European Economic Association

Bayesian New Neoclassical Synthesis (NNS) Models: Modern Tools for Central Banks Author(s): Frank Smets and Rafael Wouters Source: Journal of the European Economic Association, Vol. 3, No. 2/3, Papers and Proceedings of the Nineteenth Annual Congress of the European Economic Association (Apr. May, 2005), pp. 422-433 Published by: The MIT Press on behalf of European Economic Association Stable URL: http://www.jstor.org/stable/40004985 Accessed: 03/05/2010 22:16 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=eea. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

European Economic Association and The MIT Press are collaborating with JSTOR to digitize, preserve and extend access to Journal of the European Economic Association.

http://www.jstor.org

BAYESIANNEW NEOCLASSICALSYNTHESIS (NNS) MODELS: MODERN TOOLS FOR CENTRALBANKS Frank Smets EuropeanCentralBank

Rafael Wouters NationalBank of Belgium

Abstract This paperdiscusses the advantagesof Bayesian New Neoclassical Synthesismodels as tools for monetarypolicy analysisandforecasting.The combinationof a sound,microfoundedstructurewith a good probabilisticdescriptionof the observeddatamakesthose models suitablefor investigatingthe structuralsourcesof businesscycle fluctuations,for analysingoptimalmonetarypolicy responsesto those developmentsand for makingeconomic projectionsconditional on various policy assumptions.The paper gives two examples of such analysis. (JEL:E40, E5O,C11)

1. Advantages of the Bayesian NNS Methodology Following the work by King and Wolman (1996), Goodfriend and King (1997), Rotemberg and Woodford (1997), Chari, Kehoe, and McGrattan(2002), and others, a new generation of sticky-price-and-wage DSGE models (the New Neoclassical Synthesis or NNS models) has become very popular in monetary policy analysis.1 These models combine the rigor of the Real Business Cycle (RBC) approach, which is characterised by the derivation of behavioral relationships from the optimizing behavior of agents (households, firms, governments) subject to technological and budget constraints and the specification of a well-defined equilibrium concept, with the tractable introduction of nominal rigidities (typically Calvo or Taylor-typecontracts), which imply a nontrivialrole for monetary policy. In contrast to RBC models, those models typically also include a number of other shocks than the traditionaltechnology shock. The result of the New Neoclassical Synthesis is a macroeconomic model, which in structure is quite comparable with traditional reduced-form AS/AD macro models. For example, the simplest model can be written in the form of a forward-looking IS equation, a New Keynesian Phillips curve and a monetary Acknowledgments: The views expressed are solely our own and do not necessarily reflect those of the European Central Bank or the National Bank of Belgium. E-mail addresses: Smets: frank.smets @ccb.int; Wouters: [email protected] 1. See Woodford's (2003) seminal textbook for a thorough and comprehensive analysis.

Journal of the European Economic Association April-May 2005 © 2005 by the European Economic Association

3(2-3):422^33

SmetsandWouters BayesianNew NeoclassicalSynthesisModels

423

policy reaction function.2 However, in contrast to many reduced-form models, these models also embed a structuralinterpretationof the reduced-formparameters and shocks in terms of the "deep"parametersthat govern tastes, technology, and institutional constraints and a consistent treatmentof expectations formation and stock-flow dynamics. An example of an NNS model that also includes a monopolistically competitive labour market and a capital accumulation sector is derived in some of our recent work (Smets and Wouters 2003a, 2003b). The Smets-Wouters model contains three main types of agents. Households consume, work, set wages, and invest; firms hire labour and capital, produce goods and set the prices of those goods; and the central bank sets the short-term interest rate in response to the deviation of inflation for the inflation target and a theoretically consistent outputgap. The model contains a relatively large number of real and nominal frictions such as monopolistic competition in goods and labour marketswith sticky nominal prices and wages, partialindexation of prices and wages, costs of adjustment in capital accumulation, external habit formation and variable capital utilization and fixed costs. Finally, the Smets-Wouters model also contains a relatively large set of structuralshocks: two "supply" shocks (a productivity and a labor supply shock); three "demand"shocks (a preference shock, an investment-specific technology shock and a government consumption shock); three "cost-push" shocks (a price markup,a wage markupand an equity premium shock); and, finally, two monetarypolicy shocks (a temporaryinterest rate shock and a persistent inflation target shock). From a policy point of view, the advantages of the microfoundations are many. First, information about deep structural parameters (e.g., coming from microstudies) can be used to calibrate/estimate the model. This is particularly useful when time series are short or regime changes have taken place, as for example has been the case following the introduction of the euro in 1999. For example, ongoing research on price setting at the microlevel in the euro area is very useful to form priors about the degree of price stickiness that should be incorporatedin the NNS models.3 Second, the cross-equation restrictionsimplied by the structuralmodel help to identify the type of shocks that hit the economy. In particular,it becomes easier to identify whether shocks are of a permanentor a temporarynature.For example, the permanentincome hypothesis suggests that permanentshocks to income should have large effects on consumption, whereas the effects of temporaryshocks should be small.

2. See, for example, Clarida, Gall, and Gertler (1999). 3. See, for example, Aucremanne and Dhyne (2004) and other ECB Working Papers produced under the heading of Eurosystem's Inflation Persistence Network.

424

Journalof the EuropeanEconomicAssociation

Third, these models are less subject to the Lucas critique and therefore more suitable for policy analysis. At a minimum, one has a better feel for which parameters are likely to be policy invariant.For example, it is probably reasonable to assume that the degree of risk aversion or other preference parametershave been relatively stable over time and across countries. On the other hand, more ad-hoc features like the degree of indexation of wages may have adjusted over time. A structuralinterpretationof the parameters allows distinguishing between those parameters both in estimation and policy analysis. Finally, DSGE models are more appropriatefor normative analysis. The utility of the agents can be used as a model-consistent measure of welfare to evaluate alternativepolicies. For example, models with nominal rigidities have been used to make a strong case for price stability in King and Wolman (1996) and Rotemberg and Woodford (1997).4 However, in order to reap the full benefits of such a structuralmodel for monetary policy analysis, the model also needs to be consistent with the data. Many topical monetary policy questions involve not only a qualitative answer, but also a quantitative one. It is therefore important that the model can match the time series properties of the many macroeconomic data series that central bankers use to base their policy decisions on. In the RBC tradition, the most common empirical approachis to calibratethe model in orderto match a selected numberof moments in the data. In our previous work (Smets and Wouters2003a, 2003b) we have demonstratedthat the currentgeneration of NNS models, as has been developed by Christiano, Eichenbaum, and Evans (2005), is rich enough to capture the full data-generating process of the main macro series, provided one allows for a sufficient number of structural shocks. The "many structural shocks" approach used in Smets and Wouters (2003a) corresponds closely to the "real world" of monetary policy analysis. Central bank economists typically spend a lot of time and effort figuring out what are the sources driving current business cycle and inflation developments. This approach faces, however, two challenges. First, from a policy perspective there is a need to link the structural shocks mentioned above to other observable macro variables. For example, the so-called markup shocks identified in Smets and Wouters (2003a) could stand in for many different types of disturbances such as changes in taxes, changes in the degree of competition or relative price shocks in flexible-price sectors that are not modeled such as food and energy.5 Second, from a research point of view the "many structuralshocks" approachneeds to be reconciled with the long history of business cycle analysis, which startsfrom the premise that most macro variables are predominantlydriven by only a few factors.6 One advantage of the 4. A recent welfare application to a model similar to Smets and Wouters (2003) is performed in Laforte (2003) and Onatski and Williams (2003). 5. See De Walque, Smets, and Wouters (2004). 6. For a recent analysis, see, for example, Giannoni, Reichlin, and Sala (2004).

Smets and Wouters Bayesian New Neoclassical Synthesis Models

425

empirical methodology discussed below is that the researchercan explicitly test which shocks are importantand which ones are not. In order to bring the NNS model to the data, Smets and Wouters (2003a, 2003b) apply a full-system Bayesian likelihood estimation method. In a first step standardprocedures are used to calculate the likelihood of the linearised rational expectations model. In a second step, the likelihood is combined with a priordistribution over the structuralparametersto form the posteriordensity function, which is simulated using a Monte Carlo Markov Chain (MCMC) algorithm, such as the Metropolis-Hastings algorithm. The advantages of the second (Bayesian) step are manifold. First, it formalises the use of priorinformationcoming from microeconometric or previous macrostudiesand therebyallows one to make a direct link with the calibration tradition. De facto, the maximum likelihood estimates will be pulled towards the values that are a priori plausible. Second, the use of priors makes the estimation algorithm of the highly restricted model much more stable by restricting the search to plausible values and by introducing "curvature"in the objective function.7 From a practicalpoint of view, the MCMC methods have made the estimation of largersimultaneousmodels tractable.8Third,the Bayesian analysis provides a frameworkfor evaluating misspecified and non-nested models, based on their out-of-sample prediction performance. As discussed in Sims (2003), one should, however, be awareof potentialproblems with Bayesian model comparison techniques in case the set of models considered is quite sparse.9 Fourth, the MCMC methods deliver a full characterizationof the parameter and shock uncertainty.This is particularlyuseful in a policy context where the risk analysis surroundingthe inflationprojectionplays an importantrole. In Smets and Wouters(forthcoming), we illustratethis use by calculating inflation and deflation risk measures as proposed by Killian and Manganelli (2003) for the euro area. Finally, the Bayesian estimation outputand the marginaldata density in particular provides the appropriateinput for model averagingand Bayesian decision making undermodel uncertaintyas discussed in Brock, Durlauf, and West (2003) and Del Negro and Schorfheide (2004a). 2. The Smets-Wouters Model: Two Applications In the rest of this short paper, we briefly review two applications of the Bayesian NNS model estimated in Smets and Wouters(2003a, 2003b). The first application is based on De Walque, Smets, and Wouters (2004) and investigates one possible 7. The implications of the relatively tight priors in Smets and Wouters (2003a) on the posterior parameterestimates have been examined by Onatski and Williams (2003). 8. See Sims (2003). 9. Del Negro and Schorfheide (2004b) consider a continuum of models going from an unrestricted VAR to a structuralNNS model to address this problem.

426

Journalof the EuropeanEconomicAssociation

explanationfor the (implausibly) high degree of nominal price stickiness found in Smets and Wouters(2003a, 2003b). This application illustrateshow the Bayesian estimation techniques can be used to test the relevance of various frictions in a DSGE model. The second application is an extension of Smets and Wouters (forthcoming), in which we illustratethe use of Bayesian NNS models as a tool for understandingeconomic developments and for making conditional inflation and output projections. We apply two different methods for calculating projections conditional on a future interest rate path and show the importanceof structurefor interpretingthe results.

2.1. The Importance of Nominal Price Rigidities Smets and Wouters (2003a) find that from an empirical point of view the most importantfriction is the degree of nominal price stickiness. Given the assumed structureof the goods market and the hypothesis that the marginal cost for each of the monopolistically competitive firms is the same across firms, the estimated degree of price stickiness points to an average contract length of more than two years. Although these estimates are very similar to those derived using singleequation estimation methodologies (e.g., Gall and Gertler 1999; Gall, Gretler,and Lopez-Salido 2001 ; and Eichenbaumand Fischer 2003), such contractlengths are implausible compared with the evidence derived from microdata.10This points to a misspecification of the underlying model of the goods market.As pointed out by Gall and Gertler(1999), Woodford (2003), Weinke and Sveen (2003), Eichenbaum and Fischer (2003), and Coenen and Levin (2004), one possible source of misspecification is the assumption of identical marginalcosts across firms. Those papers show that introducingreal rigidities in the form of firm-specific capital will introduce firm-specific upward marginal cost curves and reduce the elasticity of inflation with respect to changes in real marginal costs. The intuition is simple: when firms realise that deviations of their own price from the prices set by other firms may lead to relatively large changes in relative demand and marginal cost, they will be more reluctant to change their price relative to those of other firms. Such real rigidities can therefore substitute for the nominal rigidity in accounting for a relatively flat estimated Phillips curve. In De Walque, Smets, and Wouters (2004), we investigate this hypothesis by comparing the baseline Smets-Woutersmodel with a model with Taylorcontracts in which the mobility of capital is restricted across firms.11Figure 1 summarises the main results. It plots the marginal data density as a function of the assumed 10. This micro evidence typically points to average contract durations of one year or less. 11. In the Taylor contracting model, the contract length is fixed and staggered across firms. All the other features of the model are kept the same.

Smets and Wouters Bayesian New Neoclassical Synthesis Models

427

Figure 1. The log data density for a Taylor contracting model with different contract lengths (with and without mobile capital). Source: de Walque, Smets, and Wouters (2004). Notes: Vertical axis: log data density. Horizontal axis: Taylor contract length. Baseline refers to the Calvo baseline model estimated in Smets and Wouters(2003a). MK refers to Taylorcontracting models with mobile capital. NMK refers to Taylor contracting models with firm-specific capital.

Taylor contract length (from one to ten quarters) under the two hypotheses of perfect capital mobility (MK line) and no capital mobility (NMK line). The horizontal line shows the marginaldata density of the baseline Smets-Wouters model with Calvo contracts and perfect capital mobility.12 A number of results are worth noting. First, under the assumption of perfect capital mobility across firms modeling price rigidities through Taylor ratherthan Calvo contractsgenerally leads to a deteriorationof the empirical fit of the model. Consistent with the baseline results, the best-fitting Taylor contract model is one with an implausibly long contract length of eight quarters.Second, introducing firm-specific capital in the Taylorcontractingmodel uniformly leads to a betterfit of the model. Moreover, consistent with the papers mentioned above, the contract length that maximizes the empirical fit in this case falls to three quarters,which is much more in line with the micro evidence discussed above. This model slightly outperformsthe baseline Smets-Wouters model. Finally, imposing flexible prices in the goods market (corresponding to a contract length of one quarter) leads to a very significant deterioration of the marginal data density of the model. Hence, in spite of the presence of nominal wage rigidities in the Smets-Wouters model, nominal price rigidities remain an important friction to incorporate in NNS models. One feature of the results not reportedhere is that the elasticity of 12. These estimates are performed using euro area data.

428

Journalof the EuropeanEconomicAssociation

substitutionbetween the monopolistically supplied goods is estimated to be very large. Consistent with the intuition mentioned above, this helps slowing down the response of newly set prices to changes in marginal cost. However, as argued in Coenen and Levin (2004), such a high degree competition may be at odds with other micro evidence suggesting that adding additional features such as a varying elasticity of demand (as in Eichenbaum and Fischer 2003) is necessary. The exercise reported above illustrates how the Bayesian methodology can be used to distinguish between various hypotheses regardingthe frictions and the shocks that govern the dynamics of the macroeconomy, while taking into account evidence from the micro literature.Recent papers, such as Adolfson et al. (2005) and Gall and Rabanal (2004) provide additional examples of such an empirical strategy.

2.2. Conditional Projections: Two Alternative Methods In Smets and Wouters (2004), we illustrate the use of the Smets-Wouters model as a projection model and illustrate how monetary policy interventions affect the posterior distributionaroundthe inflation forecast and the correspondingrisk measures. In this section we extend this analysis by stressing the importance of using structuralmodels for producing conditional projections. In particular, we compare two methods for deriving projections conditional on an interest rate path derived from interest rate futures and show how they can yield very different results that will be difficult to interpretwithout the benefit of a structuralmodel. 13 The first approachhas been advocated by Leeper and Zha (2003) and consists of adding unexpected policy shocks over the projection horizon with the goal of mimicking the alternativeinterest rate path. This approachrequiresat a minimum that the monetary policy shocks are identified in the projection model. Figure 2 provides an illustration of this method for euro area inflation and output growth projections at the end of 2003 using the Smets-Woutersmodel.14 The solid lines show the unconditional projections of output growth, annual inflation and the short-termnominal interest rate over a two-year forecast horizon (2004-2005). 15 This unconditional forecast is derived under the assumption that the central bank follows the estimated reaction function. The broken lines show the projection conditional on an interest rate path consistent with the one implied by interest

13. Recently, the Bank of England shifted the focus in its Inflation Reports to such conditional forecasts. See Inflation Report, August 2004, Bank of England. 14. For this exercise, the model was reestimated using data up to the last quarter of 2003. The projection horizon runs over 2004 and 2005. 15 . Each of the lines correspondto the 25 , 50, and 75 percentiles, taking into account both parameter and shock uncertainty.

Smets and Wouters Bayesian New Neoclassical Synthesis Models

429

Figure 2. Predictive density for output growth, inflation, and the short-term interest rate: unconditional (full line) versus conditional (broken line) projections using the Leeper and Zha method. Notes: Projections based on the Smets-Wouters (2003a) model reestimated to take into account data up to the third quarterof 2003.

rate futures at the end of 2003. A number of observations are worth making. First, the unconditional interest rate forecast is quite a bit higher than the one implied by interest rate futures. As a result, we need to add quite significant and persistently negative policy shocks to the baseline forecast in order to reproduce the interest rate path implied by the futures. As arguedby Leeper and Zha (2003), whether such a procedure is reasonable will depend on whether such a sequence of policy shocks is consistent with historical experience.17 Second, as expected, the much more gradual tightening by the central bank in the alternativescenario leads to an upward shift in the projection distributionfor both output growth and inflation. It also leads to a clear widening of the 90% confidence band aroundthe inflation projection in particulartowards the end of the projection horizon. It is now interestingto compare these results with those derived from an alternative method (the minimum entropy method) recently applied by Robertson, Tallman, and Whiteman (2002) and Cogley, Morozov, and Sargent (2003). This method to construct conditional projections consists in finding the forecast distribution that is closest to the unconditional forecast distribution (in a KLIC sense) subject to the constraint that the expected interest rate follows a certain 16. Note that after the two-year forecast horizon, the estimated policy reaction function is again assumed to kick in. 17. One shortcoming of the Leeper and Zha method is that the series of shocks necessary to pin down the future interest rate path will be different for every simulation underlying the projection distribution.

430

Journalof the European Economic Association

Figure 3. Predictive density for growth, inflation and interest rate: unconditional distribution (full line) versus tilted distribution with the futures rates imposed by minimum entropy (broken line).

path.18 As argued in Robertson, Tallman, and Whiteman (2002), the minimum entropy method can be used to incorporate in the forecast exercise additional information that is not captured by the model. One advantage compared to the previous method is its flexibility. The method can be applied to any unconditional forecast distribution without the need for a structuralmodel. Moreover, as the method is based on a tilting of the forecast distribution,there is also no need to exactly enforce the constraint. However, the lack of a structuralinterpretationis also a weakness of the approach.This can be illustratedby our application to the euro area as shown in Figure 3. As shown in the third panel of Figure 3, imposing the constraint that in expected terms the projected interest rate is consistent with the path implied in the interest rate futures, leads to downward shift in the projection distribution of the interest rate. As mentioned before, one difference with the Leeper and Zha (2003) method is that the interest rate forecast distributiondoes not have to collapse to a single projection. However, the implications for the inflation and output growth forecast distribution are quite different. In contrast to the results presentedin Figure 2 the distributionfor inflation shifts downwardin line with the lower interest rates, while the forecast distributionfor outputmoves up slightly.19 18. KLIC stand for the Kullback- Leibler Information Criterium,which is a measure for calculating the distance between two distributions. 19. Note that the high value of the KLIC criterium (not reported), which summarises the distance between the two distributions, indicates that the constraint on the expected interest rate path implies a relatively unlikely given the posterior distribution of the model forecast.

Smets and Wouters Bayesian New Neoclassical Synthesis Models

431

Figure 4. Predictive density for the exogenous processes of the forecast: unconditional distribution (full line) versus tilted distributionwith the future rates imposed by minimum entropy (broken line).

How can this be explained? To answer this question, one again needs a structural model to link the shifts in the distribution of the endogenous variables to shifts in the distributionsof the structuralshocks. Figure 4 shows the implied shifts in the distribution of the structuralshocks. In a statistical sense, the more gradual interest tightening expected by financial market is found to be most consistent with an upward tilting of the forecast distributionof the exogenous productivity and labour supply processes. By giving a higher probability to positive productivity and labour supply developments, the interest rate forecast distribution is shifted downwards,but so is the inflation forecast distribution.The output growth distribution, on the other hand, is shifted upwards. Shifts in the distributions of the other shocks appearto be less important.One structuralinterpretationof the interest rate path implied by the futures is therefore that the financial marketshad a relatively optimistic outlook regarding the future supply developments driving the economy at the end of 2003. In sum, two main conclusions can be drawn from this exercise. First, using minimum entropy methods to combine information from private sector expectations with the predictions from a structuralprobabilisticmodel such as the SmetsWoutersmodel may help interpretingwhat may drive those marketexpectations. Second, the exercise also illustrates the importance for monetary policy decision making to construct independentprojection exercises based on structuralmodels.

432

Journalof the EuropeanEconomicAssociation

Such projections can serve as a benchmark against which other information can be evaluatedin terms of its underlying assumptions and correspondinglikelihood.

3. Conclusions In this short paper we discussed the advantages of Bayesian NNS models as tools for monetary policy analysis and forecasting. The combination of a sound, micro founded structure with a good probabilistic description of the observed data makes those models suitable for investigating the structuralsources of business cycle fluctuations, for analyzing optimal monetary policy responses to those developments and for making economic projection conditional on various policy assumptions. The paper gives two examples of such analysis.

References Altig, D andLawrenceJ Christiano,MartinEichenbaum,andJesperLinde(2002). "Technology Shocks and AggregateFluctuations."Workingpaper,NorthwesternUniversity. Aucremanne,Luc and EmmanuelDhyne (2004). How FrequentlyDo Prices Change/ Evidence Based on the Micro Data UnderlyingBelgian CPI."NBB WorkingpaperNo. 43. Brock, William, Steven Durlauf,and KennethWest (2003). Policy Evaluationin Uncertain Economic Environments."BrookingPaperson Economic Activity, 2003:2. Chan, Varadarajan V., PatrickJ. Kehoe, and Ellen R. McGrattan(2002). "StickyPrice Models and the Business Cycle: Can the ContractMultiplierSolve the PersistenceProblem?" Econometrics 68, 1151-1179. Christiano,LawrenceJ., MartinEichenbaum,and CharlesEvans (2005). "NominalRigidities and the Dynamic Effects of a Shock to MonetaryPolicy."Journal of Political Economy, 113, 1-46. Clarida,Richard,Jordi Gall, and Mark Gertler(1999). "The Science of MonetaryPolicy: A New KeynesianPerspective."Journalof EconomicLiterature,37, 1661-1707. Coenen,GunterandAndrewLevin (2004). "StaggeredPriceContractsandInflationDynamics Revisited."Workingpaper,EuropeanCentralBank. Cogley, Timothy,Sergei Morozov and Thomas J. Sargent(2003). "BayesianFan Chartsfor U.K. Inflation:Forecastingand Sources of Uncertaintyin an Evolving MonetarySystem." Workingpaper,Bank of England. Del Negro, Marcoand FrankSchorfheide(2004a). "Priorsfrom GeneralEquilibriumModels for VARs."InternationalEconomicReview,45, 643-673. Del Negro, Marco and FrankSchorfheide(2004b). "MonetaryPolicy with PotentiallyMisspecifiedModels?"Workingpaper,Universityof Pennsylvania. De Jong, David N., Beth F. Ingram,and CharlesH. Whiteman(2000). A Bayesian Approach to Dynamic Macroeconomics."Journalof Econometrics,98, 203-223. De Walque, Gregory,Frank Smets, and Rafael Wouters (2004). Price Setting m General Equilibrium:AlternativeSpecifications."Workingpaper,EuropeanCentralBank. bichenbaum,Martin,and Jonas Fischer (2003). Testing the Calvo Model ot Sticky Prices. Workingpaper,FederalReserve Bank of Chicago. Gall, Jordi and Mark Gertler (1999). Inflation Dynamics: A Structural Econometric Approach."Journalof MonetaryEconomics,44, 195-222.

Smets and Wouters Bayesian New Neoclassical Synthesis Models

433

Gali, JordiandMarkGertler,andDavid Lopez-Salido(2001). "EuropeanInflationDynamics." EuropeanEconomicReview,45, 1121-1 150. Gali, Jordi and Pan Rabanal(2004). Technology Shocks and Aggregate Fluctuations:How Well Does the RBC Model Fit PostwarU.S. Data?"NBERMacroeconomicsAnnual.MIT Press. Giannom, Domenico, Lucrezia Reichhn, and Luci Sala (2002/2004). TrackingGreenspan: Systematicand UnsystematicMonetaryPolicy Revisited.CEPRDiscussion Papers3550. Goodfriend,MartinandRobertG. King (1997). "TheNew Neoclassical Synthesisandthe Role of MonetaryPolicy."In NBERMacroeconomicsAnnual 1997, edited by Ben S. Bernanke and Julio J Rotemberg.MIT Press. Killian, Lutz and Simone Manganelli(2003). "TheCentralBank as a Risk Manager:Quantifying and MeasuringInflationRisks."Workingpaper,EuropeanCentralBank. King, RobertG and AlexanderL Wolman(1996). InflationTargetingin a St. Louis Model of the 21st Century."NBER WorkingPaper5507, NationalBureauof Economic Research. Laforte Jean-Philippe(2003). "ComparingMonetaryPolicy Rules in an EstimatedGeneral EquilibriumModel of the U.S. Economy."Workingpaper,PrincetonUniversity. Leeper, Eric M and Tao Zha (2003). Modest Policy Interventions. Journal of Monetary Economics,50, 1673-1700. Onatski,Alexei andNoah Williams(2003). "EmpiricalandPolicy Performanceof a ForwardLooking MonetaryModel."Workingpaper,ColumbiaUniversity. Robertson,JohnC, andEllis W.Tallman,andCharlesH. Whiteman(2002). "ForecastingUsing RelativeEntropy."WorkingPaper2002-22, FederalReserve Bank of Atlanta. Rotemberg,Julio J. and Michael D. Woodford(1997). "AnOptimization-BasedEconometric Frameworkfor the Evaluationof MonetaryPolicy: ExpandedVersion."NBER Technical WorkingPaper233. Sims, ChristopherA. (2003). ProbabilityModels tor MonetaryPolicy Decisions. Working paper,PrincetonUniversity. Smets, Frankand Rafael Wouters(2003a). "AnEstimatedDynamic StochasticGeneralEquilibriumModel of the EuroArea."Journalof the EuropeanEconomicAssociation, 1, 11231175. Smets, Frankand Rafael Wouters(2003b). "Shocksand Frictionsin U.S. Business Cycles: A Bayesian DSGE Approach."Workingpaper,EuropeanCentralBank. Smets, Frankand Rafael Wouters(2005). ComparingShocks and Factions in US and Euro Area Business Cycles: A Bayesian DSGE Approach."Journalof AppliedEconometrics. Smets, Frankand Rafael Wouters(2004). "Forecastingwith a Bayesian DSGE Model: An Applicationto the EuroArea."Journalof CommonMarketStudies,42, 841-867. Weinke,Lutz and Tommy Sveen (2003). Inflationand OutputDynamics with rirm-Owned Capital."Workingpaper,UniversitatPompeuFabra. WoodfordMichaelD. (2003). Interestand Prices: Foundationsof a Theoryof MonetaryPolicy. PrincetonUniversityPress.