An analysis of trading strategies in eleven European stock markets

Abstract

In recent years, the validity of the weak form efficient market hypothesis (EMH) has been called into question as several studies have uncovered evidence that technical trading rules have predictive ability with respect to both developed and emerging stock market indices. This study analyses the forecasting power of 2 of the most popular trading rules using index data for a selection of 11 European stock markets over the January 1991 to December 2000 period. The findings indicate that the emerging markets included in this paper are informationally inefficient; these markets displayed some degree of predictability in their share returns, although the developed markets did not. Furthermore, the results point to large differences in the performance of the rules examined; while small size filters consistently outperformed the buy-and-hold strategy in the emerging markets examined even after the consideration of transaction costs, the performance of the moving average rules was erratic and varied dramatically from market to market.

Keywords:

• Trading rules
• emerging markets
• market efficiency

Notes

1. For example, studies by Fama and Blume (1966), Dryden (1970) and Sweeney (1988, 1990) focus on the profitability of filter rules in either the USA or UK, while Jennergren (1975), Huang (1995) and Hunter (1998) focus on the viability of the filter technique in Sweden, Taiwan and Jamaica, respectively. Studies which concentrate on the profitability of the moving average rule include Ratner and Leal (1999) and Gunasekarage and Power (2001). Other studies have adopted a wider focus by examining different classes of trading rule (the moving average rule and the trading range breakout rule). However, these studies have tended to focus on the applicability and profitability of these rules in a single stock market, typically either the USA or UK. For example, see Brock et al., (1992), Hudson et al., (1996) and Bessembinder and Chan (1998).

2. Van Horne and Parker (1967) also examined the performance of a modified version of this rule where the moving average was weighted to place greater emphasis on more recent prices. However, this modification earned less profit than its unweighted counterpart in every instance. A follow-up article by Seelenfreund et al. (1968) employed a quadratic approach to smooth prices but again reached the same result that the buy-and-hold approach dominated the trading rule, especially once commission costs were taken into account.

3. Noting that the shorting part of the filter strategy was loss-making in the Fama and Blume study, Sweeney's rule considered only long equity positions, thus raising the measured return to the filter. Of course, in doing so, the transaction costs involved in shorting are eliminated. A second innovation in the Sweeney strategy relates to the dataset. Rather than including all 30 Dow Jones Industrial shares in the analysis, Sweeney's approach was to include only those shares that demonstrated potential. In this respect, the success of his strategy lay in the fact that the ‘winner’ shares in the Fama and Blume study persisted as winners in later periods.

4. These studies also examined the trading range breakout rule. Only the results for the moving average rule are documented here in order to provide a comparison with the analysis of this study.

5. On examining the Hong Kong stock market however, Coutts and Cheung (2000) found that both the moving average rule and the trading range breakout rule offered potentially profitable trading strategies, exclusive of transaction costs, over very short periods of time.

6. The results for the moving average strategy relate to the fixed length moving average rule.

7. The authors also noted that break-even costs are not homogenous across markets. For example, they computed that, for the emerging markets of Malaysia, Taiwan and Thailand, break-even costs were higher, at an average of 2.32%.

8. Specifically, the following eleven indices were examined: the Hex index (Finland), the CAC-40 (France), the DAX index (Germany), the Athens SE index (Greece), the BUX index (Hungary), the ISEQ index (Ireland), the MIB Storico index (Italy), the BVL index (Portugal), the Madrid SE index (Spain), the ISE National index (Turkey), and the FTSE All Share index (UK). These indices were selected on the grounds that they are comprehensive indices which track the movements of many or all of the shares traded on the respective markets. They therefore reflect a broad industrial base, and as such, are ideal for investigating the applicability and validity of trading rules.

9. The inclusion of the Athens bourse in the Morgan Stanley Capital International index series of developed markets on 31 May 2001 finally marked the promotion of Greece from emerging to developed market status, following its entry to the euro-zone in January 2001. However, the country's promotion from an emerging to a developed economy has highlighted weaknesses in the market, as international fund managers have shown less interest in Greece since its promotion; forecasts that Greece would attract inflows of between $500 m to $1.5 bn from developed market funds have proved wildly optimistic, while emerging market funds have withdrawn an estimated $1 bn, triggering a sharp fall in stock market prices (Financial Times, 2001a, b).

10. Portugal has recently been upgraded from emerging to developed market status. In particular, indices such as Morgan Stanley's MSCI Europe indices have recognised Lisbon as a developed market since December 1997, while the IFC promoted the market out of the emerging category in April 1999 (Financial Times, 1998, 1999a). Nevertheless, the market is labelled as ‘emerging’ in this study due to its perception among investors as belonging to the emerging category (Financial Times, 1999b).

11. The size of the filter is of crucial importance to the profitability of this trading rule. In particular, the more stringent the filter, the fewer the losses that are made, but also the smaller are the gains from any move that exceeds the filter size. There is therefore a trade-off between the reliability of information and the cost of information; the more stringent the filter, the higher the reliability, but the more of the move that is sacrificed in identifying it both in getting in and getting out. For example, as the filter size is increased, the number of transactions is reduced and losses on small moves are eliminated. However, gains on large moves are also reduced, and some moves which would yield gains with a small filter will yield losses with a large filter (Alexander, 1964).

12. Transaction cost data for each of the 11 sample markets were obtained from Domowitz et al. (2000). This paper provides estimates of one-way equity trading costs for 42 countries. Direct costs of trading, such as broker costs and taxes, were estimated to be, in basis points, 27.9 for Finland, 22.8 for France, 24.3 for Germany, 58.2 for Greece, 74.8 for Hungary, 106.0 for Ireland, 26.3 for Italy, 43.8 for Portugal, 32.5 for Spain, 41.0 for Turkey and 39.3 for the UK.

13. Similarly, it is assumed that the investor can only buy after a sell transaction.

14. The analysis reported in this paper was undertaken in foreign currency terms. That is, the paper adopts a UK perspective, rather than a local perspective, in order to measure the gains that could accrue to UK institutional investors who employ technical trading rules to exploit any predictability in share prices. This approach is not without its limitations. First, the analysis in this paper implicitly assumes that after each sell transaction, the investor exits the local currency until the next buy signal. Such a strategy would have the effect of increasing transaction costs, thus eroding some of the gains available from employing the trading rules examined here. Second, exchange rate movements by themselves could trigger a trade. For example, if the Turkish lire experienced a period of depreciation relative to the pound, this could trigger a sell even though the share price itself could be experiencing modest gains. However, the analysis was also undertaken in ‘local’ returns unadjusted for currency movements against sterling. The results from this further analysis were not different in character from those arising from using the sterling-adjusted returns which are reported here. In particular, with the exception of Turkey where the magnitude of the gains was much bigger when using local currency returns, both the sign and the size of the return to each trading strategy were broadly similar. Results based upon local currency returns are available from the authors on request. Thanks go to an anonymous referee for making this point.

15. Although the profit from the filter rule was significantly greater than the profit from the buy-and-hold strategy only for the 1% filter in Italy.

16. Jennergren and Korsvold (1975) document that this statement holds well for the Swedish market; using both serial correlations and runs tests, they found that Swedish shares displayed larger deviations from randomness than both the US and UK markets.

17. The profitability of large-size filters (35%, 40%, 45% and 50%) was also investigated. However, in many instances, no signals were triggered. For this reason, the results from these strategies are not reported in the paper, but are available from the authors on request.

18. One possible explanation for this difference in results could be the time period studied. In particular, while this study focuses on the recent 10-year period 1991–2000, previous studies have examined earlier and much longer time periods. For example, Brock et al. (1992) examined the 90-year period 1897–1986, Hudson et al. (1996) focused on the 1935–1994 period, and Bessembinder and Chan (1995) studied the period 1975 to 1989.

19. Mauchly's test of sphericity was used to test the assumption of homogeneity of covariance which underlies the use of mixed ANOVA. The results for this test showed that sphericity could not be assumed for the moving average rules. A more conservative test (the Greenhouse-Geisser test) was therefore used to obtain the values reported in panel B of Table 8.

20. Further work undertaken by the authors (and not reported here) suggests this may be the case. In particular, plots of the estimated marginal means of each country by type showed large differences in the performance of the moving average rules. In particular, the plots showed that while the performance of moving average rules in emerging markets was markedly different from that of the developed markets, the performance of the rules for the large and small developed markets was broadly similar. That few differences exist in the performance of these rules across the different developed markets was further supported by the results from an ANOVA in which (i) small and major developed markets were combined into one group; and (ii) small developed markets were omitted from the analysis. Results from both of these analyses revealed that type of country was significant at the 5% level. These results are available on request from the authors.