Evaluation of Norges Bank's projections.

• Author: Jore, Anne Sofie
• Position: Statistical Data Included

This article analyses Norges Bank's projections for 1998, published in autumn 1996 and autumn 1997 respectively. Compared with earlier articles on this subject, we go one step further in the analysis by examining projections with a two-year horizon. It is also important to be able to analyse economic developments further ahead because decisions on economic policy will influence the economy more than one year ahead. The evaluation focuses on the contributions from erroneous assumptions concerning exogenous variables, such as public demand and externally generated inflation. A comparison with two-year projections from Statistics Norway is also included.

All in all, our forecast errors for 1998, presented in autumn 1997, were smaller than the forecast errors for earlier years. There is no clear evidence that the projections would have been substantially better if we had known actual movements in exogenous variables in advance. The forecast errors increase for some variables and are reduced for others.

In principle, one would expect the forecast errors in the projections presented at the end of 1996 to be greater than the errors in the projections presented at the end of 1997. This is confirmed for important real variables such as mainland demand and employment. For wage and price inflation, however, the forecast errors are smallest in the earliest projections.

The projections for 1998 show a larger forecast error in the projection for consumer price inflation than in earlier years as a result of lower-than-expected imported price inflation. Towards the end of 1998 it was evident that consumer price inflation was lower than implied by exchange rate movements. In the consumer price equation, we have since 1999 used an expanded import-weighted exchange rate index, which includes the currencies of several countries in Asia. The actual forecast error is largely due to the fact that the effects of the crisis in Asia were erroneously evaluated in two ways. First, the crisis had a surprisingly strong effect on international prices. Second, the depreciation of Asian currencies contributed to a stronger effective krone exchange rate than implied by traditional exchange rate indices.

Analyses of forecast errors are an important part of the work aimed at making the projections more accurate. At the same time, continuous efforts are made to improve the analyses, in the form of both short-term cyclical analyses and further development of the model. The analysis of forecast errors in the projections for 1998 confirms that there is a potential for improving the RIMINI model. Various types of shock, such as international financial turbulence and fluctuations in the oil price, will nevertheless continue to be a source of uncertainty in economic developments.

Norges Bank's analyses of developments in the Norwegian economy are published in the inflation reports four times a year. Projections for developments in the Norwegian and international economy are an important basis for the formulation of monetary policy. In addition, the analyses are used as a basis for advice on the orientation of economic policy in general. The macroeconomic model RIMINI, developed in Norges Bank's Research Department, has been the principal tool for the Bank's analyses since 1994. RIMINI is an econometric model with nearly 400 equations. About 70 of the equations are estimated on the basis of historical data, while the remaining equations are definitional relationships.

Norges Bank aims to produce the best possible projections for the Norwegian economy. It is important that errors are revealed in order to improve the model and the way in which the model is used. This in turn will result in more accurate projections. We also wish to compare Norges Bank's projections with those of other institutions.

Norges Bank places considerable emphasis on transparency and the availability of its forecast work. This work also includes analyses of earlier projections. The projections are based on a model that is publicly known, and the Bank's use of the model is published. The purpose is to provide others with the basis for evaluating how we have arrived at our projections and how accurate they are. Systematic evaluation also places greater demands on consistency and documentation of the projections in the Inflation Report, which in itself will improve the quality of the analysis.

Norges Bank intends to publish analyses of its projections annually. So far, such analyses have been published in articles in Economic Bulletin 1998/1 (Jore 1997) and 1999/2 (Jore 1999). In addition to detailed analyses of Norges Bank's projections, these articles presented summary measures of forecast errors for the Ministry of Finance and Statistics Norway, showing that the three institutions' projections were almost equally accurate. The articles also showed that the amplitude of the cyclical upturn was considerably underestimated by all the institutions. In an article published in Penger og Kreditt 1996/1 (Madsen 1996), Norges Bank's projections for the years 1987-1994 were compared with those of other institutions. This article also concluded that Norges Bank's forecasts were about as accurate as projections from other institutions.

The article in Economic Bulletin 1999/2 primarily analysed the projections for 1997. The projections for 1998 were examined briefly on the basis of preliminary national accounts figures for 1998. This article is based on revised national accounts figures published in September and provides a more thorough analysis. This time the analysis has been expanded to include forecast errors in the projections presented at the end of 1996.

Finally, we compare Norges Bank's projections two years ahead with corresponding projections from Statistics Norway.

Sources of forecast errors

The macroeconomic model RIMINI has been the main tool for Norges Bank's projections since 1994. In the model, important economic relationships are represented by quantified empirical relationships. The model also ensures consistency in that demand equals supply in the various markets.

There are important sources of forecast errors in an economic model. The model's coefficients are quantified on the basis of historical data. There are uncertainty intervals around each coefficient, and the interaction between many equations in a model increases the uncertainty around each variable. Changes in the functioning of the economy may not be captured in the quantification of coefficients. Finally, there are areas where the model does not sufficiently take into account important economic...