Stock return predictability in emerging markets

Abstract

The stock return predictability still be a puzzle in the financial economics society that have not yet solved for several decades. Some of them believe that they are predictable, and that stakeholders are able to ensure opportunities to allocate their assets in advance while others disagree and believe in stock market efficiency. In spite of the fact that a number of researchers that have recognized the predictive model as fact, there are still some doubts in terms of econometric issues. Econometricians generally agree that the predictive variable has a local–to-unity property that has a significant correlation with stock returns in an infinite set of a given sample and the other issue is the near unit root characteristic of stock returns’ stochastic volatility. Both of the issues end up with an over-reject characteristic of standard hypothesis testing form of predictive regression.

Due to the previously mentioned econometrics issues, the CJP approach will be applied to the predictability of stock returns as well as to testing to rectify the issues. The CJP approach will utilize the change of time method to rectify the nonstationary of stochastic volatility and the nonparametric instrumental variable estimator known as the Cauchy estimator to fix the endogeneity problem of covariates.

To investigate the stock return predictability with the mentioned econometrics issues, this research applies stock return data from Stock Exchange of Thailand (SET) to extract stochastic volatility and testing of local-to-unity property of it. Consequently, the time change method is applied to generate robustness hypothesis testing for unit root or near unit root stochastic volatility of stock returns. The last step applies the nonparametric Cauchy estimator to make a set of instrument variable of covariate for the stock return prediction

Empirical results show that the stock return significantly generates local-to-unity and near local-to-unity of stochastic volatility. The time change method can be applied to resolve the local-to-unity of stochastic volatility problem by using stochastic stopping time of the process or volatility time to replace the calendar time, which will make stochastic volatility in each new period become stationary across all observations. The Cauchy estimator was applied for random sampling based on volatility time and shows no support for predictability at any frequency for the selected predictors, such as the dividend–price and earnings–price ratios.

This research concludes that it seems clear that stock returns cannot be predicted by dividend–price and earnings–price ratios if the characteristics of the data are properly checked and managed.