Measuring Monetary Policy Shocks in a SmallOpen Economy.∗

Giuseppe De Arcangelis and Giorgio Di Giorgio.†

September 2000

Abstract

This paper presents different specifications of a structural VAR modelwhich are useful to identify monetary policy shocks and their macroeconomiceffects for the Italian economy in the 90s. The analysis is based on a detailedinstitutional description of the functioning of the domestic market for bank re-serves. In this setting, we try to establish if monetary policy shocks are betteridentified using exchange rates or foreign exchange reserves as a conditioningvariable for the small open economy framework. Our analysis confirms theview that the Bank of Italy has been targeting the rate on overnight inter-bank loans in the 90s. This is coherent with either proposed modeling choices.Therefore, we interpret shocks to the overnight rate as purely exogenous mon-etary policy shocks and study how they impact the economy.

1 Introduction

Many recent studies have been devoted to analyze the effects of monetary pol-icy shocks on output and inflation. Most of the studies focusing on the short-runresponse of the economy to monetary policy have been conducted by using themethodology of Structural Vector Autoregression (SVAR).1 This approach has theadvantage of imposing a minimal set of theoretical restrictions on the model to betested, thereby allowing for a close-to-pure statistical investigation of the time seriesproperties of the variables included in the analysis. In this literature, a monetarypolicy shock is identified with the residual of an equation regressing a monetary

∗We thank two anonymous referees for useful comments and suggestions on an earlier versionof this paper; Alberto Baffigi and Giorgio Valente provided precious help with the structuralstability analysis. Giorgio Di Giorgio gratefully acknowledges financial support from CNR andthe hospitality and excellent research environment provided by the Department of Economics andBusiness of the Universitat Pompeu Fabra in Barcelona.

†Giuseppe De Arcangelis: Dipartimento di Scienze Economiche, Universita degli Studi di Bari,email: g.dearcangelis@caspur.it. Giorgio Di Giorgio: Universita LUISS Guido Carli, email:gdg@luiss.it.

1See Bernanke and Blinder (1992), Christiano and Eichenbaum (1992), Sims (1992), Strongin(1995), Leeper, Sims and Zha (1996), Clarida and Gertler (1997), Bernanke and Mihov (1997,1998),Christiano, Eichenbaum and Evans (1998) and Bagliano and Favero (1998). Canova (1995) andWatson (1994) provide a general methodological discussion.

1

policy instrument on a set of variables that are considered relevant for the decisionsof the central bank. Hence, monetary policy shocks are defined as statistical inno-vations and represent a purely exogenous component of policy. Impulse responsesof different macroeconomic variables to these shocks are not subject to the rational-expectation critique2 and can then be interpreted as the empirical dynamics beyondthe comparative statics exercises developed in standard equilibrium analysis. Adrawback of this analysis is that the impulse responses do not consider the “endoge-nous” component of monetary policy, such as any feedback rule linking policy tothe state of the economy.3 Indeed, this endogenous component might well be evenmore important than the exogenous one. However, it is hard to identify and singleout the original source of the many shocks hitting the economy and to which policyresponds. Hence, the consequences of these endogenous changes in monetary policyare more difficult to interpret as they combine the effect of the original shocks andthat of the policy reaction.

Most of these studies have adopted a closed economy framework. Identifyingmonetary policy shocks in an open economy context presents some additional dif-ficulties. These are usually due to the simultaneous reaction between interest andexchange rates innovations, which in turn can be responsible for the emergence ofnew empirical puzzles, as the one of an impact depreciation of the exchange ratefollowing a monetary policy contraction in the domestic country. Bagliano, Faveroand Franco (1999) provides a complete discussion of the topic by estimating a two-country model including Germany and the US.

In this paper we present different model specifications that can be used to identifymonetary policy operating regimes and monetary policy shocks in a small openeconomy. More precisely, we focus on the Italian economy in the 90s and try toestablish if monetary policy shocks are better identified using exchange rates orforeign exchange reserves as a conditioning variable for the small open economyframework.4 In our sample, we have two periods of quasi-fixed nominal exchangerates (1989.06-1992.09; 1996.11-1998.04) and one of free floating (1992.10 - 1996.10).Given the limited span of the subperiods and the monthly frequency of the data, wetreat the whole sample as one of managed floating of the Lira and propose differentmodel specifications to check whether the identification of the central bank operatingregime and of the monetary policy shocks is robust enough.

Our methodology is based on De Arcangelis and Di Giorgio (1998), which in turnextended to a small open economy the research strategy introduced by Strongin(1995) and further developed by Bernanke and Mihov (1998) for the US. Moreprecisely, we give a structural content to the VAR models by linking econometricanalysis with the institutional knowledge of how the market for banks reserves (i.e.,the market in which monetary policy is actually conducted) works in Italy. Inour estimated models, indeed, identification hinges on a detailed description of theoperating procedures used by the Bank of Italy. The advantage of this procedure is

2See Keating (1990) for a discussion on SVAR and rational-expectations econometric models.3See Clarida, Gali and Gertler (1999a) for a detailed review of such issues.4In our econometric exercise we also include a foreign (German) short-term interest rate. Pre-

vious applications of VAR analysis to Italy are Bagliano and Favero (1995), Buttiglione and Ferri(1994), De Arcangelis and Di Giorgio (1998, 1999), and Gaiotti, Gavosto and Grande (1997),Gaiotti (1999).

2

that it allows for a direct test of different model alternatives that are nested in thesame specification, without imposing a priori one identification mechanism. Thecorrect measure of a monetary policy shock is then selected by the data itself.

The paper is organized as follows. In section 2, we briefly motivate our interestfor the Italian economy, while in section 3 we discuss how it is possible to combinevector autoregression techniques with institutional analysis. In section 4 we firstestimate in a larger sample the model proposed in De Arcangelis and Di Giorgio(1998); we also extend it to include a foreign (German) short-term interest rate. Insection 5 we study a modified version of the model that explicitly focuses on therole of foreign exchange reserves. We summarize our conclusions in section 6.

2 Motivation

We believe that Italy provides an interesting case study for different reasons.First, important institutional changes have affected, in the last decade, both

the degree of economic and political independence5 of the Bank of Italy, as well asthe environment in which monetary policy is decided and implemented. In termsof central bank independence, between 1992 and 1994 a series of laws were passedgiving the central bank Governor the exclusive responsibility to set and change themonetary policy instruments (the discount rate and the reserve coefficient, as wellas the growth rate of the monetary base). In particular, any source of direct andpermanent financing of the Treasury deficit via high-power money was prohibited.6

As regards the modus operandi of monetary policy, this had already been substan-tially modified by the introduction, at the end of the 80s, of a series of importantinstitutional changes in the money market. A screen-based market for TreasuryBills (MTS) was opened in 1988, the one for interbank deposits (MID) in 1990.The mandatory reserve regime was also gradually reformed (starting in 1990) so asto allow banks to average provisions in the maintenance period. Other importantchanges have been the abolition of the floor price on T-Bills auctions (1988) and anew discipline in terms of fixed-term advances.7

Second, in the 90s, Italy achieved a stable reduction in the inflation rate. Thiswas accompanied by a lower-than-average rate of economic growth with respect toprevious decades. Economic growth was also considerably lower, in this period, thanin other European countries that were facing similar (although smaller in magni-tude) problems in terms of fiscal adjustment required to meet the Maastricht budgetcriteria. It is important to assess the role that monetary policy played in generatingthese macroeconomic outcomes.

Third, Italy can be viewed as a good example of a “small” open economy. Mostof the studies on the US and Germany are conducted either in a closed economyframework (e.g., Bernanke and Mihov, 1997 and 1998) or with a two-country model

5Political independence is defined as the ability of the central bank to establish its policy targetswithout government interference. Economic independence is defined as the ability of the centralbank to autonomously activate its instruments in order to reach the monetary policy goals. SeeGrilli, Masciandaro and Tabellini (1991).

6See Passacantando (1996). Price stability was explicitly stated as the primary target of mon-etary policy only in 1998.

7See Gaiotti (1992) and Sarcinelli (1995).

3

(Bagliano, Favero and Franco, 1999). In the Italian case, neither framework isappropriate. Moreover, we will be interested in evaluating the role of modeling theexchange rate regime. In the period under study, this variable can be looked at underdifferent perspectives: as a monetary policy intermediate target, as a transmissionchannel of the monetary impulses and as an information variable monitored by thecentral bank. The different roles performed by this variable might have affectedthe operating procedures that the central bank used in order to conduct monetarypolicy.

3 Methodology

From an empirical point of view, it is essential to be able to isolate purely exogenouspolicy shocks. In the case of monetary policy, if we just focused on monetary policy“actions”, including feedback responses of policy to many other possible shockshitting the economy, we would never come up with a measure of the macroeconomiceffects originated by a purely exogenous change in monetary policy. We would onlydescribe the mixed effects originated by different heterogeneous shocks and by howpolicy makers react to these shocks. Even though this reaction might in practiceaccount for most of ordinary monetary policy interventions, we would still like toinvestigate the effects of a monetary policy shock isolated both by other kinds ofshocks in the economy and by the endogenous changes that these other shocks mightpush. Moreover, the responses to unforecastable (and structural) innovations are lesssubject to the Lucas’ critique.

Structural VARs have been useful in pursuing this strategy. In a structuralVAR, after the estimation of the unrestricted vector autoregression (i.e., with nocontemporaneous interactions among the variables), the econometric identificationof economically meaningful (i.e., structural) innovations occurs in a second stagewhere reasonable constraints must be introduced.8 These constraints are typicallydesigned as restrictions on the contemporaneous influence among fundamental (i.e.,non-structural) and structural innovations, where the latter are assumed to be mu-tually and serially uncorrelated. Basically, the analyst has to make a number ofidentifying assumptions in order to be able to estimate the reaction function of thecentral bank, including assumptions on what variables are monitored and on whatkind of interaction the exogenous policy shock has with variables in the reactionfunction of the monetary authority (the so called endogenous component” of pol-icy). The main identifying assumption is that policy shocks have to be orthogonalto variables in the reaction function of the central bank.9 Hence, within the systemof equations in the VAR, policy shocks can be estimated as the residuals in thelinear regression of the central bank instrument on the variables in the central bankreaction function. According to this assumption, the monetary policy instrumentchanges following a contemporaneous innovation to the variables in the informa-tion set of the central bank; while these latter variables are constrained to have

8See Amisano and Giannini (1997, chap. 1). Technically, the estimation of a structural VARwith Choleski decomposition can be developed equation by equation with ordinary least squares,by using the properties of a Wold causal chain.

9Christiano, Eichenbaum and Evans (1998) call this the Recursiveness Assumption.

4

no contemporaneous reaction to a change in the policy instrument. Obviously, thisassumption can be sensibly maintained when the observation period is one month.It is less acceptable when the VAR deals with yearly or even quarterly data.10

Structural VARs where one variable is assumed to be the monetary policy in-strument have been estimated for the US economy by Bernanke and Blinder (1992),who used the Fed Funds rate, or by Christiano and Eichenbaum (1992), who selectedthe nonborrowed reserves aggregate. Bernanke and Mihov (1997, 1998) have gener-alized this approach by considering a vector of policy variables, instead of just onepolicy variable. In the first stage, the estimation of an unrestricted VAR generatestwo subvectors of innovations, one related to nonpolicy variables (uy,t) and one topolicy variables (up,t)

11

R(L)

[yt

pt

]=

[uy,t

up,t

]where R(L) is a matrix of polynomials in the lag operator L and R(0) = I; yt isthe vector of nonpolicy variables and pt is the vector of policy variables.

In the estimation of the orthogonalized, economically meaningful (structural)innovations in the second stage, a recursive causal block-order is assumed from theset of nonpolicy variables to the set of policy variables. Moreover, the recursivecausal order is also established for the nonpolicy variables in yt. In terms of therelationship between the fundamental innovations, uy,t and up,t, and the structuralinnovations, νy,t and νp,t, which are mutually and serially uncorrelated, this implies:[

A1,1 0A2,1 A2,2

] [uy,t

up,t

]=

[B1,1 00 B2,2

] [νy,t

νp,t

]where A1,1 is lower-triangular and B1,1 is diagonal so that there is a Wold recur-

sive (causal) ordering among the nonpolicy variables in yt. Moreover, A2,1 is a fullmatrix so that there is a Wold block-recursive (causal) ordering between nonpolicyand policy variables.

Building on previous work by Strongin (1995), the vector of policy variablescontains aggregates and interest rates characterizing the market for bank reserves.As a matter of fact, monetary policy is effectively conducted through the market forbank reserves. The idea is then that, in order to correctly identify a monetary policyshock, it could be useful to model the different operating procedures of the centralbank according to appropriate constraints in the relationship between up,t and νp,t.Hence, the core of the analysis focuses on the shape that the matrices A2,2 and B2,2

must take for the different operating procedures to work properly. This requireslinear and nonlinear constraints on the elements of those two matrices. A test foroveridentifying restrictions can finally be applied to check whether the constraintsimplied by the different regimes are rejected by the data. Impulse response functionsof policy and nonpolicy variables to monetary shocks are used to further checkwhether the identified monetary-policy innovations can be plausibly qualified so.12

10An alternative identifying assumption is to exclude contemporaneous reaction of the policyinstrument to variables in the central bank information set while allowing the latter to contempo-raneously respond to changes in the policy instrument. See Leeper, Sims and Zha (1996).

11Bold lower-case (capital) letters indicate vectors (matrices).12In this empirical approach to monetary policy, nonstationarity of the data is not generally

5

4 A model of the Italian Economy

In this section we summarize and re-estimate in a larger sample the model in De Ar-cangelis and Di Giorgio (1998). This model uses the empirical framework describedin section 3 and extends it to a small open economy framework in order to studythe conduct of monetary policy in Italy in the 90s.13

We define different monetary regimes according to the necessary constraints im-plied on the central bank operating procedures in the domestic market for bankreserves. In the VAR model we include two nonpolicy variables: Italian consumerprices and an index of the Italian industrial production. Moreover, in some spec-ifications, the German call money rate is also included. These nonpolicy variablesaffect the estimation as they enter in the reaction function of the central bank.

The small open economy feature is modeled by including the exchange rate vis-a-vis Germany in the analysis (besides the German call money rate when includedin the nonpolicy block). As we have both periods of quasi-fixed exchange rates andof free floating in the sample, we chose to consider the real exchange rate in ourVAR model. On one side, this has the advantage of guaranteeing some variabilityeven with targeted nominal exchange rates. On the other side, it shows a quasiidentical short-run pattern given the existence of nominal rigidities and it is surelya better variable to look at in terms of the monetary transmission mechanism.14 Inthe model, the exchange rate is listed last, after the block of policy variables. Indeed,although the exchange rate is clearly a nonpolicy variable, we cannot exclude thecontemporaneous reaction of the exchange rate to innovations in the policy variables,in particular to innovations in the short-term rate.

Hence, the relationship among fundamental and structural innovations can besummarized as follows: A1,1 0 0

A2,1 A2,2 0a3,1 a3,2 1

uy,t

up,t

ur,t

=

B1,1 0 00 B2,2 b2,3

0 0 b3,3

νy,t

νp,t

νr,t

emphasized and cointegration analysis not undertaken. A first justification is that the data maybe quasi-nonstationary ; in fact, the presence of unit roots in the time series cannot be tested withhigh power (see, for instance, Campbell and Perron, 1991). Moreover, even though unit rootsmay characterize the data, Sims, Stock and Watson (1990) show that most traditional, standardasymptotic tests are still valid if the VAR is estimated in levels.

The neglecting of cointegration constraints is motivated by the following considerations. First,the analysis is generally focused on short-run constraints and the short-run dynamic responseof the system. When cointegration constraints are excluded, this only implies that the long-runresponses of some variables are not constrained and might follow a divergent path. However, theshort-run analysis is still valid. Second, Sims, Stock and Watson (1990) proved that standardasymptotic inference is not affected even when the variables included in the VAR in levels arecointegrated. Finally, although FIML estimates are no longer efficient if cointegration constraintsare not included, they still remain consistent. Hence, the lower efficiency in the estimates canbe justified by the objective difficulty in the economic interpretation of some of the cointegrationconstraints showed by the data.

13See Bernanke and Mihov (1997) and Clarida and Gertler (1997) for similar studies on monetarypolicy in Germany.

14Kim and Roubini (1995) and Clarida and Gertler (1997) also used the real exchange ratealthough they did not have any problem of quasi-fixed exchange rate regime in their studies andcould have chosen the nominal one.

6

where ur,t and νr,t are respectively the fundamental and structural innovation relatedto the real exchange rate; a3,1 and a3,2 are full (row) vectors and the (column) vectorb2,3 represents the possible correlations between the structural innovations in themarket for bank reserves (including a possible monetary policy-induced variable)and the structural innovations in the exchange rate.

The core of our identification strategy is in the lower-right corner of the system,which models the market for bank reserves and explicitly considers the role of theexchange rate: [

A2,2 0a3,2 1

] [up,t

ur,t

]=

[B2,2 b2,3

0 b3,3

] [νp,t

νr,t

](1)

4.1 The Italian Market for Bank Reserves

A description of how the Italian market for bank reserves works and of the operatingprocedures of the Bank of Italy is given in Buttiglione, Del Giovane and Gaiotti(1997).15 Here we simply explain our model equations. In terms of innovations, wespecify the demand for total reserves as:16

uTR = −αuOV + σdνd (2)

where uTR is the innovation in total reserves, uOV is the innovation in the overnightinterest rate and νd is the unit-variance, orthogonal innovation in the demand fortotal reserves (i.e., an indicator of the shifting in the demand for total reserves);σd is a measure of the standard deviation of the structural shock assigned to thisequation. We use the rate on overnight loans since this has recently become themost important interest rate in the market for bank reserves.

We then divide the total amount of bank reserves in the sum of two aggregatesthat we define, in line with the US literature, as borrowed and nonborrowed reserves.We define fixed-term advances as borrowed reserves.17 The nonborrowed reservesaggregate thus includes the item Anticipazioni Ordinarie18 and all open-marketoperations. Our sorting can be motivated by the fact that Anticipazioni Ordinarie:a) were of limited amount, established by the central bank; b) should have rationallybeen used first as the least-costly source of finance;19 and c) could in principle becanceled by the Bank of Italy with short notice.20

15See also De Arcangelis and Di Giorgio (1998).16We omit the time subscript t to ease notation.17Fixed-term advances (Anticipazioni a Scadenza Fissa) is an explicit standing facility that could

be automatically drawn by banks and on which a penalty rate is applied. This “ceiling” rate inthe money market was established by the central bank analogously to the discount rate. The twoofficial interest rates defined a corridor that normally contained the every-day fluctuations of allother money-market rates.

18A credit line that banks could activate with the Bank of Italy, providing a limited amount oflow-cost finance. The discount rate (tasso ufficiale di sconto), usually a floor for money marketrates, was the cost paid by banks on the amount of credit effectively drawn from this line.

19In reality, though, this credit line is never completely used since many cash managers keep aportion of the line as a buffer stock. However, the unused credit is quantitatively negligible.

20A similar choice was made in Bernanke and Mihov (1997) for Germany.

7

In terms of innovations, the demand for fixed-term advances can be expressedas a positive function of the spread between the overnight rate and the rate onfixed-term advances:

uFTA = −β(uiFTA− uOV ) + σbνb (3)

where uFTA is the innovation in the fixed-term advances and uiFTAis the innovation

in the interest rate on fixed-term advances; νb is a unit-variance, orthogonal shockrelated to the borrowed-reserves component (i.e., a measure of the shift in the de-mand for fixed-term advances) and σb is the standard deviation of the structuralshock related to this equation. Since the rate on fixed-term advances has alwaysbeen changed discretely and according to monetary-policy decisions during the pe-riod under investigation, we set uiFTA

= 0 and consider only innovations in theovernight interest rate as a determinant of the demand for fixed-term advances.21

We then model the central-bank direct intervention in the market for bank re-serves by specifying how it supplies nonborrowed reserves, namely:

uNBR = φdνd + φbνb + φrνr + σsνs (4)

where νr is the unit-variance, orthogonal innovation in the real exchange rate (tobe defined in the last equation), whereas νs is the own (unit-variance and orthog-onal) innovation in monetary policy, a measure of the exogenous component of themonetary policy stance; σs is the standard deviation of this measure.

The latter equation represents the operating reaction function of the monetaryauthorities in the market for bank reserves. An innovation in the supply of nonbor-rowed reserves is designed to offset structural innovations in the demand for totalreserves (νd), in the demand for borrowed reserves (νb) and in the exchange rate(νr); uNBR could be also induced by an exogenous innovation in monetary policy.

The three innovations on the quantitative variables are not independent sinceuTR ≡ uFTA + uNBR. Hence, uNBR can be replaced in the last equation by thedifference between the innovation in total reserves and the innovation in the demandfor fixed-term advances. As a result, by substituting equations (2) and (3) in (4),the latter can be rewritten as follows:22

−(α+ β)uOV = (φd − σd)νd + (φb + σb)νb + φrνr + σsνs (5)

Finally, the equation for the exchange rate23 establishes that its innovation canbe affected by the innovations in all the other variables included in the VAR. Wecan thus write

ur + γ′TRuTR + γNBRuNBR + γOV uOV = σrνr

21The same strategy, with respect to the discount rate, has been applied by Bernanke and Mihov(1998) in their study of US monetary policy. Alternatively, Bernanke and Mihov (1997) consideredan additional equation for the rate on fixed-term advances when studying German monetary policy.

22The following is the correct formulation, given that our notation is slightly different from theone in Bernanke and Mihov (1998). In De Arcangelis and Di Giorgio (1998) the correspondingequation contained a mistake which however did not affect the results of the estimation.

23In our definition, an increase (decrease) in the exchange rate represents a real depreciation(appreciation) of the Lira with respect to the DM.

8

or equivalently, since uTR ≡ uFTA + uNBR,

ur + γTRuTR + γFTAuFTA + γOV uOV = σrνr (6)

where obviously γTR = γ′TR + γNBR and γFTA = −γNBR.

Equations (2), (3), (5) and (6) provide the specified version of (1) for our model.In extensive form:

1 0 α 00 1 −β 00 0 −(α+ β) 0γTR γFTA γOV 1

uTR

uFTA

uOV

ur

=

σd 0 0 00 σb 0 0

(φd − σd) (φb + σb) σs φr

0 0 0 σr

νd

νb

νs

νr

(7)

The order condition for the identification of the complete VAR model is violatedsince 21 variances and covariances are available from the first stage of the estimationand these are not sufficient to obtain the 23 parameters included in the structuralform, whose lower-right corner is 6). Identification can be achieved by imposing ap-propriate constraints that reflect different operating procedures of monetary policy.We describe two possible monetary regimes based on these operating procedures,one based on the control of the overnight rate and the other on the control of thenonborrowed reserves aggregate24. Technically, these regimes correspond to two dis-tinct overidentified structures that can be tested in order to select which one, ifany, is accepted by the data. In our framework, overidentification requires fixing 3parameters.

(1) OV Regime. In this policy regime the monetary authorities offset all ex-ogenous shifts in the market for bank reserves (i.e., all the structural innovationsνd and νb) so as to control the overnight rate. In terms of parameter constraintsthis means that φd = σd and φb = −σb. Moreover, the authorities are also assumednot to allow the supply of nonborrowed reserves to respond to innovations in theexchange rate: i.e., φr = 0. Therefore, eq. (5) becomes:

−(α+ β)uOV = σsνs

and the relative estimated indicator of monetary policy shocks (i.e., σsνs) is propor-tional to the estimated innovation in the overnight interest rate.

(2) NBR Regime. In this quantitative regime the monetary authorities are as-sumed to offset exogenous shifts in the market for bank reserves in order to controlthe total amount of nonborrowed reserves. In this case, the central bank’s operatingprocedures imply φd = 0, φb = 0 and φr = 0 as the appropriate parameter con-straints. The estimated indicator of monetary policy shocks then coincides with theinnovation in the nonborrowed components of reserves:

24See De Arcangelis and Di Giorgio (1998) for other regimes as well as for an analysis of just-identified structures.

9

uNBR ≡ uTR − uFTA = σsνs

4.2 Data, Estimation and Results

We estimate the model between June 1989 and May 1998. The sample does notinclude the 80s as in this period monetary policy in Italy could not be describedas following a market-based approach. We also excluded the last months of 1998,following the announcement of which “countries” would have entered the EMU.

All data are monthly. The variables included in the VAR are (from top tobottom): the consumer price index and the industrial production index as nonpolicyvariables; total bank reserves, fixed-term advances and the overnight interest rateas policy variables; finally, the real exchange rate between Italy and Germany asthe last variable. The price index, the industrial production index and the realexchange rate have been log-transformed. Total reserves and fixed-term advanceshave been normalized by dividing them by the 18-month (past) moving average oftotal reserves in order to use the level relationship among total reserves, fixed-termadvances and nonborrowed reserves.25 Since the Italian banking system experiencedsome relevant changes in the reserve requirement ratios in the sample period, weused the adjusted series for total bank reserves offered by the Bank of Italy.26 Theovernight rate is in levels. Further details on the data are given in the Appendix.

The number of lags employed is six and the estimation is based on the FIMLmethod. As shown in Table 1, the results in De Arcangelis and Di Giorgio (1998)are confirmed. In particular, the regime based on controlling the overnight rateis accepted, while the overidentifying test rejects the hypothesis that the Bank ofItaly in the 90s was targeting the nonborrowed reserves aggregate. Table 1 showsestimation results for both the whole sample period and the subsample following theexit of the Italian Lira from the “hard EMS”.27 Although the estimated elasticitiesare different, the conclusions of the overidentifying tests are robust to the change inthe estimation period.28

25Bernanke and Mihov (1998) claim that this kind of strategy is more appropriate than theshort-run normalization proposed by Strongin (1995).

26The only available data on total reserves (adjusted for the numerous changes in the requiredreserves ratio) and fixed-term advances refer to the maintenance period of required reserves thatdoes not coincide with the calendar month (the maintenance period goes from the 15th of onemonth to the 14th of the next one). All the other data are instead referred to the commoncalendar month. This mismatch may cause some problems in the relationship between the reserveaggregates and the other data within the month. The problem may appear particularly seriousfor the relationship between reserves aggregates and the overnight rate, on which it hinges ouridentification procedure. However, when we compared our time series for the interest rate withanother series that matches the reserve aggregate period, we found very high correlation in bothlevels and in first differences between the series. Hence, the bias introduced by the mismatch doesnot seem to be operatively very relevant, although methodologically important.

27Although technically the Lira rejoined a quasi-fixed exchange rate regime in November 1996,the new ERM was characterized by such wide bands that it could be regarded as quite similar tomanaged floating.

28In order to check more formally for possible instability in the estimated coefficients of the

10

Table 1: Estimation Results for the Benchmark Model (Standard errors in paren-thesis; bold face (italics) indicates 95% (90%) significance)

Regime Overid.Test (prob.) α β

(Period 1989:6–1998:5)OV 0.630 -0.149 0.486

(0.110) (0.110)NBR 2.11e-12 0.824 1.437

(0.167) (0.165)

(Sub-Period 1993:3–1998:5)OV 0.371 1.201 -0.360

(0.650) (0.317)NBR 3.55e-5 23.333 -0.111

(12.014) (0.341)

The impulse responses to the monetary policy shocks in the two estimatedregimes are shown in Figures 1 and 2.29 These pictures provide further supportto the OV regime.

In Figure 1, after a monetary contraction, output falls at impact. The slowdownin economic activity lasts for about two years, but it is statistically significant onlyin the period included between the 8thand the 18th month after the shock. Althoughwe have not included a commodity price index, we find no evidence of a pricepuzzle.30 The response in terms of the price level is not significantly different fromzero.31 The model does not exhibit a liquidity effect either. As a matter of fact,although fixed-term advances (FTA) increase considerably at impact and in thefirst 3 months after the shock, total reserves do not show any significant dynamics.Hence, nonborrowed reserves fall, exactly as predicted by the standard liquidityeffect. Finally, the response of the exchange rate is never statistically significant atour confidence level. Overall, these IRFs seem coherent with textbook predictionsof the effects of a monetary policy shock. We interpret them as a signal of correct,

reduced-form VAR, we performed standard stability test (Hansen, 1992). The results detect noevidence of instability in the equation coefficients: all statistics are below 0.1, with just two ex-ceptions (0.14 and 0.27) that are still well below the 5% critical value (0.470). Among the sixequations, only the one for the interest rate shows instability in the variance, which might be dueto the presence of ARCH effects.

29The impulse response functions are relative to the whole sample. They have been constructedby designing a restrictive monetary shock leading to a 100 basis-point increase in the overnightrate for all the monetary regimes. Responses in output, prices and the exchange rate are percentdeviations from the corresponding variables’ values before the shock. Dashed-line bands refer tothe 95% confidence interval and are computed by the delta method (Hamilton, 1994, and Amisanoand Giannini, 1997).

30In the smaller sample estimated in De Arcangelis and Di Giorgio (1998), the puzzle was stillpresent, although only temporary (the first 4 periods) and much less intensively than in previousVAR analysis.

31The response of prices in European countries has already been documented to be less pro-nounced than in the US (Smets, 1997), when not totally counterintuitive (Sims, 1992). An expla-nation for the lower price sensitivity to monetary policy shocks in Europe might be related to thepresence of stronger nominal rigidities.

11

although still parsimonious, specification of the statistical model.32

The IRF shown in Figure 2 are relative to a monetary policy shock in the NBRregime. They are clearly less appealing, when not intuitively totally wrong. Fol-lowing the same monetary contraction, the reduction in output is practically neverstatistically significant and prices show a persistent increase. The response of the ex-change rate indicates a real depreciation following (for about one year) the monetarycontraction.

In order to check for the robustness of our results, we extend our model to includethe German call money rate as an additional variable in the nonpolicy block. Giventhe exogeneity of the German rate with respect to all the other domestic variables,we list it first in the nonpolicy block.33 Estimation results are shown in Table 2.Once again, the OV regime is confirmed by the data, both in the whole sample andalso in the subsample 1993:3–1998:5.34

5 An alternative specification using Foreign Ex-

change Reserves

The small open economy feature of Italy might as well be captured by using foreignreserves as an alternative to the exchange rate. This would correspond to assumingmanaged floating of the Lira over the whole sample under consideration. Althoughsuch a working hypothesis is obviously an approximation, it is not much strongerthan the previous assumption of free floating of the real exchange rate vis a visthe D-mark. In this section we will test whether such an alternative specificationproduces similar or different results in terms of the identification of the central-bankoperating regime and of the exogenous monetary policy shocks.

We replace (the log of) the exchange rate with (the log of) foreign reserves inthe policy block of the VAR model. The relevant equations become (2) and (3) plus

32We computed the Forecast Error Variance Decompositions (FEVD) of both the price leveland output. This exercise (available on request from the author) provides information about thequantitative importance of each of the identified structural shocks in explaining the variability ofour nonpolicy variables. Exogenous monetary policy shocks account (significantly at the 95% level)for about 15 to 20% of the output volatility one year after the shock, while their contribution toprice volatility is almost null. Exactly the opposite is true for exogenous shocks to the real exchangerate: these have been particularly effective on price dynamics. While, exogenous monetary policyshocks were important in determining output dynamics, the sharp reduction in inflation of the90s might instead be explained by either the endogenous component of monetary policy or by thedynamics of the exchange rate. Indeed, we have also estimated a simple forward looking monetarypolicy rule a la Clarida, Gali and Gertler (1998 and 1999b) for Italy in the 90s and obtained thatthe coefficient on deviations of inflation from target was significantly higher than 1 (1.3), while theone on output stabilization was not statistically different from zero. We interpret this finding assuggesting that the endogenous component of Italian monetary policy in the 90s, as embedded insuch a simple monetary policy rule, was focusing on inflation and not on output stabilization.

33See also Chinn and Dooley (1997).34IRFs relative to monetary policy shocks look very similar to the ones of the benchmark model

and are available from the authors upon request. In particular, in the OV regime prices decreaseafter 4 months, but the response is still of a small magnitude; output decreases, although thestandard errors of the estimated IRFs are slightly higher. Moreover, a German restrictive monetaryinnovation raising the German rate by 100 basis points increases the Italian rate by more than 200basis points after one month and monotonically declines after seven months.

12

Figure 1: OV Regime: Responses to a Monetary Contraction (1% Increase in the OV Rate)

Response of the Price Level to nu_MS

0 5 10 15 20 25 30 35 40 450.15

0.10

0.05

0.00

0.05

0.10

0.15

0.20

0.25

0.30Response of Output to nu_MS

0 5 10 15 20 25 30 35 40 45-1.00

-0.75

-0.50

-0.25

0.00

0.25

Response of Total Reserves to nu_MS

0 5 10 15 20 25 30 35 40 450.6

0.4

0.2

0.0

0.2

0.4Response of Fixed-Term Adv. to nu_MS

0 5 10 15 20 25 30 35 40 45-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

Response of the Int. Rate to nu_MS

0 5 10 15 20 25 30 35 40 450.4

0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2Response of The Real Exch. R to nu_MS

0 5 10 15 20 25 30 35 40 45-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

13

Figure 2: NBR Regime: Responses to a Monetary Contraction (1% Increase in the OV Rate)

Response of the Price Level to nu_MS

0 5 10 15 20 25 30 35 40 450.25

0.00

0.25

0.50

0.75

1.00Response of Output to nu_MS

0 5 10 15 20 25 30 35 40 45-1.50

-1.25

-1.00

-0.75

-0.50

-0.25

0.00

0.25

0.50

0.75

Response of Fixed-Term Adv. to nu_MS

0 5 10 15 20 25 30 35 40 451.0

0.5

0.0

0.5

1.0

1.5

2.0Response of the Int. Rate to nu_MS

0 5 10 15 20 25 30 35 40 45-0.50

-0.25

0.00

0.25

0.50

0.75

1.00

1.25

1.50

1.75

Response of NBR to nu_MS

0 5 10 15 20 25 30 35 40 453.0

2.5

2.0

1.5

1.0

0.5

0.0

0.5

1.0Response of The Real Exch. R to nu_MS

0 5 10 15 20 25 30 35 40 45-3

-2

-1

0

1

2

3

4

5

14

Table 2: Estimation Results for the Model with the German call money rate andthe exchange rate (Standard errors in parenthesis; bold face (italics) indicates 95%(90%) significance)

Regime Overid.Test (prob.) α β

(Period 1989:6–1998:5)OV 0.237 -0.186 0.421

(0.111) (0.108)NBR 1.43e-15 1.302 0.942

(0.151) (0.180)

(Sub-Period 1993:3–1998:5)OV (estim. with 4 lags) 0.654 -0.814 0.617

(0.674) (0.296)NBR (estim. with 4 lags) 0.0047 0.570 -110.90

(0.320) (286.05)

the following two that replace, respectively, equations (5) and (6) above:

uNBR − φFuFR = φdνd + φbνb + φfrνfr + σsνs (8)

uFR + γ1uTR + γ2uFTA + ψuOV = σfrνfr (9)

In this model foreign reserves are allowed to react contemporaneously to funda-mental innovations in all policy variables, exactly as done before for the exchangerate. In addition we explicitly take into account the possibility that the centralbank sterilizes either fundamental or structural innovations to foreign reserves whenchoosing the supply of nonborrowed reserves. We still have 21 variances and covari-ances from the first stage estimation, but now 24 parameters are to be estimated. Inorder to overidentify and estimate the model, we thus need at least 4 short-run re-strictions. We start by studying the two monetary regimes analyzed in the previoussection: an interest rate targeting regime based on the control of the overnight (OVregime) and a quantitative regime controlling the supply of nonborrowed reserves(NBR regime). But we do also estimate two modified versions of these regimeswhere innovations in foreign reserves do also partly contribute to identify exogenousmonetary policy shocks. In the modified OV regime, the identified monetary policyshocks will correspond to a linear combination of the fundamental innovations inboth the overnight rate and foreign reserves. In the modified NBR regime, monetarypolicy shocks will be identified as the component of innovations to nonborrowed re-serves which is orthogonal to fundamental innovations in foreign reserves. Formally,the four regimes are identified by the following constraints.

• OV Regime: φd = σd, φb = −σb and φF = φfr = 0

• NBR Regime: φd = φb = φF = φfr = 0

15

Table 3: Empirical Results for the Model with Foreign Reserves (Standard errors inparenthesis; bold face indicates 95% significance)

Regimes Ov. Test (Prob) α β ψ φF

OV 0.794 -0.167 0.425 0.075 0.00(0.112) (0.112) (0.015)

NBR 8.1e-014 0.832 1.446 -0.087 0.00(0.171) (0.174) (0.013)

modified OV 0.700 -0.259 0.257 0.00 1.384(0.130) (0.131) (0.560)

modified NBR 0.613 -0.305 0.286 0.353 26.462(0.126) (0.126) (0.065) (8.179)

• Modified OV Regime: φd = σd, φb = −σb and ψ = φfr = 035

• Modified NBR Regime: φd = φb = φfr = 0 and γ1 = γ2 = 036

The estimation results are presented in Table 3.The only model which is clearly rejected by the overidentification test is the

NBR Regime. The model based on controlling the overnight rate is accepted withthe highest probability. It does also produce estimates of the parameters α and βwhich are on line with our previous results and similar impulse response functions(see Figure 3 ). In particular, output falls for about two years after a monetarycontraction, while the price level does not significantly react. A modified version ofthe OV regime where monetary policy shocks are identified by a linear combinationof innovations to the overnight rate and to foreign reserves is also accepted, and thegenerated IRFs (see Figure 4 ) are basically identical, although the parameter α issignificantly of the wrong sign in this model. Finally, even though also the modifiedversion of the NBR regime is accepted, its IRFs are considerably less satisfactory,as we do not obtain any significant output response and we find evidence of both aprice and a liquidity puzzle (see Figure 5).

Overall, our results seem to reinforce our previous conclusion that a model ofinterest rate targeting was indeed followed by the Bank of Italy in the 90s. Thedata support our findings under different choices in terms of the macroeconomicvariable used to characterize the small open economy framework, exchange ratesversus foreign reserves.37

35We cannot estimate simultaneously ψ and φF as the model turns out to be not identified inthe rank condition.

36In this case, we can directly estimate both ψ and φF as the model is identified in the rankcondition.

37We also tried to estimate a 7-variable VAR including both the exchange rate vis a vis Germanyand foreign reserves, but serious convergence problems were detected. Instead, a 7-variable VARincluding the German call money rate (as at the end of Section 4.2) and foreign exchange reserveswas estimated for the “pure” OV and NBR regimes: again, only the OV regime is not rejected bythe overidentifying test.

16

Figure 3: OV Regime when including Foreign Reserves: Responses to a Monetary Contraction (1% Increase in the OV Rate)

Response of the Price Level to nu_MS

0 5 10 15 20 25 30 35 40 45-0.16

-0.08

0.00

0.08

0.16

0.24

0.32

0.40

0.48Response of Output to nu_MS

0 5 10 15 20 25 30 35 40 45-1.0

-0.8

-0.6

-0.4

-0.2

-0.0

0.2

0.4

Response of Foreign Res. to nu_MS

0 5 10 15 20 25 30 35 40 45-15.0

-12.5

-10.0

-7.5

-5.0

-2.5

0.0

2.5

5.0Response of Tot. Bank Res. to nu_MS

0 5 10 15 20 25 30 35 40 45-1.0

-0.8

-0.6

-0.4

-0.2

-0.0

0.2

0.4

Response of Fixed-Term Adv. to nu_MS

0 5 10 15 20 25 30 35 40 45-0.32

-0.16

0.00

0.16

0.32

0.48

0.64

0.80

0.96Response of the Int. Rate to nu_MS

0 5 10 15 20 25 30 35 40 45-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

17

Figure 4: Modified OV Regime: Responses to a Monetary Contraction (1% Increase in the OV Rate)

Response of the Price Level to nu_MS

0 5 10 15 20 25 30 35 40 45-0.2

-0.1

0.0

0.1

0.2

0.3

0.4Response of Output to nu_MS

0 5 10 15 20 25 30 35 40 45-1.12

-0.96

-0.80

-0.64

-0.48

-0.32

-0.16

0.00

0.16

0.32

Response of Foreign Res. to nu_MS

0 5 10 15 20 25 30 35 40 45-12.5

-10.0

-7.5

-5.0

-2.5

0.0

2.5

5.0Response of Tot. Bank Res. to nu_MS

0 5 10 15 20 25 30 35 40 45-0.72

-0.54

-0.36

-0.18

-0.00

0.18

0.36

0.54

Response of Fixed-Term Adv. to nu_MS

0 5 10 15 20 25 30 35 40 45-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0Response of the Int. Rate to nu_MS

0 5 10 15 20 25 30 35 40 45-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

18

Figure 5: Modified NBR Regime: Responses to a Monetary Contraction (1% Increase in the OV Rate)

Response of the Price Level to nu_MS

0 5 10 15 20 25 30 35 40 45-0.16

0.00

0.16

0.32

0.48

0.64Response of Output to nu_MS

0 5 10 15 20 25 30 35 40 45-2.5

-2.0

-1.5

-1.0

-0.5

0.0

0.5

1.0

Response of Non-Borr. Res. to nu_MS

0 5 10 15 20 25 30 35 40 45-2.0

-1.5

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0Response of the Int. Rate to nu_MS

0 5 10 15 20 25 30 35 40 45-0.75

-0.50

-0.25

0.00

0.25

0.50

0.75

1.00

1.25

1.50

Response of Tot. Res. to nu_MS

0 5 10 15 20 25 30 35 40 45-1.5

-1.0

-0.5

0.0

0.5

1.0

1.5Response of Foreign Res. to nu_MS

0 5 10 15 20 25 30 35 40 45-42

-36

-30

-24

-18

-12

-6

0

6

19

6 Conclusions

This paper presents different model specifications that are useful to identify mone-tary policy shocks in a small open economy framework. By focusing on the Italianeconomy in the 90s, it shows that correct identification can be obtained by linkingeconometric analysis with a detailed description of the institutional and operativeprocedures used by the central bank to intervene in the domestic money market.Our results can be summarized as follows.

In the 90s, monetary policy in Italy has been based on an interest rate targetingregime, with the central bank controlling the rate on overnight loans. This conclu-sion holds using both exchange rates and foreign exchange reserves as conditioningvariables for the small open economy framework under study. Moreover, our resultsare confirmed even when including a short-term foreign interest rate (the Germancall money rate). Shocks to the overnight rate, or the residual of a regression ofthis rate on a set of variables in the information set of the central bank, can thenbe interpreted as purely exogenous monetary policy shocks, as opposed to the en-dogenous component of policy which might result from a feedback rule linking themonetary policy instrument to different state variables.

Our impulse response analysis is broadly coherent with the expected effect of amonetary policy shock on output and inflation.

Appendix: Data Sources

All the data used in the analysis and the relative sources are listed:

• Italian price level: Italian CPI (1990=100), line 64 in International MonetaryFund, International Financial Statistics, CDROM version (IFS);

• industrial production index: line 61 in IFS;

• total bank reserves: period average, series adjusted for the change in the re-quired reserves ratio; January 1987–December 1987 series S275391M, in Bancad’Italia, Base Informativa Pubblica(BIP) April 1997; January 1988–May 1998applied growth rates contained in series S393004M, BIP November 1998;

• fixed-term advances (Anticipazioni a Scadenza Fissa): period average, seriesS987727M in BIP;

• overnight interest rate: from Datastream;

• nominal Lit/DM exchange rate: computed as a cross rate from the dollarexchange rates contained in IFS;

• German price level: German CPI (1990=100), line 64 in IFS;

• Foreign Reserves: cumulated variations in Foreign Reserves starting in 1970:1(plus an additional component of 15000 in order to avoid negative numbersand take the log transformation); variation in Foreign Reserves is series 08645in BIP.

20

• German call money rate: line 60b in IFS;

The real exchange rate has been obtained by dividing the product of the nominalexchange rate times the German price level, by the Italian price level.

References

[1] AMISANO G. - GIANNINI C. (1997), Topics in Structural VAR Econometrics,Springer Verlag, Berlino.

BAGLIANO F.C. - FAVERO C.A. (1995), “Canale creditizio del meccanismodi trasmissione della politica monetaria ed intermediazione bancaria. Un’analisiempirica”, w.p. n. 91 CEMF Paolo Baffi, Universita Bocconi.

BAGLIANO F.C. - FAVERO C.A. (1998), “Measuring monetary policy withVAR models: an evaluation”, European Economic Review.

BAGLIANO F.C. - FAVERO C.A. - FRANCO F. (1999), “Measuring MonetaryPolicy in Open Economies”, IGIER working paper n.133.

BERNANKE B. - BLINDER A. (1992), “The Federal Funds Rate and theChannels of Monetary Transmission”, American Economic Review, 82.

BERNANKE B. - MIHOV I. (1998), “Measuring Monetary Policy”, QuarterlyJournal of Economics.

BERNANKE B. - MIHOV I. (1997), “What Does the Bundesbank Target?”,European Economic Review, April.

BUTTIGLIONE L. - DEL GIOVANE P. - GAIOTTI E. (1997), “The Role ofDifferent Central Bank Rates in the Transmission of Monetary Policy”, Bancad’Italia, Temi di Discussione n. 305.

BUTTIGLIONE L. - FERRI G. (1994), “Monetary Policy Transmission viaLending Rates in Italy: Any Lessons from Recent Experience?, Banca di Italia,Tema di Discussione n. 224.

CAMPBELL J. - PERRON P. (1991) “Pitfalls and Opportunities: WhatMacroeconomists Should Know About Unit Roots”, NBER Macroeconomic An-nuals, p. 141-201.

CANOVA F. (1995), “Vector Autoregressive Models: Specification, Estima-tion, Inference, and Forecasting”, in Handbook of Applied Econometrics ed. byPesaran, M. H. - M. Wickens, Blackwell.

CHINN M. - DOOLEY M. (1997), “Monetary Policy in Japan, Germany andthe United States: Does One Size Fit All?”, NBER Working Paper n.6092, July1997.

CHRISTIANO L. - EICHENBAUM M. (1992), “Identification and the Liq-uidity Effect of a Monetary Policy Shock”, in Political Economy, Growth andBusiness Cycles ed. by A. Cukierman - Z. Hercowitz - L. Leiderman, MIT Press.

CHRISTIANO L. - EICHENBAUM M. - EVANS C. (1998), “Monetary PolicyShocks: What Have We Learnt and to What End”, NBER w.p.6400.

21

CLARIDA R. - GERTLER M. (1997), “How the Bundesbank Conducts Mone-tary Policy”, in Reducing Inflation: Motivation and Strategy, ed. by Romer C.- D. Romer, University of Chicago Press for NBER.

CLARIDA R. - GALI J. - GERTLER M.(1998), “Monetary Policy Rules inPractice: Some International Evidence”, in European Economic Review.

CLARIDA R. - GALI J. - GERTLER M. (1999a), “The Science of MonetaryPolicy: a New Keynesian perspective”, Journal of Economic Literature.

CLARIDA R. - GALI J. - GERTLER M. (1999b), “Monetary Policy Rulesand Macroeconomic Stability: Evidence and Some Theory”, forthcoming inQuarterly Economic Journal.

DE ARCANGELIS G. - DI GIORGIO G. (1998), “In Search of Monetary PolicyMeasures: The Case of Italy In The 90s”, in Giornale degli Economisti ed Annalidi Economia, 2.

DE ARCANGELIS G. - DI GIORGIO G. (1999), “Monetary Policy Shocks andTransmission in Italy: a VAR Analysis”, Universitat Pompeu Fabra, Economicsand Business Department, Working Paper n. 446.

GAIOTTI E. (1992), “L’evoluzione dei metodi di controllo monetario ed ilmodello mensile della Banca di Italia”, mimeo.

GAIOTTI E. (1999), “The Transmission of Monetary Policy Shocks in Italy,1967-1997 ”, Banca d’Italia, Temi di Discussione n.363.

GAIOTTI E. - GAVOSTO A. - GRANDE G. (1997), “Inflation and MonetaryPolicy in Italy: Some Recent Evidence”, Banca d’Italia, Temi di Discussionen.310.

GRILLI V. - MASCIANDARO D. - TABELLINI G. (1991), “Political and Mon-etary Institutions and Public Finance Policies in the Industrialised Democra-cies”, Economic Policy.

HAMILTON, J. (1994), Time Series Analysis, Princeton University Press.

HANSEN, B. (1992), “Testing for Parameter Instability in Linear Models”,Journal of Policy Modeling, 14, pp. 517-33.

KEATING, J. W. (1990), “Identifying VAR Models under Rational Expecta-tions”, Journal of Monetary Economics, 25, p.453-76.

KIM S. - ROUBINI N. (1995), “Liquidity and Exchange Rates: A StructuralVAR Approach”, paper presented at the CEPR Workshop on “Monetary Policyand Exchange Rates in Europe”, Bonn Feb. 10-11, 1995.

LEEPER E. - SIMS C. - ZHA T. (1996), “What Does Monetary Policy Do?”,Brookings Papers on Economic Activity, 2.

PASSACANTANDO F. (1996), “Building an Institutional Framework for Mon-etary Stability: the Case of Italy (1979-94)”, BNL Quarterly Review, 196.

SARCINELLI M. (1995), “Italian Monetary Policy in the 80’s and 90 s: theRevision of the Modus Operandi”, BNL Quarterly Review, 195.

SIMS C. (1992), “Interpreting the Macroeconomic Time Series Facts: the Ef-fects of Monetary Policy”, European Economic Review, 36.

22

SIMS C. - STOCK J. H. - WATSON M. W. (1990), “Inference in Time SeriesModels with Some Unit Roots”, Econometrica, 58, pp. 113-44.

SMETS F. (1997), “Measuring Monetary Policy Shocks in France, Germanyand Italy: the Role of the Exchange Rate”, Swiss Journal of Economics andStatistics, 133, pp. 597-616.

STRONGIN S. (1995), “The Identification of Monetary Policy Disturbances:Explaining the Liquidity Puzzle”, Journal of Monetary Economics, 35.

WATSON M. W. (1994), “Vector Autoregression and Cointegration”, in Hand-book of Econometrics vol. IV.

23