Chapter

Chapter 4. Nuts and Bolts of a Forecasting and Policy Analysis System

Author(s):
Tobias Adrian, Douglas Laxton, and Maurice Obstfeld
Published Date:
April 2018
Share
  • ShareShare
Show Summary Details
Author(s)
Rania Al-Mashat Kevin Clinton Douglas Laxton and Hou Wang

Inflation-forecast targeting involves using a wide range of information in order to obtain the best forecasts for inflation and the economy. The staff must extract the underlying pressures on inflation and the economy from data that may be conflicting and noisy. —D. Laxton, D. Rose, and A. Scott (2009)

Central bank staff need to be able to organize the efficient and thorough provision of economic information to enable monetary policy committees to make good decisions. That is the purpose of the Forecasting and Policy Analysis System (FPAS). It is intended to help monetary policy committee meetings focus squarely on the strategic, medium-term outlook. Forecast meetings with policymakers run the risk of getting sidetracked into technical details or wiggle in the latest numbers that have little importance in the grander scheme. The FPAS facilitates the big picture, employing a macroeconomic model with standard properties that broadly accord with the view of the monetary policy committee on how the economy works.

The main purposes of the forecast output are these:

  • A model-based, macroeconomic forecast. The forecast provides an economically coherent view of the short- to medium-term outlook, with a baseline and alternative scenarios. Whereas the main forecast exercise would be quarterly (the periodicity of the national accounts), there would be updates for each monetary policy committee meeting. The most crucial part of the forecast is an endogenous future path for the short-term interest rate.
  • Measures of uncertainty. The baseline forecast includes model-consistent confidence intervals for key variables (shown as bands or fan charts). These reflect a normal range of variability of random shocks implied by the historical data as well as parameter uncertainty.
  • Risk assessments. The output of the forecasting team includes risk assessments to the baseline forecast in the form of alternative scenarios for specific assumed shocks, and options for the policy rule.

In view of the complexity of the economy, the large volume of relevant data, and the need to draw on expertise across departmental lines, production of the forecast and the related analyses make substantial demands on resources. The FPAS works on a calendar determined by the announced policy-decision dates of a particular monetary policy committee. Speed and accuracy are crucial, so that policymakers may base decisions on the most up-to-date information available. This requires a streamlined, robust production system—even then, the forecast takes several weeks to produce.

The Forecasting and Policy Analysis System

An effective system has the following components:

  • Full-time forecast team. The demands on staff are ongoing in practice. Policymakers often ask for quick model-based responses to questions arising from current developments or policy debates, in addition to scheduled forecasts and updates. All this can be handled efficiently only by a team with full-time responsibility. There are, however, advantages to rotating staff between the forecasting team and other groups—to avoid burnout, to broaden work experience, to develop human capital, and to improve workforce flexibility.
  • Core quarterly forecasting model. This core model should be small enough that forecasts with updated or alternative assumptions can be quickly and efficiently produced. This core model contains a standard transmission mechanism for monetary policy, and the forecast path for the short-term interest rate is endogenous.
  • A suite of satellite and ancillary models. Satellite models would take results generated by the core model as inputs to derive projections of sectoral detail of interest to policymakers. A central bank would normally employ many ancillary models, results from which may be used as inputs to shape the dynamics of the core model, or the assumptions that underlie baseline or alternative scenarios. For example, sectoral specialists often use single-equation indicator models in their near-term monitoring.
  • Schedule of deadlines and meetings for each forecast exercise. An example would be this sequence:
    • Issues meeting with representation from senior management to define the main influences in the upcoming outlook and to clear up technical issues regarding data, models, and the like.
    • Near-term forecast meeting.
    • First round projection incorporates the most recent data and allows diagnosis of previous forecast errors.
    • Several iterative rounds incorporate new model modifications, new external assumptions, and new, monitored, near-term forecast.
    • Forecast meeting presents baseline and risk assessments to monetary policy committee members—who may request alternative scenarios.
    • Monetary policy committee policy decision meeting.
    • Writing deadlines for the internal forecast report and for the published monetary policy report, which includes key variables from the forecast.
    • A “postmortem” to assess the preceding process and discuss avenues for improvement.
  • Near-term (current and next quarter) forecasting subsystem. Expert judgment easily outperforms simple model extrapolations of GDP components, for short-term forecast accuracy. Judgmental (that is, monitored) short-term forecasts exploit simple empirical correlations, for example, leading indicator models, as well as special factors and nonquantitative knowledge bearing on how current developments are likely to unfold. The forecasting team would use a monitored near-term forecast to set initial conditions for the model-based medium-term forecast. Over horizons longer than a year, model-based forecasts are more useful, because they include the effects of endogenous policy responses and other macroeconomic feedback.
  • Source of external assumptions about the international economy. These might be taken from the published forecasts of the IMF or other international institutions. The international departments of some central banks, however, develop their own views of the external outlook—for example, the Bank of Canada closely watches the US economy and commodity prices because of their dominant importance for the Canadian outlook.
  • Reporting database and network. The forecast team updates the database weekly. It circulates a brief reporting package that highlights revisions to previous data and the new numbers. The arriving data are incorporated into updates of the monitored near-term forecast. Everyone involved in the process, across all departments, has immediate access to the relevant information, including the monitoring of the near-term forecast.
  • Adequate information technology. This would include software for model building and simulation, for database updates and automated reports, and for the archiving of vintage models and databases. In view of staff turnover and the risk of loss of institutional memory, archives of model documentation are also essential.
  • Explicit accounting for revisions from one forecasting exercise to another. These allow decomposition of the revisions into the various con-tributing factors:
    • Modifications of the model
    • Revisions to historical data
    • Revised views about long-term equilibrium values of variables like potential output, the real interest rate, and the real exchange rate
    • Unexpected changes in exogenous variables

Internal gains from a well-structured system are substantial. Staff input directly affects policy advice (most directly, in the form of the forecast interest rate path). As a result, staff members are better motivated and, knowing the direction of policy, work more effectively to serve the needs of policymakers. This helps keep the focus on the strategic medium-term issues at meetings with policymakers. Moreover, as staff members become better attuned to policymaker thinking, over time they develop ways to improve the process (for example, through better presentation materials or improved models). The system implies frequent communications both horizontally (across divisions) and vertically (between staff and senior management).

The Core Model

The core forecasting model varies in size, complexity, and theoretical specification from one central bank to another. In addition, the structure and calibration reflect the different economic features of each country. One common requirement, however, is that under simulation the model must exhibit standard macro-economic properties that are plausible to policymakers. A second is production efficiency. The model should not be so large or complex that maintenance absorbs a lot of resources or that the derivation of forecasts consumes a lot of time. This argues for a model of modest size, using readily available and easily updated data.

At a minimum, the model produces forecast paths for inflation (core and headline), for GDP growth, for the output gap (actual GDP minus potential or full-employment GDP), for the short-term interest rate (the policy instrument for the purposes of the model), and for the exchange rate.

The simplest core model is organized around four key behavioral equations: an aggregate demand function for output, an expectations-augmented Phillips curve for inflation, a modified form of uncovered interest parity—which allows for variable country risk premiums—for the exchange rate, and a monetary policy reaction function for the interest rate. Other equations include identities, definitions, auxiliary analytical and reporting equations, statistical processes for trend variables (potential output and trends in the real exchange rate, interest rate, and country risk premium), and equations for international variables.

The aggregate demand equation explains the cyclical movements in output as a function of the real interest rate, real exchange rate, and foreign demand fluctuations. Higher real interest rates make borrowing for current consumption and investment costlier; this results in reductions in private outlays and overall demand. World demand is a key external driver of open economies. Movements in the real exchange rate also affect demand (both domestic and foreign) for domestic output by changing the price of locally produced goods relative to the rest of the world.

The Phillips curve embodies the dynamics of core consumer price inflation. The latter refers to the average rate of increase in sticky prices, which move over time rather than jump immediately, in response to pressures of supply and demand. In practice, consumer price inflation is defined to exclude energy and unprocessed food items, whose prices are subject to high short-term volatility. Over the cycle, according to the Phillips curve, it is output gaps—excess demand or excess supply—that drive core inflation. But with a modern expectations-augmented curve there is no lasting trade-off between output and inflation. In the long term, under inflation targeting, a policy reaction function ensures that the inflation rate converges to the fixed official target. The equation would also capture certain short- to medium-term factors. These include intrinsic persistence in inflation deriving from rigidities in adjustments of wages and prices, pass-through from changes in the real exchange rate (which affect the prices of imported consumer goods directly), and in noncore components (energy and food prices).

Uncovered interest parity has the exchange rate adjust to current and expected interest rate differentials (that is, the differences between the short-term rate of interest at home and abroad) and to investor perceptions captured in a risk premium. The longer a given interest differential is expected to persist, the greater the effect on the current exchange rate. For short-term forecasts, the country risk premium may for simplicity be set exogenously, based on judgment and on considerations not incorporated directly in the model. However, for long-term simulations, the model should have an endogenous risk premium that responds to changes in external and government debt. For emerging market economies, the evolution of the real exchange rate would also be affected by differential productivity trends.

The monetary policy rule (or reaction function) describes the systematic behavior of policymakers. The policy instrument, which is to say the short-term interest rate, reacts to actual and anticipated deviations in the inflation forecast and to domestic excess demand or excess supply (the output gap). The policy rule embodies some degree of interest rate smoothing, such that changes in the rate are typically distributed over time and not executed suddenly in one step. This gradualism is a consequence of the uncertainty associated with economic analysis, economic forecasts, and the evaluation of economic cycles; it also helps clarify the intent of the central bank’s policy actions.1 Notwithstanding the smoothing, the policy rule guarantees that core inflation will return to the target rate in the medium term following any shock. It is therefore a crucial foundation for the nominal anchor to the economy.

One may distinguish several transmission channels for the effects on domestic output and core inflation in such a model. The real interest rate affects the output gap, with a distributed lag, both directly (the internal channel) and through its impact on the exchange rate (the external channel). The exchange rate channel in the model has three distinct aspects: (1) direct, through imported goods in the consumer price index basket; (2) indirect, through prices of intermediate imported goods; and (3) expenditure switching, where the real exchange rate redirects spending toward, or away from, domestic production and therefore affects the output gap. In turn, changes in the output gap imply medium-term variations in the core rate of inflation.

Expectations play an important role in all of this, as is evident from the discussion in the previous chapter. Variations in the expected medium-term inflation rate affect the real interest rates at which households and firms borrow and save, and the behavior of the exchange rate. In addition, the aggregate demand equation may contain a forward-looking element, relating current investment and durable consumption decisions to expectations of future output.

The core model can be adapted to inflation-targeting regimes that are in the process of establishing credibility following years of inflation drift. For monetary policy in this phase of development, a credibility-building process may be added to the standard model, allowing the central bank to earn credibility over time if it succeeds in achieving announced targets—or vice versa if targets are missed.

Typical models embody model-consistent forward-looking expectations such that current behavior anticipates the model’s own predictions for future periods. This allows analysts to explore complicated interactions between the private sector and the authorities. For example, simulation experiments can illustrate in numerical terms the implications of a policy credibility problem or the results of alternative strategies in situations where the reaction of the public’s expectations to a policy action play a key role in policy transmission.

Numerical calibration of the coefficients of a core FPAS model draws on a wide range of evidence and theory. Calibration methodology contrasts with the traditional econometric approach to macro modeling, which was to estimate each equation individually in an attempt to uncover the data-generating process.2 The objective of calibrating the model is a structure that in simulation yields economically plausible results and corresponds broadly with the data.

A close fit to historical data is not a requirement. Theoretical priors may take precedence over empirical estimates in the calibration of an equation if the estimates imply untenable properties for the whole system. This is not a trivial risk in view of the well-known conceptual difficulties of identification and estimation for systems of equations, not to mention the vagaries of the available data series, which are often short and affected by structural changes (Berg, Karam, and Laxton 2006).

Dynamic Stochastic General Equilibrium Methodology

A relatively small core model on the above lines has been used in many central banks to good effect, and it may still be adequate in most countries. However, there has been a trend in the more advanced inflation targeters toward more sophisticated, dynamic stochastic general equilibrium (DSGE) models, which are based on explicit optimizing behavior. These models require a much heavier input of specialized resources, and policymakers may find that their complexity presents a problem. The approaches can be blended as there is no conceptual inconsistency in their design. For example, DSGE models can be used to explore the longer-term implications of changes to government budgets, taking into account the dynamics of the public debt. Or they may be applied to certain key sectors (a case in point is agriculture in India). DSGE models are also useful for studying trends in real exchange rates or other unobservable variables like potential output.3 The results would go as inputs to shape the dynamics of forecasts and simulations in the core model.

While the core model might initially abstract from some important macro-financial linkages and nonlinearities, it will be important to incorporate these features over time. Standard DSGE models now include financial shocks as well as financial accelerators that provide much stronger propagation mechanisms.4Chapter 7, for example, emphasizes the importance of incorporating financial information in both the baseline forecast and confidence bands. Other models have been developed to include important nonlinearities for doing monetary policy analysis.5

Looking ahead, as technical resources improve and policymakers demand for rigorous analysis increases, one might expect increased use of DSGE methodology in core FPAS models. Simpler approaches will continue to be used for various purposes in the suite of models, including for the core model in some cases. The likelihood is that the implementation of inflation-forecast targeting will continue to rely on a pragmatic blend of approaches to modeling. Moreover, the core model, regardless of the exact methodology employed, will continue to embody macroeconomic principles that can be understood by a broad, informed audience.

References

    AlichiA.R.Al-MashatH.AvetisyanJ.BenesO.BizimanaA.ButavyanR.FordN.GhazaryanV.GrigoryanM.HarutyunyanA.HovhannisyanE.HovhannisyanH.KarapetyanM.KharaishviliD.LaxtonA.LiqokeliK.MatikyanG.MinasyanS.MkhatrishviliA.NurbekyanA.OrlovB.PashinyanG.PetrosyanY.RezepinaA.ShirkhanyanT.SopromadzeL.TorosyanE.VardanyanH.Wang and J.Yao. Forthcoming. “Estimates of Potential Output and the Neutral Rate for the U.S. Economy.” IMF Working PaperInternational Monetary FundWashington, DC.

    • Search Google Scholar
    • Export Citation

    ArgovE.N.EpsteinP.KaramD.Laxton and D.Rose. 2007. “Endogenous Monetary Policy Credibility in a Small Macro Model of Israel.” IMF Working Paper 07/207International Monetary FundWashington, DC.

    • Search Google Scholar
    • Export Citation

    BergA.P.Karam and D.Laxton. 2006. “A Practical Model-Based Approach to Monetary Policy Analysis—Overview.” IMF Working Paper 06/80International Monetary FundWashington, DC.

    • Search Google Scholar
    • Export Citation

    ChristianoL. J.R.Motto and M.Rostagno. 2014. “Risk Shocks.” American Economic Review104 (1): 2765.

    ClarkP.D.Laxton and D.Rose. 2001. “An Evaluation of Alternative Monetary Policy Rules in a Model with Capacity Constraints.” Journal of Money Credit and Banking33 (1): 4264.

    • Search Google Scholar
    • Export Citation

    FreedmanC.M.KumhofD.LaxtonD.Muir and S.Mursula. 2009. “Fiscal Stimulus to the Rescue? Short-Run Benefits and Potential Long-Run Costs of Fiscal Deficits.” IMF Working Paper 09/255International Monetary FundWashington, DC.

    • Search Google Scholar
    • Export Citation

    LaxtonD. and P.Pesenti. 2003. “Monetary Policy Rules for Small, Open, Emerging Economies.” Journal of Monetary Economics50 (5): 110946.

    • Search Google Scholar
    • Export Citation

    LaxtonD.D.Rose and A.Scott. 2009. “Developing a Structured Forecasting and Policy Analysis System to Support Inflation-Forecast Targeting (IFT).” Bank of Canada Working Paper 09/65Ottawa.

    • Search Google Scholar
    • Export Citation

    LaxtonD.D.Rose and R.Tetlow. 1993. “Monetary Policy, Uncertainty and the Presumption of Linearity.” Bank of Canada Technical Report No. 63Ottawa.

    • Search Google Scholar
    • Export Citation

    WoodfordM.2003. “Optimal Interest Rate Smoothing.” Review of Economic Studies70 (4): 86186.

1

Woodford (2003) justifies interest rate smoothing on the grounds of improved signal extraction. Incremental changes spread over time allow the public to discern more easily the intent of policymakers. A given change in the policy rate would then result in more predictable changes in longer-term interest rates. In contrast, frequent large changes in the policy rate would create noise and would not reliably be followed by the desired changes in longer-term rates.

2

The two approaches are not irreconcilable. Bayesian methodology provides a synthesis.

3

See Laxton and Pesenti (2003) for a model that can be used to study trends in the real exchange rate caused by productivity catch-up in the traded-goods sector. Alichi and others (forthcoming) provide a multivariate filter for estimating potential output and the neutral rate for the US economy.

4

Freedman and others (2009) and Christiano, Motto, and Rostagno (2014) provide structural DSGE models with financial accelerators.

5

See Clark, Laxton, and Rose (2001) for a model with capacity constraints, and Argov and others (2007) for a model with endogenous policy credibility. See Laxton, Rose, and Tetlow (1993) for a discussion of the implications of falsely presuming linearity in the Phillips curve.

    Other Resources Citing This Publication