Monetary spillovers from the ECB : reasons to adjust?

(1)

University of Amsterdam

Faculty of Economics and Business

Monetary Spillovers from the ECB:

Reasons to Adjust?

Mehmet Emre Demirkiran 10525742

Supvervisor: Dr. K. Mavromatis University of Amsterdam Second reader: N. Leefmans, MS University of Amsterdam Word count: 9,060

(2)

This document is written by Mehmet Emre Demirkiran who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Signed: Date:

(3)

List of Figures

1 Hodrick-Prescott filters on GDP . . . 12

2 Standardized GDP figures in levels . . . 13

3 CPI inflation rates . . . 15

4 Main monetary policy rates . . . 16

5 Impulse-response functions of Turkey . . . 25

6 Impulse-response functions of Poland . . . 26

7 Impulse-response functions of Hungary . . . 27

(5)

List of Tables

1 LM Test Statistics for Turkey . . . 36

2 LM Test Statistics for Poland . . . 36

3 LM Test Statistics or Hungary . . . 37

4 LM Test Statistics for the Czech Republic . . . 37

5 VAR Lag Order Selection Criteria for Turkey . . . 38

6 VAR Lag Length Criteria for Poland . . . 39

7 VAR Lag Length Criteria for Hungary . . . 40

(6)

Abstract

In a rapidly integration world, spillover effects occur all the time. Although the financial markets are well-integrated, especially in the West, and the goods markets find ever new ways to further integration, a study on the monetary spillover effects from large economies on smaller ones has only recently found curiosity among scholars; in particular, those spillover effects stemming from the European Central Bank.

In this study, I research the monetary spillover effects from the European Central Bank. Therefore, I employ a structural Vector Autoregressive Model to be able to produce a struc-tural analysis by way of identifying the monetary spillovers from the ECB. By producing so-called Impulse-Response Functions, I have found results that indicate some (minor) monetary spillovers, but monetary spillovers nonetheless.

Finally, I interpret my results and give some advises for policies on the basis of these interpretations.

1 Introduction

The Second Wave of globalization led to a decades long post-WWII integration of trade and finance markets worldwide and in particular the European Union with the EMU as the most advanced and prominent form. Naturally, the increasing interconnectedness between economies leads to increased spillovers of public policy among economies during normal and non-normal times. Crises might even bring about a more pronounced effects of these spillovers as the interconnectedness might affect the macroeconomy more heavily during times of increased volatility. As a response to the worsening macroeconomic and monetary variables during the Global Financial Crisis (GFC) and the Sovereign Debt Crisis (SDC) of the EU, the ECB set forth a monetary policy that was expansionary. The ECB primarily used conventional methods until the Zero Lower Bound (ZLB) was hit. However, in this (increasingly) open world, the policies of the ECB are not without effects elsewhere.

After the ZLB was hit, the ECB implemented unconventional monetary policy (UMP). Such UMPs — e.g. quantitative easing (QE) — aims at boosting monetary and macroeconomic variables by lowering interest rates such to boost (private) investments in the (real) economy. Besides the direct interest rate channel, the re-balancing of portfolios for short and medium term assets can also create positive output effects. The pass-trough of CMPs and UMPs gets transmitted through several variables, among which the prices of the sovereign bonds of the Eurozone economies is one. As such, over the past decade, the spread of the yields between the different Eurozone economies have been lowered, as well as their volatility. Several years further in time, the macroeconomic variables in the Eurozone show signs of recovery in domestic, i.e. Eurozone, economies. So far, policymakers have different opinions about the success of the program and its contribution to the economic recovery. Some view the ECB’s endeavor as successful in affecting the current macroeconomic variables as well as future expectations. Others, such as DNB president Knot, remain skeptical due to, inter alia, the increased balance sheet of the ECB, (additional) difficulties in the pensions systems, and/or failure of QE to achieve its intended results. However, these analyses consider domestic aspects of the shock, i.e. a country-specific view.

(7)

Whereas the domestic impact of the ECB’s monetary policy is researched at large, the knowl-edge on the international spillovers of the ECB’s monetary policy lag behind (Potjagailo, 2017). Unsurprisingly, this difference is partly due to the fact that the ECB has a mandate for the Euro-zone and not for foreign economies. The impact of ECB monetary policy on member economies is simply more interesting and more relevant. Nevertheless, there is an increasing body of literature on the effects of (U)MP of larger economies on smaller ones. This body of research argues that the interconnectedness of financial sectors and economies by increasing trade flows will leave spillover effects, and that the extent of these effects should be considered by policymakers to improve the efficacy of their policies. In this line, Kucharcukova, Claeys, and Vasicek (2016) have researched the spillovers of the ECB’s monetary policy outside the Eurozone but within the Single Market.

Similarly, Halova and Horv´ath (2015) study the spillover of the ECB on Central Europe as well,

but focus on UMPs.

Specifically, countries that are within the Single Market or who have a (relative) large trade volume with the members thereof, will be affected by the ECB’s MP. On one side, these foreign economies can get affected due to changes in the macroeconomic activity of the Eurozone as a result of QE. On the other hand, they might get affected by changes in financial variables. Since capital markets are liberalized to a great extent, spillovers of ECB MP are at first expected in these variables. Exchange rates and investment flows between Eurozone and non-Eurozone economies could move along in the same direction.

A great disadvantage for the non-Eurozone countries is that the ECB has no mandate to adjust their policies based on the economic situation of these countries. However, as they are affected by ECB policies, they need to consider the policies of the ECB in their forecasts in order to improve policies for the future.

In this thesis, I wish to study the spillover effects of MP of the ECB on these neighboring economies. Specifically, I wish to investigate the effect of the main policy rate of the ECB on several variables of the sample countries. My methodology is a structural vector autoregression (SVAR). This method is not uncommon for this type of research, as other have done so as well (see (Potjagailo, 2017, pp. 129-30) for a more detailed analysis). I will use time-series data from January 2003 until December 2017. These data will be gathered and/or interpolated to a monthly frequency

The structure of this paper is as follows. Firstly, the next section presents the recent literature on the topic. It starts with the theoretical literature and then discusses the empirical literature that is now available on the subject matter. Section 3 describes the data that is used in this research as well as some stylized facts on these data. After the data section, the fourth section will present the empirical model. This section firstly describes a reduced-form VAR model and then turns over to the structural VAR model. The results are presented in the 5th section and ordered on their sample country. Afterwards, the 6th section discusses the results and provides some policy recommendations. A conclusion is given in section 7.

(8)

2 International spillover of monetary policy

This section will discuss the existing literature on the international transmission of monetary policy for the empirical analysis presented later on in this thesis. Firstly, theoretical transmission channels are discussed after which a presentation of current findings found so far and presented in empirical literature will be given. For the theoretical transmission channels, two models will be used to explain possible causal relations with respect to the spillover of monetary policy; the Mundell-Fleming-Dornbusch and open-economy DSGE models.

2.1 Theoretical transmission channels

Potjagailo (2017, pp. 128-9) presents a good overview of two theoretical models that explain spillovers of monetary policy to other economies. The first one is the traditional textbook Mundell-Flemming-Dornbusch model. The other a standard open-economy DSGE model. Additionally, there is a difference in effect depending on exchange rate regime (flexible or fixed).

2.1.1 Traditional Mundell-Fleming-Dornbusch models

In the traditional models, expansionary foreign MP shocks increase the foreign demand for domestic goods as the foreign output and national income increase after the shock. This shock leads to an increase in the demand from foreigners for domestic goods. The increase of demand for domestic goods leads the home country to export to the foreign country. Ceteris paribus, this additional (net) export leads to a rise in the output of the home country. The channel through which the increase of exports happens is the open-economy IS curve. Put differently, the expansionary foreign MP shock leads to an income absorption effect or demand channel effect on the domestic market.

However, when we allow time to move on and the economy to regain its equilibrium, the Mundell-Fleming-Dornbusch model predicts that the domestic currency will appreciate. Not only because the foreign currency will depreciate in value due to the foreign monetary expansion and the increase of foreign currency thereof; but also because the rise in (net) exports from the home country leads to an appreciation of the domestic currency due to the denomination of the additional exports in home currency. Eventually, this can lead the domestic trade balance to deteriorate and decrease domestic output via the expenditure switching effect as the price of imports will become cheaper (see Dornbusch (1980); Gali and Monacelli (2005); Potjagailo (2017)).

So far, the analysis is for a small open economy with a flexible exchange rate. However, when countries peg their currency against (in this case) the euro, theory suggests that domestic output should move similarly with foreign output. This is because the increased foreign demand increases foreign output and the trade channel connects the increasing foreign demand to the domestic economy. A pegged currency will, contrary to the abovementioned, disrupt the economy in less volatile manners, depending on the peg. However, in countries with flexible exchange rates, the adjustment of the exchange rate counteracts the previously mentioned demand channel and thus restricts the a priori knowledge on the direction of spillovers. These spillover effects are ambiguous.

(9)

Gali and Monacelli (2005) contend that the effect in case of a flexible exchange rate depends on the degree of openness to trade and the elasticities of substitution between domestic and foreign goods and inter temporal substitution that captures the economy’s consumption smoothing preference via intratemporal trade.

Standard open-economy DSGE models capture spillover effects over trade flows. Strong financial integration and decreasing or outright abolishment of trade barriers can transmit monetary policy internationally. A drop in the foreign interest rate of a large open economy can lead to a decline in the domestic interest rates. This is suggested to occur indirectly via a decline in the world interest rates (Svensson and Van Wijnbergen, 1989). The immediate effect of a decreasing foreign interest rate of a large open economy is that the financial products denominated in foreign currency become cheaper and credit demand increases. Bernanke and Gertler (1995) argue that credit-constraint borrowers benefit from this in particular as they may access previously inaccessible credit.

As in the traditional model, an appreciation of the domestic currency will also lower the value of debt held domestically and denominated in foreign currency. This leads to positive effects on the indebtedness of a country. Moreover, the creditworthiness of debtors also increases, as their debt decreases in value. In reality, we have seen reversed examples as indebted countries with high amounts of foreign denominated debt fail to pay off debt when a sudden depreciation of their domestic currency occurs (e.g. Latin American countries). A specific example of integration is the ownership of domestic banks by foreign banks. Especially for Europe, where bank-based loans are relatively important, foreign monetary expansion can have positive effects on domestic lending in proportion to foreign ownership of the domestic banking industry. Put differently, easier access to money abroad could flow through subsidiaries to home. In this case, domestic output can follow the same path as foreign output (Devereux and Yetman, 2010).

The high degree of economic integration in the European Union and the (increasingly) open trade policies of its neighbors suggest that the abovementioned effects will be relatively large on the European continent. Continuing integration and centralization of economies as well as harmonizing policies between Eurozone as well as non-Eurozone countries predict that the spillovers will become larger. Moreover, since some economies are relatively large and not within the Euro, different configuration will have different effects.

2.2 Empirical literature on monetary spillovers

The empirical literature on monetary policy spillovers is limited. Among the few studies that exist, a large part focuses on the monetary spillovers of the US Fed. Moreover, the lack of UMPs makes it harder to measure the spillovers of UMP on other economies. Nevertheless, as the GFC has provided a pressure for central banks to implement UMPs, there is an increasing literature on the subject. Sadly, until now, this literature has not sufficiently researched the effect of the ECB’s UMPs.

(10)

sovereign bond rates of advanced economies by 20 to 80 basis points compared to 100bp in the US. Deconstructing the channels, they find that the fall is mainly attributable to the signaling channel in the US and Canada and to portfolio restructuring in Australia, Germany, and Japan.

Fratzscher, Duca, and Straub (2016) detect a positive boost to global equity markets and a lowering of credit risk between banks and sovereigns as a result of ECB UMPs among the members of the G20.

With respect to the empirical methods employed, almost all studies make use of a VAR model. Feldkircher and Huber (2016) apply a Bayesian Global VAR (GVAR) model. Jannsen and Klein (2011) use a SVAR model and find that MP shocks have an increasing effect in five Western non-Euro economies. Liu, Mumtaz, and Theophilopoulou (2011) and Mumtaz and Surico (2009) both use a factor-augmented VAR (FAVAR) model and find that the UK output increases despite and appreciation as a consequence of international monetary expansions and that the significance

decreases to zero over time, respectively. Feldkircher (2015) and Halova and Horv´ath (2015) also

apply a GVAR model and analyze the transmission mechanism of interest rate shocks from the Eurozone to a set of non-Euro countries. They find that small economies react stronger as suggested by the theory on small-open economies. Gambacorta, Hofmann, and Peersman (2014) make use of a panel VAR. They focus on the ZLB and the effectiveness of UMPs. They find that an exogenous increase in the balance sheet of a central bank at the ZLB leads to a temporary rise in economic activity and consumer prices. These effects seem similar as those of CMPs. Despite heterogeneity among countries, these effects seem not significantly different between them.

(11)

3 Data and Stylized Facts

The goal of this section is to present the data and its sources. It provides an overview of the indicators that are used in this research. Additionally, this section presents some stylized facts that the data show. This is done because the economies in the sample show some differences; even though they are largely similar.

3.1 Sample and variable selection

The data of this research is gathered via Thomson Reuters’ Datastream software. For this research, it is relevant to use data going back to the formal introduction of the Euro; this has been done so by abolishing the two-currency system on January 1, 2002. However, as in most researchers, I encountered issues in finding data going back to the formal introduction of the Euro. One variable does not provide data from January 2002 onward and to keep my sample consistent for all countries, I chose a different starting date. Therefore, I have chosen January 2003 as the start of my sample and December 2017 as its end.

As the empirical model further in this chapter shows more formally, the variables at the basis of this study are (1) the output gap of the country in question, (2) its inflation rate [year-on-year rate in percentages], (3) its main central bank policy rate, and (4) its effective exchange rate in real terms. The countries for which I gathered data are the Eurozone countries (aggregated to one or directly seen as one whole), the Czech Republic, Hungary, Poland, and Turkey. The reason for this choice is because these countries are situated in Central and/or Eastern Europe and all have no formal introduction of the Euro within the sample period.

A consideration was made to add the fourth member of the Visegrad group, namely Slovakia. But since Slovakia joined the Eurozone within the sample period, addition of this country was omitted to keep things practical. Moreover, the exclusion of Slovakia will most likely not affect the conclusion of this paper, because the conditions for Slovakia entering the Euro was to align its monetary policy outcomes with that of the ECB.

Additionally, there are other European countries that do not use the Euro as their currency. These countries have different exchange rate regimes with respect to the Euro. Sweden has a flexible exchange rate regime. Contrary to Sweden, Denmark has its currency pegged to the Euro. Another example is Switzerland, which had to abandon the peg of its Franc with the Euro in 2015 due to pressures from capital fleeing to safe havens. Nevertheless, the choice was made to exclude these countries. Although each and individual case is interesting in and of itself, these countries are all advanced economies with a long tradition of economic integration. The countries of the sample have a peculiarity in that they have recently started to open; they are transition and/or emerging market economies.

(12)

3.2 Data

With respect to obtaining the output gap, I first collected data on output that was not seasonally adjusted. This data is standardized by Thomson Reuters and comes in quarterly frequency. Because it has been standardized by Thomson Reuters, it is directly ready for use—all data on GDP were retrieved from the same source. However, to obtain the data with monthly intervals, I used EViews. Therein, the function of interpolation helps to generate high frequency data from low frequency data. The method I chose was cubic spline. Other methods are available as well. Fernandez (1981) provides a method for interpolation in which other time-series data are used as proxy; most often in short-run models, the industrial production index is used. However, I chose to use EViews cubic spline as this was the more practical option to work with.

Having obtained a monthly series for the variable of GDP, a logarithmic transformation was done. Thereafter, a Hodrick-Prescott filter was used to separate the cycle from the trend and thus get the gap of GDP. As I am working with monthly data, the value of the lambda was set at 14,400. The results are shows in figure (1).

For the variable of inflation rates, searches in Datastream resulted in various central banks and other national institutions that generate and distribute this data. Specifically, the variable here is the general CPI rate that is not seasonally adjusted and based on year-on-year calculations. (The choice has been made to keep the data unadjusted as it was not possible to consistently find seasonally adjusted data for all variables.)

Just like the inflation rate, the central bank main policy rates data was gathered in Datastream. Similar to the inflation rates, the data goes back to the formal introduction of the Euro.

For the exchange rates, I gathered data in Datastream from the Bank for International Set-tlements. Specifically, the choice was made to use the broad interpretation of the real effective exchange rates as these were available for all countries. These data are in index form. An impor-tant consideration for the addition of this variable is the fact that we want to answer a question that includes small and open economies. These features require us to account for the effects of the exchange rates in transmitting spillovers and/or affecting the domestic (small) economy either positively or negatively.

3.2.1 Additional variables

Because of differing characteristics in economies, additional variables may help identifying monetary spillover effects. For example, the Turkish lira is stylized by a continuing decline in value against the dollar and other major currencies since 2004. This increases the foreign-denominated debt of the Turkish private sector and may lead to a higher percentage of non-performing loans, bankruptcies,

or lower investments. The interest rate channels partially affect the value of Turkish

foreign-denominated debt. Therefore, adding this debt may provide a better image on how (large) the spillovers work.

(13)

-.3 -.2 -.1 .0 .1 .2 .3 2.8 3.2 3.6 4.0 4.4 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 L_CZECH_GDP Trend Cycle

Hodrick-Prescott Filter (lambda=14400)

(a) Hodrick-Prescott filter on Czech GDP

-.15 -.10 -.05 .00 .05 .10 .15 7.6 7.8 8.0 8.2 8.4 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 L_EURO_G DP Trend Cycle

(b) Hodrick-Prescott filter on Eurozone GDP

-.3 -.2 -.1 .0 .1 .2 .3 2.8 3.0 3.2 3.4 3.6 3.8 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 L_HUNG ARY_GDP Trend Cycle

(c) Hodrick-Prescott filter on Hungarian GDP

-.3 -.2 -.1 .0 .1 .2 .3 3.6 4.0 4.4 4.8 5.2 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 L_PO LAND_GDP Trend Cycle

(d) Hodrick-Prescott filter on Polish GDP

-.4 -.2 .0 .2 .4 4.0 4.4 4.8 5.2 5.6 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 L_TURKEY_G DP Trend Cycle

(e) Hodrick-Prescott filter on Turkish GDP

(14)

3.3 Stylized Facts

In this section, several stylized facts will be presented to give preliminary interpretations to the data. Figure 2 shows the GDP of the sample countries excluding the Eurozone.

0 50 100 150 200 250 300 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 CZE CH_GDP TURKEY_GDP POLAND_G DP HUNGARY_G DP

Figure 2: Standardized GDP figures in levels

The first observation that is made is the difference in the size of the respective economies. We can distinguish a group of two smaller countries and two larger countries. Although I have excluded the graph of the economy of the Eurozone—to keep this graph at a reasonable scale—, it is without

doubt that these four economies can be classified as smaller economies vis-`a-vis the Eurozone.

Another dynamic that we observe is the reduction of the size of the economies around the start of the GFC. While the two smaller economies show a similar development in their years prior to the GFC, Turkey and Poland show another dynamic. Of course, Turkey has grown faster than Poland— this might not be the case in per capita terms—, but its economy shrank relative more during the GFC as we see that the spread between the two economies gets smaller. Nevertheless, with respect to the output gap, we see from figure (1) that the Turkish economy had a large movement in its cycle; thus it is reasonable to state that the Polish economy got through the GFC more stable.

(15)

Finally, figure (2) shows that the economies of the Czech Republic, Hungary, and Poland are back on track; they exhibit signs of recovery and renewed growth after the stagnation due to the GFC. The Turkish economy, however, does not show this trend. Actually, in the sample period taken, we observe that the Turkish economy had its peak around 2013 and is moving downwards. This statement is also corroborated by figure (1). The said economies that are getting back on their growth track see an upward moving trend shown with the Hodrick-Prescott filter. The Turkish economy exhibits no such trend.

Figure (3) shows the CPI inflation rates of the countries. I have set the sample a little earlier to highlight the high inflation rates of the end-1990s. Again, we can distinguish two groups (exclud-ing the Eurozone): one group consist(exclud-ing of Turkey and another group with the Czech Republic, Hungary, and Poland. However, these four economies show a large drop in their inflation rates. At the basis of this development lies the change in monetary policy objectives. All central banks have adopted an inflation targeting framework. With the overall drop in inflation rates, it is possible to see that these economies are changing and developing into open-market economies.

The central banks’ main policy rates are presented in figure (4). All central banks have lowered their main policy rate during the GFC. Furthermore, two central banks have hit the ZLB, namely the central banks of the Czech Republic and the Eurozone; however, the Czech central bank has already started to increase its main policy rate during the course of 2017. Additionally, when the ZLB was hit, the central banks of the Czech Republic and the Eurozone used different tools to achieve their inflation target. The European Central Bank has resorted to unconventional monetary policies, such as quantitative easing, to achieve its inflation target. The Czech central bank, however, intervened in the foreign exchange markets and made use of the exchange rates with the Euro as a supplementary tool besides their 2-week repo rate. Therefore, it reduced the value of the Czech korunas and, as a means of forward guidance, it committed to this strategy for a certain amount of time.

(16)

-1 0 1 2 3 4 5 98 00 02 04 06 08 10 12 14 16 euro_cpi_yoy 0 20 40 60 80 100 98 00 02 04 06 08 10 12 14 16 turkey_cpi_yoy -4 0 4 8 12 16 98 00 02 04 06 08 10 12 14 16 poland_cpi_yoy -5 0 5 10 15 20 98 00 02 04 06 08 10 12 14 16 hungary_cpi_yoy -4 0 4 8 12 16 98 00 02 04 06 08 10 12 14 16 czech_cpi_yoy

(17)

0 1 2 3 4 5 2004 2006 2008 2010 2012 2014 2016 euro_main 0 10 20 30 40 50 2004 2006 2008 2010 2012 2014 2016 turkey_main 1 2 3 4 5 6 7 2004 2006 2008 2010 2012 2014 2016 poland_main 0 2 4 6 8 10 12 14 2004 2006 2008 2010 2012 2014 2016 hungary_main 0 1 2 3 4 2004 2006 2008 2010 2012 2014 2016 czech_main

(18)

4 Empirical model

The model that is used for this study is presented in this section. I use a simple empirical model to search and identify possible monetary spillovers. Specifically, I use a VAR model. Such a model describes the development of a set of variables over a certain time period. Therefore, the VAR model uses time-series data. What follows is a representation of the different steps that are taken to derive the structural VAR from a reduced VAR. To these belong the identification and optimal lag choice.

4.1 Reduced form

The reduced-form VAR represents the model without any structural restrictions for identification and is denoted by a VAR(p) model. This model is a VAR with p lags. Equation (1) is the formal representation of such a model:

yt= G0+ G1yt−1+ . . . + Gpyt−p+ et (1)

where yta k × 1 vector of k variables in the system, G0is a vector of constants, Gpwith p = 1, . . . , p

representing k×k matrices of coefficients belonging to the endogenous variables of the system, and et

a k times1 vector of white noise innovations. These white noise innovations are especially important for VARs. First of all, they are important because they need to be serially uncorrelated; that is, there should be no autocorrelation in the residuals. Secondly, with the use of these white noise innovation, we can derive impulse-response functions by structuring this vector.

For the VAR to not have issues with serial autocorrelation, the expectation of these white noise innovations should equal 0 with the following variance-covariance structure:

E[ete0τ] =

n Ω, if t = τ

0 otherwise (2)

Important to stress is that this reduced-form VAR does not have any restrictions. Therefore, structural analysis cannot be conducted on the basis of this reduced form. For us to be able to structurally analyze the monetary policy spillovers, a structural factorization of the VAR(p) is is necessary (see section 4.4).

4.2 Stationarity

To be able to conduct an analysis using the VAR(p) mode, it is important to have VAR(p) model that is stationary. The conditions for this stationarity are similar to those of univariate AR pro-cesses. Indeed, a VAR is a system of AR equations.

We can rewrite equation (1) by isolating the error term and use calculus to obtain a Moving-Average (MA) representation:

(19)

yt(In− G1yt−1− . . . − Gpyt−p) = G0+ et (3)

A(L)yt= G0+ et (4)

where In is the identity matrix and A(L) = Ik−Pp_i=1GiLi denotes the autoregressive lag

polyno-mial with Li as the lag operator to the ith degree (Stock and Watson, 2015, pp. 631-2).

For the VAR to be stationary as system, the characteristic polynomial should lie outside the unit circle; or within when the roots are inverted as with EViews. Formally, this is shown in equation (5):

det(In− G1yt−1− . . . − Gpyt−p) = 0 (5)

In short, this is the reduced-form VAR with et representing the reduced-form innovations for

time t.

4.3 Optimal Lag Choice

An important element of VAR(p) models is the optimal length of lags. Several criteria exist to calculate the optimal lag choice in VAR(p) models. Stock and Watson (ibid., p. 596) provide two criteria to choose the optimal lag length for VAR(p) models:

BIC(k) = lnhSSR(k) T i + kln(T ) T (6) AIC(k) = lnhSSR(l) T i + k2 T (7)

where BIC is the Bayesian Information Criterion (also known as the SIC or Schwarz Information Criterion), AIC is the Akaike Information Criterion, k the number of coefficient (including the intercept), SSR(k) the sum of squared residuals, and T the total number of observations in the time-series. Additionally, other methods exist as well. In EViews, the Hannan-Quinn Information Criterion is also provided.

Both for the BIC as the AIC, a two practical considerations have to be taken into account when it comes to choosing the optimal lag length. Firstly, all the candidate models must be estimated over the same sample. Put differently, different estimations based on different lag lengths must be estimated in the same sample and may not be subject to any change in the number of observations or the observations itself (ibid., pp. 596-7). Secondly, the estimation of the optimal length of lags in a VAR(p) model is subject to significant changes depending of the number of lags, p. Although it is at the discretion of the researcher, it is advised to start with a relative large lags and reduce

(20)

account the stability of the VAR models, their Information Criteria, and their LM Test Statistics for Autocorrelation, these lag lengths came out as the best configurations.

4.4 Structural Form

Using structural VAR models over unrestricted VAR models has its advantages. One reason to choose for a structural VAR over a reduced VAR is to identify the shocks and effects of the variables on each other in the system. A reduced VAR cannot identify independent shocks as its variance-covariance matrix is not restricted. To obtain a structural VAR model, I first rewrite equation (1):

Ayt= β0+ β1yt−1+ . . . + βpyt−p+ ut (8)

where Gp has been replaced by βp, A is a full rank matrix of coefficients that captures the

con-temporaneous effects of the variables’ relations in the vector y, and utthe structural shocks to the

system.

To get back from the structural VAR to the reduced VAR, we can multiply equation (8) by A−1

and with the fact that A−1A = In we obtain the reduced VAR model again:

A−1Ayt= A−1β0+ A−1β1yt−1+ . . . + A−1βpyt−p+ A−1ut (9)

yt= G0+ G1yt−1+ . . . + Gpyt−p+ et (10)

where A−1βp= Gp.

As with the reduced form, the structural VAR can be rewritten as a MA representation:

A(L)yt= ut (11)

with A(L) = A0− βpLp.

A more explicit representation of the structural VAR is given by equation (12):

A             yt πt it y_t∗ π∗_t i∗_t reert             = c + p X i=1 βi             yt−i πt−i it−i y∗_t−i π∗_t−i i∗_t−i reert−i             +             uy,t uπ,t ui,t uy∗_,t uπ∗_,t ui∗_,t ureer,t             (12)

where yt is the output gap, πt is the CPI inflation rate, it is the main policy rate of the central

(21)

with an asterisk and are in this case the Eurozone. The variable c is a k × 1 vector measuring the constants; that is, k = 7. Furthermore, A is a lower triangular matrix to identify the shocks. The next subsection will discuss the identification of the structural VAR model.

4.5 Identification

In order to identify the independent and residual shocks in the reduced VAR, we need to structure the vector of the dependent variable and the vector of the shocks in equation (1). As such, we transform the reduced and unrestricted VAR into a structured VAR in which we can identify the structural shocks to the system. We can write dynamic of the identification more formally:

A × e = B × u (13)

where A is the identification matrix, e are the residual (shocks), B the matrix that structures the residuals shocks, and u the structural shocks. These structural shocks satisfy the conditions for a VAR analysis, because the coefficients off the main diagonal are restricted to be zero. We can see this formally in equation (15).

To identify the independent shocks, I chose to use the Cholesky decomposition. This decompo-sition is a recursive factorization of the system in the VAR with a standard unit triangular matrix for matrix A and diagonal matrix for B.

Thus, matrix A in equation (13) is:             1 0 0 0 0 0 0 a21 1 0 0 0 0 0 a31 a32 1 0 0 0 0 a41 a42 a43 1 0 0 0 a51 a52 a53 a54 1 0 0 a61 a62 a63 a64 a65 1 0 a71 a72 a73 a74 a75 a76 1             (14) and matrix B:             b11 0 0 0 0 0 0 0 b22 0 0 0 0 0 0 0 b33 0 0 0 0 0 0 0 b44 0 0 0 0 0 0 0 b55 0 0 0 0 0 0 0 b66 0 0 0 0 0 0 0 b77             (15)

(22)

(see equation [1]). This means that the GDP gap of the small-open economy is not affected by any other variable and that its real effective exchange rate is affected by all other variables. As such, we can create a structural analysis for an open economy. Additionally, because this kind of structuring is often used, econometric software, such as EViews, standardly have the option to select this kind of restrictions.

Thus, the ordering of the variables and their effects is as follows. There is no contemporaneous

nor other effect on the domestic GDP gap. The domestic inflation rate is contemporaneously

affected by the domestic GDP gap and its own contemporaneous effect is restricted to 1. The effect of the domestic main policy rate is contemporaneously affected by the GDP gap and the inflation rate, while its own effect is restricted at 1. The same counts for the foreign economy, namely the Eurozone, but with the difference that all the mentioned domestic variables do posit a contemporaneous effect on the Eurozone. The foreign variables are themselves contemporaneously restricted by 1. Lastly, the real effective exchange rate is also contemporaneously restricted to 1 for itself; however, the GDP gaps, the inflation rates, and the main central bank policy rates of both the domestic and foreign economies have an effect on the real effective exchange rate.

(23)

5 Results

In this section, I will present the results of my analyses. Conducting time-series analyses while deploying structural VAR(p) models allows us to investigate the responses of variables to impulses. The functions that generate these results are called impulse-response functions. As the question at hand here concerns monetary spillovers from the ECB to neighboring open and small economies within the EU Customs Union, the variable that is used as the impulse in said functions is the ECB’s main policy rate. Impulse-response functions delineate the response of the variables at interest generated from an innovation shock to the impulse. In our case, the variables at interest respond to one Cholesky standard deviation innovation.

Referring back to the stability condition of (structural) VAR(p) models, the stability condition practically translates in responses that “die out” over time. For this research, I limited the responses to 24 periods; that is, 24 months or 2 years. While the results in this section do not show the eventual “dying out”, regenerating the impulse-response functions at higher periods do result in the eventual “dying out” of the impulse.

The following part will present the impulse-response functions; however, aside from some de-scriptive comments, it will refrain from discussing the results in this section.

5.1 Impulse-Response Functions of Turkey

The impulse-response function to Cholesky one standard deviation innovation of the Turkish vari-ables with the main policy rate of the ECB as the impulse is shown in figure (5). The upper-left graph shows the response of the Turkish GDP gap to the innovation of the main policy rate of the ECB. We see that the gap does reacts very cautiously and positively; however, after the first two periods, the gap responds negatively to the innovation from period 3 until period 9. After the first negative wave, the GDP gap moves upward and starts to react positively after nine months. The graph shows that the GDP gap changes its reaction over time, either positive or negative, but moves toward a positive reaction after a period of close to two years (21 months to be precise).

The upper-right graphs depicts the response of the inflation rate in Turkey. This rate responds negatively to an innovation in the ECB’s main policy rate. Nevertheless, unlike the GDP gap, the inflation rate starts to react positively relative sooner and does not move back below the x-axis. The change to a positive response happens at the 6th period where it takes a value of 0.000.

Unlike the three other responses, the main policy rate in Turkey starts reacting positively to an innovation in the Eurozone’s main policy rate. This response is shown in the lower-left graph. Moreover, it does not move below the x-axis and thus remains reacting positive as far as 24 months in the future. However, from period 5 to period 6, we observe a decline in the response from a value of 0.42 to 0.37 respectively. Additionally, we can see that band of the standard errors of the Turkish main interest rate response also become positive and remain so. We also observe that the

(24)

positively in magnitude and remains positive. We can see this response in the lower-right graph.1

5.2 Impulse-Response Functions of Poland

The Cholesky one standard deviation innovation results with respect to the Polish economy are presented in figure (6). Contrary to the Turkish economy, we see that the original response of the Polish GDP gap is negative to the innovation. Similarly, after its initial negative reaction period of 6 months, the GDP gap of Poland starts to react positively and changes its reaction, either positive or negatively, over time.

The response of Poland’s inflation rate is similar to that of Turkey, namely negative at first and positive thereafter. This change happens at the 12th period, i.e. after a year. Just like the Turkish inflation rate, the reaction stays positive. However, we can see a linear upward movement in Poland’s reaction, which is contrasting with the Turkish reaction—which slopes downward after around 11 periods.

With respect to the main interest rate of Poland, we see that the response is positive through all of the 24 periods ahead. From period 4 to period 6, a small decline occurs; nevertheless, the overall response remains positive. Additionally, we observe a growing trend in the response of the Polish main rate and a significantly different scaling of the y-axis compared to the Turkish response.

Contrary to the Turkish real effective exchange rate, the Polish exchange rate responds positively to the innovation.

5.3 Impulse-Response Functions of Hungary

The Hungarian impulse-response functions are shown in figure (7). As with the Turkish GDP gap, the Hungarian GDP gap responds very cautiously but positively to the innovation. However, the length of its positive reaction is longer when compared to the response of the Turkish GDP gap. Similar to the previous two countries, the Hungarian GDP gap changes its response between positive and negative, according to this study, as we move further in time (after three quarters).

Hungarian CPI inflation responds very cautiously negatively on the innovation, but quickly changes this into a positive response (after one quarter). Going forward to the two years ahead, we can see a gradual growth in the positive reaction of the inflation on the innovation.

With respect to the response of the main policy rate of the Hungarian central bank, we observe a similar initial positive response as with the previous two countries. Nevertheless, the main policy rate of Hungary differs from the previous two policy rate responses. That is, all three responses stop growing after the first 3 to 6 months. However, the Hungarian response almost moves back to 0, i.e. having a value of 0.03 at the 7th period. Nevertheless, the response gradually grows larger as it reaches 0.33 in the 24th period.

Focusing on the real effective exchange rate of the Hungarian currency, we see that the exchange rate responds minimally. In period 7, the response is 0.0001 and in period 16 this takes the value

(25)

of 0.0012. From period 16 to period 24, the response grows and ultimately reaches the value of 0.0074.

5.4 Impulse-Response Functions of the Czech Republic

Figure (8) shows the impulse-response functions for the Czech Republic. We can see that the Czech response to the innovation with respect to the GDP gap is negative. The In this respect, the Czech economy is similar to the Polish economy. Moving further in time, after seven periods the Czech GDP gap starts to respond positively on the innovation and it changes this positive response after period 23 where it is valued 0.0001.

Inflation-wise, the Czech economy responses negatively to the innovation until the 6th period. However, just like the other countries, the Czech economy eventually starts to react positively to the innovation and remains so until the 24th period.

With respect to the response of the main policy rate of the Czech Republic, a unique feature is observed. The Czech main policy rate gradually grows to a value of 0.141 without a decline until the 20th period. This value reaches 0.124 at the 24th period.

Just like Poland and Hungary, the Czech real effective interest rate initially responds positively to the innovation and this response turns into a negative response after 3 periods. The 10th period sees another change in the response as the real effective exchange rate starts to react positively to the innovation.

(26)

-.02 -.01 .00 .01 .02 2 4 6 8 10 12 14 16 18 20 22 24

Response of L_TURKEY_GDP_GAP to EURO_MAIN

-.4 -.2 .0 .2 .4 .6 2 4 6 8 10 12 14 16 18 20 22 24

Response of TURKEY_CPI_YOY to EURO_MAIN

-0.2 0.0 0.2 0.4 0.6 0.8 1.0 2 4 6 8 10 12 14 16 18 20 22 24

Response of TURKEY_MAIN to EURO_MAIN

-.010 -.005 .000 .005 .010 .015 .020 2 4 6 8 10 12 14 16 18 20 22 24

Response of L_TURKEY_REER to EURO_MAIN Response to Cholesky One S.D. Innovations ± 2 S.E.

(27)

-.012 -.008 -.004 .000 .004 .008 .012 .016 2 4 6 8 10 12 14 16 18 20 22 24

Response of L_POLAND_GDP_GAP to EURO_MAIN

-.4 -.2 .0 .2 .4 .6 2 4 6 8 10 12 14 16 18 20 22 24

Response of POLAND_CPI_YOY to EURO_MAIN

.0 .1 .2 .3

Response of POLAND_MAIN to EURO_MAIN

.000 .004 .008 .012 .016

Response of L_POLAND_REER to EURO_MAIN Response to Cholesky One S.D. Innovations ± 2 S.E.

(28)

-.02 -.01 .00 .01

2 4 6 8 10 12 14 16 18 20 22 24

Response of L_HUNGARY_GDP_GAP to EURO_MAIN

-.2 .0 .2 .4 .6 .8 2 4 6 8 10 12 14 16 18 20 22 24

Response of HUNGARY_CPI_YOY to EURO _MAIN

-.2 .0 .2 .4 .6 .8 2 4 6 8 10 12 14 16 18 20 22 24

Response of HUNGARY_MAIN to E URO_MA IN

-.010 -.005 .000 .005 .010 .015 2 4 6 8 10 12 14 16 18 20 22 24

Response of L_HUNGARY_REER to EURO_MAIN Response to Cholesky One S.D. Innovations ± 2 S.E.

(29)

-.010 -.005 .000 .005 .010 .015 2 4 6 8 10 12 14 16 18 20 22 24

Response of L_CZECH_GDP_GAP to EURO_MAIN

-.2 -.1 .0 .1 .2 .3 .4 2 4 6 8 10 12 14 16 18 20 22 24

Response of CZECH_CPI_YOY to EURO_MAIN

.05 .10 .15 .20 .25

Response of CZECH_MAIN to EURO_MAIN

-.004 .000 .004 .008 .012

Response of L_CZECH_REER to EURO_MAIN Response to Cholesky One S.D. Innovations ± 2 S.E.

(30)

6 Discussion and Policy Recommendations

Now that I have concluded my analysis and retrieved the results from my data set, I will continue with the discussion of the results. The results show some monetary spillover effects and this section provides a discussion on the main findings of this research. The structural VAR analyses and their impulse-response functions show some similarities and some differences between the four sample economies.

Let me start with the discussion on the responses of the GDP gaps on the innovation. The GDP gap has percentages as its unit. This means that in the impulse-response graph, one innovation in the main interest rate of the ECB, also in percentages, will show the change of the GDP gap in percentages.

The overall response to the GDP gaps show similarities between Poland, the Czech Republic, and Turkey. They show a different response to the Hungarian GDP gap, which responds positively. In terms of effects on the economies, the GDP gaps decline initially in the said three economies. This means that when there is a negative gap, this gap becomes more negative; or, when there is a positive gap, this gap will move toward equilibrium. In any case, the GDP declines. Traditional models on foreign monetary policy shocks predict that the foreign demand (i.e. the Eurozone) for domestic goods will increase when the foreign central bank pursues an expansionary monetary policy. Another way to state this is that the GDP of the domestic country will increase.

In my research, the Cholesky one standard deviation innovation leads to an increase in the foreign main policy rate. Inverting the dynamics of the theory described in section 2, we must observe a reduction in the value of the GDP gap. This means that the overall GDP gets reduced when the European Central Bank increases its main policy rate and it is in accordance with the traditional models for small and open economies on foreign monetary policies. The logic behind this dynamic is also observed in my sample. An increase in the ECB’s main policy rate attracts capital and when this capital comes from small and open economies, a reduction in their GDP finds place.

One argument to describe this dynamic is that the contractionary foreign monetary policy slows down the foreign economy. This is because foreign consumers have to pay higher interest rates when they take out loans to buy goods or, rather than consuming their capital, they will increase their savings as the received interest rates on their savings will be more favorable. As a result, the demand for the goods in the domestic market declines. Graphically, this means that the open-economy IS curve in the domestic market moves to the left.

Furthermore, foreign as well as domestic investors might perceive the conditions of the risk in the small-open economies differently. Indeed, they may want to take risks when the interest rate of the large foreign economy is at or near zero. We observed this during the aftermath of the GFC when capital in advanced economies started to move toward and allocate in emerging market economies as the returns were higher in the latter group. Inverting this logic, a non-zero foreign interest rate means that saving capital in the foreign market will generate returns. As a result, foreign investors may want to return back to their respective economy due to the new cost-benefit

(31)

calculations. Similarly, domestic investors may allocate their capital to this larger foreign economy as well. They might want to have some capital stored in less volatile economies; meaning that they search for a safe haven.

The graphs that represent the impulse and responses on the inflation rates from an innovation in the main interest rate of the ECB are to be read similarly to the ones of the GDP gaps. As both variables have percentages as their units and no other processing of the variables occurred, the responses are denoted in percentages.

The analysis on the inflation rates is similar to the analysis on the GDP gaps. As the foreign monetary policy institution pursues a contractionary monetary policy, the foreign demand for do-mestic goods will show a decline. This means, ceteris paribus, that the overall demand for dodo-mestic goods will decline. In other words, the inflation rates will see a reduction. Most economies in my sample show signs of a reduction in their inflation rates. We can see, for example, that the Turkish inflation rate declines with 0.1 in about three months after the shock. Together with the movement of the Turkish GDP gap, the inflation rates start to rise again.

Regarding the main policy rates of the domestic economies, all economies show a positive re-sponse to the innovation. This rere-sponse seems odd when looking at the inflation rate dynamics. The domestic inflation rates decline immediately after the shock as the domestic economies gets slowed down. Increasing interest rates at this moment will lead to an additional slowing down of the economies.

However, some empirical literature has found some evidence that point toward a similar dynamic. Iacoviello, Navarro, et al. (2018) investigate the spillovers of the monetary policy rate of the United States after this rate started to increase again. In their paper, they refer to two other papers that have researched the matter of monetary policy spillovers. As such, both Rey (2016) and Miranda-Agrippino and Rey (2015) show that changes in the interest rates of so-called “core” countries can trigger a global financial cycle. This cycle, in turn, and regardless of whether the affected countries have flexible or fixed exchange rate regimes, may generate positive global spillovers in variables such as main interest rates.

Standard open-economy DSGE models (see section 2.1) describe these effects. For a financial cycle to have effect, strong financial integration is necessary and this is the case in the EU Customs Union. Members of this customs union have abolished many or all trade barriers and these includes for the financial barriers. Again, by inverting the logic in section 2.1, we can describe the effects of the main policy rates. When the logic is inverted, section 2.1 describes that the immediate effect of an increasing foreign interest rate of a large open economy is that the financial products denominated in that foreign currency become more expensive and credit demand decreases. In turn, this may increase the credit demand in the domestic economy and to slow this increase down, the central bank may increase its policy rates. In sum, the response of the domestic main policy rate to an increase in the foreign main policy rate is not unobserved in other studies.

(32)

models predict that the domestic currency will appreciate during an expansionary foreign monetary policy shock of a large economy. That means, vice versa, that a contractionary foreign monetary policy shock should lead to a depreciation of the domestic currency in real effective terms. The results of this study show that this is the case for Turkey. Figure (5) shows that the Turkish exchange rate declines by about 0.002 per cent. This response is not at all large. The Czech real effective exchange rate also declines, but not initially. After a period of appreciation follows a period of depreciation. This depreciative period ends after 10 months and turns into an appreciative period. With respect to the other two economies, an initial and consistent appreciation of the currency in real effective terms finds place. Although the effects are very small—measured in tenths of a percentage—they exists nevertheless.

The traditional open-economy models predict that exchange rates will appreciate over time after an expansionary foreign monetary policy shock. In our case, this is also true, but then after a contractionary foreign monetary policy shock. As aforementioned, Gali and Monacelli (2005) point our that the effect with flexible exchange rates regimes depend on the degree of openness to trade and the elasticities of substitution between domestic and foreign goods and the substitutions, both intra- as well as intertemporally, that capture an economy’s consumption smoothing preferences via intratemporal trade. In other words, a priori knowledge on how the exchange rates will move in in a flexible exchange rate regime remains ambiguous. The ambiguity stemming from such exchange rate regimes come from the fact that the exchange rates adapt to the conditions and movements in the economy.

Another important element of the ambiguity of the exchange rates may come from the sample selection. The economies of the Czech Republic, Hungary, and Poland are in a process of integration in the European Single Market. The three economies are former transition economies and all four countries are still significantly behind their Western peers. Therefore, one explanation why these countries exhibit appreciating real exchange rates is due to the promising nature of their future. Also, their development requires capital. Often, domestic capital markets are too small to supply the needed capital. As such, foreign capital markets are called upon to provide the necessary funds. Such investments in (long-term) development may supply a continuous stream of foreign capital (among others in the form of FDI). This may explain an initial decline in some countries, but a continued appreciation over time after the impulse.

Usually, it is advised to keep the exchange rate from fluctuating too much. Moreover, a con-tinuing appreciating exchange rate may gradually impede the development of domestic (export) markets. However, the effects in this research are very small. The effort of intervening in the for-eign exchange markets and thereby risking to lose the trust from economic agents with regard to the flexible nature of the exchange rate regime is therefore ill-advised. However, although it occurred recently and lies, therefore, outside the sample period, the Turkish central bank did intervene to keep the Turkish lira from dropping. The year 2018 shows that the Turkish lira lost roughly 25% of its value. The tools that the Turkish central bank uses is its main policy rate, which has increased to about 17 percentage points.

(33)

innovation is relative small for the GDP gaps in the first few months. When the time proceeds, the effects remain small. The largest effect is seen in the Czech and Polish economies. However, these effects remain very small. Therefore, the GDP gaps remain relative unaffected.

The responses of the inflation rates show that there is an increase in the inflation rates over the longer horizon. It is important to keep an eye on the inflation rates as all countries have committed themselves to an inflation targeting regime. Over the longer horizon, the inflation rates start to increase and looking at figure (3), it seems that particularly Turkey has troubles in lowering inflation to levels of its Western peers. We also see that the inflation rates of the Czech Republic, Hungary, and Poland have risen over the past year. For a fact, the Turkish inflation rate is getting a lot higher lately due to large (public) investments and show signs of overheating. Whereas the ECB has announced to not yet increase its main policy rate in 2018, 2019 will bring this expected rise. As such, if the Turkish economic and monetary authorities do not respond in time, they will invite the economy to a meltdown as seen in during the 2000-1 financial crisis of the country.

(34)

7 Conclusion

This study sought to investigate possibilities of a monetary spillover from the European Central Bank on neighboring economies. For this question, it used time-series data ranging from January 2003 until December 2017. The countries that this study researched were the Czech Republic, Hungary, Poland, and Turkey. These countries are all members of the EU Customs Union and are, therefore, (legally) integrated.

As for the empirical method, this research made us of a structural vector autoregressive model as is common among other scholars in academia and other respectable institution. Before obtaining the structural VAR model, a reduced-form VAR model was described as a baseline model. The difference among the two is that a structural VAR model structures the reduced-form VAR model in order to be able to perform structural analysis. My choice for this structural factorization was the Cholesky decomposition, which is widely used among researchers who employ structural VAR models. With a structural VAR model, it is possible to derive Impulse-Response functions and see how an economy reacts to a structural shock; in this case a structural shock to the main policy rate of the European Central Bank. A description of this model is found in section 4.

Section 5 present the results of this research and gives descriptive commentary. While there are some differences among the economies, the results, nevertheless, show that overall the economies exhibit similar responses to the impulse of a structural shock from the ECB’s main policy rate.

The discussion that followed in section 6 found that the responses are, by and large, in ac-cordance with theoretical literature on the subject matter. The discussion that I provided was mainly on the background of the traditional open market-economy models that are also known as the Mundell-Fleming-Dornbusch models. The responses from the respective GDPs are just like the model predicts. A contraction in the foreign monetary policy rate reduce foreign demands in domestic goods and therefore lowers the GDP of the domestic country. This is also reflected by the responses of the inflation rates. As the GDP sees a reduction, the overall economy slows down and thereby belongs a reduction in the inflation rates. However, the Hungarian case is slightly different in that the response of the GDP and of inflation are positive and very small in the initial periods after the impulse. Nevertheless, they do move together. Discussion on the other two variables show a discord with theory.

As this research was limited in time and resources, it is advised to investigate this subject matter further. Particularly, several issues obscure the validity of my results. One such this is that I did not account for the shadow interest rates for the main policy rates. Another thing is that interpolation was behind my capabilities. Therefore, I advice other researchers to take account of these matters.

(35)

References

Bauer, M. D. and C. J. Neely (2014). “International channels of the Fed’s unconventional monetary policy.” In: Journal of International Money and Finance 44, pp. 24–46.

Bernanke, B. S. and M. Gertler (1995). “Inside the black box: the credit channel of monetary policy transmission.” In: Journal of Economic perspectives 9.4, pp. 27–48.

Devereux, M. B. and J. Yetman (2010). “Leverage constraints and the international transmission of shocks.” In: Journal of Money, Credit and Banking 42.s1, pp. 71–105.

Dornbusch, R. (1980). Open economy macroeconomics.

Feldkircher, M. (2015). “A global macro model for emerging Europe.” In: Journal of Comparative Economics 43.3, pp. 706–726.

Feldkircher, M. and F. Huber (2016). “The international transmission of US shocks—evidence from Bayesian global vector autoregressions.” In: European Economic Review 81, pp. 167–188. Fernandez, R. B. (1981). “A methodological note on the estimation of time series.” In: The Review

of Economics and Statistics 63.3, pp. 471–476.

Fratzscher, M., M. L. Duca, and R. Straub (2016). “ECB unconventional monetary policy: market impact and international spillovers.” In: IMF Economic Review 64.1, pp. 36–74.

Gali, J. and T. Monacelli (2005). “Monetary policy and exchange rate volatility in a small open economy.” In: The Review of Economic Studies 72.3, pp. 707–734.

Gambacorta, L., B. Hofmann, and G. Peersman (2014). “The effectiveness of unconventional mon-etary policy at the zero lower bound: A cross-country analysis.” In: Journal of Money, Credit and Banking 46.4, pp. 615–642.

Halova, K. and R. Horv´ath (2015). International spillovers of ECB’s unconventional monetary

policy: The effect on Central and Eastern Europe. Tech. rep. IOS Working Papers. Iacoviello, M., G. Navarro, et al. (2018). “Foreign Effects of Higher US Interest Rates.” In: Jannsen, N. and M. Klein (2011). “The international transmission of euro area monetary policy

shocks.” In:

Kucharcukova, O. B., P. Claeys, and B. Vasicek (2016). “Spillover of the ECB’s monetary policy outside the euro area: How different is conventional from unconventional policy?” In: Journal of Policy Modeling 38.2, pp. 199–225.

Liu, P., H. Mumtaz, and A. Theophilopoulou (2011). “International transmission of shocks: A time-varying factor-augmented VAR approach to the open economy.” In:

Miranda-Agrippino, S. and H. Rey (2015). “US monetary policy and the global financial cycle.” In: NBER working paper 21722.

Mumtaz, H. and P. Surico (2009). “The transmission of international shocks: a factor-augmented VAR approach.” In: Journal of Money, Credit and Banking 41.s1, pp. 71–100.

Neely, C. J. et al. (2010). The large scale asset purchases had large international effects. Federal Reserve Bank of St. Louis, Research Division.

(36)

Rey, H. (2016). “International channels of transmission of monetary policy and the Mundellian trilemma.” In: IMF Economic Review 64.1, pp. 6–35.

Stock, J. H. and M. W. Watson (2015). Introduction to Econometrics: Global edition. Pearson Education Boston, MA.

Svensson, L. E. and S. Van Wijnbergen (1989). “Excess capacity, monopolistic competition, and international transmission of monetary disturbances.” In: The Economic Journal, pp. 785–805.

(37)

8 Appendices

VAR Residual Serial Correlation LM Tests

Null Hypothesis: no serial correlation at lag order h Sample: 2003M01 2017M12

Included observations: 180

Lags LM-Stat Prob

1 67.04953 0.0442 2 100.1192 0.0000 3 86.59821 0.0007 4 87.02798 0.0007 5 42.31707 0.7390 6 81.16034 0.0026 7 50.24879 0.4237 8 47.36555 0.5396

Probs from chi-square with 49 df.

Table 1: LM Test Statistics for Turkey

Lags LM-Stat Prob

1 97.02045 0.0001 2 78.31277 0.0049 3 72.63699 0.0158 4 73.38493 0.0136 5 81.40813 0.0025 6 53.16344 0.3170 7 59.52152 0.1443 8 60.35174 0.1283

(38)

Lags LM-Stat Prob

1 60.14349 0.1321 2 55.58377 0.2407 3 83.78191 0.0014 4 64.55652 0.0673 5 53.41073 0.3086 6 50.25718 0.4234 7 49.32147 0.4603 8 41.89663 0.7540 9 46.79878 0.5628

Table 3: LM Test Statistics or Hungary

Lags LM-Stat Prob

1 80.44465 0.0031 2 103.8993 0.0000 3 94.08391 0.0001 4 90.38625 0.0003 5 67.44970 0.0413 6 63.82920 0.0757 7 60.05728 0.1338

(39)

VAR Lag Order Selection Criteria Sample: 2003M01 2017M12 Included observations: 178

Lag LogL LR FPE AIC SC HQ

0 −736.9748 NA 1.01e-05 8.359267 8.484394 8.410009 1 936.2154 3195.981 1.20e-13 -9.890060 -8.889050 -9.484124 2 1291.106 649.9686 3.86e-15 -13.32704 -11.45014 -12.56590 3 1462.563 300.5317 9.80e-16 -14.70296 -11.95018 -13.58663 4 1663.063 335.6674 1.81e-16 -16.40520 -12.77654* -14.93368* 5 1715.174 83.14412 1.78e-16 -16.44016 -11.93561 -14.61344 6 1802.721 132.7958 1.19e-16 -16.87327 -11.49284 -14.69136 7 1903.836 145.4232 6.88e-17 -17.45883 -11.20251 -14.92172 8 1962.642 79.95005 6.51e-17* -17.56901 -10.43681 -14.67671 9 1995.079 41.54814 8.43e-17 -17.38291 -9.374821 -14.13541 10 2056.131 73.40016 8.07e-17 -17.51833 -8.634358 -13.91564 11 2123.594 75.80145* 7.38e-17 -17.72578 -7.965926 -13.76790 12 2177.198 56.01234 8.11e-17 -17.77750* -7.141763 -13.46442

* indicates lag order selected by the criterion

LR: sequential modified LR test statistic (each test at 5% level) FPE: Final prediction error

AIC: Akaike information criterion SC: Schwarz information criterion

(40)

0 −167.6823 NA 1.64e-08 1.940914 2.065085 1.991260 1 1595.240 3369.141 8.82e-17 -17.10267 -16.10931 -16.69991 2 1927.210 608.6115 3.81e-18 -20.24678 -18.38422 -19.49159 3 2107.359 316.2615 8.92e-19 -21.70399 -18.97224 -20.59638 4 2325.988 366.8111 1.37e-19 -23.58876 -19.98781* -22.12873 5 2368.728 68.38393 1.50e-19 -23.51920 -19.04906 -21.70676 6 2483.046 174.0174 7.44e-20 -24.24496 -18.90563 -22.08009 7 2572.957 129.8704 4.91e-20 -24.69952 -18.49099 -22.18223* 8 2638.919 90.14805 4.29e-20 -24.88799 -17.81026 -22.01827 9 2694.710 71.90895* 4.26e-20* -24.96345* -17.01653 -21.74131

AIC: Akaike information criterion SC: Schwarz information criterion HQ: Hannan-Quinn information criterion

Table 6: VAR Lag Length Criteria for Poland

(41)

0 −326.2914 NA 9.57e-08 3.703238 3.827408 3.753584 1 1291.236 3091.275 2.58e-15 -13.72485 -12.73148 -13.32208 2 1636.298 632.6133 9.65e-17 -17.01442 -15.15187 -16.25924 3 1788.692 267.5354 3.08e-17 -18.16324 -15.43149 -17.05563 4 2004.965 362.8579 4.85e-18 -20.02183 -16.42088* -18.56180* 5 2051.806 74.94556 5.06e-18 -19.99784 -15.52770 -18.18539 6 2155.705 158.1573 2.83e-18 -20.60783 -15.26850 -18.44296 7 2240.997 123.2002 1.96e-18 -21.01108 -14.80255 -18.49379 8 2319.430 107.1915* 1.49e-18* -21.33811* -14.26039 -18.46840 9 2350.646 40.23437 1.95e-18 -21.14051 -13.19360 -17.91838

(42)

0 59.50169 NA 1.32e-09 -0.583352 -0.459182 -0.533006 1 1708.286 3151.010 2.51e-17 -18.35873 -17.36537 -17.95597 2 2069.877 662.9169 7.80e-19 -21.83197 -19.96941 -21.07678 3 2280.198 369.2302 1.31e-19 -23.62442 -20.89267 -22.51681 4 2479.691 334.7054 2.48e-20 -25.29657 -21.69562* -23.83654 5 2538.260 93.71025 2.27e-20 -25.40289 -20.93275 -23.59044 6 2649.347 169.0985 1.17e-20 -26.09274 -20.75341 -23.92787* 7 2724.275 108.2295 9.15e-21 -26.38083 -20.17231 -23.86354 8 2786.207 84.64019 8.35e-21 -26.52452 -19.44680 -23.65481 9 2846.527 77.74547* 7.89e-21* -26.65029* -18.70338 -23.42816

Table 8: VAR Lag Length Criteria for the Czech Republic

Monetary spillovers from the ECB : reasons to adjust?

University of Amsterdam

Faculty of Economics and Business