Portfolio optimisation is a multi-objective optimisation problem (MOP), where an investor aims to optimise the conflicting criteria of maximising a portfolio's expected return whilst minimising its risk and other costs. However, selecting a portfolio is a computationally expensive problem because of the cost associated with performing multiple evaluations on test data ("backtesting") rather than solving the convex optimisation problem itself. In this research, we present ParDen, an algorithm for the inclusion of any discriminative or generative machine learning model as a surrogate to mitigate the computationally expensive backtest procedure. In addition, we compare the performance of alternative metaheuristic algorithms: NSGA-II, R-NSGA-II, NSGA-III, R-NSGA-III, U-NSGA-III, MO-CMA-ES, and COMO-CMA-ES. We measure performance using multi-objective performance indicators, including Generational Distance Plus, Inverted Generational Distance Plus and Hypervolume. We also consider meta-indicators, Success Rate and Average Executions to Success Rate, of the Hypervolume to provide more insight into the quality of solutions. Our results show that ParDen can reduce the number of evaluations required by almost a third while obtaining an improved Pareto front over the state-of-the-art for the problem of portfolio selection.
翻译:组合组合优化是一个多目标优化问题(MOP ), 投资者的目的是优化使投资组合预期回报最大化的矛盾标准,同时尽量减少风险和其他成本。 然而,选择组合是一个计算成本高昂的问题,因为对测试数据进行多重评价(“回测试”),而不是解决 convex优化问题本身的成本。 在这一研究中, 我们提出ParDen, 将任何歧视性或基因化机器学习模型作为一种算法, 作为一种替代方法, 以缓解成本高昂的计算后测试程序。 此外, 我们比较替代计量算法的性能: NSGA- II、 R-NSGA- II、 NSGA- III、 R-NSGA-III、 U- NGSGA-III、 U- NSGA- III、 MO- CMA-ES和 COMO- CMA-ES。 我们用多目标性能指标来衡量绩效, 包括代代间距离加、 逆向后代代代距离加和超音量。 我们还考虑元指标、 成功率和平均执行到成功率率, 等替代算算算法的替代算算算算算法, 的替代算算算算算法, 将更深入地展示前数, 以显示改进的组合组合的成绩评估,, 以显示比先期的成绩, 以获得更前数。