Stock market correlation and investor attention

• Main Author: Symitsi, Efthymia
• Published: University of East Anglia 2017
• Subjects: 658
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.738649

This thesis deals with three separate problems in �nance related to covariance. First, I assess the forecasting performance of popular multivariate GARCH, hybrid implied and realised covariance models in terms of statistical and economic criteria. I perform a rigorous analysis across major equity indices using di�erent forecasting horizons, market regimes, loss functions and tests. A Vector Heterogeneous Autoregressive speci�cation is the best among competing models. Less complex models that rely on high-frequency data yield superior forecasts and reduce the portfolio risk. Hybrid estimators that combine optionimplied and high-frequency information also have merit when option-implied volatilities are corrected for the volatility risk-premium. During �nancial turmoil the ranking does not change signi�cantly but forecast accuracy deteriorates. Second, I investigate comovement in investor attention as a determinant of excess stock market comovement proposing a novel proxy, \co-attention". Co-attention is estimated as the correlation in demand for market-wide information across stock markets approximated by the Google Search Volume Index (SVI). My results reveal signi�cant co-attention driven to some extent by correlated news and fundamentals. Most importantly, I �nd that coattention is positively related to excess comovement. This e�ect is more pronounced in developed economies and during recessions. I fail to document signi�cant e�ects of correlated news supply on stock markets, lending support to the idea that information demand governs investing decisions. Co-attention is not only induced through international investors, but domestic investors as well. My results provide evidence of attention-induced �nancial contagion in unrelated economies. However, international investors' co-attention appears to facilitate volatility transmission indirectly across markets. Third, I solve the optimal budget allocation problem across keywords for paid search adiv v vertising accounting for the risk induced by maintaining a portfolio of volatile and correlated keywords. In a mean-variance context, I maximise the growth rates in keyword popularities. Advertising costs and conversion rates are shown to be irrelevant. I demonstrate practical implementation using readily available data from Google Trends database estimating averages, variances and co-variances as growth rates in SVIs. Based on keyword sets for major sectors, I form e�cient frontiers consisting of optimal combinations of keywords. Optimal keyword portfolios o�er statistically higher risk-adjusted performance against portfolios constructed using popular heuristics. A proposed heuristic based on risk-adjusted performance reduces the computational cost and provides competing results.