Regression based approaches to optimize active asset allocation
Project Description
Active, tactical asset allocation approaches are applied portfolio-strategies in order to achieve additional gains on the markets, where the assets are invested. Based on various indicators the weights of the assets are regularly rebalanced - or even new asset classes are inserted - to generate a positive alpha. The type of indicators underlying the allocation system moves on a wide range, but in quantitative models econometric equations often serve as primary signals providers.
The proposed dissertation introduces a regression based tactical asset allocation system with the declared goal to construct a model that outperforms the respective passive, strategic asset allocation along with a lower or equal volatility. Contrary to mainstream theories the assumptions state that various short term market inefficiencies appear continuously and it is possible the exploit them with the help of macroeconomic, fundamental and technical variables. The accordingly generated econometric equations deliver predicted returns for the selected stock market based on historic time-series of the exogenous (macroeconomic, fundamental and technical) factors.
The main difference to classical models is the sophisticated rolling regression that dynamically recalculates the equation before forecasting for each period. Unlike standard econometric models where fixed variables are determined through the process, in the dissertation approach the possible influencing factors are selected into a pool that provides the variables for a stepwise regression. Hence a secure structure is obtained, which monitors changes in the market condition and tries to fit the model to the newly appearing environment, or warns priori to unreliable predictions. Furthermore the inclusion of technical indicators is expected to raise the explanatory power of the asset allocation.
The proposed dissertation introduces a regression based tactical asset allocation system with the declared goal to construct a model that outperforms the respective passive, strategic asset allocation along with a lower or equal volatility. Contrary to mainstream theories the assumptions state that various short term market inefficiencies appear continuously and it is possible the exploit them with the help of macroeconomic, fundamental and technical variables. The accordingly generated econometric equations deliver predicted returns for the selected stock market based on historic time-series of the exogenous (macroeconomic, fundamental and technical) factors.
The main difference to classical models is the sophisticated rolling regression that dynamically recalculates the equation before forecasting for each period. Unlike standard econometric models where fixed variables are determined through the process, in the dissertation approach the possible influencing factors are selected into a pool that provides the variables for a stepwise regression. Hence a secure structure is obtained, which monitors changes in the market condition and tries to fit the model to the newly appearing environment, or warns priori to unreliable predictions. Furthermore the inclusion of technical indicators is expected to raise the explanatory power of the asset allocation.
