Correlation between stock and bond returns is of immense importance as it plays a vital role in investors’ diversification and asset allocation decisions. The purpose of this thesis is to explain the driving forces behind volatilities and correlation of stock and bond returns and to investigate how this knowledge can be used to form portfolios that outperform traditional asset allocation strategies.
This paper takes a forwarding-looking approach by using analyst forecasts of macroeconomic variables to predict future realized volatilities and correlation of stock and bond returns. Variables from relating literature that have exhibited predictability of co-movement of returns are identified and analyst forecasts of these variables are used in predictive regressions. The factors used to forecast volatilities and correlation are implied stock market volatility, inflation rate, short and long rate, as well as change in corporate profits and change in real GDP. Mean consensus as well as dispersion in analysts’ forecasts of the abovementioned economic variables are applied in the analysis. To scrutinize the predictive power of the forecast variables we are controlling for historical data on the same variables.
An in-sample predictability analysis reveals that predictive regressions using analyst forecasts outperform, in terms of adjusted R2, models using historical data only, but a specification including both historical and forecast variables perform even better. Additionally, the single best in-sample predictor of future realized volatilities and correlation is simply the lag, that is, the previously realized value. Results show that all mean consensus and dispersion variables are statistically significant in predicting volatilities and stock-bond correlation except mean consensus of corporate profits, and dispersion in forecasts of real GDP growth.
Next, an out-of-sample predictability analysis is conducted to examine how minimumvariance portfolios formed using analyst forecasts perform compared to two simple benchmark strategies. Our results show that a strategy using both historical data and analyst forecasts yield the best performance, which is in line with the results from the in-sample analysis. Almost all proposed strategies perform statistically better than the equally-weighted benchmark portfolio, however all strategies fail to provide evidence for statistical and economic outperformance compared to a simple moving-average strategy. Additionally, out-of-sample, the use of analyst forecasts seems to unlock some predictability, as strategies excluding the lag perform better. Several robustness checks confirm that analyst forecasts of macro variables can be used to improve investors ex-ante allocation of wealth between stocks and bonds compared to an equally-weighted strategy but fails to outperform a simple movingaverage strategy.
|Educations||MSc in Finance and Investments, (Graduate Programme) Final ThesisMSc in Accounting, Strategy and Control, (Graduate Programme) Final Thesis|
|Number of pages||122|