Discovery Of Multi-component Portfolio Strategies With Continuous Tuning To The Changing Market Micro-regimes Using Input-dependent Boosting
Free (open access)
127 - 146
V. V. Gavrishchaka, O. V. Barinova, A. P. Vezhnevets & M. A. Monina
Recently proposed boosting-based optimization offers a generic framework for the discovery of compact and interpretable portfolios of complimentary trading strategies with stable (non-resonant) performance over a wide range of market regimes and robust generalization abilities. Inherent complexity control allows the framework to work with very large pools of heterogeneous base strategies with well-established properties. However, in its current version, the framework outputs a collection of dynamic strategies with fixed parameters and constant weights defining capital allocations. This excludes any adaptive regime adjustment or switching on the portfolio level for additional profitability from regime-specific patterns. In this work we extend boosting-based optimization framework by including capability to discover portfolio strategies with continuous and adiabatically smooth adjustment to the current market microregime. Such regime adaptivity is naturally provided by the input-dependent boosting. The proposed generalization preserves clarity and interpretability of the original framework since the dynamic base strategies of the multi-component portfolio and their optimal parameters remain fixed. However, the weights of the base strategies are adaptively varied in time according to the implicit rule discovered by boosting. Operational details of the new framework and encouraging results are illustrated using real market data. More rigorous theoretical foundation for the general concept of the boosting-based optimization is also outlined. Keywords: adaptive boosting, ensemble learning, regime switching, trading strategies, portfolio optimisation.
adaptive boosting, ensemble learning, regime switching, trading strategies, portfolio optimisation.