Model-Based Monetary Policy Analysis and Forecasting
Montevideo, Uruguay, August 26 – 30, 2019
The course took place in the Central Bank of Uruguay. It was jointly organized by CEMLA, the IMF and the Central Bank of Uruguay. The main objectives of the course were the presentation, discussion, analysis and implementation of New Keynesian semi-structural models. Such models are part of the Forecasting and Policy Analysis System (FPAS) recommended by the IMF. It is worth mentioning that the FPAS is a scheme to think about economic policy within a central bank rather than a model itself. The FPAS uses a set of models.
The course was divided into several theoretical sessions, delivered by Santiago Acosta-Ormaechea and Jorge Restrepo, and five practical sessions, delivered by Juan Pablo Medina. Their implementation was mostly in IRIS, a package in Matlab. All instructors have worked at the IMF, albeit Juan Pablo Medina is nowadays in the Adolfo Ibañez University in Chile.
In the course, a team from each central bank delivered an economic juncture presentation, also explained the decision process for monetary policy and how they use the type of models presented in the course for policy decision purposes. If it was the case that their central banks did not use these models, they explained what steps they could take to implement them and commented on some of the possible challenges they could face.
Next, we describe some of the main points from the course. The various monetary regimes in several economies were presented. The regimes in emerging economies were underscored. The range of exchange rate regimes go from the fixed-exchange rate, dollarized economies to free floaters. While the types of regimes are balanced in terms of number of economies, when one considers the economic size of the countries that have floating exchange rates, they surpass those with a fixed-exchange rate.
The semi-structural model was presented. There was some emphasis made on the importance of a model having micro-foundations. The main rationale for doing so is to address the so-called Lucas Critique. In a nutshell, this means that models should be invariant to policy modifications. In other words, if one uses a reduced form model to make a policy recommendation, there is the possibility that the model would change as a consequence of the policy, making it useless to assess the policy itself and its full implications. In other words, structural models are invariant to policy changes. The semi-structural feature of the models, however, was given a pragmatical justification. The use of semi structural models requires much discipline.
The different monetary policy transmission channels were explained and how such channels are captured by the model. Of course, the analysis of policy transmission channels is at the heart monetary policy. Their correct identification is crucial for the formulation of a policy response. Their identification is challenging given that they can be simultaneously present and generally interact with each other.
The relevance on considering the different components of the consumer price index were underlined. In particular, the relative importance of such components depending on the economy being considered was explained. Thus, as typical question is how to respond to a commodity shock. The answer depends, among other factors, on the weight that such commodity has within the consumer basket and well as the relative importance of such a commodity as an input to the economy in question.
There were explanations on the preparation of data before using it to model, the estimation of gaps, and long-run trends for the model. The estimation of gaps and long-run trends requires judgment in particular in the case of emerging market data, given two considerations. First, the fact that there is a short history of time series in EMEs. Second, the statistical properties of such time series. Some are likely non-stationary.
The Kalman Filter was explained and how it can be used as a possible method to estimate the gap of a given variables. The Kalman Filter has been used in areas different such as in the implementation of the GPS, and it is nowadays widely used in macroeconomics, for example. The central idea is that one observes (more generally, measures) a variable with some noise. Thus, one is interested in accounting for such a measuring the variable accounting for the presence of the noise. It is important to underscore that the variable (without the noise) follows an equation of its own. Thus, the typical model in the Kalman Filter context entails an observation equation and a state-transition (hidden) model.
The calibration of models was briefly considered. In particular, the advantages and disadvantages of calibrating a model. This was done relative to other common approaches such as econometric estimation.
Much emphasis was made on the importance of examining and presenting alternative scenarios and their implications on the predictions of the model. For example, central banks provide their predictions on several variables. The well-known dot plot from the Fed is an example. It is important to underscore that such predictions might be conditioned on different assumptions. In the case of the dot plot from the Fed, such predictions are conditioned on the policy that the president providing the forecast judges as the adequate one. Similarly, there are some fan charts on macroeconomic variables that are conditioned on the expected policy rate judged optimal or on the policy rate expected by the market.
Valuable explanations were given on how to present alternative scenarios and, especially, how to present their implications. This is related to the type of scenarios used, in particular in terms of their conditionality.
The instructors shared their experience and knowledge on the implementation of some of these models and their implication on the economic policy of some of these central banks. Participants found this session particularly illustrative and the learned much about their colleagues’ experiences in a candid atmosphere.
From the coding perspective of the course, much of it was about having participants obtaining hands-on experience on executing codes in Matlab, with an emphasis on how changing some parameters made results different and, more importantly, why.
In addition, as one of the contributions of CEMLA, the paper “TIIE-28 Swaps as Risk-Adjusted Forecasts of Monetary Policy in Mexico”, was presented. It is in the spirit of the course. How introducing a model implication in a forecast can improve and provide economic content to it.
Mr. Santiago Acosta-Ormaechea, Economist, Western Hemisphere Division, Institute for Capacity Development (ICD)
Mr. Jorge Restrepo, Senior Economist,Western Hemisphere Division, ICD;
Mr. Juan Pablo Medina, Expert, ICD
Mr. Santiago García-Verdú, CEMLA
Lecture: Monetary Policy Modernization, Forecasting and Policy Analysis System (FPAS) and the Course Overview
- Frameworks for Monetary Policy;
- Overview of Transmission Channels;
- FPAS Components: Databases, Monitoring and Reporting, Short- and Medium-Term Forecasting, Communication and Decision Making
Lecture: Introduction to a Small New Keynesian Model for Policy Analysis
Long Run Trends and Steady State;
A Brief Overview of Calibration
Workshop: Software and Model Codes
- Suite of Codes;
- Impulse Response Functions
Lecture: CPI Components and their Relative Prices
Mr. Jorge Restrepo
- Core, Food, and Energy: Inflation and Relative Prices;
- Shocks to CPI Components, First and Second-Round Effects, Policy Responses;
- Transitory and Permanent Shocks in Relative Prices
Lecture: Alternative Exchange Rate Regimes
- Exchange Rate Management;
- The Exchange Rate as an Operational Target;
- FX Interventions
Workshop: Data Transformation, Interpretation and Model Calibration
- Data Preparation, Univariate Filtration;
- Initial Conditions, Implications for the Inflation Outlook;
- Preliminary Calibration
Workshop: Model Properties: Different Policy Regimes
- Transmission of Shocks in the Model
Lecture: Estimating Long-Run Trends and Gaps
- PPP, and the Balassa-Samuelson Effect; UIP;
- Identifying Trends; Univariate Filtration Methods: Hodrick-Prescott;
- The Multivariate (Kalman) Filter
Workshop: Analysis of Trends and Model Calibration
- Understanding the Multivariate Filter;
- Model Filtration and Recalibration
Lecture: Scenarios: Risk Analysis under Uncertainty
- Macro Scenarios for Policy Dialogue;
- Baseline Forecast and Alternative Scenarios
Workshop: Scenario Formulation and Policy Analysis
- Baseline Forecast;
- Design of Alternative Scenarios;
- External Environment
Monitoring and Reporting; Short- and Medium-Term Forecasting; Communication and Decision Making in Participant’s Countries; Steps to Improve Processes
Presentation of Group Projects
- Key Takeaways from this Course;
- Extensions: Calibration vs. Estimation