基于机器学习的智能TWAP和VWAP算法的研究及应用

TWAP与VWAP算法为两类较常见的经典交易算法.传统的VWAP算法在TWAP算法的基础上，大多使用预测日内成交量分布的方法指导算法下单.传统成交量分布的预测效果严重依赖于市场交易惯性，但交易量分布受到日内诸多突发因素的影响，导致算法对市场突发状况的应对能力较弱.本文对传统TWAP与VWAP算法进行改进，利用滚动的1分钟粒度高频实时资金博弈数据，基于Logistic分类器训练量价模型，以该预测结果为入参构建最优化期望执行均价模型，求出当下各个价格档位对应委托数量的最优解.通过相对高频的分钟级价格预测机制，保证算法实时跟踪市场行情走势并寻求相对优势的交易机会.该算法经测试可以稳定地跑赢市场均价，具备推广应用的可行性.

Research and Application of Intelligent TWAP and VWAP Algorithms Based on Machine Learning

TWAP and VWAP are two classic trading algorithms. The traditional VWAP algorithm is based on the TWAP algorithm and mostly uses the method of predicting the intraday volume distribution to guide the algorithm to place orders. The effectiveness of traditional trading volume distribution prediction relies heavily on the maintenance of market trading pattern, but the trading volume distribution is affected by many intraday uncertain factors, resulting in the algorithm''s weak ability to respond to market emergencies. This paper improves the traditional TWAP and VWAP algorithms, uses rolling one-minute high-frequency real-time financial data, trains the price-and-volume model based on the Logistic classifier, then substitutes short-term predication results into the optimization model to generate optimal order price and quantity. The high-frequency real-time one-minute price prediction mechanism can ensure that the algorithm follows the market condition and seeks relatively favorable trading opportunities. This algorithm has been tested to outperform the market average and has the feasibility of promotion and application.