Three Studies on Stock-Bond Correlation

Abstract

This thesis consists of three chapters, each of which explores separately the explanation, prediction and application of the time-varying stock-bond correlation. In the first chapter titled 'Stock-Bond Correlation and Economic States', I explore the state-dependent nature of the stock-bond correlation with four state variables -- monetary policy, equity cash flow innovation, expected inflation and stock market uncertainty. First, I study the association between each state variable and the stock-bond correlation conditional on whether the other state variables are at their historical high or low levels. Then, I further define states according to the estimated regimes of all state variables and analyse the variation of the stock-bond correlation across these states. I find that the long spell of negative stock-bond correlation in recent years is mainly associated with economic states in which monetary policy is accommodative and expected inflation is below average, and that equity cash flow innovation and stock market uncertainty have only a modest influence on this correlation. These findings are consistent with the view that the variation in discount rates plays a prominent role in determining asset prices -- and hence correlation -- under a discounted cash flow framework. In the second chapter titled 'Forecasting the Co-movement between Stocks and Bonds with Economic States', I examine whether the state-dependent regression model from Chapter 1 improves predictive accuracy and market timing ability relative to other predictive models, specifically linear regression models and time series statistical models, including the random walk, the exponentially weighted moving average and the autoregressive integrated moving average models. This assessment of predictive power also sheds light on competing theories in explaining the stock-bond correlation. I find that the state-dependent regression model has smaller predictive errors and a higher chance of forecasting the correct sign of stock-bond co-movements across a range of investment horizons. This finding suggests that the prediction of asset return correlation may be improved through studying the conditional association with the economic and financial fundamentals. In the third chapter titled 'Volatility-managed Portfolios Incorporating Conditional Correlations', I explore the application of the state-dependent stock-bond correlation from Chapter 1 to a portfolio management strategy called 'volatility targeting'. This strategy is often used to mitigate tail risks and drawdowns for long-short stock portfolios by dynamically scaling the up/down exposure to risky assets according to their conditional variances. Recent research demonstrates that volatility targeting improves Sharpe ratios and delivers positive alphas for volatility-managed portfolios in the equity market, but less so when bonds are included. I propose a new dynamic scaling approach that incorporates conditional correlations in addition to conditional variances into portfolio formation, which I call 'covariance-managed portfolios'. I show that these portfolios perform well in a stock-bond context under a variety of situations. The findings of this chapter highlight the benefits of incorporating conditional correlation in the portfolio management process. Show more