Date of Degree
Finance and Financial Management
Asset Allocation, Bayesian, Forecast
The first half of this dissertation consists of two essays addressing dynamic asset allocation problem by exploring time-varying volatility and covariance between different assets.
In the first essay, I propose a time-varying Bayesian approach based on autoregressive models. To allow a parsimonious specification while improving predictive power, I specify a step function that considerably decreases the number of parameters to be estimated. To reduce data dimensionality, I use orthogonal portfolios instead of correlated assets in estimation and forecast. Finally, a Bayesian estimation is applied to dynamically update coefficients and error variance. I combine Bayesian time-varying autoregression with step function restriction in the univariate forecast of return and volatility of orthogonal portfolios. Using a daily rebalancing portfolio of four asset classes, this approach generates Sharpe ratios above 2 under a range of specifications within the dataset.
In the second essay, I implement a new approach that dynamically rebalance portfolio based on forecast of asset returns under coefficient uncertainty and time-varying conditional covariance in a multivariate setting. I incorporates Principal Component Analysis in a vector autoregressive form multivariate Bayesian Dynamic Linear Model to forecast multivariate asset return and covariance. This approach can be applied to data with large dimension. It combines forecasting of return vector and covariance matrix all in one model. I use a daily rebalancing portfolio of five asset classes to show the improved portfolio performance as measured by the goodness of fit measures and the ex-post Sharpe ratio compared to several competing approaches.
The second half of the dissertation studies the intraday price responses to overnight movements of individual stocks and index exchange-traded funds. I find that the overnight returns are followed by a reversal during the first half-hour of intraday trading, especially concentrated in the first ten minutes. Such reversal is significant in both time series of index and cross-section of individual stocks. The reversal effect implies market mispricing at the open and investors correcting the prices during the first half-hour. The effect leads to profitable trading opportunities. By forming a long-short portfolio based on overnight performance, investors earn a daily premium of 16.7 basis points. The results are robust after controlling for market risk, reversal from lag intraday interval, and different market conditions.
Feng, Yalan, "Dynamic Asset Allocation: A Bayesian Approach" (2014). CUNY Academic Works.