Dynamic portfolio optimization according to market states and diversification using hierarchical clustering
Abstract
The choice of assets and the state of the market are essential factors to consider in constructing an optimized portfolio for investments. Developments in the research of portfolio optimization techniques introduced models that consider factors such as market state and asset selection. We then explore and integrate these portfolio optimization methods and techniques into our research. In this study, we propose a portfolio optimization method which optimizes a portfolio given the market state and selects assets based on asset correlation to ensure the diversification of assets in the portfolio. We use machine learning techniques in modelling market states and in the portfolio optimization process. Different market states have different statistical distributions since markets go through bull and bear states, where the value of assets in the market tend to rise or fall respectively. To model the non-stationary nature of the market, we cluster time periods and classify them into market states based on the statistical properties of the stock returns. Market states have varying statistical distributions of returns, so we train a portfolio optimizer in each market state. To optimize the portfolio, we use the Hierarchical Agglomerative Clustering method to group correlated assets together. Afterwards, we select lowly correlated assets with high historical Hierarchical Momentum to construct our portfolio. The constructed portfolio then had a higher Sharpe ratio and mean return than the 1n equally weighted portfolio with comparable variances when tested in sample. By selecting lowly correlated assets with high Sharpe ratios and considering the market state at a given investment period, we then show that a diverse portfolio with high returns can be generated with the proposed method.
