Comparison of Monetary Policy Actions and Central Bank Communication on Tackling Asset Price Bubbles—Evidence from China’s Stock Market

We examine the different effects of monetary policy actions and central bank communication on China’s stock market bubbles with a Time-varying Parameter SVAR model. We find that with negative responses of fundamental component and positive responses of bubble component of asset prices, contractionary monetary policy induces the observed stock prices to rise during periods of large bubbles. By contrast, central bank communication acts on the market through expectation guidance and has more significant effects on stock prices in the long run, which implies that central bank communication be used as an effective long-term instrument for the central bank’s policymaking.


Introduction
The 2008 financial crisis and the high volatility of China's stock market in recent years highlight the importance of financial stability. There has been a hot debate on whether asset price bubbles have wider economic consequences and whether monetary policy should react to it [1]. Research of Liu et al. [2] proves that as an emerging market, China's stock market has larger volatility than stock market of the US (a mature market). However, there have been limited studies on how different kinds of monetary policy instruments should react to it. In recent years, the emergence of unconventional monetary policy has caused widespread concern among academia and industry, bringing the comparison of conventional and unconventional monetary policy on the table. Previous studies mention that transmission efficiency of conventional monetary policy is lower in developing than in developed countries because of the financial market incompleteness. Whereas central bank communication, one of the unconventional monetary policies, can enhance the predictability of monetary policy and guide the market effectively by expectation management, and play a central role in macroeconomy supervision [3]. In this paper, we try to tackle the problems of whether monetary policy should react to asset price bubbles in China and what instrument is more efficient in doing this by analyzing different effects of conventional and unconventional monetary policy on asset prices using data of China's stock market, a developing market with unconventional monetary policy used frequently in the past two decades.
In the first place, there are two problems we have to tackle-identification of the bubble, and the endogeneity problem. Previous studies usually identify the bubble by extracting the fundamental component from the asset prices, but the frequently used discounted cash flow method of calculating the fundamental component is not suitable to China's stock market with few dividends. So we adopt the residual income valuation model developed by Feltham and Ohlson [4](F-O model) in this paper. Further, we construct a Time-varying Parameter Structural Vector Autoregression (TVP-SVAR) model of three variables, the monetary policy indicator, the observed asset prices and the fundamental component of asset prices, to identify the bubble by distinguishing different impacts of monetary policy on the observed asset prices and the fundamental component.
The endogeneity problem means that monetary policy and the asset price bubble not only influence each other, but also both can be affected by many other factors. Previous studies have used different methods to overcome the endogeneity problem, such as the use of high-frequency data [5][6], the identification technique based on heteroskedasticity [7][8][9], instrument variable method and event study. In this paper, we set monetary policy shocks to be exogenous in the TVP-SVAR model to overcome the endogeneity problem.
We use quarterly data in this paper. Therefore, when constructing the communication indicator, it is necessary to give weights to communication events in different months in a quarter because of their timeliness. To be specific, the effect of central bank communication is quite timely because of its high frequency. Whereas the accumulation and bursting of a bubble are relatively longer processes. Reconciling the two of them is another issue needed to be addressed. To do this, we give weights to communication events chronologically to differentiate the orders and effects of them by the method of variance decomposition. There are some previous studies on central bank communication putting weights on different code words according to their intensity [5][6][7]. To the best of our knowledge, using weights on the timing of a communication event is the first try in this field. This paper has three contributions. First, it is the first time to compare the different effects of monetary policy actions and central bank communication on asset price bubbles in China, which is not only a comparison of direct and indirect, but also of conventional and unconventional monetary policy instruments. Second, taking account of the timeliness of central bank communication, we give weights to communication events chronologically to differentiate the orders and effects of them, which makes the communication indicator more applicable. Third, we use the F-O model based on residual income to calculate the fundamental component of stock prices, which is more suitable for China's stock market with few dividends.
Through impulse response analysis of the TVP-SVAR model, we find that there are positive responses of the observed asset prices to contractionary monetary policy actions during periods of high bubbles. This is based on the fact that the fundamental component of asset prices has negative responses to monetary policy shocks whereas the bubble component has positive responses. Different from these results of monetary policy actions, central bank communication has more significant long-term effects on asset prices, mainly through its role of expectation management.

Monetary Policy Actions
Whether monetary policy should take account of asset prices is one of the most discussed policy issues in recent years. Akram et al. [5], Christiano et al. [6], Borio [7] argue that there will

Link of Monetary Policy Actions and Central Bank Communication
As two monetary policy instruments, monetary policy actions and central bank communication have certain connections. Central bank communication is considered to be another way of monetary policy transmission before the actual monetary policy actions, which is called the 'market-expectations channel' of monetary policy transmission by Neuenkirch [26] and Gertler and Karadi [27]. Through this channel, communication enhances the predictability of monetary policy, shortens the transmission lag and improves the efficiency of monetary policy actions [28].
Some studies explore further the conditions on which central bank communication can function better. Brand et al. [22] and Jansen and De Haan [29] argue that communication is effective only on the condition that it is consistent with monetary policy actions.
To conclude, there are limited researches studying the differences of the two monetary policy instruments in affecting stock prices and controlling bubbles, especially in the background of China's undeveloped stock market with higher fluctuations. In this paper, we use data of China's stock market and monetary policy to give a first try in this field.

Method and Data Method
The methodology of this paper is based on the theory of rational bubbles raised by Blanchard and Watson [30] and the analysis of Galí and Gambetti [11]. The theory of rational bubbles suggests that asset prices can be decomposed into a bubble component and a fundamental component, and with the hypothesis of investor's rational expectations and rational actions, there could exist asset price bubbles even under the condition of no arbitrage. Investors would buy the assets undervalued, expecting to sell them at a higher price, thus push up the asset price gradually. According to this theory, the fundamental component responds negatively to monetary policy shocks, whereas the bubble component grows at the rate of the interest rate. As analyzed by Galí and Gambetti [11], we can figure out the existence of bubbles when the observed asset prices rise in response to an interest rate increase. As to central bank communication, a similar analysis can be applied.
Based on the analysis above, we build the following three-variable Time-varying Parameter Structural Vector Autoregression (TVP-SVAR) model (1) with two lags (determined based on the highest marginal likelihood), in which we consider time-varying of both the parameters and the variance covariance matrix of the model (stochastic volatility) as in Primiceri [31]. Where is a 3×1 vector of the observed variables at time t. f t is the fundamental component of asset prices. m t represents monetary policy indicator, and p t is the observed asset prices. a t is a 3-dimensional column vector of intercepts. B t , p (p = 1, 2) is a 3×3 matrix of the autoregressive coefficients. Residual ε t follows the distribution N(0, O t ), where O t is the variance covariance matrix. The subscript t of a t , B t , p (p = 1, 2) and O t shows that they are all time-varying, and O t indicates the stochastic volatility of the model. Following Primiceri [31], we assume that the VAR coefficients follow the random walk: We define the covariance Where A t is a lower triangular matrix with diagonal elements of 1, H t is a diagonal matrix. We assume the vector of non-zero and non-one elements of matrix A t (stacked by rows) evolves as random walk: α t = α t−1 + μ α,t . Let h t be the vector of the diagonal elements of H t and we assume it evolves as geometric random walk: In h t = ln h t−1 + μ h,t . Further, all the innovations in the model, ε t , μ β,t , μ α , t , and μ h , t are assumed to be mutually uncorrelated at all leads and lags and follow the jointly normally distribution below.  Where I 3 is a 3-dimensional identity matrix, S β , S α , and S h are positive definite matrices.
To overcome the endogeneity problem, we set monetary policy shocks as exogenous to the observed asset prices, which means that monetary policy indicator doesn't respond contemporaneously to shocks of the observed asset prices. To be specific, we set the simultaneous response coefficients of monetary policy indicator to asset price indicator in the coefficient matrix (covariance matrix) to be 0 in the TVP-SVAR model.
The estimated coefficients are multiplied by allowing random variation of the parameters. So we refer to Primiceri [31] to apply the Markov Chain Monte Carlo (MCMC) methods in the context of a Bayesian inference to estimate the model, in which Gibbs sampling procedure is used to obtain the joint posterior distribution of the parameters in the model. We assume that the initial states for β t , α t , ln h t , and the hyper parameters S β , S α , and S h are independent of each other. The priors for the hyper parameters S β , S α , and S h , are assumed to be distributed as independent inverse-Wishart. The priors for the initial states of β t , α t , and ln h t are assumed to be normally distributed.

Discussion about the Exogenous Assumption
The exogenous assumption of monetary policy shocks to the observed asset prices is the key of our model. In this part, we give a discussion about the rationality of its application in China. For the central bank of China (the People's Bank of China, the PBOC), there are mainly four ultimate objectives, economic growth, price stability, full employment and balance of payments. Inflation is the key factor determining monetary policy decisions, whereas financial stability is just one of the factors it accounts for, which has been communicated by authorities of the PBOC in public for several times. Only when the financial situation affects or contains enough information of the trend of inflation, can it affect policy decisions. We try to justify this assumption and explain it through Figs 1-4. Figs 1 and 2 show trend of the benchmark lending rate and its relationship with the Shanghai Composite Index and the CPI (detailed descriptions of the data used in this paper will be stated in Section "Data"). We can see from the figures that interest rate adjustments mainly follow the trend of the CPI, and there is no apparent causal link between interest rate adjustments and the stock market index. To be more specific, we give detailed explanations of the three periods with interest rate adjustments of high-frequencies, 1993-1994, 2007-2008, 2010-2011. During the year of 1993, the PBOC raised the interest rate considering the hyper-inflation problem during that time, even though the stock market index is hovering around the low point of 1000. The CPI went down since late 1994, after when the interest rate was also turned down gradually. Not until then did the stock market realize a recovery.
Period 2007-2008 is also a clear example. It is almost seven months between the two interest rate rises on August, 19, 2006 andMarch, 18, 2007, during which time the stock market index rose from 1598.02 to 2930.48 for about 83%. We can see that even with the apparent bubble accumulating, the stock market instability has not become the causation of the PBOC to raise the benchmark interest rate. It is only when there appears instability in the inflation, the CPI, which is 2.7% in February, 2007 rising from 2.2% of January, 2007, that the monetary policy decision changed, followed by another five interest rate rises. We can further confirm the exogeneity of monetary policy to the stock market by the fact that the stock market index turned down since October, 2007 from the top of 6092.06, however there was another interest rate rise after that on December, 21, 2007, considering that the CPI is still high in November, 2007 at 6.9%.
Similar situation happened during middle of 2010 and middle of 2011. There were a series of interest rate rises when the CPI rose up whereas there were no apparent bubbles in the stock market.
To conclude, monetary policy does not react to the stock market in China, which makes the exogenous assumption applicable in this paper, even with quarterly data.
When it comes to central bank communication, we can get the same conclusion. Therefore, we can justify our assumption that monetary policy, not only conventional, but also unconventional monetary policy, reacts to the inflation index and is exogenous to the situation of the stock market.

Data
For the three variables in the TVP-SVAR model, we use the logarithmic form of the benchmark lending rate log(r t ) as the indicator of monetary policy actions and the communication index c t as the indicator of central bank communication. Also, we give the results of the overnight interbank lending rate as the monetary policy actions indicator for robustness test. The logarithmic difference form of the Shanghai Composite Index Δ log(Index t ) is applied as the observed asset prices and the logarithmic form of the fundamental component log(funda t ) is Our sample period is 1994Q1-2013Q4 for the estimation of monetary policy actions, 2003Q1-2013Q4 for the estimation of central bank communication, considering the data availability. We also give results of monetary policy actions during period of 2003Q1-2013Q4 in Section "Empirical Results of Central Bank Communication", to make the results comparable. The sample contains the several ups and downs of China's stock market in recent years, including the accumulation, expansion, and bursting of bubbles, thus can be an appropriate representative.
Data resources of this paper are as follows. We get the benchmark lending rate from the website of the PBOC, the Shanghai Composite Index, net assets and return of equity of the stock market (to calculate the fundamental component) from the Wind financial database.

Measurement of Monetary Policy Actions
We select the benchmark lending rate as the monetary policy indicator. There are two reasons why we think of it as appropriate. First, the intermediate target of China's monetary policy has transformed gradually from quantitative indicators to price-based indicators, and the lending rate can fully represent monetary authorities' intentions. Second, the applicability of lending rate as monetary policy indicator in China has been affirmed by the work of Feng and He [32] and Ouyang [20] on monetary policy transmission effects. We also use the more market-oriented monetary policy indicator, the overnight interbank lending rate, to verify the robustness of our results.

Measurement of Central Bank Communication
There are mainly two forms of the PBOC communication. One is written communication, thus reports regularly published by the Monetary Policy Committee (MPC). The other form is oral communication, including press conferences of the PBOC, statements, speeches, and interviews made by members in the MPC. We get written communication events through the official website of the PBOC and the oral communication events through the Internet by keyword searching. When searching the keyword, we only focus on oral communications of the governor of the PBOC, Mr. Zhou Xiaochuan, and nine successive deputy governors, whose tenures should be accounted for to make sure the communication events we obtain happened within their tenures. The search commands we use are the name of the governors together with the term 'monetary policy' to extract all relevant events, and we only select each event once on the basis of its first recording to avoid repeat counting. The oral communication events are summarized in Table 1. Summary and classification of both oral and written communication are shown in Table 2.
The construction of central bank communication indicator is a key to our estimation. There are generally two methods to build this indicator, 'content analysis' [15,33] and 'narrative approach' [34][35][36], both define a dummy variable according to the event's information content. For example, Ehrmann and Fratzscher [33] define +1 as tightening monetary policy, 0 as neutral inclination, and -1 as easing inclination. Narrative approach is different from We define communication events following Ehrmann and Fratzscher [33] in this paper. To keep in line with the work in 'content analysis', we also try to reduce the chance of  We use quarterly data in this paper, so the sequence of the communication event is an important factor in the construction of the quarterly communication indicator. Effects of events happen in the first month in a quarter are definitely different from those of the third month, and it would be biased to just add assignment of each communication event in a quarter to get the quarterly indicator. Therefore, we use the method of variance decomposition to determine the weights that should be put on the communication events happening in the first, second, and third month in each quarter to reflect this timeliness.
We set a VAR model with two lags (derived by AIC) in the following form.
Where 3 ] is a vector of the short-term interest rate ir t --the overnight interbank lending rate--and communication indicators of the three months in a quarter, c t,1 , c t,2 , and c t,3 , which are calculated by adding the assignments of the events within each month. Through the establishment of the VAR model, we get results of the variance decomposition, in which the contributions of the shock of c t,1 , c t,2 , and c t, 3 to the fluctuation of ir t are the weights that should be put on the communication events happening in the first, second, and third month. The results we get are 0.4299, 0.0825 and 0.4876, respectively. Therefore, the final communication indicator is constructed as the weighted sum of the three month indicators.
There are some other studies giving weights to variables according to the sequences that we can refer to. For example, Del Negro et al. [37] study the relationship between forward guidance and the federal funds rate, in which they provide different penalties on deviations of the interest rate path over the short, intermediate, and long horizons to explain the forward guidance puzzle.

Calculation of the Fundamental Component of Stock Prices
We use data of stocks in the Shanghai Stock Exchange (SSE) to calculate the fundamental component of asset prices, in consistent with the use of the Shanghai Composite Index in our model. We apply the residual income valuation model developed by Feltham and Ohlson [4] (F-O model) to calculate the fundamental of stock prices [38,39].
Where V t is the price of the fundamental at time t. BV t is the net worth at time t, which we use the averaged net assets per share of SSE. N is the number of periods investors can predict on the stock market, representing investors' prediction ability. We assume N to be 10 years. ROE is the return on equity, which we apply the average return on equity of the previous 10 years of SSE. ρ is the risk-free rate, which we adopt data of the 10-year Treasury bond yield. The data source is Wind financial database. We calculate the fundamental prices during period of 1993-2013 and find our results quite consistent with those obtained by Chen et al. [40] using the dynamic residual income valuation model, demonstrating the applicability of F-O model and the accuracy of our calculation. Further, we get the approximate bubble size by excluding the fundamental component from the average stock price of SSE (data source: Wind financial database). Fig 5 plots the Shanghai Composite Index and the bubble, from which we can see that the bubble moves closely with the stock index and can be a good fit for the several ups and downs of China's stock market. Fig 6 shows the approximately calculated bubble component proportion in asset prices. We can see that the percentage of the bubble component is over 60% even in the bear market, which indicates the excessive bubble problem in China's stock market.

Empirical Results of Monetary Policy Actions
To compute the posterior estimates of the parameters, we apply the Gibbs sampling procedure by drawing M = 10,000 samples after the initial 1,000 samples are discarded, which is determined by the convergence diagnosis (CD test) results shown in Table 3. We estimate the TVP-SVAR model for different lag lengths and the optimal lag number is 2 based on the highest marginal likelihood. Table 3 and Fig 7 are the main results of the key parameters in the model. Table 3 presents the posterior mean, the standard deviation, critical values of the 95% confidence interval, the convergence diagnosis results of Geweke [41], and the inefficiency factor of the parameters, respectively. The null hypothesis of CD test is the convergence of the posterior distribution, and it cannot be rejected proved by the results here. The inefficiency factor is quite low, indicating the efficient sampling of the parameters and state variables, on the basis that the more the unrelated samples (number of sampling/inefficiency factor), the higher the sampling efficiency. Fig 7 shows the autocorrelation equation, the sample path and the posterior density of the parameters. We can see from the correlation diagram that the correlation drops gradually and tends to be stable, which means the sampling method applied can obtain some relatively low correlated samples. The sample paths also have stable trends. Fig 8 displays the impulse responses of stock market index and the fundamental component to interest rate shocks for one-to three-year horizons over time in the TVP-SVAR model. We also present the impulse responses in the simple VAR to compare. We find that for the simple VAR model, the patterns of responses of stock prices to a tightening monetary policy have changed little over time and reduce to 0 around the year of 2004. In comparison, the impulse responses vary significantly over time in the case of TVP-SVAR model. The responses of the fundamental component to interest rate shocks are mainly negative, consistent with the consensus that the fundamental component decreases to a contractionary monetary policy. Nevertheless, the stock market index rises above its initial value in response to a monetary policy shock during the initial periods and stays positive for the majority of the sample periods. Although this result is not in accordance with what would be expected from the conventional view that asset prices will decrease following a monetary policy shock, it can be explained by the theory of rational bubbles discussed in Section "Method" According to this theory, the bubble component grows at the rate of interest rate. There are negative responses of the fundamental component and positive responses of the bubble component to the rise of interest rate, and the stock market index will also react positively when there is higher proportion of bubble component in the asset prices. This table reports the coefficient results of the model: the posterior mean, the standard deviation, critical values of the 95% confidence interval, the convergence diagnosis results of Geweke [41], and the inefficiency factor of the parameters. The estimates are multiplied by 100. doi:10.1371/journal.pone.0166526.t003

Impacts of Monetary Policy Actions and Central Bank Communication on Asset Price Bubbles
At the same time, we notice that movements of stock market index's responses to monetary policy shocks are always in step with the trend of China's stock market and the accumulation and bursting of the bubbles. Our sample period starts from the year of 1994, a period with high bubbles, when it displays the highest positive response of the stock market index to monetary policy shocks in the impulse response figure. This means that with high bubbles, monetary policy actions are not effective in controlling asset prices. In the following periods, bubbles burst and the figure shows that the impulse response goes down accordingly. After that, the China's stock market embraces a rapid expansion following the reform of listed companies' shareholder structure in 2006, when we can see in the figure that the response climbs up and leads a positive trend again.
Based on the above evidences, we can make the preliminary judgment that results we get from the impulse responses of stock market index to monetary policy actions indicate the    We consider the key reason is the effective guidance of investors' expectation about the future monetary policy by central bank communication. Our results are in line with Gürkaynak et al. [21], who define central bank communication as the 'future path' factor of monetary policy and has relatively long-term effects on the economy. Sarno and Taylor [42] and Fratzscher [17] argue that central bank communication is able to reduce heterogeneity in the market's information and expectations, and induce asset prices to move closely to the underlying fundamentals, which is called the 'coordination channel'. Through this channel, central bank communication has longer-lasting effects because it changes the dynamics of the financial markets. In general, we suggest on the basis of the results here that central bank communication be used as an effective long-term instrument in the monetary policy toolkit.
In fact, panel [b] of Figs 11 and 12 is another robustness test for the conclusions we get in Section "Empirical Results of Monetary Policy Actions". It applies a shorter sample period (2003-2013) for the estimation of asset prices and monetary policy actions and the impulse response results surely confirm the validity and accuracy of the results above.

Concluding Remarks
In this paper, we compare the different effects of monetary policy actions and central bank communication on China's stock price bubbles using a three-variable TVP-SVAR model with stochastic volatility. We get the following conclusions. Although central bank communication does not have significant short-term effects on asset prices either, it can restrain the excessive expansion of asset prices in the long run without causing drastic fluctuations in the stock market. In contrary to the effects of monetary policy actions, central bank communication is effective in controlling asset prices with high bubble proportion. According to our analysis, the key factor is its effective guidance to investors' expectation about the future monetary policy. The central bank can make better use of communication as a long-term instrument.