Asymmetric Dependence, Persistence and Firm-Level Stock Return Predictability

We examine whether excess returns can be predicted using information about asymmetric dependence (AD) of firm returns. AD is a significant predictor in all the stock markets examined; it can predict future excess returns up to 15 months ahead in the USA, UK and Australia. Our results are not biased by potential persistence of AD, as we find no evidence of autocorrelation in the firm AD across all markets. We further test whether one time series of AD from any of the stock markets considered is useful in forecasting AD from other stock markets. We find significant spillover effects of AD between the US and Australian stock markets. The level of AD in the USA influences future levels of AD in Australia. Consistent with the concept of efficient markets, few firm-specific characteristics have been shown to significantly predict excess firm-level returns in out-of-sample empirical tests. In this chapter, we provide striking out-of-sample evidence that asymmetric dependence (AD) has been able to predict firm-level excess returns through a significant historical sample period (1959-2015). We provide evidence of return predictability using AD information from multiple financial markets and across different asset classes. We analyse the power of AD to predict stock returns of US equities, US real-estate investment trusts (REITs), Australian listed equities and equities listed in the UK. We find that AD measured by the Alcock and Hatherley (2017) adjusted J statistic (J^Adj) is a significant predictor of future excess returns in all the tested financial markets. We also explore whether AD can predict stock excess returns in longer horizons and use future excess returns from the next 3 to 15 months. We reveal that the firm-level AD predicts returns of up to 15 months ahead in the USA, UK and Australia. J^Adj is strongly significant in our predictive regressions with a high t-statistic. However, the traditional tests of stock return predictability may falsely reject the null hypothesis too frequently in the presence of persistent regressors (Mankiw and Shapiro, 1986; Elliott and Stock, 1994; Stambaugh, 1999; Campbell and Yogo, 2006). In order to validate our results, we explore the persistence of our chosen AD metric by testing the autoregressive properties of the J^Adj time-series processes using standard econometric tests. We also focus on the higher-order autoregressive (AR) components to study the patterns and cyclical behaviour of the autocorrelations of AD. We do this for AD, as well as for upper-tail and lower-tail AD autoregressive processes. We find no evidence of a significant persistence of AD in any of the financial time series analysed. No persistence increases the validity of our predictability results. We also explore migration probabilities of the two distinct types of AD: Lower-tail asymmetric dependence (LTAD) and upper-tail asymmetric dependence (UTAD). The information about the type of AD is crucial for an investor, as it completely changes the outlook on return prediction. All our predictive regressions suggest that LTAD (UTAD) is associated with positive (negative) future excess returns. We find that if a listed firm exhibits LTAD today, it is more likely to be LTAD in 12 months ahead. This holds across all the financial markets examined. This higher migration probability of LTAD may be explained by a higher prevalence of LTAD. Finally, we explore the autoregressive properties of AD in a multivariate setting using the vector autoregression (VAR) model. We once again do not find any sign of a significant serial autocorrelation of AD. We also focus our attention on spillover effects of AD and examine whether there are any connections between the aggregate levels of AD across the four financial markets. We test for Granger causality and explore the dynamics in the AD levels between the markets using the VAR model. We find that there are substantial spillover effects between US and Australian equities. Specifically, the aggregate level of AD in the USA can predict future levels of AD of Australian-listed firms. The first section of this chapter presents the empirical results of return predictability, using AD as a predictor, controlling for common factors, such as Capital Asset Pricing Model (CAPM) ß, size, value, idiosyncratic risk, momentum, coskewness and cokurtosis. In the second part of this chapter, we focus on exploring the persistence of AD measure. At the end of this chapter, we analyse spillover effects of the aggregate levels of AD from all the four financial markets.