This article examines whether financial variables are useful as leading indicators of the output gap and mainland GDP growth. Financial variables may be leading indicators either because they (a) are priced on the basis of expectations, (b) affect the economy with a lag or (c) are published earlier and more frequently than GDP figures. Moreover, they are not subject to significant revisions. We find that house prices, equity prices, credit growth, money growth, real exchange rates, real short-term interest rates and the difference between long- and short-term interest rates can serve as leading indicators of GDP growth and/or the output gap. The output gap is most strongly correlated with growth in domestic credit to enterprises (lagged 0-4 quarters) and cyclical fluctuations in equity prices (lagged 2-5 quarters). We include effects of equity prices and enterprise credit in an econometric forecasting model of GDP. The model takes into account that equity prices and credit growth may influence each other and that changes in GDP may feed back to financial variables. The model fits well and has stable coefficients.

1 Introduction

Norges Bank sets the key rate on the basis of output and inflation forecasts. In the projection process, the Bank assesses how the key rate will influence these variables during the projection period. A solid assessment of both the current economic situation and developments in the next few quarters is essential to making sound projections for economic developments over a longer period. The short-term analysis is based primarily on current statistics and other information about the economic situation, including information from Norges Bank's regional network (2) and other surveys (3). The Bank also uses several models to project GDP growth in the next few quarters. (4)

The variables monitored by the Bank include developments in credit, money, house prices, equity prices, market rates and exchange rates. This article examines whether such financial variables (5) can be useful as leading indicators of GDP growth and the output gap. (6)

A number of arguments support the use of financial variables as predictors of the output gap and GDP growth in the next few quarters. First, measures of most financial variables are fairly accurate and they are not subject to significant revisions. Second, financial variables may be leading indicators of developments in the real economy. This may be because they are priced on the basis of expectations, because they affect the economy with a lag or because they are published earlier and more frequently than GDP figures. In efficient markets, equity prices, market rates and exchange rates are set continuously. Data on credit, money and house prices are updated monthly. House price figures are updated immediately after month-end, whereas data on credit and money are updated with a lag of roughly one month. By contrast, the national accounts are only published quarterly, with a lag of more than two months, and may be revised extensively (see e.g. Bernhardsen et al., 2006).

We discuss the data and possible relationships between financial variables and the real economy in Sections 2 and 3. In Section 4, we use a simple correlation analysis to assess whether financial variables can function as leading indicators of GDP growth and the output gap. In this analysis, we only consider the correlation between the output gap/GDP growth and one financial variable at a time. Since several of the financial variables appear to lead GDP growth and/or the output gap, we expand the analysis by estimating a model using several explanatory variables for GDP growth (Section 5). The model also takes into account that the financial variables may influence each other and that GDP may have feedback effects on the financial variables.

2 Financial variables as indicators and choice of data

2.1 Financial variables as indicators

The relationships between financial variables and the real economy are complex. Financial variables and the real economy may be driven by the same underlying forces, but they may also influence each other. Moreover, it may be difficult to differentiate between cause and effect. There is reason to believe, however, that some financial variables may be leading indicators of GDP growth and the output gap. In that case, it may be useful to employ these financial variables in forecasting.

We use correlation analysis and econometric methods to assess whether financial variables can function as leading indicators (information variables) of GDP growth and the output gap. This approach can be related to Astley and Haldane (1995) who write:

"The logic of information variables is that they need not have any well-defined structural relation with the final targets: they need only possess systematic, leading indicator information over them.... Of course, some of our results mar indeed have structural content."

Husebo and Wilhelmsen (2005) used correlation analysis to examine whether 30 macroeconomic variables lead, lag or coincide with the output gap. However, they do not consider any financial variables other than interest rates and exchange rates.

Our analysis can also be related to empirical studies of relationships between asset prices, interest rates and output growth (see e.g. Goodhart and Hofmann (2000), Mayes and Viren (2001) and English et al. (2005)). These studies show that asset prices can provide information about developments in output and prices. In the first study, the authors find that real equity prices, real exchange rates and real short-term interest rates are significant right-hand-side variables (with one lag) in a model for forecasting the output gap in Norway. English et al. (2005) also include different measures of credit and money to predict developments in output and prices.

2.2 The data

The output gap is estimated as mainland GDP at constant prices as a percentage of potential output. We use the same measure of the output gap that was presented in Inflation Report 1/06. In section 3, we also present gaps for private consumption, housing investment and mainland business fixed investment. These gaps are estimated as the real value of these variables (adjusted for seasonality and noise) as a percentage of the variables' estimated trends. The trends are estimated using a Hodrick-Prescott filter ([lambda]=40000).

Table 1 presents an overview of the financial variables examined in this article. The series for credit, money, house prices and equity prices have been deflated by the CPI-ATE (consumer prices adjusted for tax changes and excluding energy products). In our examination of potential relationships between financial variables and the real economy in section 3 and in the correlation analysis in section 4, we have adjusted GDP and the financial variables (except interest tares) for noise and seasonality (7) to ensure that these factors do not influence results and conclusions. We have also made seasonal adjustments and filtered out noise in the CPI-ATE. We employ the four-quarter rise in the CPI-ATE (unadjusted) to estimate real short-term interest tares. Thus, we measure all the financial variables in real terms, with the exception of the difference between 5-year nominal government bond yields and 3-month nominal money market rates.

In sections 3 and 4, we use four-quarter growth in aggregate figures for real credit and real money. We include both the level of the series and the four-quarter rise in real house prices and real equity prices. We de-trend the level series to express cyclical developments. The trend in real house prices seems to fluctuate over time. We have estimated this trend using a Hodrick-Prescott filler ([lambda]=40000). The real equity prices, on the other hand, appear to rise by a constant percentage over time, which is the same as saying that the logarithm of real equity prices has a linear trend. We have estimated the trend of the logarithm of real equity prices using the linear least square method. Finally, we have estimated a real house price gap and real equity price gap which express real house prices and real equity prices as a percentage of trend. We also include the level of the real exchange rate and its four-quarter rise. Since the real exchange rate is stationary, we have not de-trended the level series.

The econometric analysis in section 5, however, is based solely on unadjusted variables, i.e. variables that have not been de-trended or adjusted for noise or seasonality. Instead, we control for such factors by including a linear trend in the model, by including seasonal dummies and by allowing the inclusion of variables that are lagged several quarters.

We confine the correlation analysis in section 4 to the period 1993-2005. This is because it is likely that the relationships between the real economy and financial variables have changed over time, making information from earlier periods less relevant for forecasting future developments. Figures for the 1980s are influenced by the liberalisation of money, credit and capital markets and other economic policy changes. Moreover, there was a banking crisis in Norway in the period 1988-1993. Since 1993, the economic situation has been more stable. It is therefore likely that the relationships between the real economy and financial variables have been more stable since 1993 than over a longer period.

Nevertheless, we use data from 1990 when we estimate a simultaneous equation model in section 5. The background for this is that we use a model with several variables and lags, and therefore need somewhat longer data series (i.e. several degrees of freedom) to estimate fairly precisely the coefficients in the model. This may be justified by the fact that we can take structural breaks into account in an econometric study, thus benefiting from data for a somewhat longer period.

3 Potential relationships between financial variables and the real economy

This...