PROGRAM | Economics

By: Hao Jiang Chair: Siyan Wang Co-Chair: Paul Laux

ABSTRACT

In this dissertation, I propose a method for incorporating dynamically-updated estimates of the parameters of assets’ return distributions into a risk parity portfolio optimization process to achieve better investment outcomes. Specifically, I combine Dynamic Conditional Correlations (DCC) / Asymmetric Dynamic Conditional Correlation (ADCC) Generalized Autoregressive Conditional Heteroscedasticity (GARCH) models, indexes of macroeconomic uncertainty, and the risk parity approach to portfolio construction. My method exploits the rich dynamic statistical relationships among asset returns, conditional volatilities, and correlations, to develop dynamic portfolio optimal weights. The approach is validated in two empirical applications, one focusing on the dynamic allocation of an investment portfolio across major U.S. asset classes, and the other focusing on Exchange-Traded Fund (ETF) construction with individual stocks in the US energy sector.
Procedurally, I first design a daily macroeconomic news-based uncertainty index, the Surprise Strength Indicator (SSI). Along with an off-the-shelf macroeconomic uncertainty index, Economic Policy Uncertainty (EPU) Index, these indexes augment multivariate DCC and ADCC-GARCH models. Both macroeconomic uncertainty indices enter the return, conditional variance, and correlation equations as exogenous variables. I find strong evidence that the ADCC-GARCH model augmented with the SSI index has the best predictive power for the returns and volatilities in both aggregate asset case and individual energy stock case, especially during the crisis periods. The significance of the exogenous uncertainty indices in explaining correlations is time-varying and is plausibly sensitive to the financial crises.
In my next step, I design an asset allocation approach based on the predictions from my DCC/ADCC GARCH models. Specifically, I use the out-of-sample forecasted variance-covariance matrix as input in a risk parity procedure to find the optimal portfolio weights. The risk parity procedure unbinding the dependency of the self-determined weights distribution or historical distribution of asset returns, as the subjective opinions and expected returns are more difficult to be estimated accurately than the expected variance-covariance matrix. Also, the risk parity procedure gives higher weights to safe assets, which provides the risk-averse investors protections during the crisis.
Evidence from applying my approach to US asset class data over 2007 to 2017 shows clear benefits versus the benchmarks, such as 1/N equal-weighted portfolios, single asset market-level indexes, and portfolios based on the risk parity approach using as input the predictions from DCC/ADCC-GARCH models without the macroeconomic uncertainty indexes. Specifically, my approach produces higher realized returns and lower realized risks out of sample. When my portfolio management approach is used to construct passively and actively managed energy stock portfolios, my passively managed portfolio significantly outperform the existing benchmark portfolios with a lower tracking error. My actively managed portfolios can use a much smaller set of stocks to achieve higher returns, lower risks, higher Sharpe ratios, and comparable information ratios than popular energy mutual funds from Vanguard and Fidelity, and consequently reduce computational burden and transaction costs. Overall, my findings provide significant implications for policymakers and portfolio managers in terms of the stability of the financial systems, asset allocation decisions, and designing portfolio management strategies.

Back >