Sources of Predictability of European Stock Markets for High-technology Firms

Abstract

The paper reports on studies of return predictability of stock indexes of blue-chip firms and high-technology firms in Germany, France and the UK during the second half of the 1990s. Return predictability was measured in terms of first-order autocorrelation coefficients, and evidence was found for the return predictability of stock indexes of high-technology firms, but not for the return predictability of stock indexes of blue-chip firms. These findings suggest that a candidate for explaining the economic sources of the return predictability of these stock indexes of high-technology firms is transaction costs in the form of the costs of gathering and processing information in new technological fields.

Keywords:

• Stock markets
• return predictability
• high-technology firms

Acknowledgements

Financial support from the European Commission: DG Research in cooperation with DG ECFIN and DG ESTAT (Contract No. SCS8-CT-2004-502642) is gratefully acknowledged. We thank three anonymous referees for most helpful comments. We also thank Giovanni Urga and partici-pants of the first research meeting and the first conference of the FINPROP consortium, and participants of the Global Finance Conference 2005 for helpful comments. The usual disclaimer applies.

Notes

¹For recent surveys of the empirical literature on the efficient markets hypothesis, see Fama (1991) and Cochrane (1999). For recent evidence on the efficiency of the German stock market for smaller high-technology firms, the so-called Neuer Markt, see Bohl and Reitz (2006).

²In addition, efficient stock markets for high-technology firms are important for the development of other financial market segments such as venture capital markets (Black and Gilson, 1998; Bascha and Walz, 2001).

³Studying the implications of incomplete information for capital market equilibrium has a long tradition in the finance literature. See, for example, Merton (1987).

⁴Datastream does not disseminate stock market data beyond two decimals. This may give rise to rounding errors that could result in spurious return predictability when stock prices are low. However, as we shall report in Section 2.3, our empirical results suggest that returns predictability was relatively constant in the case of the Nemax50 and the Nouveau Marché, and declined over time in the case of the Techmark100. Because stock prices significantly increased in 1999 and in the first months of 2000 and, thereafter, started declining, the predictability of returns should have increased over time if predictability were spurious because of rounding errors.

⁵It is important to note that the Neuer Markt was closed at the end of 2002. For this reason, the Deutsche Börse has begun on 24 March 2003 to calculate the TECDAX index. This index comprises the 30 largest high-technology firms in terms of capitalization and trading volume in the so-called Prime Standard. For calculating this index, only firms are considered that are not included in the DAX30. Because the closing of the Neuer Markt implies a break in our index of the Nemax50, we used a shorter sample period 1/1/1998–31/12/2002 to check the robustness of our estimation results. The estimation results for the shorter sample period are similar to those we report in this paper. They are not reported, but are available from the authors upon request.

⁶To estimate the model, we used the algorithm proposed by Harvey, Ruiz and Sentana (1992). We used Gauss 3.6 to estimate the model. To this end, we used the computer programs described in Kim and Nelson (1999). Harvey (1992) and Kim and Nelson (1999) provide detailed descriptions of time-varying parameter models.

⁷Because the sampling distribution of the parameters is nonstandard, care must be taken when conducting tests for significance (see Harvey 1992, p. 236). If the point estimate of a parameter is zero, then the corresponding coefficient is a constant, and conventional statistical theory can be used to conduct tests for significance. If the point estimate of a parameter is nonzero, then the corresponding coefficient varies and its significance can be graphically analyzed.

⁸One can either use the filtered or the smoothed estimates of the model’s coefficients to measure the predictability of returns. The difference between the two lies in the information set one uses (Kim and Nelson, 1999). Filtered estimates are based on information available up to period \textit{t}. Smoothed estimates are based on all available information in the entire sample. We report filtered estimates because, in any given period \textit{t}, a stock market participant can only use information up to time t for making inferences about the time-varying predictability of returns.

⁹We also analysed a TARCH effect when we estimated the time-varying parameter model described in Section 2.2. However, when we estimated a time-varying parameter model that features a TARCH effect, we encountered convergence problems in the case of some indexes. For this reason, we estimated a time-varying parameter model without a TARCH effect.

¹⁰We used WINRATS 6.1 to estimate Equations (4) and (5). We used the simplex algorithm to start, and the BFGS algorithm to finish the maximization of the loglikelihood function of the model.