Abstract:
We use a quantile machine learning (random forests) approach to analyse the predictive
ability of newspapers-based macroeconomic attention indexes (MAIs) on eight major fundamentals
of the United States on the realized volatility of a major commodity-exporting emerging stock market,
namely South Africa. We compare the performance of the MAIs with the performance of a news
sentiment index (NSI) of the US. We find that both fundamentals and sentiment improve predictive
performance, but the relative impact of the former is stronger. We document how the impact of
fundamentals and sentiment on predictive performance varies across the quantiles of the conditional
distribution of realized volatility, and across different prediction horizons. Specifically, fundamentals
matter more at the extreme quantiles at short horizons, and at the median in the long-run. In addition,
we report several robustness checks (involving sample period and alternative definitions of realized
volatility), and indicate that the obtained results for South Africa also tend to carry over to other
emerging countries such as, Brazil, China, India, and Russia. Our results have important implications
for investors with volatility being an input for portfolio allocation decisions. In addition, with
stock market variability also capturing financial uncertainty, its accurate prediction based on US
fundamentals and sentiment also has a role in policy design to prevent possible collapse.
