discussion, this paper instead uses a wider range of specific variables and analyzes them individually to understand more clearly which variables were affected more directly and strongly by QE. Moreover, Girardin and Moussa (2011) did not examine the expectations channel or the portfolio substitution channel specifically, while this paper carefully considers these channels by inspecting actual data.  Using Japanese data after 1995, Schenkelberg and Watzka (2013) examined QE using a SVAR model to identify the effects of shocks in the zero-lower bound economy. They found that QE shocks lowered Japanese long-term interest rates and increased output and prices. We note that our study departs from this in that we analyze a number of additional variables. More precisely, Schenkelberg and Watzka (2013) analyzed the BOJ current account balances, CPI, industrial production, 10-year government bond yields, and exchange rates. Indeed, one of the main contributions of our paper is to obtain new evidence on the effects of the QE by specifying a wider range of variables in our analysis. 8 Although Honda (2014)   , we also investigate the monetary base, the index of the capacity utilization ratio, the absolute unemployment rate, the short-and medium-term credit spreads, and the Nikkei 225 stock index price (the Nikkei). Third, Honda (2014) used standard VAR models instead of the VECMs that we employ in this study. The advantage of VECMs is that they consider the cointegrating relationship between the monetary base and the other economic and financial market variables. 9 We also note that we check the results from VECMs by MSDR models, Bayesian VAR models, and causality analyses.
In summary, there is at present a lack of rigorous comprehensive empirical research concerning the effects of QE in Japan from 2001 to 2006. In response, we examine QE by considering the cointegrating relationships between the monetary base and other economic and financial market variables. Following this, and based on the theoretical discussion in the next section, we also attempt to identify the channel through which QE worked.

Theoretical discussion-effects and channels
This section discusses the potential effects and channels of QE, which work when short-term policy interest rates are at very low levels or even zero. The first is the expectations channel. This means that central banks can affect asset prices and economic activity by firstly influencing market participants' expectations of future short-term interest rates, as suggested by such studies as Svensson (2001), Eggertsson andWoodford (2003), andWoodford (2003). As  noted, these studies find that even when the nominal policy interest rate is zero or near zero, a central bank can stimulate the economy by committing to the public to maintain the short-term rate at a low level for even longer than previously expected. They also suggested that if this commitment is credible, it should affect other asset prices and stimulate real economy by changing market expectations.
The second is the portfolio substitution channel, as suggested by Tobin (1969), Brunner and Meltzer (1973), and Andrés, López-Salido and Nelson (2004). Specifically, if money is an imperfect substitute for other financial assets, large increases in the money supply will result in investors' portfolio rebalancing, and this will increase some non-money asset prices and decrease others. Consequently, through these portfolio substitution effects, economic activity will be stimulated, even in a lower-bound interest rate economy.
With Japanese QE after March 2001 facing a virtually zero-interest rate economy, by increasing its current account balances, the BOJ continued to inject more liquidity than needed to ensure a zerocall rate. As for the commitments in this policy, the BOJ at first promised that until the annual CPI change rate became stable at zero or above, it would continue QE. Afterwards, in October 2003, the BOJ supplied a more detailed description of its commitment. That is, it would continue QE (i) until the latest annual change in core CPI became zero or above and expected to continue as such in the future, and (ii) until the future core CPI inflation rate was not expected to fall below zero again. Hence, we understand that the BOJ's conditional commitments are equivalent to a commitment to continue monetary easing until certain economic conditions are met, a policy discussed in the situation where the expectations channel is expected to work.
Moreover, we note that the BOJ's act of setting a high reserve target is more visible, and thus the commitment may be more credible than a purely verbal promise regarding future monetary policy. Once again, we regard the targeted current account balances as a clear visible policy signal, as suggested by ; thus, the so-called signaling effect is also expected in the context of Japan's QE. Furthermore, with this form of QE executed from 2001 to 2006, the BOJ's current account balances increased roughly sevenfold over these five years, and the monetary base expanded from about 13 to 22% of nominal GDP over the same period. Given these unusual money injections, we also expect the portfolio substitution effect on non-money assets in Japanese financial markets, and through the effect, the Japanese economy would be stimulated.
Based on this theoretical discussion, we conduct a careful empirical analysis to clarify not only the effectiveness of this Japanese QE program but also the channels and mechanisms governing this QE during the policy period.

Data
This section describes the data and variables used in our analysis. In this study, we investigate the effects of the QE executed in Japan from March 2001 to March 2006. Accordingly, we mainly use Japanese monthly data over an identical period to analyze the direct and exclusive effects of the QE program. All data are from QUICK Corp. and the BOJ.
Our first variable group comprises macroeconomic variables. First, MB is the monetary base in Japan. Our interest is mainly on the effect of increases in MB through QE on the other variables. Second, IP is the most representative index of industrial production in Japan. Thus, IP represents overall productivity. Third, ICU is the capacity utilization ratio. This ratio indicates the degree of utilization of plant and equipment, and hence a higher ICU identifies greater economic activity in Japan. Lastly, UNE is a variable representing Japanese labor market conditions, as measured by the absolute unemployment rate in Japan. A lower UNE suggests an improvement in Japanese labor market conditions. Our next group of variables reflects Japanese credit market conditions. First, SCSD is the shortterm credit spread, being the difference between the short-term Nikkei bond index yield and the two-year Japanese government bond yield. Second, MCSD is the medium-term credit spread, which equals the medium-term Nikkei bond index yield minus the five-year Japanese government bond yield. 10 We also consider that these credit spread variables are proxies for the risk aversion of market participants, such that lower values of SCSD and MCSD are associated with decreases in financial market risk aversion.
The final variable relates to the Japanese stock market. Namely, our variable NK indicates the Nikkei 225 stock price index (the Nikkei). The Nikkei is a well-known Japanese stock market index, mostly comprising large firms. We also consider that the prices of the Nikkei reflect the market expectations as to the future stock market and economic conditions. 11 Figure 2 plots the six economic and financial variables other than MB, with the shaded area designating the QE period. As shown, IP, ICU, and NK all generally increased during this period. In contrast, UNE, SCSD, and MCSD all decreased in general. Table 1 provides descriptive statistics for our variables, including the mean, median, maximum, minimum, standard deviation, skewness, and kurtosis. As shown, NK is right (or positively) skewed as the value of skewness is around one, while MB and ICU are left (or negatively) skewed. For the variable NK, the kurtosis value is more than three, suggesting fat-tailed (or leptokurtic) distributions with relatively many extreme observations. In contrast, the values of the other six variables are all less than three, indicating thin-tailed (or platykurtic) distributions with relatively few extreme observations.

Unit root and cointegration tests
Before estimating the VECMs, we perform unit root and cointegration tests. Table 2 presents the results of the augmented Dickey-Fuller unit root tests. As shown, we conduct tests on both the levels and first differences of our seven variables. Table 2 shows that in levels, all variables have unit roots, while their first differences do not contain unit roots. Therefore, all seven variables are nonstationary in levels but stationary in first differences. Table 3 shows the results of Johansen's cointegration tests (Johansen, 1991(Johansen, , 1995 for the bivariate combinations of MB and the remaining six variables. Table 3 details the number of cointegrating relations for (i) no constant term in the cointegrating equation (CE) or in the VAR model (hereafter "model type 1") and (ii) a constant term in the CE but no constant in the VAR model ("model type 2"). "Lag" in Table 3 indicates the lag order of each VAR model. The test results in Table 3 suggest that all six bivariate combinations have cointegrating relations. More specifically, Table 3 indicates that both the trace and maximum eigenvalue statistic tests suggest that in model type 1, MB and IP (Panel A) and MB and ICU (Panel B) have one cointegrating relation. Panels A and B also show that the lag orders of the VAR models for MB and IP and MB and ICU are two.  In this figure, IP denotes the Japanese industrial production index; ICU means the capacity utilization ratio index in Japan; and UNE is the absolute unemployment rate in Japan.
In addition, SCSD denotes the Japanese short-term credit spread and this variable is constructed by subtracting the two-year Japanese government bond yield from the short-term Nikkei bond index yield; MCSD means the Japanese medium-term credit spread and this variable is constructed by subtracting the five-year Japanese government bond yield from the medium-term Nikkei bond index yield; and NK denotes the Nikkei 225 stock price index in Japan. This figure is drawn for the period from January 2000 to December 2007 and the shaded area indicates the period of the Japan's QE. Furthermore, eight lines in this figure mean the date when the BOJ raised its targeting amount of current account balances.
Furthermore, the above two cointegration tests suggest that in model type 2, MB and UNE (Panel C), MB and SCSD (Panel D), MB and MCSD (Panel E), and MB and NK (Panel F) also have one cointegrating relation. The above four panels indicate that the VAR lag order of MB and UNE is four, the lag orders of MB and SCSD and MB and MCSD are six, and that of MB and NK is two. The details of the VECMs, as specified based on these results, are in the next section. denotes the Japanese industrial production index; ICU means the capacity utilization ratio index in Japan; and UNE is the absolute unemployment rate in Japan. In addition, SCSD denotes the Japanese short-term credit spread and this variable is constructed by subtracting the two-year Japanese government bond yield from the short-term Nikkei bond index yield; MCSD means the Japanese medium-term credit spread and this variable is constructed by subtracting the five-year Japanese government bond yield from the medium-term Nikkei bond index yield; and NK denotes the Nikkei 225 stock price index in Japan. represents the amount of the monetary base in Japan (trillion yen); IP denotes the Japanese industrial production index (point); ICU means the capacity utilization ratio index in Japan (point); and UNE is the absolute unemployment rate in Japan (percent). In addition, SCSD denotes the Japanese short-term credit spread and this variable is constructed by subtracting the two-year Japanese government bond yield from the short-term Nikkei bond index yield (percent); and MCSD means the Japanese medium-term credit spread and this variable is constructed by subtracting the five-year Japanese government bond yield from the medium-term Nikkei bond index yield (percent). Finally, NK denotes the Nikkei 225 stock price index in Japan (yen).

VECM analysis of the QE effects
This section documents the details of the VECMs estimated in this study. Based on the results in Panels A-F in Table 3, we specify the VECMs as follows. 12 First, for MB and IP, we apply a VECM(2) with (1) where ΔMB t−p is the first difference of MB, and the lag order is p (the same hereafter); ΔIP t−q is the first difference of IP, with lag order q.
For MB and ICU, we specify a VECM(2) with a CE, CE mb−icu = MB t−1 + mb−icu ICU t−1 : where ΔICU t−q is the first difference of ICU, with lag order q.
For MB and UNE, we employ a VECM(4) with a CE, where ΔUNE t−q is the first difference of UNE, with lag order q.
For MB and SCSD, we specify a VECM(6) with a CE, CE mb−scsd = MB t−1 + mb−scsd SCSD t−1 + h mb−scsd : (2)  (Johansen, 1991(Johansen, , 1995   where ΔSCSD t−q is the first difference of SCSD, with lag order q. For MB and MCSD, we use a VECM(6) with a CE, CE mb−mcsd = MB t−1 + mb−mcsd MCSD t−1 + h mb−mcsd as follows: where ΔMCSD t−q is the first difference of MCSD, with lag order q.
Finally, for MB and NK, we specify a VECM (2) where ΔNK t−q is the first difference of NK, with lag order q.

Empirical results from the VECMs
This section documents our empirical results. In what follows, we explain the estimation results for the VECMs specified above. We then discuss the impulse response functions for our six variables given an increase in the monetary base. We note that these VECM analyses employ exactly the same QE period to clarify the direct and exclusive effects of Japan's QE during the QE period. Panels D and E of Table 4 provide the results for the monetary base and bond market variables. The results show that the CEs of MB and SCSD and MB and MCSD are statistically significant in explaining the evolution of ΔSCSD and ΔMCSD, respectively. Moreover, the four-period lag of ΔMB statistically significantly explains the evolution of ΔSCSD with a negative sign (Panel D), and the (8)

Estimation results for the VECMs
four-period lag of ΔMB also statistically significantly explains the evolution of ΔMCSD with a negative sign (Panel E). Thus, our VECMs well capture the linkages between MB and SCSD and MB and MCSD. Hence, we can see that QE in Japan lowered short-and medium-term credit spreads, thereby decreasing risk aversion in Japanese bond markets.
Finally, Panel F of Table 4 displays the results for the monetary base and stock market variable. The results show that the CE of MB and NK is statistically significant in explaining the evolution of ΔNK. Moreover, the one-and two-period lags of ΔMB statistically significantly explain the evolution of ΔNK (Panel F) with all positive signs. Thus, our VECM again captures effectively the relations between MB and NK. Therefore, the results show that QE in Japan drove up stock prices, thereby improving expectations of the stock market and future economic conditions in Japan.

Analyzing impulse responses from the VECMs
In the previous subsection, we demonstrated that our bivariate VECM modeling of the direct effects of the increase in the monetary base on various economic and financial market variables was highly successful. In this subsection, we further examine the direction of these QE effects by analyzing the impulse response functions. Figure 3 depicts   Japan; IP denotes the Japanese industrial production index; ICU means the capacity utilization ratio index in Japan; and UNE is the absolute unemployment rate in Japan. In addition, SCSD denotes the Japanese short-term credit We next consider the responses of the Japanese bond market variables. Panels D and E show that SCSD and MCSD respond negatively to increases in the monetary base. Therefore, the above findings again indicate that QE reduced short-and medium-term credit spreads, decreasing risk aversion in Japanese bond markets. Finally, with regard to the Japanese stock market variable, Panel F reveals that NK responds positively to a positive shock in the monetary base. Thus, as these results show, QE pushed up stock prices and improved market expectations in Japan. Overall, the results from the various impulse response analyses demonstrate that from 2001 to 2006, Japan's QE had a clear positive effect on the economy and financial markets in Japan.

Applying the MSDR models
In this section, we further investigate the effects of QE from 2001 to 2006 in Japan using MSDR models. 13 For our robustness check, we use the annual percentage change (growth) rates of IP, ICU, UNE, and NK and the credit spread variables of SCSD and MCSD. Using these and two-regime MSDR models, we conduct robustness checks of our results in the previous section. In addition, to identify the effects of the QE policy in the period after its end, our robustness checks employ a longer sample period from January 1985 to November 2015, which includes the financial crisis periods represented by the US Lehman shock. 14 In setting up the models, we assume that each of these four change rates and two credit spread variables follows a process that depends on the value of an unobserved state variable, s t . We assume that there are M = 2 regimes, and hence when s t = m, for m = 1 or 2, the analyzed variable is in state or regime m in period t. More specifically, the two-regime MSDR model for the variable y is specified as follows: where t follows an independent and identically distributed (iid) standard normal distribution. In addition, the expressions of the mean, (m), and the volatility, (m), in model (13) indicate that and are regime dependent.
In the Markov-switching models, the probability of transitioning from regime i in period t − 1 to regime j in period t is expressed as follows: and in our case of M = 2, the matrix of transition probabilities, p(t) can also be written as follows:

Evidence from the MSDR models
This subsection describes the results of the application of the MSDR models. We first present the estimation results for the MSDR models in Panels A-F in Table 5. All panels include the estimation results for the higher and lower regimes for the economic and financial market variables. As these results present, all model intercepts and the volatilities for the six variables are statistically significant at the 1% level. Therefore, we can see that all our MSDR models are very accurate.
Based on the results of our impulse response analyses, Figure 4 displays the probabilities that the annual percentage changes in IP, ICU, and NK are in the higher state, because these three variables respond positively to a positive shock in MB. In addition, we also present the probabilities that the (13) y t = (m) + (m) t , (14) P(s t = j|s t−1 = i) ≡ p ij (t), (15) annual percentage changes for UNE and the credit spread variables SCSD and MCSD are in the lower state, as these three variables respond negatively to an increase in MB.

Table 5. Estimation results of the two-regime Markov-switching dynamic regression models: the cases of the Japanese economy and financial markets from January 1985 to November 2015
Notes: This In terms of the economic variables, the growth rates of IP, ICU, and UNE are in the favorable regime during and well after the end of QE in March 2006 until around the Lehman shock in September 2008. Thus, these results suggest that productivity and labor market conditions in Japan kept recovering even after the end of QE. As for the Japanese bond market variables, SCSD and MCSD are in the lower regime state long after the implementation of QE. The credit spreads of SCSD and MCSD are in the lower state even after the end of QE; and they switched to the higher state around the Lehman shock in the US. These results suggest that QE decreased short-and medium-term credit spreads in Japan and continued to decrease risk aversion in Japanese bond markets, even after QE.
Finally, with regard to the stock market variable, after executing QE, the growth rate of NK moved to, and remained in, the higher regime even after the end of QE. The growth rate of NK remained in the higher state until the period immediately after the Paribas shock in August 2007. Thus, QE in Japan stimulated and continued to drive up stock prices even after the end of the QE program. Therefore, the above additional empirical analyses using MSDR models, which employ a longer sample period of January 1985 to November 2015, confirm the robustness of all our positive QE effects derived from the VECMs and impulse response functions.  IP denotes the Japanese industrial production index; ICU means the capacity utilization ratio index in Japan; and UNE is the absolute unemployment rate in Japan. In addition, SCSD denotes the Japanese short-term credit spread; MCSD means the Japanese mediumterm credit spread; and NK denotes the Nikkei 225 stock price index in Japan.

Results from the Bayesian VAR models
To check further the robustness of our results, we next estimate Bayesian VAR models (Doan, Litterman, & Sims, 1984;Litterman, 1986) and plot the impulse responses from the models in Figure 5. 15 This Bayesian VAR analysis also employs exactly the same sample period of the QE program to clarify the precise and direct effects of Japan's QE during the QE period. The results are almost identical to those exhibited in Figure 3. As to the responses of the economic variables, Panels A and B show that IP and ICU respond positively to the monetary base increases. Panel C also indicates that UNE responds negatively to a positive shock in the monetary base. These results therefore again imply that from 2001 to 2006, QE improved economic productivity and labor market conditions in Japan.
We next check the responses of the Japanese credit spread variables. Panels D and E show that SCSD and MCSD respond negatively to the increase in the monetary base. Slightly differently from the results in Panels D and E of Figure 3, the responses of SCSD and MCSD in Figure 5 are rather rapid and smooth. Hence, the above results again indicate that QE reduced short-and medium-term credit spreads and decreased the degree of risk aversion in Japanese credit markets. Finally, as for the Japanese stock market variable, Panel F of Figure 5 again exhibits that NK responds positively to a positive shock in the monetary base. Thus, the results again indicate that through QE, Japanese stock

Results of causality analysis
We next employ Granger causality analysis, with the results shown in Table 6. Again, this Granger causality analysis also uses exactly the same QE period to clarify the exclusive and direct effects of this Japanese QE during the QE period. Panel A provides the results of causality tests from MB to other six variables, while Panel B shows the results of the causality tests from these other six variables to MB. The results in   ), respectively. Moreover, IP denotes the Japanese industrial production index; ICU means the capacity utilization ratio index in Japan; and UNE is the absolute unemployment rate in Japan. Further, SCSD denotes the Japanese shortterm credit spread; MCSD means the Japanese mediumterm credit spread; and NK denotes the Nikkei 225 stock price index in Japan.

What channel worked?-Interpreting the results
The next important question is through what channel did QE affect Japanese financial markets and the Japanese economy? In order to respond to this question, we conduct several specific Granger causality tests and present the results in Table 7. The results show that there are unidirectional causal effects from SCSD to NK, from MCSD to NK, from SCSD to UNE, from MCSD to UNE, from UNE to NK, from NK to ICU, and from NK to IP. We provide a more careful interpretation later, but these results statistically indicate that the (i) decreases in credit spreads led to stock price increases, that the (ii) decreases in credit spreads led to declines in the unemployment rate, that (iii) declines in the unemployment rate led to stock market increases, and that (iv) stock market recoveries led to improvements in industrial production and capacity utilization in Japan.
In addition, we provide further evidence of the recovery of market expectations during QE in Japan. Figure 6 plots the changes in the short-term economic survey of enterprises in Japan. 16 Specifically, Panel A shows the year-on-year percentage changes in terms of future business conditions, and Panel B displays future business condition survey changes from the previous quarter. Both panels provide details of the changes in expectations of manufacturing, nonmanufacturing, and all-industry firms in Japan. This figure clearly suggests that market expectations clearly recovered  during the QE in Japan, linked to the exogenous MB increases by this BOJ's QE as seen in Figure 1. This supports our interpretation of the effectiveness of the expectations channel in this Japanese QE program. We consider that this linkage between Japan's QE and expectation improvements was driven by the BOJ's commitment and forward guidance in the QE. Namely, in this QE, as we documented in the Theoretical discussion section, the BOJ committed to continuing to increase current account balances until the Japanese core CPI stably remains positive for the foreseeable future.
Therefore, taking into consideration the expectation recovery suggested in Figure 6, we interpret our empirical results as follows. First, (i) through QE, risk aversion in bond markets decreased and stock markets rose because of the recovery in financial market expectations; second, (ii) following the decline in risk aversion in Japanese bond markets, the recovery of expectations in corporations decreased unemployment rates; and third, (iii) the stock market recovery associated with changes in market expectations boosted industrial production and capacity utilization. Hence, we consider that the recovery of the Japanese economy during the QE period was mainly through the expectations channel.
We now turn to discuss the next possibility; namely, the portfolio substitution channel. To investigate this, we further examine funds flow data in several economic sectors. 17  Given the context of our analysis, we focus on the fund flows of stocks and corporate bonds. To start, Figure 7 shows that the fund flows into stocks increased only in the overseas sector (Panel F), whereas in other economic sectors, they decreased or were almost unchanged during the QE period. As to corporate bond fund flows, these increased in banks only very slightly, whereas in all other sectors, they decreased. Thus, during QE, only the overseas sector bought Japanese stocks, and almost no sector bought Japanese corporate bonds. Hence, we cannot find the evidence suggesting that portfolio rebalancing by investors boosted stock prices and decreased corporate bond yields in the period of QE.
In light of these results, we suggest that the primary channel for QE in Japan was not the portfolio substitution channel but the expectations channel. Figure     Moreover, unlike previous papers, our study implemented a more comprehensive analysis with regard to the effects of QE in Japan using a wider range of economic and financial variables. That is, we examined the effects of QE on (i) productivity, (ii) labor market conditions, (iii) credit spreads, and (iv) equity index prices in Japan. From the viewpoint of data analysis, the use of factor analysis is also interesting, as in Girardin and Moussa (2011). They constructed and analyzed three kinds of extracted factors: a real activity factor, a price factor 18 , and an interest rate factor. This is an interesting method because it enables the analysis of much more information through examining the extracted factors. We instead used a wider range of specific variables, which we individually analyzed in order to yield much more clear evidence in terms of the QE's effects and works. This approach enables us to understand more clearly which variables were affected more directly and strongly by QE. Clarification of these issues is one of the main contributions of this study.

Contribution
Furthermore, our study also examined the direct effects of QE while taking into consideration the cointegrating linkages between the policy variable, the monetary base, and the other variables using bivariate VECMs. Bivariate VECMs are highly effective in identifying the direct connection between QE and various economic and financial market variables during the policy period. This is a significant advantage of this study. We note that deriving robust evidence of the effectiveness of QE is important for future monetary policy and research. Careful repeated robustness checks using univariate two-regime MSDR models, Bayesian VAR models, and causality tests is another of the main contributions of this study.
Finally, the most important contribution of this study is the clarification of the mechanism and transmission channel of Japan's QE. Our additional analysis of causality, survey, and funds flow data suggests that regarding Japan's QE, the main effective channel was not portfolio substitution but expectations. 19 We note that stock prices and credit spreads both strongly relate to expectations regarding the future evolution of financial markets and the macroeconomy. We consider that our careful attempt to inform our empirical analysis with theory regarding the potential QE effect mechanism is our most significant contribution.

Implications
This subsection discusses several implications of our results by comparing them with those of other studies. For example, Schenkelberg and Watzka (2013) documented that the success of the QE experiment in Japan was only temporary. In early 2007, only one year after the end of QE in Japan, the subprime crisis happened and the following Lehman shock in the US negatively affected the Japanese economy. Thus, it is not easy to assess whether the actual effects of the QE were only temporary. IP denotes the Japanese industrial production index; ICU means the capacity utilization ratio index in Japan; and UNE is the absolute unemployment rate in Japan. In addition, SCSD denotes the Japanese short-term credit spread; MCSD means the Japanese mediumterm credit spread, and NK denotes the Nikkei 225 stock price index in Japan. Moreover, the arrows in this figure indicate unidirectional causal relations.
To measure the duration of the effectiveness of QE in the real world, we consider that the univariate MSDR approach adopted in this paper is superior to the application of MSVAR models elsewhere. Univariate MSDR models enable us to specify more precisely the time points for regime changes because of their higher estimation accuracy when compared with multivariate models. Nevertheless, the application of the Markov-switching factor-augmented VAR model estimated by Girardin and Moussa (2011) is technically interesting. Our application of the univariate MSDR models demonstrated that industrial production and the capacity utilization ratio in Japan remained in the higher regime beyond the Paribas shock and until around the Lehman shock.
Moreover, our MSDR analyses also show that the unemployment rate in Japan remained in the lower regime beyond the Paribas shock and until just before the Lehman shock. These results imply that the above economic productivity and unemployment rate variables do not support the suggestion that the success of the QE experiment in Japan was temporary. Therefore, it is very important to monitor the duration of these effects and the exit strategy for the QQE program currently conducted by the BOJ. Furthermore, our empirical results suggested that the recovery in expectations was a key factor for stimulating economic activities and improving labor market conditions. Hence, we stress the crucial role of expectations management in practical monetary policy, as suggested by such theoretical studies as Krugman (1998) 20 and Eggertsson and Woodford (2003). Thus, the most important implication of our study is that in unconventional monetary policy, the key is whether successful expectations management is accomplished. This is a crucial implication for the current QQE by the BOJ and the monetary policy in Europe, for example.
We also note that in the case of Japanese QE from 2001 to 2006, the targeted amount of BOJ current account balances was a clear policy target and that this setting of a visible signal was effective, not only for obtaining the signaling effect  but also for achieving better expectations management. We consider that policy regime changes (e.g. Sargent, 1982) through clear signals and credible commitments supplied by central banks are also important for successfully managing expectations in conducting monetary policy.

Summary and conclusions
In this paper, we investigated the effects of Japanese QE, which was executed by the BOJ from March 2001 to March 2006. We empirically examined the effects of this unconventional monetary policy on a range of variables in Japan including economic activity, labor and credit market conditions, and stock market evolution. The findings and contributions from our VECMs, MSDR models, Bayesian VAR models, causality tests, and other important data analyses are as follows.
• First, QE improved Japanese credit market conditions by lowering short-and medium-term credit spreads. We suggest that this was because of favorable declines in risk aversion through the QE program.
• Second, QE increased stock prices, as measured by the Nikkei 225 stock price index in Japan. We consider this means that market expectations for future economic conditions and financial markets recovered through QE.
• Third, QE improved labor market conditions in Japan. In particular, the Japanese unemployment rate remained lower for much longer, even after the end of QE. Our MSDR model suggested that the improved regime continued until just before the Lehman shock in the US.
• Fourth, QE had a positive effect on economic activity in Japan. Japanese industrial production and the utilization of plant and equipment increased both during and after the end of the QE program. The recovery in these measures continued until about the time of the Lehman shock in the US.
• Fifth, other than VECMs, MSDR models, Bayesian VAR models, and Granger causality tests also supported all of the above positive effects of the Japan's QE. We therefore consider our findings to be robust.
• Sixth, our empirical study clarified the mechanism and channel for Japan's QE. Additional analyses suggested that the primary channel was expectations. Moreover, as stock prices and credit spreads both strongly relate to expectations regarding future financial markets and the macroeconomy, our results demonstrate the importance of expectations, as reflected in asset prices, for monetary policy to be effective.
• Finally, our empirical inspections revealed the working mechanism and channel for Japan's QE, which will provide valuable feedback for the theoretical research. Our results suggest that building a theoretical monetary policy model including the effects of market expectations changes will be an important future research topic in economic theory.
As already emphasized, there are few comprehensive empirical studies of the effects of QE in Japan from 2001 to 2006. Thus, the rigorous and comprehensive empirical evidence derived from this study is important for deepening our understanding of the actual effects of unconventional monetary policies on the economy and financial markets. As we noted, we analyzed the QE effects by using exactly the same QE period except in the MSDR analyses to reveal the exclusive and direct effects of Japan's QE during the QE period. We again emphasize that this is one of the most noteworthy features of this study.
Overall, our empirical evidence shows the importance of expectations management, the signaling effect, and regime changes through monetary policy. Hence, we consider that our findings will be useful in designing future monetary policies in a rapidly changing and highly integrated global economy. As such, our evidence concerning Japan's experience of QE from 2001 to 2006 will also be of benefit for other countries. This likewise suggests important research for the future, including the effects of Japanese QQE, LSAPs in the US, and QE in Europe.