Machine Learning Methods In Algorithmic Trading

算法交易中的机器学习方法

• 批准号: 2517936
• 金额: --
• 依托单位: Imperial College London
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2017
• 资助国家: 英国
• 起止时间: 2017 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2517936
• 关键词: Machine; Learning; Methods; Algorithmic; Trading

In the context of the new trends in trading business, the aim of this thesis is to contribute in developing automated trading algorithms for acquiring explicit trading signals; Long or Short (Buy or Sell respectively) positions in Futures products with liquid underlying assets SP500, DAX 30 and German Bunds. The algorithms are based on two contemporary classification Machine Learning techniques:SVM (Support Vector Machines)Feed Forward Multilayer Perceptron (MLP) Neural Network (NN) The parameters of the above methods are optimized using a heuristic method known as Differential Evolution (DE), which is a promising evolutionary algorithm able to produce an optimized classification method faster and more accurate than a classical grid search.Raw high frequency data, also known as intraday data, are employed with the aim to generate normalised technical indicators which will be used as data set in our algorithms. The defined data set comprises to train and optimise (via DE) the SVM and NN classification methods, using Python as the appropriate programme language.Out of sample data are used in order to calculate the profit and loss outcome and hence to evaluate the accuracy of the algorithmic outputs, comparing and contrasting the advantages and disadvantages of each method.

在交易业务新趋势的背景下，本文的目的是为开发用于获取明确交易信号的自动交易算法做出贡献；对标的资产为标普500指数、DAX 30指数和德国国债的期货产品进行多头或空头（分别买入或卖出）头寸。这些算法基于两种当代分类机器学习技术：SVM（支持向量机）前馈多层感知器（MLP）神经网络（NN）上述方法的参数使用一种称为差分进化（DE）的启发式方法进行优化，这是一种有前途的进化算法，能够比传统的网格搜索更快、更准确地产生优化的分类方法。原始高频数据，也称为日内数据，用于生成归一化的技术指标，这些指标将用作我们算法中的数据集。定义的数据集包括训练和优化（通过DE）支持向量机和神经网络分类方法，使用Python作为适当的程序语言。使用样本外的数据来计算利润和损失结果，从而评估算法输出的准确性，比较和对比每种方法的优缺点。