© 2014 Research Academy of Social Sciences
http://www.rassweb.com 96
International Journal of Empirical Finance
Vol. 2, No. 2, 2014, 96-105
Alternative Beta Risk Estimators in Emerging Markets: The Case of
Tunisia
Habib Hasnaoui
1
Abstract
In this paper, we use the sample selectivity model to estimate the systematic risk for Tunisian stocks. This
approach is applied in the case of extreme thin trading where data are censored due to the presence of zero
returns. The approach is a two-step procedure: a selectivity component which deals with the discreteness in
the observed data and a regression component which applies to the non zero return data. In addition, this
study compares the new beta estimate to the standard OLS beta and the Dimson Beta. The results reveal that
on average, the selectivity model corrects for the general downward bias in OLS betas more suitably ten the
Dimson correction. Our approach is more appropriate to deal with the presence of zero return observations
associated with extreme thin trading situations in emerging markets.
Keywords: Censored data, asynchronous trading, thin trading.
1. Introduction
The increasing globalisation of the world’s financial markets has led to a greater emphasis on the pursuit
of the benefits of international diversification. In turn, this has led to consideration of a broader range of
capital markets as possible investment opportunities. One such alternative is the Maghreb region in order to
provide an alternative source of capital to firms from traditional banking systems, Hearn (2011). Specifically,
the Tunisian market has benefitted from the European Neighbourhood Policy (European Commission
website, 2010) that has facilitated the attraction of foreign investments through the provision of assistance in
improving regulation and corporate governance. Following its inclusion in a number of renowned emerging
market benchmark indices including Morgan Stanley Capital International (MSCI), Standard and Poors and
Financial Times Stock Exchange FTSE, international investor awareness of the Tunisian Stock Exchange
TSE has increased. However, the determination of an appropriate risk measure for individual stocks is a key
issue for investors. The Capital Asset Pricing Model (CAPM) and the corresponding systematic risk (beta)
seems to be one such alternative. But according to Pereiro (2001), the use of the CAPM in the context of
emerging markets is problematic given the illiquidity patterns and the small size of the markets. One
alternative is to use different measures, such as that emanating from the downside risk model (D-CAPM)
suggested by Estrada (2002) whose results clearly illustrate that the CAPM beta understates the risk relative
to the downside risk measure.
The empirical evidence has shown that in a univariate setting, thin trading makes the standard realised
variance estimator biased and inconsistent, Griffin and Oomen (2011). Several approaches were proposed to
deal with this problem as data sub-sampling, Zhang et al.,(2005), the kernel-based auto covariance
adjustments, Barndorff- Nielsen et al., (2008a) or the pre-averaging methods to correct the variance-
covariance structure, Jacod et al.,(2009) and Podolskij and Vetter (2013). The additional problem of non-
synchronous trading is encountered in a multivariate setting, Fisher (1966) and Epps (1979). To overcome
this problem, two conceptually different approaches have been suggested in the literature. The first mitigates
non-trading biases by incorporating lead and lag autocovariance terms into the realised covariance estimator
based on synchronized returns, Scholes and Williams (1977), Dimson (1979) and Cohen et al.,(1983). The
1
Higher Institute of Management of Tunis, Tunisia