Chapter 3. Empirical processes and results
3.1 European Union Cross-Countries Evidence: Panel Vector Autoregression
In this estimation, we aim to clarify the relations between heterogeneities of individual country inflation response to the balance sheet policy since late 2008. We use the panel VAR model with stochastic volatility to outline macroeconomic variables to balance sheet policy shock over time.10 Our empirical approach follows Gambacorta et al. (2014).
The advantages of this model are, firstly to identify common economic factors to all economies in the economic depression after financial crises, and secondly to use mean group estimator that allows cross-country heterogeneity and does not require the identical structures and dynamics of the economies in the VAR model.11 This is particularly important because it allows us to consider different responses across countries to the shock of balance sheet policy measures.
The Panel vector autoregression (VAR) model is estimated as
10 As Gambacorta et al. (2014) indicated, structural VAR have been extensively used to analyze the macroeconomic effects of monetary policy. For instance, structural VAR are applied in Bernanke and Blinder (1992), Strongin (1995), Bernanke and Mihov (1998), and Christiano et al. (1999) to study the United States; Peersman and Smets (2003), the euro area.
11 However, this model does not capture the cross-country spillover effects of balance sheet policy measure. In that case, the balance sheet policy shocks for those countries, which not adopting unconventional monetary policy, originate from the open operations of national central banks.
‧
1. RGDP𝑖,𝑡：the nature log of seasonally adjusted real GDP,
2. CPI𝑖,𝑡：the nature log of seasonally adjusted consumer price index;
3. TA𝑖,𝑡：the nature log of seasonally adjusted central bank total assets;
4. SI𝑖,𝑡：the nature log of seasonally adjusted stock market price index.12 𝛼𝑖 is a vector of constants for each European Union country 𝑖, 𝐴(𝐿)𝑖 is a matrix polynomial of the lag operator L, and 𝐵𝑖 is the contemporaneous impact matrix of mutually uncorrelated disturbances 𝜀𝑖 for countries 𝑖 = 1, … , N.
The specification of this model simplifies and explains the features in the period conducting balance sheet policy, which includes: (i) the aggregate output and prices to capture the macroeconomic dynamics, (ii) the assets holdings by the central banks at the zero lower bound, and (iii) the major stock market index of each country to capture the financial market turmoil.13 In addition, we expand the analysis of Gambacorta et al.
(2014) by examining balance sheet expansions of central bank in a broader set of sample countries and across a broader span of sample period from January 2008 to May 2017, and twenty-eight European Union countries, i.e., namely Germany (BD), Belgium (BG), Bulgaria (BL), Croatia (CT), Cyprus (CP), Czech (CZ), Denmark (DK), Estonia (EO), Spain (ES), Finland (FN), France (FR), Greece (GR), Hungary (HN), Ireland (IR), Italy (IT), Latvia (LV), Lithuania (LN), Luxembourg (LX), Melta (MT),
12 All the data in this paper is gathered from the Data Stream. About decomposing quarterly real GDP data in order to parallel available monthly monetary data, we use Eviews software (time-series frequency conversion function) to convert quarterly data to monthly data, and choose cubic type of low to high frequency method.
13 Haldane et al. (2016) demonstrated that QE has an upward pressure on the price of corporate bonds and equities, while put a downward pressure on the exchange rate. Furthermore, there is one stream of literature after the crisis proves that unconventional monetary policy surprise affected the stock returns. (Glick and Leduc 2012, Wright 2012, Rogers et al. 2014)
‧
Kingdom (UK).14 Data were taken from national central banks as well as Datastream. We use data of European union countries whichconsist of 19 Euro countries and 9 non-euro countries in this dissertation.15 The domestic economies of non-euro countries have high degree of economic interdependence with the ECB monetary policy.16As Gambacorta et al. (2014) suggested, a balance sheet policy shock is identified as an exogenous alteration to central bank balance sheet. The variables are setas policy-makers conduct the monetary measures. In order to accomplish this, we use a mixture of zero and sign restriction on the impact matrix B of equation (1). We adopted the sign identification restrictions which has been proposed to produce the impulse responses by means of implied signs.17 In our model, we disentangle real economy from balance sheet policy and other financial shocks without imposing a notable impact from the responses of macro-variables by using the mixture restrictions, in order to leave the output and inflation responses open as they are the research inquiries in this estimation. The sign identification assumptions are outlined in Table 3. Conformed to the reference model, we
14 Bluwstein and Canova (2016) state the onset of ECB unconventional monetary in Dec. 2007 when ECB proceeding €271.6 billion reciprocal currency agreement, whereas our sample period starts from Jan. 2008 to Jun. 2017. We find that most of UMP literatures use early 2008 as the beginning of the sample date period.
15 From official website of EU, the non-euro area member countries are Bulgaria, Croatia, Czech Republic, Hungary, Poland, Romania, Sweden. Member countries negotiate an opt-out from the EU legislation are UK and Denmark.
16 See literatures, such as Dibooglu and Kutan (2001), Fidrmuc and Korhonen (2003), Laxton and Pesenti (2003), Fidrmuc (2004), and Golinelli and Rovelli (2005).
17 The mixture of zero and sign restrictions have been conducted by Canova and De Nicolo (2002) and Uhlig (2005) before. By using mixture restrictions, the permissible impulse function is decreased;
consequently, it enhances the identification of balance sheet policy shock.
‧
assume the identifying sign restrictions as following. First, it is assumed that impact of shocks to the central bank balance sheet on output and consumer prices is only one lagged. While the same period impact on output and consumer prices is restricted to be zero, the alteration to both variables are allowed to have an instantaneous impact on the balance sheet and stock market volatility.18
Secondly,recent literatures suggested that open market operation of securities purchase, such as conducting balance sheet policy, would increase the price of equities, thereby we assume that the balance sheet policy shock would lift the stock market index.19 These restrictions are bound for only one month after the balance sheet policy shock. Based on the usual lag-length selection criteria, one lags of the endogenous variables are used in the estimations.20
Table 3
Zero and sign restrictions of a balance sheet policy shock
Our model is estimation in following steps. First, each equation of
18 This assumption, as Gambacorta et al. (2014) indicated, is common in monetary transmission studies.
It resolves monetary policy shocks from real economy disturbances, such as aggregate supply and demand shocks, and does not push macro-variables to respond in certain directions.
19 Meltzer (1995) suggested the open market buying of securities would make more base money to be held, thereby wealth owners buy the existing bonds and real capital for additional money.
20 As Abrigo and Love (2016) suggest, we first run panel VAR lag order selection on estimated sample, the selection standard is based on the model Bayesian information criteria (BIC), Akaike information criteria (AIC) and Hannan-Quinn information criteria(HQIC).
Output
‧
reduced-form in the VAR model is estimated at each single country level taking into consideration the correlation among the residuals of the same endogenous variable across 28 countries, i.e. the correlation between all countries output residuals, between all price residuals, between all stock market index residuals, and between central bank balance sheet residuals.
Second, we identify the balance sheet policy shocks of each individual economy by using the identification of restrictions in Table 3. Followed by the model designed of Gambacorta et al. (2014), since the shocks in equation (1) are mutually orthogonal, 𝐸(𝜀𝑡𝜀′𝑡) = 𝐼 , the variance-covariance matrix Ω of an individual country VAR system is equal to 𝐵𝑄𝑄′𝐵′ , where B is the Choleski decomposition of Ω, and Q an orthonormal matrix of the form:
Q = [
with QQ’=I. Due to B is the Choleski factor of Ω, variations to output and prices will affect balance sheet and stock market immediately, while the contemporaneous impact of the third and fourth shocks in the system on output and prices is restricted to be zero. As the result, one of them is a variation to the stock market index and the other an exogenous shift to the central bank balance sheet, which disentangled by the sign restrictions.
Then, we draw a random 𝜃 in the range [0, 𝜋], where the 𝜃 is the same for all countries, and generate the corresponding impulse function for each individual economy:
𝑅𝑡+𝑘 = 𝐴(𝐿)−1𝐵𝑄(𝜃)𝜀𝑡 (3)
The requirement of Gambacorta et al. (2014) is that the two remaining
‧
shocks satisfy the sign restrictions for all countries simultaneously with the condition 𝑅𝑡+𝑘𝑆𝐼 ≥ 0 and 𝑅𝑡+𝑘𝑇𝐴 > 0, and then keep the draw. Otherwise, the draw is rejected. However, we relax the condition of satisfying the sign restriction for all countries to for mean group in order to have enough impulse functions. In accordance with the setting of Gambacorta et al.
(2014), we have to repeat the draw by the bootstrapping until having 5000 mean group impulse response functions, and average the impulse response function from the individual economies to get a mean group impulse response function of European Union countries. Even though we relax the requirement, our sample countries are too many to completing the draw.21 Consequently, we record the 16thand 84th percentiles of this practice in the figures.22 The mean group PVAR impulse function was shown in the Figure 7. The impulse function indicates that the overall European Union balance sheet policy shocks enlarges the central bank balance sheet about 0.06% which in line with reference model, and diminish to 0.01% about 35 months in our model.23
In addition, output growth and inflation reply to the shocks with a slight upsurge, and the summits are 0.008% and 0.006% respectively. On the other hand, stock market index growth, denoted the financial system, is more sensitive to the balance sheet policy shocks with an instantaneous
21 As a result, we take off the data group of Czech and Estonia for completing 5,000 impulse response functions for the rest of countries.
22 As Gambacorta et al. (2014) suggested, the impulse response bands should not be understood as conventional confidence bands since equation (3) reflects model uncertainty by the draw of θ; as well as the sampling uncertainty by the bootstrapping draw. However, the sign restrictions literatures, such as Giordano et al. (2007) and Bénétrix and Lane (2009), usually consider the 16th and 84th percentiles of the impulse respond distribution as the confidence interval.
23 Gambacorta et al. (2014) indicate that the ECB unconventional monetary policy shock is
characterized by an increase in the euro area central bank balance sheet of about 2.5% that diminish to 0.4% after about 4 months from their impulse function.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
upright response, and its summit is about 0.3%. Our mean group PVAR results are in line with literatures, which revealing both output growth and inflation are affected by the shocks, but extremely small.24
24 See literatures, such as Baumeister and Benati (2010), Gertler and Karadi (2011), Kapetanios, Mumtaz et al. (2012), Gambacorta et al. (2014), Pesaran and Smith (2016), and Weale and Wieladek (2016).
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Note: Horizon is monthly, and the shadow area is the 16th and 84th bootstrap percentiles.
Figure 7 The mean group impulse function
To inspect the individual country impulse responses, each country has balance sheet increase for answering the shocks, however with divergent degree. The advance economies of euro area, i.e., Germany, France, and Netherlands, response to the balance sheet policy measures with higher
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
increase than the mean group Panel VAR impulse response and also fade out faster (about 2 to 20 months) than the mean group impulse function.
Those euro countries which suffer sovereign debt crises, i.e. Cyprus, Spain, Greece, Ireland, Italy and Portugal, have dissimilar increase of central bank balance sheet assets responding to shocks, some are higher while some are lower, but their impulse responses do not fade out, yet maintaining higher than the benchmark impulse response. The opt-out countries of European Union, UK and Denmark, UK has 8% increase while Denmark with 7%
increase responding to the shock, and their impulse responses disappear with faster speed, about 10 months in UK and one month in Denmark.
In this study, we focus on the inflation effects. The inflation responds to balance sheet policy shock with an 0.15% increase in the mean group impulse function, reaching the peak after 17 months and returning to baseline after 35 months.25 The inflation effects of balance sheet policy are indeterminate from the mean group impulse response, where the rise of inflation is insufficient. The individual country impulse function of our PVAR results reveals the heterogeneous inflation response cross-country.
The inflation impulse responses of advanced economies are divergent. The inflation effect of balance sheet policy shock to Germany is at the range from 0.1 to -0.1%, whereas inflation effect of Germany is negligible but lasts more than 35 months. In Belgium, the impulse function indicates that CPI declines to -0.8% in response to the balance sheet policy shocks,
25 Gambacorta et al. (2014)) indicates that the impact on prices is a temporary effect with a peak increase of 0.08 % while the impact of interest rate shocks on the price level is found to be very sluggish with a peak only after about 2 years or even later.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Note: Horizon is monthly, the dotted area is the 16th and 84th bootstrap percentiles of individual country, and the shadow area is the 16th and 84th bootstrap percentiles of mean group.
Figure 8 The central bank asset impulse functions of European Union countries
Germany Belgium Bulgaria Croatia Cyprus
Denmark Spain Finland France
Hungary Ireland Italy Latvia
Luxembourg
Melta Netherland
s
Austria
Portugal Romania Sweden Slovenia
United Kingdon
Greece
Lithuania
Luxembourg
Poland
Slovak Republic
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Note: Horizon is monthly, the dotted area is the 16th and 84th bootstrap percentiles of individual country, and the shadow area is the 16th and 84th bootstrap percentiles of mean group.
Figure 9 The inflation impulse functions of European Union countries
Germany Belgium Bulgaria Croatia Cyprus
Denmark Spain Finland Greece
Hungary Ireland Italy Latvia Lithuania
Luxembourg Melta Netherland
s
Austria Poland
Portugal Romania Sweden Slovenia
United Kingdon
France
Slovak Republic
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Note: Horizon is monthly, the dotted area is the 16th and 84th bootstrap percentiles of individual country, and the shadow area is the 16th and 84th bootstrap percentiles of mean group.
Figure 10 The output growth impulse functions of European Union countries
Germany Belgium Bulgaria Croatia Cyprus
Denmark Spain Finland Greece
Hungary Ireland Italy Latvia Lithuania
Luxembourg Melta Netherlands Austria Poland
Portugal Romania Sweden Slovenia
United Kingdon
France
Slovak Republic
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Note: Horizon is monthly, the dotted area is the 16th and 84th bootstrap percentiles of individual country, and the shadow area is the 16th and 84th bootstrap percentiles of mean group.
Figure 11 The stock market impulse functions of European Union countries
Germany Belgium Bulgaria Croatia Cyprus
Denmark Spain Finland Greece
Hungary Ireland Italy Latvia Lithuania
Luxembourg Melta Netherland
s
Austria Poland
Portugal Romania Sweden Slovenia
United Kingdon
France
Slovak Republic
‧
nearby country – Luxembourg has similar reaction as its CPI decreases to 0.4% about 10 months later. The CPI hits the bottom about 15 months later, and return to the baseline about 35 months. On the contrary, the impulse response of France and Netherlands are different. The impulse response of France attains the peak of 0.2% about 7 months and return to the baseline around 20 months later. Netherlands reaches the peak of 0.6% about 15 months, and return to baseline about 25 months later.
The countries undergoing European sovereign debt crisis results in three completely different cases. The inflation impulse responses of Spain and Italy lower to -0.3% after hitting the bottom for 5 months, whereas Italy hits the bottom after 10 months. Cyprus and Ireland respond to balance sheet policy shocks with a short upsurge after 1 to 3 months, then the responses lower to -0.4% and -0.6% respectively and gradually. For Greece and Portugal, they respond to the shocks with a rise of CPI to the peak of 0.65% after 15 months and 35 months respectively. About the recently joined member countries, Latvia was notable that its inflation replies to the shocks with an escalation to 2%, the highest upsurge response among the EU countries.26 Lithuanian CPI answers to the shocks with an increase of 0.8% about 12 months later, while Malta with an increase of 0.4% after 5 months. Slovenia and Slovakia respond with decadent inflation levels, where Slovenian bottom is -0.6% and Slovakian is -0.4%.
The replies of UK and Denmark, owning to their opt-out of the single currency agreement, are dissimilar. The inflation level of UK respond to
26 From official website of ECB, we define the recently joined euro area countries as those entry euro area after the onset of euro at Jan. 1st 2002. Slovenia adopted euro at Jul. 11rd 2006. Then, Cyprus and Malta adopted euro at Jul. 10st 2007. Later, Slovakia and Estonia joined at Jul. 8th 2008 and Jul. 13th 2010 respectively. At the last, Latvia and Lithuania respectively joined euro area at Jul. 9th 2013 and Jul. 23th 2014.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
shocks with an upsurge to 0.25% after 4 months and return to baseline after 10 months, while Denmark responding the shocks with a short ambiguous upsurge and decreasing gradually to -0.1% about 20 months later. For the newly EU member country, Bulgaria answer the shocks with an instant upsurge to 0.5% and return to zero about 10 month later, while the peak of Romania is 0.4% and lasting for more than 35 months.27 The central European Countries, i.e., the Czech Republic, Hungary and Poland, conduct interest rate policy and do not adopt euro. The Polish inflation response decrease to -0.5% about 10 months later, after the shocks imposed, while the Hungary react to the shocks with an indefinite inflation rise.
It is notable that the results of inflation impulse function revealing that the balance sheet responses of non-euro area countries to the balance sheet policy shocks are no less than euro countries. Although the mean group impulse function in figure 9 exhibits the response of overall national central bank assets to balance sheet policy shock are 0.06% in the peak, the non-euro country responses, from the individual EU country impulse functions, are no less than euro country responses. Even the non-euro countries which did not adopt balance sheet policy have the same degree of responses. As we mentioned early, balance sheet policy is “conventional”
open market operation of central banks. The results uncover that the central bank assets of non-euro countries also have instant responses after the crisis.
27 From official website of EU, we define the newly EU member countries are Bulgaria, Romania and Croatia. Bulgaria and Romania joined at Jan. 1st 2000, and then Croatia join at Jul. 1th 2013.
‧
3.2 European Union Cross-Countries Evidence: Panel Data Results
Next, we use panel data model to further look into the relations between inflation and variables, which constituting the PVAR model in the preceding section. We prefer panel data in fixed-effect model for several reasons. First, we aim to investigate the cross-countries heterogeneous responses to the expansionary of monetary base, a cross-sectional estimation is more adequate. Second, a fixed-effect model is more sufficient to detect the individual attribute of the cross-section observations to one specific focus which is allowed to be correlated with the explanatory
Next, we use panel data model to further look into the relations between inflation and variables, which constituting the PVAR model in the preceding section. We prefer panel data in fixed-effect model for several reasons. First, we aim to investigate the cross-countries heterogeneous responses to the expansionary of monetary base, a cross-sectional estimation is more adequate. Second, a fixed-effect model is more sufficient to detect the individual attribute of the cross-section observations to one specific focus which is allowed to be correlated with the explanatory