Lecture-01-课堂标注版.pptx

机器学习与量化交易实战第一讲机器学习 与 量化交易 实战重点为什么要做算法交易交易系统的开发与设计时间序列分析策略建模及其优化方法策略评价与回测风险管理交易策略的实现交易策略的执行This is a team work, but you need to know the big picture and see which role suits you.The Big Picture And How The Course Is Organized Office Hours:8-10 pm 周二／周四Weekly ReadingsHomeworkExtra Projects For Motived Students.算法交易综述Algorithmic Trading Without Bullshit何谓算法交易Algorithmic Trading利用自动化平台，执行预先设置的一系列规则完成交易行为。算法交易：优势历史数据评估执行高效无主观情绪输入可度量评价交易频率算法交易：劣势成本技巧算法交易流程大前提：基于某种平台：提出假设建立模型回测验证执行交易交易策略的来源市场微观结构研究（for HFT mostly)基金结构套利(fund structure arbitrage)机器学习／人工智能机器学习流程机器学习流程一个例子Four paradigms of machine learningConnectionismSymbolismFrequentistsBayesianA (super) Brief History of A.I58-69 NNs, logics 70s WINTER80-95 NNs95-10 SVMs and statistical learning10s-currentMachine Learning in a nutshellDataModel Objective FunctionOptimizationMachine Learning TradingLimit Order Book Modeling Price-based Classification ModelsText-based Classification ModelsReinforcement Learning4 Key Factors that makes magic happensGood Model and Efficient Training AlgorithmsHardware( GPU/CPU)(high quality) DATAPlatform(keras/tensorflow/sklearn)A little aside: 深度学习CNN for spatial data (空间）LSTM for temporal data（时间）A little aside: 强化学习A little aside: 强化学习A little aside: 自然语言处理第二个例子Key ProblemHow to define the input features?特征工程特征选择交易策略的评估策略基本假设 Sharp Ratio杠杆频率风险W/L模型复杂度最大亏损(Maxium drawdown)Benchmarking回测何谓回测？ 将交易策略在历史数据中进行合理验证的过程。科学的回测十分重要(大部分人死在回测上）回测的意义策略筛选策略优化策略验证错误的回测方法很多情况下，回测结果不错，实盘交易不尽如人意。造成的偏差原因主要有：乐观主义偏差。(special look back region) 时间旅行。程序BugTrain/Val/Test set幸存者误差工具和语言PythonSklearnPandasAnd more…量化交易：从工程的角度量化交易：从工程的角度Event Event QueueDataHandlerStrategyPortfolioExecutionHandlerBacktestNext Time一些金融基本知识以及Python…This is a team work,,, but you need to know the big picture and see which one suits you.The content is self-contained. And is my own design.强调的是自动化的过程行为背后的原理无所谓。周易也可以量化交易Books33