|
It is important to note that the OECD's projections are based on a mixture of information from a number of economic and statistical models, as well as expert judgement.
Where are we in the business cycle?
One of the most difficult and most important steps in the projection process is the assessment of the current state of the business cycle in each country and also in the world as a whole. Here, macroeconomic model simulations are important for "setting the scene" at the start of each projection round. The first step is to look at what is new since the last projections were produced - such as changes in commodity prices (in particular the oil price), exchange rates and interest rates, fiscal trends and whether economic activity has developed differently in the recent past from what had been expected. With this new information, and using the previous set of projections as a starting point, the multi-country quarterly model, INTERLINK, provides preliminary estimates of the impact that this new information may have on the projections. This is a useful way to understand how our view of the world economy should change in the light of the new information, and of ensuring that changes in perspective can be made in a way that is consistent across countries.
The "scene-setting" also takes account of the projections from separate econometric models that make use of high frequency indicators to provide estimates of GDP growth in the major seven OECD economies in the two quarters following the last quarter for which data has been published. The objective is to help reduce the information gap and provide a starting point for the forecasting exercise which is consistent with recent high-frequency data. The Canadian model developed in Mourougane (2006) is a short term indicator based model and predicts quarterly GDP by efficiently exploiting all available monthly information. To this aim, monthly forecasting equations are estimated using the GDP series published every month by Statistics Canada as well as other monthly indicators. The procedures are automated and the model can be run whenever major monthly data are released, allowing the appropriate choice of the model according to the information set available. The most important gain from this procedure is for the current quarter forecast when one or two months of GDP data are available, with all monthly models estimated in the paper outperforming a standard quarterly autoregressive model in terms of size of errors. The use of indicators also appears to improve forecasting performance, especially when an average of indicator based models is used. Real time forecasting performance of the average model appears to be good, with an apparent stability of the estimates from one update to the next, despite the extensive use of monthly data. For the six remaining countries, the models developed in Sédillot and Pain (2003) combine information found to be statistically significant from both "soft" indicators, such as business surveys, and "hard" indicators, such as industrial production and retail sales, and use is made of different frequencies of data and a variety of estimation techniques. The use of regression techniques to identify indicator series that are closely related to GDP growth over the economic cycle as a whole differs from the longstanding approach used to produce the OECD Composite Leading Indicator series. The latter are constructed using a set of 5-10 variables for each country that have been observed to be closely related to past turning points in a proxy reference series such as GDP or, more typically, industrial production. An automated procedure has been developed allowing the indicator models to be run whenever new monthly data are released.
Characteristics and performance. The main findings from the model by Sédillot and Pain (2003) are the following:
-
For current-quarter forecasts, that is forecasts made at or after the start of the quarter to which they relate, estimated indicator models appear to outperform autoregressive time series models, both in terms of size of error and directional accuracy. This suggests there are clear gains from developing such models.
-
The main gains from the monthly approach start to appear once one month of data is available for the quarter being forecast. This is typically two to three months before the publication of the first official outturn estimate for GDP. This finding is in line with other empirical studies.
-
For one-quarter-ahead projections, the performance of the estimated indicator models does not appear noticeably better than that of time series models until one to two months of information become available in the quarter preceding the one for which the forecast is being made. However there are some modest gains in terms of directional accuracy from using the indicator models.
-
The most suitable model for any given information set and fixed forecast horizon varies across countries and time. For the current quarter, models with hard indicators alone, or combining hard indicators with survey information, outperform models using only survey data. The pure hard indicator model appears the most suitable for the United States, the euro area, Canada and Japan, whereas some form of combined model, either through estimation or through a consensus of the different model forecasts, appears more suitable for Germany, France, Italy and the United Kingdom. For the one-quarter-ahead forecasts, the inclusion of hard indicator data for the quarter in which the forecast is being made appears to add little to the information provided by surveys. Survey data appear to contain especially useful information in France and Italy.
There are limits to the ability of any estimated model to forecast the quarterly rate of GDP growth precisely. Even when a complete set of monthly indicators are available for a quarter, the 70 per cent confidence band (approximately one standard error) around any point estimate for GDP growth in that quarter is found to range from 0.4 to 0.8 percentage points, depending on the country or region. The degree of uncertainty around a point estimate is also found to widen as the forecast horizon lengthens. Moreover, forecasting errors can arise for many different reasons, including revisions to the initial published data for some high frequency indicators and inaccuracies in the projections of the incoming monthly data. The overall conclusion is that it is not optimal to always employ the same single, fixed-coefficient, indicator model for each country. Instead, it is preferable to have a suite of indicator models that can be updated automatically as new data appears. The model to be chosen from this suite at any given moment depends on the information then available.
Add-factor analysis
As country desks make their own forecasts, these are compared with what the INTERLINK model predicts based on historical relationships, the difference being "residuals" or "add-factors". Large deviations in the projections from the equation results point to the need to think about why historical relationships may have broken down and whether a breakdown would be temporary or permanent. Judgments about whether and how quickly historical relationships reassert themselves can have significant implications for the "story" behind the projections. On the other hand, large deviations may just encourage a re-appraisal of the projections. The results of econometric modelling both inside and outside the OECD influence the thinking of country desks. Econometric estimates are combined with local knowledge and judgment when preparing the projections, but not in any rigid or mechanical way. Add factor analysis is perhaps most intensively used in deriving the external trade projections, which need to satisfy the overall consistency requirements that changes in global exports of goods and services equal changes in global imports or that the world current account discrepancy remains broadly constant over the projection period.
Scenarios and simulations
An important part of the Secretariat's analysis is to consider the balance of risks around the central projections, and to explore the medium-term effects of various alternative scenarios. For this reason a medium term reference set of projections is regularly produced and INTERLINK is used extensively when considering the medium term impacts of these alternative, or "what if" scenarios. Over the past few years, scenarios have been conducted to illustrate issues such as long-term prospects for government budgets; the causes of widening current account imbalances; and the interactions between labour markets, social transfers and budget consolidation. In terms of short term analysis, simulations have been conducted recently, inter alia, with the view to show possible implications of sharp corrections in stock market prices; faster than expected growth in the United States (a Boom-Bust scenario); faster growth in the OECD area due to high cyclical synchronisation; the implications of a weaker euro, and the impact of higher oil prices.
The medium-term reference scenario is highly conditional on specific assumptions about policies and economic developments in OECD and non-OECD countries and regions. It is also intended to provide insights on the possible build-up or unwinding of specific imbalances and tensions in the world economy over the medium term, and the scope for policies to assist the adjustment process. The medium-term reference scenario extends the short-term projections on a consistent basis over a five-year horizon. By construction, it does not embody a view about timing and extent of future cyclical events. Instead, it assumes a pattern of growth such that the gap between actual and potential output as well as the gap between actual and structural unemployment are broadly eliminated in all OECD countries over the medium term. Commodity prices and exchange rates are kept broadly unchanged in real terms. Monetary policies are assumed to be directed towards keeping inflation low, in line with medium-term objectives.Fiscal policies are, for the most part, assumed to be broadly unchanged i.e. the cyclically-adjusted primary balance is held approximately constant. This implicitly implies that the authorities take measures to offset underlying changes in the cyclically-adjusted primary balance, such as those stemming from expenditures related to the ageing of populations in most OECD countries. These assumptions result in falling debt-to-GDP ratios in most countries.
INTERLINK model properties
The model links together a set of small to medium-sized quarterly macro models for each OECD member country - with reduced-form relationships for trade and balance of payments of the six non-OECD country groupings and for commodity prices. A key feature of the model is its treatment of the world economy as a coherent and integrated whole. This means that developments in individual domestic economies, international trade and exchange rates and financial flows are determined simultaneously and, as far as possible, on a globally consistent basis. Considerable emphasis is therefore given to international economic linkages through trade and financial conditions.
The model's overall structure reflects the range of its practical uses in the work of the OECD Economics Department. In the context of the OECD's regular macroeconomic analysis and projections exercise, the model and the econometric analyses that lie behind it perform a variety of functions including: an aid to the construction and co-ordination of individual country projections, the production of globally-consistent trade projections and the preparation of alternative projections and scenarios through model-based simulation exercises.
Although broadly neo-classical in terms of structural specification and equilibrium properties, the dynamic adjustment of output, employment and prices to long-term equilibrium in the current version of the model can be classified as being essentially "New Keynesian", reflecting the presence of real and nominal rigidities in wage and price setting. In particular such rigidities mean that following a shock to the economy there will be a protracted period of adjustment before equilibrium is restored and hence persistence in output and employment disequilibria.
|
