The ECB interest rate rule under the Duisenberg presidency

European Journal of Political Economy Vol. 20 (2004) 579 – 595 www.elsevier.com/locate/econbase The ECB interest rate rule under the Duisenberg presidency André Fourcßans, Radu Vranceanu * Department of Economics, ESSEC Business School, PB 105, 95021 Cergy-Pontoise (Paris), France Received 23 October 2002; received in revised form 21 July 2003; accepted 20 January 2004 Available online 28 March 2004 Abstract This paper presents estimates of the European Central Bank (ECB)’s interest rate rule using monthly data for the period of the Willem F. Duisenberg presidency (from January 1999 to October 2003). Our results show that, like the US Federal Reserve, the ECB appears to be concerned with fluctuations in economic activity. It increases its short-term interest rate if inflation deviates from target, this reaction being stronger if based on future inflation and weaker if based on current inflation. The seemingly soft response to current inflation derives from a concern for exchange rate stability and may reflect the Bank’s forecast of future inflation. The ECB also appears to smooth its interventions in the money market. D 2004 Elsevier B.V. All rights reserved. JEL classification: C51; E52; E58 Keywords: ECB; Augmented Taylor rule; Monetary policy; Interest rate; Central bank 1. Introduction A monetary policy rule describes how a central bank adjusts its main policy instruments in response to changes in the macroeconomic environment. While numerous theoretical studies have analyzed monetary policy rules within a normative perspective, in recent years, economists have directed attention to investigating decision rules followed by central bankers in every day practice. Because, over time, short-term interest rates have nearly everywhere become the favored instrument of monetary policy management, empirical monetary policy rules are most often referred to as interest rate rules. * Corresponding author. Tel.: +33-1344-33183; fax: +33-1344-33001. E-mail address: vranceanu@essec.fr (R. Vranceanu). 0176-2680/$ - see front matter D 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.ejpoleco.2004.01.003 580 A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 The seminal study on monetary policy rules is by Taylor (1993). Studies have usually focused on the US economy, although other countries have been investigated and there have been international comparisons (see, for example, Clarida et al., 1998, 2000). Some studies have assessed the Bundesbank’s policy rule, which has been used as a criterion by which to evaluate the policy of the European Central Bank (ECB) (Faust et al., 2001; Surico, 2003). Other studies ask whether the US Fed’s rule would well fit the European Monetary Union (EMU) (Taylor, 1999a). Gerlach and Schnabel (1999), Doménech et al. (2002) and Gerdesmeier and Roffia (2003) derived virtual interest rate rules using consolidated data for the 1990s for countries that later formed the EMU; Surico (2003) inferred an interest rate rule for the euro area for 1997 – 2002 using consolidated data pertaining to individual countries for pre-EMU years and official aggregate series for the period thereafter. The ECB was established in June 1998 and began operations in January 1999 when the euro was launched. The objective of this paper is to contribute to understanding of ECB monetary policy management by inferring an interest rate rule from monthly data. The empirical analysis covers the period from January 1999 to October 2003 during which time Willem F. Duisenberg (the former governor of the Netherlands’ central bank) was head of the ECB.1 Our previous study (Fourcßans and Vranceanu, 2002) and also studies by Ullrich (2003) and Sauer and Sturm (2003) have investigated ECB behavior during shorter periods (from January 1999 to, respectively in the above three studies, March 2002, August 2002 and March 2003). The paper is organized as follows. Section 2 introduces the traditional theoretical framework and reviews the literature with its main empirical results. Section 3 presents the ECB policy framework and develops the empirical evidence on the ECB’s interest rate rule. Section 4 presents the conclusions and draws policy implications. 2. Interest rate rules: the standard framework 2.1. An elementary policy rule Taylor (1993) introduced a policy rule that seemed to describe well the US Federal Reserve’s choice of short-term interest rates (more precisely, of the Federal Funds target rate) from 1987 to 1992. He conjectured that the Fed pushes up short-term interest rates whenever the effective inflation rate rises above the implicit target of the policymaker, or if effective GDP rises above potential output. Formally, these assumptions can be represented in a simple linear form: it* ¼ r̄ þ pt þ ðb  1Þðpt  p̄Þ þ cyt ; ð1Þ where i*t is the target interest rate, pt is the inflation rate, p̄ is the target inflation rate, yt is the output gap, r̄ is the equilibrium real interest rate (pre-determined) and b and c are 1 Duisenberg was followed as head of the ECB by Jean-Claude Trichet, the former Governor of the French Central Bank. A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 581 given parameters. According to Taylor, a rule that matches well the actual series is obtained when inflation is measured over the previous four quarters, the output gap is proxied by the real GDP deviation from the linear trend, and parameter values are: r̄=2%, p̄=2%, b=1.5 and c=0.5. The initial model was based on a priori choice of parameters. Subsequent analyses used the standard ordinary least squares (OLS) regression model to assess values. Eq. (1) can be written in an equivalent form: it* ¼ r̄ þ p̄ þ bðpt  p̄Þ þ cyt ; ð2Þ where r̄ + p̄ = ī can be interpreted as the desired (target) nominal interest rate, obtained when inflation and output are at their target levels. The estimated coefficient b indicates the sensitivity of interest rate policy to deviations in the current inflation rate and c indicates the sensitivity of interest rate policy to the output gap. In a form ready for estimation, Eq. (1) is: it* ¼ c þ bpt þ cyt with c ¼ r̄ þ ð1  bÞp̄: ð3Þ Coefficients b and c are read directly from the regression model. If the equilibrium real interest rate can be inferred (for instance, from historical data), the constant term c allows us to determine the implicit inflation target: p̄=(r̄c)/(b1). ‘‘Contemporaneous’’ rules that explain interest rate decisions by means of current macroeconomic variables (Eq. (2)) have been criticized on the grounds that the policymaker does not know the value of these variables, for example because of publication lags (McCallum and Nelson, 1999). This criticism does not apply to ‘‘backward-looking’’ rules, where the interest rate depends on lagged independent variables. However, the original Taylor rule holds if policymakers are able to form correct (on average) expectations about the relevant variables so that a current variable is the best proxy for an expected variable. Orphanides (2001) pointed out another important pitfall of inferring interest rate rules from historical data. He proposed that the resulting model could not accurately describe the real-time decision of the central banker, given that the data are usually subject to significant revision. Correct estimates of the policy rule would build on un-revised data, such as was available at the time the decision was made.2 In another study, Orphanides (2003) infers an interest rate rule from real time data from the United States; results do not challenge in a significant way those based on actual data. 2.2. Generalized policy rules Since effects of monetary policy work their way through the economy with a lag of a few months, it has also been argued that policymakers focus on future rather than past or current behavior of prices (Batini and Haldane, 1999; Rudebush and Svensson, 1999). To address this criticism, so-called ‘‘forward-looking’’ interest rate rules have been estimated (inter alia, see Clarida et al., 2000; Surico, 2003; Gerdesmeier and Roffia, 2003). Denoting 2 This criticism is attenuated if it can be assumed that the policymaker uses the preliminary figures only to form his (correct) expectations about the ‘‘true’’ value of a given variable. 582 A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 by E[– ] the expectation operator and by It the information set at the time the interest rate is chosen (i.e., at time t), such a baseline policy rule takes the form: it* ¼ ī þ bðE½ptþk AIt   p̄Þ þ cE½ytþq AIt ; ð4Þ where i* t is the target interest rate, pt+k is the inflation rate k periods ahead (in annual rates), p̄ is the inflation target and yt+q is the average output gap q periods ahead. The constant term ī=r̄+p̄ denotes the desired nominal interest rate. This formulation encompasses backward-looking rules as a subset: One only needs to assign negative values to the time subscripts k and q depending on the desired lag and replace expectations with actual data. When k and q are set to zero, Eq. (4) becomes a version of Taylor’s contemporaneous rule (Eq. (2)). The basic specification above may not reflect all the factors that affect the setting of the interest rate. For instance, several central banks have a declared objective of exchange rate stability. Or central banks may target a money stock measure (as has been the case with the Swiss central bank and the former Bundesbank). Central bankers may pursue other direct economic goals. For instance, whether the central bank should take into account asset prices is still open to debate.3 The policy rule may thus include one or several of other variables. In that case, an augmented Taylor rule might take the more general form: it* ¼ ī þ bðE½ptþk AIt   p̄Þ þ cE½ytþq AIt  þ lE½HtþS AIt ; ð5Þ where H is a vector of variables other than inflation and the output gap (at time t+S ) and l is the respective vector of coefficients. It has also been argued that central bankers may be concerned that overly abrupt changes in interest rates may disrupt bond and equity markets. If this is so, the policymaker would smooth changes in interest rates such as to reach the target i*t after a more or less lengthy period. The effective interest rate chosen by the central bank, it, would follow a law of motion: it ¼ CðLÞit1 þ ð1  qÞit*; ð6Þ . .+qnLn1 is a lag-polynomial with positive coefficients; dynamic where C(L)=q1+q2L+. P stability imposes q ¼ nj¼1 qj < 1. Most often, empirical studies build on a simplified version of Eq. (6), according to which the effective interest rate that the central bank chooses at time t is a weighted average of the target rate and of the interest rate prevailing in the previous period: it ¼ qit1 þ ð1  qÞit*; with qa½0; 1: ð7Þ For q=0, the adjustment is instantaneous; the larger q, the slower is the adjustment. Eq. (7) is equivalent to: itit1=(1q)(i*i t t1), that is, within the interval (quarter or month, 3 In a national context, the political cycle may also influence the policymaker in different ways. For example, Berger and Woitek (2001) show that the Bundesbank accommodated politically related changes in money demand rather than adapting monetary policy for political ends. The ECB should be less affected by political objectives given that EMU member countries do not have identical political cycles. A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 583 depending on the data frequency), the central bank will implement only (1q) of the desired change in the interest rate. Most US studies, in general based on quarterly data, found that q=0.8, that is, within a quarter, the Fed would implement 20% of the desired change, an indication of significant policy inertia (McCallum and Nelson, 1999; Gertler, 1999; Clarida et al., 2000). Rudebusch (2002) expresses some skepticism about this result, since, in his view, no central bank will be so slow in reacting to changes in economic conditions; he argues that the smoothing in the policy interest rate might reflect a possible misspecification of the model by not taking into account serial correlation of the exogenous shocks. Combining Eq. (5) with the dynamics of the interest rate given by Eq. (7) yields a rather general form of the monetary policy rule: it ¼ ð1  qÞðī  bp̄Þ þ qit1 þ ð1  qÞbE½ptþk AIt  þ ð1  qÞcE½ytþq AIt  þ ð1  qÞlE½HtþS AIt ; ð8Þ where the effective short-term interest rate it depends on the target interest rate i*, t which itself depends on the relevant economic variables (past, current or future). 2.3. Macroeconomic stability and the interest rate rule Several theoretical analyses emphasise the essential role played by the b coefficient in the stability of the macroeconomic system (Kerr and King, 1996; Bernanke and Woodford, 1997; Taylor, 1999a; Clarida et al., 2000). These studies are based on simple macroeconomic dynamic models with three main equations: an IS curve, linking the output gap to real interest rates; a Phillips curve whereby the inflation rate is positively related to the output gap; and a monetary policy rule. When a shock pushes inflation above the target, the central bank increases its interest rate according to the policy rule. If b<1, the increase is not strong enough to bring about a higher real interest rate, demand is stimulated, and, via the Phillips curve mechanism, inflation is further enhanced. If on the contrary b>1, the strong response of the central bank brings about an increase in the real interest rate, which tempers demand and inflation. A similar logic applies to output sensitivity, with c>0 stabilising and c<0 destabilising the economic system. Empirical studies claim that the successful anti-inflationary campaign in the US in the 1980s was due to a structural change: over time, the Fed pushed the b coefficient above 1. Indeed, if many empirical estimates of b over the 1960s and 1970s are close to 0.8, they rise to 1.5 –2 for the late 1980s and the 1990s (Judd and Rudebush, 1998; Taylor, 1999b; Gertler, 1999; Clarida et al., 1998, 2000). In Europe, prior to 1999, monetary policy was managed by national central banks. However, because of the influence of the Bundesbank’s monetary policy on neighboring central banks, several economists tried to estimate a virtual interest rate rule for the now euro member countries during the 1990, building on weighted aggregate data; some of these studies also cover the ECB period until early 2002 (Gerlach and Schnabel, 1999; Doménech et al., 2002; Gerdesmeier and Roffia, 2003; Surico, 2003). The studies found a b parameter in the range of 1.6– 1.9. However, results from such studies with average A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 584 indicators should be interpreted cautiously, given the would-be statistical bias towards stability (Arnold and De Vries, 2000). To study the response of the central bank to changes in economic activity, most studies have used as a proxy the real output gap, that is, the difference in levels between current and potential output.4 The c coefficient estimated with US data over the Volcker – Greenspan period is quite stable and is in the interval [0.6, 1], depending on the model. The studies mentioned above on the virtual EMU rule found a coefficient in the range of 0.28– 1.5. However, using the real output gap as a proxy for economic activity may not be the best modeling choice, insofar as economic growth seems to be at the heart of judgments of the ‘‘right’’ economic policy or performance of a geographical zone (Walsh, 2003). Several studies have estimated Taylor rules under this alternative specification (Fair and Howrey, 1996; McCallum and Nelson, 1999; Orphanides, 2003). 3. The ECB policy rule: the empirical evidence 3.1. The ECB mission and policy instruments After this introduction to the basics of Taylor rules, we now apply the analytical framework to monetary policy management in the EMU by the European Central Bank. 3.1.1. Main objectives In keeping with Article 105 of the consolidated version of the Treaty on European Union, the primary objective of the ECB is to maintain price stability in the euro-zone.5 In 1998, the European Monetary Institute defined price stability as ‘‘a year-on-year increase in the Harmonized Index of Consumer Prices (HICP) for the euro area of below 2%’’ (ECB, 2001). In May 2003, the ECB Governing Council confirmed this definition. At the same time, the Governing Council agreed that, in the pursuit of price stability, ‘‘it will aim to maintain inflation rates close to 2% over the medium term. This clarification underlines the ECB’s commitment to provide a sufficient safety margin to guard against the risks of deflation.’’6 The HICP is a weighted average of consumer price indices of the euro member countries (Italy, France, Germany and Spain contribute to about 75% of the index). Fig. 1 plots the inflation rate, measured as the yearly relative increase in the price index from one month to the same month of the previous year. In the later months during the period, inflation was systematically above the official target, albeit not excessively. Article 105 of the Treaty on European Union also states that ‘‘without prejudice to the objective of price stability, the European System of Central Banks (ESCB) 4 Potential output, i.e., output consistent with full employment, cannot be observed. In empirical studies, potential output is set as the trend value of the output data. Yet there is little agreement on whether the trend should be linear, quadratic, follow the Hodrick – Prescott law, or something else. Taylor (1993) used a linear trend. 5 The consolidated version includes changes brought about by the Maastricht Treaty (1992) and Amsterdam Treaty (1997). See the full text at Web address: europa.eu.int/eur-lex/en/treaties/dat/EU_consol.html. 6 See ‘‘The ECB’s monetary policy strategy’’, May 8, 2003, www.ecb.int/press/03/pr030508_2en.htm. A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 585 Fig. 1. Inflation in the euro-zone (percentage change in the HCPI over the same month of the previous year). shall support the general economic policies in the Community with a view to contributing to the achievement of the objectives of the Community as laid down in Article 2.’’ According to the latter, the EU shall ‘‘promote economic and social progress and a high level of employment and (. . .) achieve balanced and sustainable development. . .’’ Several times, in particular, at the launch of the euro, ECB officials emphasized their ‘‘hands-off’’ attitude with respect to the real sector. In a representative quotation, Willem F. Duisenberg declared in 1999 that: ‘‘The ECB does not pursue an activist policy. Precise steering of the business cycle or a cyclically oriented monetary policy are not feasible and are likely to destabilize rather than stabilize the economy.’’7 Throughout the Duisenberg period, the official position of the ECB was that the best way the Bank could promote growth was to maintain price stability. However, after 2001, the ECB monthly statements began mentioning growth in the opening paragraphs, as if the Bank wished to soften its image of no concern about jobs and the economy. If the ECB has a direct economic activity goal, several indicators may convey the relevant information (such as the output gap, the unemployment rate, the output growth gap, etc.). In the empirical analysis (next section), the industrial output growth rate is used as a proxy for the output growth rate; the latter series are not available with a monthly frequency. ECB officials themselves often refer to growth of industrial output when they assess economic performance in the euro area. In order to smooth short-term fluctuations, yearly variations in the industrial output index are used (from one month to the same month of the previous year). The graph is 7 ‘‘The role of the Central Bank in the united Europe’’, speech by Willem F. Duisenberg, President of the European Central Bank, National Bank of Poland, Warsaw, Poland, on May 4, 1999. 586 A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 Fig. 2. Industrial output growth rate (percentage change over the same month of the previous year). displayed in Fig. 2. The average value throughout the period was of 1.4% (the dotted horizontal line).8 According to ECB officials’ declarations, during the period, the Bank monitored the growth rate of the monetary aggregate M3, which was taken as a good predictor of future price increases.9 Since the launch of the euro, this growth rate, the so-called ‘‘reference value’’, has been set to an annualized 4.5%. According to the Bank’s estimates, and backed by simple monetarist arithmetic, this increase in the money stock should be consistent with an inflation rate below 2%. In this context, one can reasonably assume that deviations from this value may influence the chosen interest rate, even though the central bank has always remained vague as to the meaning it gives to this indicator in the conduct of its policy. During the entire period under analysis, ECB officials argued that the Bank followed a strictly neutral stance with respect to the exchange rate. They also often announced that the best way to support the international value of the euro was by pursuing unflaggingly internal price stability. Thus, according to official position, the exchange rate is not a policy objective per se.10 8 When the ECB took charge of monetary policy on January 4, 1999, it disposed of various synthetic indicators pertaining to the EMU group of countries over the 1990s (industrial output, money stock, GDP, etc.) such as provided by Eurostat and the European Monetary Institute. These series are available on Thomson Datastream. 9 More precisely, the Bank monitored the evolution of the 3-month average growth rate of M3. See the ‘‘Annual review of the reference value for monetary growth’’ in the ECB Monthly Bulletin of December 2002. 10 See ‘‘The euro—a stable international currency’’, speech of Otmar Issing to the Hungarian Academy of Science, Budapest, February 27, 2003, www.ecb.int/key/03/sp0302227. A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 587 Fig. 3. Euro/dollar exchange rate, monthly average. However, the ECB agrees that the exchange rate represents one among many other indicators that should be included in inflation forecasts.11 The international value of the euro has a direct effect on internal prices, via import prices. In particular, euro depreciation would increase inflationary risks and vice versa. Given the low share of imports in total GDP for the euro-zone as a whole (below 15%), the direct impact can only be weak; however, depending on the pass-through effect on prices, the total impact may be more or less important. Fig. 3 describes the evolution of the euro/dollar exchange rate, defined as the price of one dollar in euros so that depreciation of the euro increases the exchange rate. The euro lost more than 20% from its launching value with respect to the dollar in the first two years. When in September 2000 one euro was traded for as less as 85 US cents, the ECB, the Fed and other main central banks coordinated their actions and bought euros. At that time, there was near general consensus that the euro/dollar exchange rate did not reflect economic fundamentals in the two regions. In the second part of 2002, the euro began appreciating against the dollar. This would have contained inflation, mainly during the period of high oil prices before the Iraq war (March 2003); this argument was used by the ECB as a justification for reducing interest rates. At the end of the Duisenberg period, the euro had regained its launching value, and many voices began criticizing an allegedly overly strong euro. Throughout the period, ECB officials opposed in public the idea of using interest rates to stabilize the exchange rate (i.e., increasing interest rates when the euro depreciates and vice versa). This declared ‘‘benign neglect’’ of the international value of the euro contrasts 11 See ‘‘Presentation of the ECB’s Annual Report 2002 to the European Parliament’’, introductory statement by Willem Duisenberg, www.ecb.int/key/03/sp030703, and ‘‘Monetary and fiscal policy in the euro area’’, speech by Willem Duisenberg, International Monetary Conference, Berlin, June 3, 2003, www.ecb.int/key/03/sp030303. See also IMF (2000). 588 A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 sharply with the public perception of the euro exchange rate against the dollar, which seemed to be considered as the main yardstick for measuring the success of the euro.12 3.1.2. Instruments The ECB uses the traditional policy instruments of minimum reserves, open market operations and standing facilities. Given the high demand for liquidity within the eurosystem, minimum reserves contribute to maintaining a shortage of liquidity. To obtain additional liquidity, banks must borrow resources from the central bank, either for a short or for a long period of time. Banks can also obtain (or place) overnight liquidity with the central bank at a predetermined interest rate (the so-called lending or deposit standing facility).13 The ECB may also intervene in the foreign exchange market. The ECB controls liquidity of the eurosystem mainly through short-term reverse-repo operations. Every week, the ECB opens a call for tenders for the euro-zone counterparts (banks and other financial institutions). It then lends cash to banks for a 2-week period (reduced to one week in 2004). Banks are asked to inform the ECB about the interest rate they are prepared to pay for every euro they borrow, knowing that those that offer to pay the higher price will be first served. Every month, the Governing Council of the ECB decides on the downward limit on interest rates, i.e., the minimum bid rate.14 According to information released by the ECB, this minimum bid rate aims at signaling the monetary policy stance to money market operators (ECB, 2001).15 However, given that the Bank controls both the minimum rate and the quantity of central money that it supplies, the information content of this indicator may be questioned. A large gap between the minimum rate and the marginal rate (at which the last euro is lent during the auction) would indicate that the quantity target prevails over the interest rate target. To date, marginal and even average interest rates on ECB short-term lending have been almost identical to this minimum bid rate. This supports the idea according to which the latter serve as a relevant indicator for analyzing monetary policy in the euro-zone. Fig. 4 is the graph of the ECB main monetary policy instrument, the bid rate on twoweek lending, since the launch of the euro. The Federal Funds target rate is also shown for comparison. The sharp (half point) reduction implemented by the ECB immediately after it began operations, on April 8, 1999, ‘‘. . .came at a time of strong disagreement about the direction of the ECB’s next move, and its size surprised almost everyone’’ (Corsetti and Pesenti, 1999, p. 306). Later, the ECB (2001, p. 80) agreed that ‘‘the assessment of monetary developments at the start of 1999 was further complicated by uncertainty over how monetary aggregates were affected during the changeover to Stage Three of EMU’’. 12 See ‘‘The single currency and European integration’’, speech delivered by Sirkka Hämäläinen, seminar on ‘‘EMU experience and prospects—a small state perspective’’, Institute for European Affairs, Dublin, October 16, 2000, www.ecb.int/key/00/sp001016. 13 On this topic, see also Ayuso and Repullo (2003) and ECB (2001). 14 Until June 2001, the ECB used a fixed rate auction, where banks borrowed reserve money at a constant pre-announced interest rate. This system was highly unstable (call for bids became as high as 100 times the allotment!) and had to be abandoned. 15 The signaling role of the minimum bid rate was emphasized by Willem Duisenberg at the ECB Press Conference on June 8, 2000, cf. www.ecb.int/key/00/sp000608.htm. A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 589 Fig. 4. ECB and Fed main interest rate instruments (annualized percentage rates). Since this move has all the properties of an internal adjustment, the subsequent empirical analysis will not take the first quarter into account.16 3.2. The econometric model We now seek to determine whether the ECB follows a simple interest rate rule. Several models are tested using monthly data for the period 1999.04 to 2003.10.17 A look at the main times series suggests that within this time period the euro-zone went through a complete economic cycle. The dependent variable is ECBRA, the annualized ECB minimum bid rate over shortterm reverse-repo operations, in percent, such as recorded at the end of the month. The main exogenous variables are:  INFH: the inflation rate, based on the euro-zone harmonized consumer price index, calculated as the yearly percentage change of the price index from one month to the same month of the previous year;  IPDEV: the deviation of the industrial output growth rate from the over-the-period average of 1.4%; the growth rate is calculated as the yearly percentage change of the industrial output index from one month to the same month of the previous year;  FXDEV: the percent deviation of the nominal exchange rate measured in euros per dollar from the average value, equal to 1.02 euros/dollar over the period. 16 Including these observations slightly deteriorates the fit of our models, but do not alter the main conclusions. 17 In general, data come from Thomson Datastream. The monthly average exchange rate comes from the Pacific database developed by Werner Antweiler. The data set can be downloaded from the web address www.essec.fr/research/papers/vranceanu/data.pdf. 590 A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 Given the above labeling, Eq. (8) can be written in a form suitable for econometric estimation, ECBRAt ¼ c1 þ c2 ECBRAt1 þ c3 INFHtþk þ c4 IPDEVtþq þ c5 htþS þ et ð9Þ where the generic h represents any relevant variable other than inflation and the industrial output growth gap. ¯ The regression coefficients relate to theoretical coefficients in Eq. (8): c1=(1q)(ībp), c2=q, c3=(1q)b, c4=(1q)c, c5=(1q)l. et is an i.i.d composite error term that captures shocks and expectation errors. Table 1 shows the coefficients of the most interesting equations. Variables are denoted by NAME(n), where n is the number of leads (or lags, for a negative number). Standard Taylor rules were estimated where the interest rate depends only on inflation and the output gap, and also augmented Taylor rules where other variables are included. As a proxy for economic activity, either the unemployment rate or the industrial output growth gap was used. Only the latter variable is statistically significant: best results were obtained with a 1-month lag (but the current variable also performs satisfactorily). In the augmented interest rate rules, the 3-month average growth rate of M3 was not significant (current and up to six lag variations in M3 growth were tested). The exchange rate deviation FXDEV was significant in all equations. Inflation was included at date t (in the so-called contemporaneous Taylor rule) and six periods ahead (in a forward-looking specification): in practice, central bankers consider that at least 6 months must pass before the effects of a change in monetary Table 1 Main equations, OLS and GMM estimates Eq. Nb. c1 ECBRA(1) INFH INFH(+6) IPDEV(1) FXDEV No. of observations Adj.-R2 F-stat LM(2) J-stat Standard contemporaneous Taylor rule Standard forward-looking Taylor rule Augmented contemporaneous Taylor rule Augmented forward-looking Taylor rule OLS GMM OLS GMM OLS GMM OLS GMM I II III IV V VI VII VIII ns 0.07 0.91*** 0.10ns – 0.05*** – 55 0.18*** 0.90*** 0.08ns – 0.03*** – 49 0.04 0.90*** – 0.17* 0.03*** – 55 0.38** 0.85*** – 0.42*** 0.03*** – 49 0.67*** 0.73*** 0.09ns – 0.06*** 0.02*** 55 0.73*** 0.70*** 0.12** – 0.07*** 0.02*** 49 0.61** 0.76*** – 0.07ns 0.05*** 0.02*** 55 0.14ns 0.78*** – 0.27** 0.05*** 0.01** 49 0.956 394 0.64 NA 0.951 NA NA 0.15 0.947 289 0.61 NA 0.939 NA NA 0.07 0.966 387 4.69 NA 0.964 NA NA 0.08 0.957 270 3.46 NA 0.952 NA NA 0.07 2 (2)=5.9. NA: not available; v0.95 ns Not significant. * Significant at 10%. ** Significant at 5%. *** Significant at 1%. ns A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 591 policy can be noticed, and more time (9– 12 months) may be required. Given the relatively short time period, we opt for this low number of leads.18 Backward looking rules estimated with 1 month lagged inflation are quite similar to those based on current inflation.19 All equations were estimated by OLS and also by the Generalized Method of Moments (GMM). As emphasized by Clarida et al. (2000), the latter method is well suited to econometric analysis of interest rate rules when the regressions are built on variables that are not known by the policymaker at the decision time, which is the typical case for forward-looking Taylor rules where the policymaker makes decisions based on expected variables. Use of the Generalized Method of Moments requires specifying as instruments variables belonging to the information set at time t and not correlated with the error term et. The instruments used here are a two-period lag of the dependent and independent variables, the lagged money growth rate and the lagged stock market index (both should convey information about future prices).20 Overall, the fit of the various regressions is good and the residual auto-correlation assumption can be rejected in keeping with the standard Ljung – Box and Breusch – Godfrey LM tests in all OLS estimates (except in V, where autocorrelation cannot be dismissed at the 5% level). The stability of the coefficients is quite high, as shown by the Cusum and Cusum of Squares tests, which do not indicate movement outside the 5% critical band. The results are easier to interpret when the policy coefficients are calculated from these equations to describe changes in the target interest rate (not the effective one). Table 2 displays the coefficients q, b, c and l derived from the GMM estimates (i.e., Eqs. II, IV, VI and VIII). Here, l stands for the sensitivity of the target interest rate with respect to the exchange rate. The q coefficient is always positive and large, implying that the ECB smoothes its interventions in the money market (cf. Eq. (7)). Compared to usual US estimates, in this model the adjustment in the interest rate takes place more rapidly, with 10% to 30% of the desired change occurring within the month (in general, US estimates show that the Fed implements 20% of the desired change within the quarter). In another study with monthly data building a ‘‘virtual EMU’’ rule during the 1990s, Gerdesmeier and Roffia (2003) found a coefficient q=0.87; Sauer and Sturm (2003), also with monthly data, found a coefficient q=0.85 for the ECB. A positive c value indicates that the ECB follows a standard ‘‘leaning against the wind’’ principle, tightening monetary policy when growth takes off and vice versa. If the 18 For their baseline estimate, Clarida et al. (2000) used a one-quarter lead; they also provide an estimate with a four-quarter lead. Sauer and Sturm (2003) test for an ECB policy rule with a 3-month lead, while Ullrich (2003) takes a 12-month lead. 19 This is unsurprising since inflation measured by the variation in the CPI from a month over the same month of the previous year contains the last 11 months of the interval which allows construction of the lagged variable. 20 The GMM estimation was carried out with the EViews econometric software, under the option ‘‘heteroskedasticity and autocorrelation consistent covariance matrix’’. The exact list of instruments contains ECBRA(2), IPDEV(2), INFH(2) M3GR(2), STOCK(1) and STOCK(2). 592 A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 Table 2 Basic coefficients of the policy rule q Standard contemporaneous Taylor rule (Eq. II) Standard forward-looking Taylor rule (Eq. IV) Augmented contemporaneous Taylor rule (Eq. VI) Augmented forward-looking Taylor rule (Eq. VIII) 0.90 0.84 0.70 0.78 b ns 0.84 2.8 0.43 1.21 c l 0.32 0.19 0.26 0.21 – – 0.08 0.05 industrial production growth rate were one percentage point below the trend value (1.4%), the ECB would reduce the short term interest rate by about 1/5 to 1/3 of a percentage point, and conversely. The estimate of c should be interpreted with caution, as the proxy, industrial production, accounts for only a third of total GDP. As a comparison, Fair and Howrey (1996) found, from US data in the period 1962 –1993, that the Fed’s sensitivity to the output growth rate was 0.22. As mentioned, a b coefficient greater than 1 is indicative of a stabilizing policy (i.e., a policy that effectively counters inflation), whereas a coefficient less than 1 implies a destabilizing policy. Previous studies focusing on ECB behavior after January 1999 did not reach consensus on this matter. Ullrich (2003) found a coefficient less than 1 and not statistically significant; Sauer and Sturm (2003), from an OLS estimate, found a significant coefficient less than 1 (b=0.51) if the industrial output gap is considered as a proxy for economic activity, or close to 1 (b=0.93) when the industrial output growth gap is included. These results contrast with our early study (Fourcß ans and Vranceanu, 2002) where b emerges as greater than 1 (b=1.16) from the estimate of a contemporaneous Taylor rule; that model, however, covered only a relatively short time period (from April 1999 to March 2002). From Tables 1 and 2, it can be seen that, in the standard contemporaneous Taylor rules, the inflation coefficient is indeed not statistically significant (but close to one). However, in the standard forward-looking specification (with inflation six periods ahead), the coefficient is significant and quite large (b=2.8). In that respect, the augmented contemporaneous rule deserves further scrutiny. This rule (from Eq. VI) is quite robust: All variables are strongly significant and the overall fit is good. In this estimate, b=0.43, which is a value close to the Sauer and Sturm (2003) estimate. Yet, this coefficient might tell only part of the story. If the central banker’s behavior is forward-looking (as would be suggested by Eq. IV) and the current (or the one period lagged) inflation rate and the exchange rate are used to forecast future inflation, both variables should be significant, in which case the impact of current inflation would be weaker. Eq. VI supports this view. In the augmented contemporaneous Taylor rule (Eq. VI), the exchange rate deviation from the over-the-period average (1.02 euro per dollar) is statistically significant: with an exchange rate 5% above this implicit target, the ECB would increase the interest rate by 0.4 percentage points. However, if depreciation of the euro leads to higher prices later, a policy of increasing the interest rate so as to reverse the currency depreciation is consistent with the goal of price stability. The same rationale applies to the contrary situation where a strong euro eases future inflationary pressures. A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 593 Fig. 5. Actual and target ECB interest rates. Fig. 5 displays the actual target rate and the fitted target rate according to the standard forward looking rule (Eq. IV) and the augmented contemporaneous rule (Eq. VI): Fitted variables do not include the smoothing effect. Eq. VIII includes both the current exchange rate and the 6-month lead inflation rate, both variables being significant. This would suggest that the central bank’s forecasting horizon might be longer than 6 months, with some of the information on future prices still being captured by the exchange rate. We add a last word about the 3-month average growth rate of M3, which was never significant despite the announced target of 4.5% (current and up to six month lagged variables were tested). Although this empirical framework is oriented towards the short run, and may not be able to uncover the influence of the money reference value, which is a medium-term concept, this finding is consistent with the ECB decision on May 2003 to underscore the role assigned to money growth; this is one indicator among many able to convey useful information on future inflation trends.21 4. Conclusions and policy implications The interest rate rule introduced by Taylor (1993) is a convenient way of looking at the behavior of central bankers. Once a monetary policy rule is inferred from the data, the rule can be incorporated into a macroeconomic model: the theoretician can then study the stability of the system, while the practitioner may use it as a forecasting device. 21 See ‘‘The ECB’s monetary policy strategy’’, May 8, 2003, www.ecb.int/press/03/pr030508_2en.htm. 594 A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 In this study, we have applied such a methodology to analyze ECB behavior between the time of the launch of the euro until October 2003, which was the challenging initial period during which Willem F. Duisenberg was president of the Bank. The empirical analysis indicates that, as does the US Fed, the ECB reacts to economic activity: when growth of industrial output deviates by one percentage point below (above) its long run value, the ECB is indicated to react by reducing (increasing) its interest rate by 1/5 to 1/3 of a percentage point. Also, as does the US Fed, the ECB reacts to variations in the inflation rate: the reaction is stronger if future rather than current inflation is considered. In a forward looking rule including only inflation and economic activity, it appears that, if future inflation deviates above (below) target by one percentage point, the ECB increases (decreases) its rate by almost three percentage points. In the estimated contemporaneous rules, the inflation coefficient falls below unity, but estimates indicate an ECB response to exchange rate deviations from its average (about parity with respect to the US dollar). This additional channel of response may have a stabilizing impact on prices, especially if exchange rate shocks propagate gradually into prices via the pass-through effect. In this contemporaneous augmented Taylor rule, the ECB would increase (decrease) its interest rate by almost a half percentage point if the euro were to depreciate (appreciate) by 5% from parity. Such a result is not consistent with the official position of the Bank, which systematically played down the role of the exchange rate in monetary policy management. These results point to an important caveat concerning empirical interest rate rules: they all share significant uncertainty about parameter estimates that are sensitive to the specification adopted. It should also be kept in mind that the analysis presented here has used ex post data and does not take into account the information available to the ECB in real time. A final remark: if Taylor rules are a convenient way to describe central banks’ behavior, it should be emphasized that this does not necessarily imply that the rules are optimal. Acknowledgements The authors are grateful to Vincent Dropsey, Jean-Pierre Indjehagopian and Allan Meltzer, as well as the participants to the XIIth International Conference of the ITFA in Bangkok, May 2002 for their helpful comments on an early version of this paper. They thank two anonymous referees for suggestions and remarks that significantly improved the quality of the paper. References Arnold, I.J.M., De Vries, C.G., 2000. Endogeneity in European money demand. European Journal of Political Economy 16, 587 – 609. Ayuso, J., Repullo, R., 2003. A model of the open market operations of the European Central Bank. Economic Journal 113, 883 – 902. Batini, N., Haldane, A., 1999. Forward looking rules for monetary policy. In: Taylor, J.B. (Ed.), Monetary Policy Rules, pp. 157 – 192. University of Chicago Press, Chicago. A. Fourcßans, R. Vranceanu / European Journal of Political Economy 20 (2004) 579–595 595 Berger, H., Woitek, U., 2001. The German political business cycle: money demand rather than monetary policy. European Journal of Political Economy 17, 609 – 631. Bernanke, B., Woodford, M., 1997. Inflation forecasts and monetary policy. Journal of Money, Credit and Banking 24, 653 – 684. Clarida, R., Gali, J., Gertler, M., 1998. Monetary policy rules in practice: some international evidence. European Economic Review 42, 1033 – 1067. Clarida, R., Gali, J., Gertler, M., 2000. Monetary policy rules and monetary stability: evidence and some theory. Quarterly Journal of Economics 115, 147 – 180. Corsetti, G., Pesenti, P., 1999. Stability, asymmetry, and discontinuity: the launch of European Monetary Union. Brookings Papers on Economic Activity 2, 295 – 372. Doménech, R., Ledo, M., Taguas, D., 2002. Some new results on interest rate rules in EMU and in the US. Journal of Economics and Business 54, 431 – 446. ECB, 2001. The Monetary Policy of the ECB. European Central Bank, Frankfurt am Main. Fair, R.C., Howrey, P.E., 1996. Evaluating monetary policy rules. Journal of Monetary Economics 38, 173 – 193. Faust, J., Rogers, J., Wright, J., 2001. An empirical comparison of Bundesbank and ECB monetary policy rules. International Finance Discussion Papers 705. Board of Governors of the Federal Reserve System, Washington, DC. Fourcß ans, A., Vranceanu, R., 2002. ECB monetary policy rule: some theory and empirical evidence. ESSEC Working Paper 02008. www.essec.fr. Gerdesmeier, D., Roffia, B., 2003. Empirical estimates of reaction functions for the euro area. ECB Working Paper 206. Frankfurt am Main. Gerlach, S., Schnabel, G., 1999. The Taylor rule and interest rates in the EMU area: a note. BIS Working Paper 73. Bank for International Settlements, Basel, Switzerland. Gertler, M., 1999. Comment on ‘‘performance of operational policy rules’’. In: Taylor, J.B. (Ed.), Monetary Policy Rules, pp. 46 – 54. University of Chicago Press, Chicago. IMF, 2000. Monetary and exchange rate policies of the euro area. IMF Staff Country Report 00/46, Washington D.C. Judd, J.F., Rudebush, G.D., 1998. Taylor’s rules and the Fed. Federal Reserve Bank of San Francisco Economic Review 3, 3 – 16. Kerr, W., King, R., 1996. Limits on interest rate rules in the IS model. Federal Reserve Bank of Richmond Economic Quarterly 82, 47 – 76. McCallum, B., Nelson, E., 1999. Performance of operational policy rules in an estimated semi-classical structural model. In: Taylor, J.B. (Ed.), Monetary Policy Rules, pp. 15 – 45. University of Chicago Press, Chicago. Orphanides, A., 2001. Monetary policy rules based on real-time data. American Economic Review 91 (4), 964 – 985. Orphanides, A., 2003. Historical monetary policy analysis and the Taylor rule. Journal of Monetary Economics 50, 983 – 1022. Rudebusch, G.D., 2002. Term structure evidence on interest rate smoothing and monetary policy inertia. Journal of Monetary Economics 49, 1161 – 1187. Rudebush, G.D., Svensson, L.E.O., 1999. Policy rules for inflation targeting. In: Taylor, J.B. (Ed.), Monetary Policy Rules, pp. 203 – 246. University of Chicago Press, Chicago. Sauer, S., Sturm, J.-E., 2003. Using Taylor rules to understand ECB monetary policy. Working Paper 1110. CESifo, Munich. Surico, P., 2003. How does the ECB target inflation? ECB Working Paper 229. Frankfurt am Main. Taylor, J.B., 1993. Discretion versus policy rules in practice. Carnegie-Rochester Conference Series on Public Policy 39, 195 – 214. Taylor, J.B., 1999. The robustness and efficiency of monetary policy rules as guidelines for interest rate setting by the European Central Bank. Journal of Monetary Economics 43, 655 – 679. Taylor, J.B., 1999. A historical analysis of monetary policy rules. In: Taylor, J.B. (Ed.), Monetary Policy Rules, pp. 319 – 341. University of Chicago Press, Chicago. Ullrich, K., 2003. A comparison between the Fed and the ECB. ZEW Discussion Paper 03-19, Zentrum für Europäische Wirtschaftsforschung, Manheim. Walsh, C.E., 2003. Speed limit policies: the output gap and optimal monetary policy. American Economic Review 93, 265 – 278.