Testing the performance of value strategies in the Athens Stock Exchange

Abstract

This study examines, for the first time consistently, the performance of value strategies in the Athens Stock Exchange (ASE) based on the price to earnings ratios, dividend yields (DYs), size (market value), market to book ratios, financial leverage ratios and systematic risk. We tested the usefulness of the above strategies, by examining the performance of portfolios of stocks formed on the basis of the above criteria, and by applying multiple regression analysis. Our univariate portfolio analysis showed that the higher returns observed in high DY stocks and low beta stocks were achieved with no additional level of risk taken. When the effect of cross-sectional correlation in the residuals of our regression model was removed, we found that only stocks with high DYs may be associated with significantly higher returns. Thus, we can conclude that except the application of the DY variable, there is little support for the argument of overperformance of value strategies even in the case of a small emerging market, such as the ASE during the period 1995–2002 examined.

Acknowledgements

We would like to express our gratitude to Sudi Sudarsanam, Nickolaos Travlos, Dimitris Malliaropoulos and an anonymous referee on an earlier version of this article for many insightful comments. We would also like to thank the participants of the EFMA Conference in Basel in July 2004 and the discussant Taras Bodnar for their contributing remarks, as well as Christos Anastassis for his assistance in some numerical calculations. Financial support from the Research Centre of the University of Piraeus is gratefully acknowledged.

Notes

¹ In the case that a company recorded losses, we decided to substitute these values with zero, since it has no economic meaning of estimating a negative PE ratio. We also excluded extreme positive or negative values of PE (extreme is defined as three SDs above the mean of the PE). For the period of time that followed the stock market ‘bubble’ in the ASE of 1999, i.e. the years 2000, 2001 and 2002 this phenomenon is intensely observed.

² In the case that a company did not pay dividend, the value is set equal to zero.

³ The construction of beta portfolios starts in year 1995 (based upon the data of three years before, i.e. 1992–1994), in order to have 36 months of data. However, we did our estimations using only one year of data (12 months) and our results did not significantly change.

⁴ To test the impact of this variable correctly, we had to exclude the companies belonging to the banking and financial sector, because they tend to have different capital structure and capital requirements from the rest of the companies.

⁵ The only difference in our definitions is that we used the fractions of Fama and French (1992) in the reverse form, so a low FINLEV means high gearing and accordingly a high FINLEV means low gearing.

⁶ Note that the number of observations in the portfolios selected by different criteria is not the same throughout the period of time examined, because there are missing observations and outliers which we had to exclude. This especially holds in the case of PE ratios in some years (e.g. 2002) when companies suffered heavy losses. Therefore, we excluded all values greater than three SDs above the mean of the PE.

⁷ Such tests were conducted by assuming equal variances of the two samples.

⁸ We considered that the F-statistic is valid under the assumptions of normal distribution of the PAARs which are in annual log form and the independent populations of the PAARs, since their correlation coefficients between the low and high groups, which we calculated, were much lower than 1.

⁹ MV is taken in its logarithmic form (MVLN).

¹⁰ FINLEV2 was included. Although it was not proved to be significant from the portfolio analysis, it showed a strong positive correlation with stock returns (ARs).

¹¹ Thanks to a useful remark made by an anonymous referee, that the DY maybe a proxy for other variables, such as the effect of the banking stocks and the large capitalization effect, we investigated the type of our sample firms which had high DYs and we found out that these firms were not necessarily banks and large firms, but they tend to belong in general to the financial services sector (the majority of these firms were investment firms in the type of mutual funds–mainly closed-end funds). Thus, we explicitly tested the influence of the financial sector in the performance of the DY variable by two ways. The first way was by carrying out again the portfolio analysis of stock selection based on the DY variable with the exclusion of the financial sector firms, and the second way was by creating a dummy variable called DUMMYFS and including it in the panel data part of our analysis. The overall results we obtained (which are available upon request) from the portfolio analysis were not essentially different from those using the entire universe of sample firms (Table 2). We found that, although, the stock returns (PAARs) of the high DY portfolios excluding the financial sector firms somehow decrease only in the years 1995–1997 in comparison with the corresponding returns of the entire universe of sample firms, their differences with the PAARs of the low DY portfolios continue to be statistically significant in these years and these differences are not accompanied by an increase in the level of total risk (proxied by their SD). Only in year 1999 (the peak year of stock market boom) the difference in PAARs is no longer statistically significant. The average PAAR for the whole period (1995–2002), although it drops to 12.6% from 17.4%, is also statistically significant at the 5% level.

¹² Equation 6a was estimated for the limited sample of firms excluding the companies of the financial sector and Equation 6b was estimated for the whole sample of firms.

¹³ The statistic we used to test for cross-sectional heteroskedasticity in the residuals was the Lagrange multiplier statistic presented in Greene (1997).

¹⁴ The GLS model with the White's (1980) correction procedure generated by E-views software program was followed to apply the cross-section weights.

¹⁵ The test for the existence of cross-sectional correlation in the residuals was carried out with the Breusch and Pagan (1980) LM test statistic:

where, r_ij  = the correlation coefficient for the residuals.

¹⁶ Since the FINLEV2 was not included in the second equation, it became feasible to increase the sample size by incorporating banks and other firms of the financial sector, which were excluded in the first sample.

¹⁷ We examined the pooled data in order to detect possible multicollinearity. Our tests indicated the absence of such a problem.

¹⁸ We would like to note that the values of R ²s which are reported in Panel A of Table 9 are based on the weighted statistics of the pooled GLS regressions (see footnote 14).

¹⁹ This investigation was a part of our sensitivity analysis previously explained in footnote 11.