Are economic tracking portfolios useful for forecasting output and inflation in Austria?*

*The opinions expressed do not necessarily reflect those of the Oesterreichische Nationalbank.

Abstract

We construct economic tracking portfolios from Austrian stock market returns, euro/dollar exchange rate changes and changes in the oil price to extract revisions of market expectations about future industrial production growth and inflation in Austria. The forecasting ability of the portfolios is evaluated in-sample and in a pseudo out-of-sample forecasting experiment. It turns out that the tracking portfolios track both target variables in-sample. The portfolios also help to forecast annual industrial production growth out-of-sample. The predictive ability of the tracking portfolios for inflation is rather low.

*The opinions expressed do not necessarily reflect those of the Oesterreichische Nationalbank.

Notes

*The opinions expressed do not necessarily reflect those of the Oesterreichische Nationalbank.

¹ In 2002, the Austrian stock market capitalization in relation to GDP was 14.8% compared to 86.0% in the US and 110.0% in the UK. Moreover, only about 7% of the Austrian population owned shares in 2002, compared to 23.4% in the US and 31.0% in the UK (OeNB, 2002).

² In 2004, for example, the share of exports of goods and services relative to GDP was about 52% and the share of imports was about 51%. Even after joining the EU, and later the EMU, about one-third of Austria's merchandise trade is with non-EU countries, primarily with Central and Eastern European countries but also with the US and Japan (OeNB, 2002). Hence, the euro/dollar exchange rate remains an important determinant for foreign trade.

³ Before 1999 the ATS/USD exchange rate was converted into a EUR/USD exchange rate.

⁴ A few words on the interpretation of regression 7a. Our focus is on the β coefficient that measures how well the ETP forecast changes in the expectation about the target variable. As mentioned, β should be 1 if the ETP captures them perfectly. On the other hand, the γ coefficient can take on any value. The reason is that is not a forecast of the lagged expectation of the target variable but a forecast of a linear combination of the lagged expectation of the ETP returns and the correlation of the lagged expectations of the target with the control variables.

⁵ The SEs in regressions 7a–7c ignore the fact that the forecasts are based on estimated parameters. This additional source of uncertainty can affect the SEs and hence the tests if the estimation sample of a forecasting model is small (Chong and Hendry, 1986). On the other hand, West and McCracken (1998) show that this additional source of uncertainty is asymptotically negligible in a recursive forecasting scheme such as ours. Hence, the HAC SEs, which are in any case only asymptotically valid, should not suffer from parameter estimation uncertainty.