Interest rate setting by the ECB in time of the sovereign debt crisis

(1)

Interest Rate Setting by the ECB in

time of the Sovereign Debt Crisis

Lisanne Oomen 10421130

BSc in Economics & Business University of Amsterdam Supervisor: Ms. S. Chan

Abstract This paper studies the behavior of the ECB on interest rate setting in time of the sovereign debt crisis. By constructing several empirical reaction functions, using real economic activity, inflation, M3 growth, and exchange rate changes as variables, the significance of these determinants is measured by an ordered probit model. The results show that the ECB only reacts to deviations in output gap and the lagged MRO rate. These outcomes will be compared to the results of Gerlach (2007) in order to argue whether the sovereign debt crisis has changed the behavior of the ECB on setting the interest rate.

(2)

2 Statement of Originality

This document is written by Student Lisanne Oomen who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is 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.

(3)

3 1. Introduction

Since the beginning of 1999, when the currency unionbecame subject to a single monetary system set by the European Central Bank (ECB), the responsibility of conducting monetary policy for the euro area is at hands of the ECB.1 Within the ECB, the Governing Council (GC), is the decision-making body and is responsible for formulating monetary policy, which includes setting the key interest rates.2 In various papers the behavior of interest setting of the ECB is studied by estimating empirical reacting functions of the Tayler variety on forecasting monetary policy. While most of these authors study the first few years of the ECB, this paper examines whether the recent European sovereign debt crisis has had an impact on the behavior of this matter.

In 2008 just as the financial crisis began, global uncertainty rose – which drove the economy of the euro area into uncertainty. After the Greek government revealed their revised budget deficit, which was twice as big as the previous estimate, the sovereign debt spreads to most of the euro area countries, initiating the greatest challenge for the European Union (EU) ever since its creation (De Santis, 2012). The fear of the spreading of this debt crisis has put a lot of pressure on policy makers within the EU. This is interesting for the ECB to measure, since it wants to preserve financial stability by implementing the right monetary policies (Alter & Beyer, 2013).

As illustrated by the financial crisis, interest rates were being pushed to zero after the occurrence of a large, contractionary shock. Especially, whenever the ECB wants to stimulate the economy, it has strong incentives to reduce interest rates as much as possible (Gerlach & Lewis, 2014). This paper discusses whether the interest rate setting behavior of the ECB has been

affected by the crisis. Using the same empirical method as Gerlach (2007), we will examine if the determinants that influence the interest rate are still the same before, as during the crisis. This is done by using an ordered probit model to understand how these factors influence the change in interest rate. I will argue that in this study the ECB reacts most to the output gap and the MRO rate when setting the new MRO rate.

1_{https://www.ecb.europa.eu/ecb/orga/escb/html/index.en.html}

(4)

4 The remainder of this paper is structured as follows. Section 2 provides an overview of related literature on studying the behavior of interest rate setting of central banks. Section 3 discusses our methodology used, from which the results are presented in Section 4. At last our results are concluded in Section 5.

2. Literature Review

Numerous authors have studied the behavior of the ECB on interest rate setting. By doing so, several researchers have estimated an empirical reaction function of the ECB on targeting the interest rate. Judd & Rudebusch (1998) for example, focus on the reaction of the Federal Reserve Bank (Fed) on economic developments, and how this has changed over time. They come to the conclusion that a reaction function derived from the Taylor rule seems to capture significant indicators of monetary policy like inflation.

Carstensen (2006) looks at the first four years of the European Monetary Union (EMU) in which he estimates the policy reaction function of the ECB. Since the ECB sets interest rates in basis points, he uses an ordered probit model. Carstensen argues that the ECB takes inflation risk very seriously, as the ECB has clarified to keep its inflation rate close to 2 percent. Carstensen uses the MRO rate as policy instrument of the ECB. He justifies his choice by saying that the MRO rate is clearly created by the ECB as an instrument to signal the stance of the monetary policy to the public. He takes the lagged MRO rate, the inflation rate, and the output gap as explanatory variables and uses data on a monthly basis. Carstensen claims that the ECB takes the risk of deflation in all earnest, as the ECB aims to keep the inflation away from zero.

Gerlach (2007) has been studying the behavior of the ECB on setting interest rate in the beginning of the ECB in 1999 to June 2006. With ordered probit techniques he estimates

empirical reaction functions for the ECB, using the ECB its monthly Bulletin to guide his choice of variables. In his work he does not only focus solely on the ECB its policy actions, which he refers to as the ECB its deeds, but he also looks at the words, which are the public statements regarding macroeconomic developments of the ECB, in order to understand how it interprets incoming data on monetary policy. In his paper, Gerlach uses real economic activity, inflation, money growth, and the rate of appreciation of the nominal effective exchange rate as variables to target the level of interest rate set by the ECB. He finds that interest rates are more closely tied to

(5)

5 economic activity than to inflation, and that the ECB reacts to M3 growth while setting the MRO rate.

In another study of Gerlach, he estimates a reaction function for the ECB using forecasts of economic growth and inflation as regressors, while looking at the ECB in 2008 (Gerlach & Lewis, 2014). Gerlach and Lewis explore the interest rate setting behavior of the ECB by creating a framework in which coefficients are allowed to change over time. In this case, monetary policy changes are estimated rather than imposed. They analyze monetary policy on a monthly basis using growth forecast, inflation forecast, the lagged interest rate, and the lagged change in interest rate as variables. They take into account that there is a possibility that there have been two

regimes in operating during their sample period, as the interest rate may have been constrained during the crisis, Gerlach & Lewis therefore employ the smooth transition model.3 They find that interest rates in time of the crisis were reduced significantly faster than in times before the crisis.

All these researchers have in common that they all derive their initial concept from Taylor, who presents a very specific and simple rule for monetary policy by the central bank. Taylor (1993) studies how econometric policy research on monetary policy rules can be used in a applied policymaking environment. According to his research, good policy rules are responding to changes in price level or real income. Taylor argues that interest rate adjustments set by the central bank, should response to: deviations in money supply, deviations of the exchange rate, deviations of the inflation rate, and deviation of real output.

Judd & Rudebusch (1998) claim that, by using the Taylor rule, their results propose a valuable way of summarizing important fundamentals of monetary policy. The way many

policymakers at central banks think about monetary policy, has been revolutionized by the Taylor rule. The Taylor rule has framed the technique of monetary policy as a methodical response to received information on economic situations (Asso, Kahn, & Leeson, 2010).

Yet not every author shares this opinion. Sauer & Sturm (2003) argue that the response on adjustments in interest rate is to large due to the fact it lacks a forward-looking perspective in the specification of Taylor.

3_{For more information see: Mankiw, Miron and Weil (1987) “The Adjustment of Expectations on to a Change in}

(6)

6 While Taylor (1993) argues the interest rate target should respond to deviations

mentioned above, it is uncertain by how much the interest should change. Sack & Wieland (2000) find that central banks tend to change interest rates in small segments. These small steps are taken in the same direction, and only rarely are these steps taken in the opposite direction. This

behavior of interest rate smoothing reflects the optimal behavior of a central bank that is

concerned with output and inflation stability, according to Sack & Wieland. Although they argue interest rate smoothing is an optimal move even if the central bank is not explicitly concerned with interest rate volatility, Gerlach (2007) finds out that there is only little evidence on interest rate smoothing by the ECB. He argues that a change in interest rate made in the past month, will reduce the change of a change being made in the next month month. It seems to Gerlach that interest rate changes are made in order to reduce further changes in the near future.

When looking at the sovereign debt crisis, Gorter et al. (2010) study whether this crisis has affected the ECB its monetary policies. They do so by starting off with the Taylor rule and update estimates reported in Gorter et al. (2008)4. They argue that monetary policy should be forward looking and should be based on expected output and inflation. For Gorter et al. it seemed that inflation was no longer the key driver for setting monetary policy. Their results show that in the of the crisis, the ECB puts more stress on preserving price stability.

Schmidt (2004) reasons that ECB has been a key player in the sovereign debt crisis that has hit the EU since 2008, along with the European Commission and EU institutional leaders. While the crisis is still raw, Orphanides (2010) argues that there are already lessons the ECB can learn from this event. He says that reviewing forward looking policy; inflation forecast and expectations, constitutes as good monetary policy. Orphanides states that central banks respond better to forecasts of future economic variables, rather than current levels of these variables. For this paper we look deeper in the work of Gerlach (2007). As we want to study whether the current sovereign debt crisis has had an effect on the behavior of the ECB on the interest setting, we use his method of working as base for our methodology. So we can compare the situation before the crisis with the current situation, during the crisis.

(7)

7 3. Methodology

In this paper we want to examine if the determinants of the interest rate are the same before, as during the crisis, we do so by studying the behavior of the ECB of interest rate setting in the timeframe of October 2008 to March 2016. The observed MRO rate of our sample is plotted in Figure 1. As you can see, the MRO rate has been decreased from approximately 4 perfect to almost zero. This can be explained by Gerlach & Lewis (2014) as they find that the ECB has strong incentives to reduce the interest rate as much as possible in order to stimulate the economy. In order to study this behavior of interest setting, we will be using an ordered probit model.5 Although the ECB makes its decisions on its monetary policy on a monthly basis, there are times that they leave the MRO rate unchanged. In fact, as is observed and will be discussed later on, the ECB does not change the MRO rate to a large extent. Whenever they do change the interest rates, they change them by discrete amounts: multiples of 25 basis points (Carstensen, 2006). For this reason, it is therefore more appropriate to estimate the behavior by using an ordered probit model that distinguishes between months in which the interest rates have been changed – raised or reduced, or have been left unchanged (Gerlach, 2004).

Figure 1 MRO rate October 2008 to March 2016 6

5_{Similar to the work of Gerlach (2007) and Carstensen (2006).} 6_{Data obtained from Datastream UvA}

0 0.5 1 1.5 2 2.5 3 3.5 4 2008 2009 2010 2011 2012 2013 2014 2015 2016 % MRO

(8)

8 An ordered probit model is a model which is used to estimate an ordinal dependent variable and its relationship with a set of independent variables. The outcome can take on more than two different values and can be ordered, which in this paper will be by size of change. The dependent variable is estimated as a linear function of the independent variables and a set of cut points. It observes the probability of an outcome of the dependent variable, that corresponds with the probability of the estimated linear function, plus a random error. The function of our ordered probit model that is estimated, is equation (4). This probability lies in a range of cut points, which are estimated for each outcome. 7 These cut points are also referred to as thresholds8; they are coefficients of the model and define where the intervals begin and end.9

In this paper, we are using data from the period of October 2008 to March 2016, using data on a monthly basis since the ECB sets the MRO rate every month. In this section, first the model is described, followed by the data that is used, and lastly the estimates will be reported and discussed.

3.1 The model

The main goal of the ECB is maintaining price stability, which is defined as year-on-year increase in the Harmonized Index of Consumer Prices (HICP). In order to pursuit this goal, the ECB set its aim to keep inflation rate below, but approximately at 2 percent. The ECB also uses a two-pillar approach on estimating the risk to price stability. These pillars are analytical

perspectives, where the ECB focuses on (1) real activity and financial conditions of the economy, and (2) analyzing monetary and credit developments with an aim to determine possible

implications for future inflation and economic growth by using broad aggregate (M3) as analysis of developments of money aggregates. 10

As it seems that the ECB adjusts its MRO rate gradually over time, rather than adjusting the MRO rate immediately to the desired level 11, the target of the ECB for the MRO rate may differ from the actual level of the MRO rate. The reason the ECB uses this smoothening while

7_{www.stata.com/manuals13/roprobit.pdf}

8_{Gerlach (2007) and Carstensen (2006) refer to thresholds in their work} 9_{www.stata.com/support/faqs/statistics/cut-points/}

10_{https://www.ecb.europa.eu/mopo/strategy/html/index.en.html}

11_{“Central banks often appear to adjust interest rates in a gradual fashion – taking small, distinct steps toward a}

(9)

9 setting interest rates, is to minimize the interest rate volatility (Sack & Wieland, 2000). Another reason the target of the ECB can differ from the actual rate, is that interest rates are set at discrete levels, as declared above (Gerlach, 2007).

Since the main goal of the ECB is maintaining price stability, it seems reasonable that the ECB its target for the MRO rate depends on: the real economic activity, inflation rate, money growth, and as Gerlach (2007) mentions; the rate of appreciation of the nominal effective exchange rate. These are denoted as: 𝑦_𝑡, 𝜋_𝑡, 𝜇_𝑡, and 𝜀_𝑡 respectively. Let 𝑖_𝑡𝑇_{be the target MRO} rate, then the target level for the MRO rate is described by the following equation:

𝑖_𝑡𝑇 = 𝛼₁𝑦_𝑡+ 𝛼₂𝜋_𝑡+ 𝛼₃𝜇_𝑡+ 𝛼₄𝜀_𝑡 (1) To allow for gradual changes in interest rates as explained by Judd & Rudebusch (1998) and considering central banks prefer interest rate smoothing (Gorter, Stolwijk, Jacobs, & De Haan, 2010), we consider the following equation:

𝑖_𝑡− 𝑖_𝑡−1= 𝛽₁(𝑖_𝑡𝑇_{− 𝑖}

𝑡−1) + 𝛽2∆𝑖𝑡−1+ 𝑒𝑡 (2) Where the constant is omitted, 𝑖𝑡𝑇− 𝑖𝑡−1 is the difference between the target rate of the ECB and the lagged MRO rate, ∆𝑖_𝑡−1 is the lagged level of change in the MRO rate, and 𝑒_𝑡 is the residual. Which explains that the MRO rate is also dependent on the lagged MRO rate and the lagged change in MRO rate.

Because the interest rates of are set in discrete steps of basis points by the ECB, we can only observe discrete level of changes. This is shown in equation (3), which is obtained by combining equation (1) and (2), where 𝛿_𝑖 ≡ 𝛼_𝑖𝛽_𝑖 and 𝑖_𝑡∗_{is the unobserved interest rate.}

𝑖_𝑡∗_{− 𝑖}

𝑡−1= 𝛿1𝑦𝑡+ 𝛿2𝜋𝑡+ 𝛿3𝜇𝑡+ 𝛿4𝜀𝑡− 𝛽1𝑖𝑡−1+ 𝛽2∆𝑖𝑡−1+ 𝑒𝑡 (3)

First of all, we expect the parameter of real economic activity to be positive, since a growth in economic activity leads to a rise in money demanded. In order to restore the

equilibrium of the demand and supply of money, the ECB will increase the interest rate so that the money demand will decrease. The parameter of inflation is positive considering the Fisher-effect; a higher expected inflation leads to an equal rise in the interest rate. We also expect the parameter of money growth to be positive, since the ECB wants to maintain price stability and aims at keeping the inflation rate close to 2 percent. A rise in money growth will namely lead to a

(10)

10 rise in inflation. In order to restrict money growth, interest rates will rise (Pilbeam, 2013). The coefficient for the appreciation of the exchange rate is expected to be negative as well. The reason for this is that appreciation of the euro can lead to inflation as effect of increasing prices. In order to maintain price stability, we expect the ECB to lower the interest rate so that investors are going to invest abroad which will lead to stagnation of the rise in appreciation of the euro. Lastly, we expect the coefficients β₁ and β₂ of the lagged MRO rate and the lagged change in MRO rate respectively, to be positive, since the ECB keeps its MRO rate unchanged most of the times and the adjustments of the MRO rate are regularly set in the same direction (Gerlach, 2007). So a decrease in the lagged MRO rate is associated with a decrease in the current MRO rate.

With the ordered probit model we can determine the probability of a certain change in the interest rate. To determine the probability of certain changes, we create five categories that will be divided by cut points. The actual change in the MRO rate that is observed ∆𝑖_𝑡, therefore will be dependent on where the unobserved variable is, relative to this set of cut points with values 𝑘𝑖. Our probit model looks as follows:

Pr(∆𝑖𝑡 = 𝑗) =

Pr (𝑘_𝑖−1< 𝛿₁𝑦_𝑡+ 𝛿₂𝜋_𝑡+ 𝛿₃𝜇_𝑡+ 𝛿₄𝜀_𝑡− 𝛽₁𝑖_𝑡−1+ 𝛽₂∆𝑖_𝑡−1+ 𝑒_𝑡 ≤ 𝑘_𝑖)

(4)

Where 𝑗 is the number of possible outcomes, 𝑘𝑖 are the cut points, and i is the number of possible outcomes.

As outlined in Table 1, we will be distinguishing the size of change of the MRO rate in five categories. These categories are: large decrease, small decrease, no change, small increase, and large increase. The reason we have chosen for these five categories, is because we want to make a distinction between the size of these changes. Considering we already know the values the MRO rate takes on in our sample, we can set the cut points ourselves. These categories will help us understand the extent of changes in the MRO rate whenever the ECB adjusts the interest rate.

(11)

11 Table 1 Actual change in MRO rate in categories

Large decrease if −0.75 ≤ 𝑖_𝑡∗− 𝑖_𝑡≤ −0.50 Small decrease if −0.50 < 𝑖_𝑡∗_{− 𝑖} 𝑡≤ −0.05 No change if −0.05 < 𝑖_𝑡∗− 𝑖_𝑡< 0.05 Small increase if 0.05 ≤ 𝑖_𝑡∗_{− 𝑖} 𝑡< 0.50 Large increase if 0.50 ≤ 𝑖𝑡∗− 𝑖𝑡≤ 0.75

What is observed in our sample, is the change in MRO rate from October 2008 to March 2016. As you can see in Table 2, smaller changes are more common than large changes. We can see that out of the ninety observations, the ECB has only changed the interest rate 16 times, which means that it left the interest rate unchanged approximately 82 percent of the times. Meaning that for the most part, the ECB has left its MRO rate unchanged.

Table 2 Changes in MRO-rate12

∆𝑖_𝑡 Freq. Percent Cum.

Large decrease 5 5.56 5.56 Small decrease 9 10.00 15.56 No change 74 82.22 97.78 Small increase 2 2.22 100.00 Large increase 0 0.00 100.00 Total 90 100.00

Comparing our results to the ones of Gerlach (2007), we see that in his study from the before the crisis, the interest rate remained unchanged 71 times out of the 89. This is roughly 80 percent. Although, it is interesting to see that while the interest rate has been left unchanged about as many times before the crisis as during the crisis, there is one outstanding difference. Before the crisis, as Gerlach (2007) points out, the interest rate has been increased 10 times and has been decreased 8 times. Looking at our results, we see that the interest rate has only been

(12)

12 increased 2 times, while it has been decreased 14 times. We can therefore argue that the

sovereign debt crisis has affected the direction and the size of the change of the MRO rate, rather than if it changes the MRO rate.

3.2 Data

I order to estimate the real economic activity, the inflation rate, the money growth, and the rate of appreciations of the nominal effective exchange rate, we first start by discussing the choice of data used.

The real economic activity is chosen as one of the indicator variables, since real economic activity has an impact on the inflation rate with a lag (Gerlach, 2004), and ensuring price stability is the main objective of the ECB. Following Gerlach (2007) we use an economic sentiment indicator (ESI), which is created by the European Commission in order to measure the real economic activity. ESI is a summary indicator that is derived for different economic sectors and used as a subjective indicator of real economic activity.13

An alternative for using ESI, is to use output gap as an indicator for measuring real economic activity. Output gap is the difference between actual Gross Domestic Product (GDP) and the target set for optimal GDP. Although this output gap is highly uncertain since it is constructed with long time lags, Taylor (1993) uses this as measure for real economic activity. According to Taylor output gap is best for explaining interest rate setting decisions taken by the central banks. However, Gerlach & Lewis (2014) argue that central banks respond more to the growth of real GDP in practice, than to output gap. A reason for this is that the trend level of output is difficult to measure with any precision in real time. Furthermore, Gerlach (2007) finds in his results that output gap appears to play no role in the analysis of the ECB on current economic conditions. For our econometric analysis we use the output gap for the euro area. Unfortunately, the obtained data on the output gap is quarterly, which we have to match with our other monthly data. Matching the data can be done by assuming the output gap to be constant over the months concerning the quarterly data, and matching the output gap to the months that belong to that specific quarter.

Besides ESI and output gap, Gerlach (2007) also looks at the expected GDP growth as an

13_{The economic sentiment indicator refers to the euro area and is based on an extensive survey of firms and} consumers.

(13)

13 indicator for real economic activity. In this paper we will use the real GDP growth forecast based on standardized ESA definition. 14_{These forecasts are reported quarterly, but since this paper} looks at all the variables on a monthly basis, we assumed the real GDP growth to be constant and calculated the grow on a monthly basis that matches the quarterly ones by using the formula given in equation (5).

𝑥 = (1 + 𝑔)3− 1 (5)

Where 𝑥 is the quarterly expected GDP growth, and 𝑔 is the growth factor of the months within the quarter.

For the overall inflation rate, we use the harmonized index of consumer prices (HICP) as measurement, which is the headline inflation in the euro area (Gerlach, 2007). As mentioned earlier on, the HICP is used to measure the inflation rate for which the ECB aims to maintain an annual level close to 2 percent.15_{The ECB regularly refers to the HICP measure while discussing} inflation pressures. Furthermore, the HICP is also been used in related work of Gerlach (2007) and Carstensen (2006).

Since the ECB is forward looking when determining monetary policy, it seems plausible that the ECB also looks at forward inflation when setting the interest rate (Gorter, Stolwijk, Jacobs, & De Haan, 2010). Since the ECB Survey of Professional Forecasters (SPF)16 provides quarterly expectations on the inflation rate of the current year, we adapted the quarterly

expectations as monthly expectations by allocating the quarterly data to the concerning months. As for the measure of money growth, we use the broad monetary aggregate (M3), considering M3 regards the highest degree of liquidity of the assets it includes. Gerlach (2007) mentions that M3 is the single most crucial determinant of monetary developments, given the emphasis the ECB puts on this matter regarding monetary policy. Although there is only little evidence on this object whether the ECB does or does not react to monetary growth, the ECB has highlighted the importance of the growth of M3 for its policy decisions (Gerlach, 2004).

Especially as the money in circulation is significant for ensuring price stability, we argue that

14_{https://www.ecb.europa.eu/stats/prices/indic/forecast/html/table_hist_rgdp.en.html} 15_{https://www.ecb.europa.eu/stats/prices/hicp/html/index.en.html}

(14)

14 using money growth is valuable in determining the interest rate setting. To do so, we used period to period changes in M3 as an indicator of M3 growth.

Lastly, following the work of Heinemann & Ullrich (2005) and Gerlach (2007), the nominal effect exchange rate (NEER) is used. This variable measures the external value of the euro against the currencies of the countries the euro area trades most with.17 In our econometric analysis we have used the weighted average of exchange rates against 19 partners of the euro area. If the index goes up, more foreign currency can be obtained with one euro, meaning an appreciation of the euro.

4. Results

Using the data described in the precious section, the estimates of equation (4) are presented in the columns 1 to 6 of Table 3. We cannot interpret the magnitude of the coefficients given in Table 3, we can only interpret the sign. In order to look at the magnitude of the coefficients, we will be looking at the marginal effects later on.

First, we will describe the difference in the models 1 to 6. As discussed in the data section, there are different methods in measuring the real economic activity. In models 1 and 2 we use the ESI as measure for real economic activity, whereas in models 3 and 4 we use the output gap, and in models 5 and 6 we use the expected GDP growth.

We also discussed two different measures of inflation, in models 1, 3 and 5 the HICP is used, and in models 2, 4 and 6 the expected inflation is used.

We expected the coefficients for real economic activity, inflation, and money growth to be positive, and the coefficient of the exchange rate to be negative. This is also supported by our results, except for the expected GDP growth as measure for real economic activity, and for the exchange rate.

What is interesting to see, is that when the expected GDP growth is used as indicator for the real economic activity, both measures for inflation are significant. Whereas the models in which ESI and the expected GDP growth is used, both measures are insignificant. The lagged MRO rate on the other hand, is significant for all different measures that are used for real

(15)

15 economic activity or inflation. We would therefore argue that the lagged MRO rate is an

important indicator for the behavior of the ECB on setting the interest rates.

As you can see in Table 3, the pseudo-R2_{is the highest when using output gap as an indicator for} real economic activity. When using this indicator, the pseudo-R2 is greater than 0.3 and has the best fit compared to the other models. Therefore, we will be looking closer to the estimates of models 3 and 4. This result is endorsed by Taylor (1993), as he argued that the output gap is best for explaining interest rate setting decisions taken by the central banks.

Interestingly, Gerlach (2007) found in his results that the psedue-R2 is much lower when using output gap. In fact, output gap is the only measure of real economic activity that is not significant in his results. Besides the output gap, the only other insignificant variable is both measures for inflation in case expected GDP growth is used as indicator for real economic activity.

In both models 3 and 4 we see that the parameter of the output gap is positive and significant, which means that with a higher output gap it is more likely that the interest rate has been increased. The lagged MRO rate is negative and significant, this means that with a higher lagged MRO rate it is more likely that the interest rate has been decreased. The rest of the variables are insignificant, suggesting that there is no reaction on inflation, money growth, the exchange rate, and the lagged change in the MRO rate. While the results of Gerlach (2007) show that only the two measures for inflation and the measure of output gap are insignificant variables.

As mentioned before, we can only interpret the sign of the coefficient, not the magnitude. Therefore, we will look at the marginal effects of models 3 and 4. Since the frequency of the observed change in interest rate is highest for category 2 (small decrease) and category 3 (no change), we will only present the marginal effects of both models for category 2 and 3. The marginal effects of the other categories of all of the models are given in the appendix.

We start off by looking at the marginal effects of model 3, in which the HICP is used as measure for inflation. The chance that the change in MRO rate falls in category 2 is 8.8 percent. In Table 4 we can see that only the marginal effects of the output gap and the lagged MRO are significant, respectively at 5 percent and 1 percent. We can interpret these results as follows: if the output gap increases with one unit, the change in interest rate is 9.1 percent less likely to be in the second category, and if the lagged MRO rate increases with one unit, the change in interest

(16)

16 rate is 18.4 percent more likely to be in the second category. Where the second category means, that the change in interest rate will be a small decrease with an amount that is smaller than -0.50 basis points, and larger or equal to -0.05 basis points. In other words, if the lagged MRO rate is increased with one unit, it is likely that the MRO rate of the current period will decrease. While it does not seem reasonable to discuss the marginal effects of the other variables since they are not significant, they are presented in Table 4 as well.

The probability that the change in MRO rate falls in category 3 is notably larger, namely 89.8 percent. Which was expected since the observed frequency of no change in interest rate was 74 times, as is shown in Table 2. In this case, if the output gap increases with one unit, it is 9.6 percent more likely that there will be no change in interest rate, while if the lagged MRO increases with one unit, it is 19.3 percent less likely that there is no change in interest rate. We can therefore say that if the MRO rate in the previous period has been raised, it is less likely that the MRO rate of the current period will not be changed. This is a result of the interest rate smoothing, since the ECB tends to change interest rates in sequences of small steps (Sack & Wieland, 2000). Therefore, a change in the MRO rate in the previous month is more likely to result in a change in the MRO rate in the current month as well.

In model 4, where the expected inflation is used as measure for inflation, the marginal effects are approximately the same, so they can be understood the same way the results of model 3 are interpreted.

Overall the only significant determinants we found in our results are the output gap and the lagged MRO rate. Therefore we argue that the ECB reacts only to changes in output gap and the lagged MRO rate in our sample.

(17)

17 Table 3 Ordered probit estimates of ECB reaction function

Model 1 2 3 4 5 6 ESI 0.0388301 (0.0267847) 0.0423606 (0.0261819) Output gap 0.6175291** (0.2858999) 0.6330418** (0.2830943) Expected GDP growth -0.1130792 (0.2284198) -0.1452954 (0.2378164) HICP 0.2893772 (0.1840042) 0.1065795 (0.2160019) 0.426752** (0.2109475) Expected Inflation 0.3761073 (0.2400279) 0.1174538 (0.2848773) 0.575848* (0.3003826) M3 4.19e-06 (4.12e-06) 3.92e-06 (4.13e-06) 3.07e-06 (4.24e-06) 2.96e-06 (4.34e-06) 5.56e-06 (4.11e-06) 5.53e-06 (4.14e-06) NEER 0.0616244 (0.11791) 0.0354584 (0.1172547) 0.0408253 (0.1210857) 0.0299124 (0.1207357) 0.050161 (0.116991) 0.0162874 (0.1150911) Lagged MRO -0.55780236** (0.2940176) -0.6404516** (0.3171987) -1.245051*** (0.314321) -1.275205*** (0.3187391) -0.8833173*** (0.2658836) -1.015018*** (0.3162836)

Lagged change MRO 0.9668716

(1.300743) 0.7088337 (1.34295) 1.621697 (1.176545) 1.584783 (1.218295) 1.959591 (1.212225) 1.787331 (1.235763) Pseudo-R2 _0.2786 _0.2784 _0.3059 _0.3052 _0.2617 _0.2578

Note: The coefficients of the variables are presented with the standard errors in brackets (). *, **, and *** denote significance at 10, 5, and 1 percent respectively in this z-test.

(18)

18 Table 4 Marginal effects after ordered probit model 3

y = Pr(ChangeMRO==2) = 0.08802351

Variable dy/dx Std. Err. P>|z|

Output gap -0.0914465 0.04319 0.034

HICP -0.0157828 0.03306 0.633

M3 -4.55e-07 0.00000 0.480

NEER -0.0060456 0.01796 0.736

Lagged MRO 0.1843729 0.06452 0.004

Lagged Change MRO -0.2401484 0.19036 0.207 Note: ChangeMRO is used in STATA and describes ∆𝑖𝑡, where

ChangeMRO==2 means that ∆𝑖𝑡 falls to category 2: small decrease.

Table 5 Marginal effects after ordered probit model 3 y = Pr(ChangeMRO==3) = 0.89803582

Output gap 0.0955519 0.04658 0.040

HICP 0.0164913 0.03432 0.631

M3 4.75e-07 0.00000 0.481

NEER 0.006317 0.01877 0.736

Lagged MRO -0.19265 0.06648 0.004

Lagged Change MRO 0.2509294 0.19625 0.201 Note: ChangeMRO is used in STATA and describes ∆𝑖𝑡, where

ChangeMRO==3 means that ∆𝑖𝑡 falls to category 3: no change.

Output gap -0.0934882 0.04289 0.029

Expected inflation -0.0173457 0.04326 0.688

M3 -4.37e-07 0.00000 0.494

NEER -0.0044175 0.01781 0.804

Lagged MRO 0.1883234 0.0673 0.005

Output gap 0.0980788 0.04612 0.033

Expected inflation 0.0181974 0.04525 0.688

M3 4.59e-07 0.00000 0.495

NEER 0.0046344 0.01869 0.804

Lagged MRO -0.1975708 0.06934 0.004

(19)

19 5. Conclusion

All in all we can conclude that the overall most important determinant which is taken into consideration while setting the MRO rate, is the lagged MRO rate. This is the only determinant that is tested significant in all our models. We therefore argue that the right way to measure real economic activity, is by using the output gap just like Taylor (1993).

Looking at the observed changes in interest rate during the sample, we argue that even during the crisis, the ECB does smooth its interest rate. However, in our results this is not an outstanding observation, since we can only recognize this when looking at the marginal effects in category 3, where there is no change observed.

Comparing our results to the results from Gerlach (2007), we argue that the ECB does not value the variables the same in times before the crisis as during the crisis when looking at the significance and the coefficients of the variables. The results of Gerlach show that all variables except the output gap and inflation are significant, while in our results is shown otherwise. A reason for this could be that present day the ECB looks at a wider set of determinants (e.g. financial stress indicators), in order to measure the economic state (Gerlach & Lewis, 2014). Another reason could be that a linear regression is not the right way to measure the relationship between the independent variables and the MRO rate. Marcellino (2002) argues that it is uncertain whether adopting linear specifications in order to evaluate the monetary policy and interest rate setting by the ECB, is the proper statistical framework to analyze this issue.

We do come to the same conclusion as Gerlach (2007) on the fact interest rates are more closely tied to economic activity than to inflation.

So overall we can conclude that the behavior of interest rate setting has changed during the crisis and that there are other variables that need to be considered when studying the behavior of the ECB on setting the interest rate.

5.1 Discussion

This study has several implications since it does not answer the questions that are raised by the conclusion. First of all; why are the measures for inflation insignificant when the main goal of the

ECB is maintaining price stability and ensuring the inflation rate to be close to 2 percent?

(20)

20

ECB on interest rate setting? Or should other variables be taking into account?

For future studies it seems therefore reasonable to look into other variables as well, especially the ones that measure the economic state in times of the crisis. We also argue that it is better to also study the announcements regarding interest rate setting, made by the ECB, similar to Gerlach (2007). By taking these announcements into account, it is more apparent to understand why the ECB did or did not react to a certain change in one of the determinants. The ECB might expect a certain effect to be temporary and assume it will only have an effect on the short run Therefore they will not respond to this in terms of setting the interest rate.

Furthermore, the crisis causes uncertainty within the economy, stability might not be something to assume so easily. We therefore argue that for further studies, it is be better to set shorter time ranges and study the reaction function of the ECB per year, for example.

(21)

21 References

Alter, A., & Beyer, A. (2013). The Dynamics of Spillover Effects During the European Sovereign Debt Turmoil. ECB Working Paper Series No. 1558.

Asso, P. F., Kahn, G. A., & Leeson, R. (2010). The Taylor Rule and the Practice of Central Banking. The Federal Reserve Bank of Kansas City RWP 10-05.

Carstensen, K. (2006). Estimating the ECB Policy Reaction Function. German Economic Review, 1-34.

De Santis, R. A. (2012). The Euro Area Sovereign Debt Crisis. ECB Working Paper Series No.

1419.

Gerlach, S. (2004). Interest Rate Setting by the ECB: Words and Deeds. CEPR Discussion Paper

No. 4775.

Gerlach, S. (2007). Interest Rate Setting by the ECB, 1999-2006: Words and Deeds.

International Journal of Central Banking, 1-45.

Gerlach, S., & Lewis, J. (2014). ECB Reaction Functions and the Crisis of 2008. International

Journal of Central Banking, 137-157.

Gorter, J., Stolwijk, F., Jacobs, J., & De Haan, J. (2010). ECB Policy-Making and the Financial Crisis. DNB Working Paper No. 272.

Heinemann, F., & Katrin, U. (2005). Does it Pay to Watch Central Bankers' Lips? The Information Content of ECB Wording. ZEW Discussion Paper No.05-70.

Judd, J. P., & Rudebusch, G. D. (1998). Taylor's Rule and the Fed: 1970-1997. Economic

Review, 3-16.

Marcellino, M. (2002). Forecasting EMU Macroeconomic Variables. IGIER Working Paper

Series n.216.

Orphanides, A. (2010). Monetary Policy Lessons from the Crisis. Central Bank of Cyprus

Working Paper No. 2010-1.

(22)

22 Sack, B., & Wieland, V. (2000). Interest-Rate Smoothing and Optimal Monetary Policy: A

Review of Recent Empirical Evidence. Journal of Economics and Business, 205-228. Sauer, S., & Sturm, J.-E. (2003). ECB Monetary Policy: How well does the Taylor Rule describe

it? CESifo Working Paper Series No.1110.

Schmidt, V. A. (2014). Speaking to the Markets or to the People? A Discursive Institutionalist Analysis of the EU's Sovereign Debt Crisis. The Britisch Journal of Politics and

International Relations, 188-209.

Taylor, J. B. (1993). Discretion versus policy rules in practice. Carnegie-Rochester Conference

(23)

23 Appendix

Table 8 Summary Statistics

Variable Obs Mean Std. Dev. Min Max

(obs) ∆𝑖_𝑡 90 -0.0466667 0.1479827 -0.75 0.25 ESI 90 96.11778 9.734989 69.3 109.1 Output gap 90 -2.420333 1.010873 -4.08 0.58 Expected GDP growth 90 -0.144703 0.8758945 -2.518294 0.4260431 HICP 90 1.266667 1.073794 -0.7 3.2 Expected inflation 90 1.33 0.901366 0.1 3.4 M3 90 19777.53 39301.23 -150566 147352 NEER 90 -0.1174522 1.438577 -3.8272 5.1367 Lagged MRO 90 0.8811111 0.8337198 0.05 4.25

Lagged change MRO 90 -0.0466667 0.1479827 -0.75 0.25

Note: for ∆𝑖𝑡 the actual observed value is used, since the change in MRO rate is been divided in categories, its

summary statistics does not make sense because the values 1 to 5 do not mean anything.

ESI -0.0011184 0.000113 0.322

HICP -0.0083347 0.00842 0.322

M3 -1.21e-07 0.00000 0.421

NEER -0.0017749 0.0036 0.622

Lagged MRO 0.0166484 0.01518 0.273

Lagged Change MRO -0.0278481 0.04344 0.521

Note: ChangeMRO is used in STATA and describes ∆𝑖𝑡, where ChangeMRO==1

(24)

y = Pr(ChangeMRO==2) = 0.09925709

ESI -0.006192 0.00436 0.156

HICP -0.0461452 0.03233 0.153

M3 -6.69e-07 0.00000 0.310

NEER -0.0098269 0.01883 0.602

Lagged MRO 0.092174 0.05556 0.097

ESI 0.0063884 0.00456 0.161

HICP 0.0476092 0.03339 0.154

M3 6.90e-07 0.00000 0.315

NEER 0.0101386 0.01945 0.602

Lagged MRO -0.0950982 0.05601 0.090

ESI 0.000922 0.00087 0.287

HICP 0.0068708 0.006675 0.309

M3 9.96e-08 0.00000 0.392

NEER 0.0014632 0.00293 0.617

Lagged MRO -0.0137242 0.01289 0.287

ChangeMRO==4 means that ∆𝑖𝑡 falls to category 4: small increase.

ESI -0.0012483 0.0012 0.299

M3 -1.16e-07 0.00000 0.437

NEER -0.0010449 0.00352 0.767

Lagged MRO 0.0188728 0.01697 0.266

(25)

y = Pr(ChangeMRO==2) = 0.0993617

ESI -0.0067473 0.00434 0.120

M3 -6.25e-07 0.00000 0.341

NEER -0.0056479 0.01868 0.762

Lagged MRO 0.1020128 0.05978 0.088

ESI 0.0070125 0.00456 0.124

M3 6.49e-07 0.00000 0.346

NEER 0.0058698 0.01941 0.762

Lagged MRO -0.1060216 0.06096 0.082

ESI 0.0009831 0.0009 0.276

M3 9.11e-08 0.00000 0.411

NEER 0.0008229 0.00277 0.766

Lagged MRO -0.0148641 0.01374 0.279

Variable dy/dx Std. Err. P>|z| Output gap -0.0140122 0.01309 0.284

HICP -0.0024184 0.00545 0.657

M3 -6.97e-08 0.00000 0.539

NEER -0.0009264 0.00284 0.744

Lagged MRO 0.0282512 0.02432 0.245

(26)

y = Pr(ChangeMRO==4) = 0.00561956

Variable dy/dx Std. Err. P>|z| Output gap 0.0099069 0.00878 0.259

HICP 0.0017098 0.00407 0.675

M3 4.92e-08 0.00000 0.543

NEER 0.000655 0.00201 0.744

Lagged MRO -0.0199741 0.01802 0.268

Variable dy/dx Std. Err. P>|z| Output gap -0.01455 0.01334 0.275

Expected Inflation -0.0026996 0.00717 0.706

M3 -6.80e-08 0.00000 0.548

NEER -0.0006875 0.00281 0.807

Lagged MRO 0.0293096 0.02528 0.246

ChangeMRO==1 means that ∆𝑖𝑡 falls to category 1: large decrease.

Variable dy/dx Std. Err. P>|z| Output gap 0.0099594 0.00889 0.263

M3 4.66e-08 0.00000 0.542

NEER 0.0004706 0.00192 0.806

Lagged MRO -0.0200622 0.01842 0.276

Variable dy/dx Std. Err. P>|z| Expected GDP growth 0.003877 0.00847 0.647

HICP -0.0146316 0.01283 0.254

M3 -1.91e-07 0.00000 0.334

NEER -0.0017198 0.00413 0.677

Lagged MRO 0.0302854 0.02325 0.193

(27)

y = Pr(ChangeMRO==2) = 0.10675158

HICP -0.0707945 0.03979 0.075

M3 -9.22e-07 0.00000 0.187

NEER -0.0083213 0.01944 0.669

Lagged MRO 0.1465348 0.0605 0.015

Variable dy/dx Std. Err. P>|z| Expected GDP growth -0.0195516 0.0402 0.627

HICP 0.0737861 0.04208 0.080

M3 9.61e-07 0.00000 0.195

NEER 0.0086729 0.02024 0.668

Lagged MRO -0.152769 0.06199 0.014

HICP 0.0116401 0.00954 0.222

M3 1.52e-07 0.00000 0.295

NEER 0.0013682 0.00329 0.678

Lagged MRO -0.0240933 0.01832 0.188

M3 -1.94e-07 0.00000 0.329

NEER -0.0005726 0.00406 0.888

Lagged MRO 0.0356863 0.02706 0.187

(28)

y = Pr(ChangeMRO==2) = 0.10552691

M3 -9.07e-07 0.00000 0.190

NEER -0.0026718 0.01889 0.888

Lagged MRO 0.1665064 0.06908 0.016

Expected GDP growth -0.0252364 0.0422 0.550

M3 9.61e-07 0.00000 0.197

NEER 0.002829 0.01998 0.887

Lagged MRO -0.1762985 0.0722 0.015

M3 1.41e-07 0.00000 0.298

NEER 0.0004155 0.00295 0.888

Lagged MRO -0.0258942 0.01998 0.195