统计与数据科学学院知识库

近年来，计算机技术对金融市场的影响不断加深。在金融市场中，数据逐渐变得丰富起来，投资者也利用计算机技术进行量化投资决策分析，旨在提高投资效率。值得注意的是，这在一定程度上降低了投资风险。本研究基于引入技术指标的预测方法，构建了VMD-INFO-KELM模型并提出了在农产品期货市场上的跨品种统计套利策略。

首先结合E-G协整检验和动态时间规整（DTW）技术，以交易活跃的8种农产品期货作为可选交易对象，从中选择最佳的统计套利组合。通过特征选择筛选出相关性高且重要程度高的技术指标，分别输入到向量加权平均优化算法（INFO）优化的KELM、RF、LSTM三个预测模型中，验证出技术指标在提高预测精度上起作用，通过制定的统计套利策略回测结果发现，其中构造的引入技术指标的INFO-KELM模型效果在回测结果中优于其他模型。

其次，基于“分解-集成”框架下，利用变分模态分解（VMD）和样本熵（SE）方法对套利品种收盘价的价差序列进行分解重构新的序列，VMD是将高度波动的原始数据分解为相对稳定的、具有周期特征的、可逻辑解释的分量。再用RF、KELM、LSTM三种机器学习方法分别对各子序列预测后进行集成，实证分析，在VMD分解下的预测效果优于单一的机器学习模型。接着，将特征筛选后的技术指标进行机器学习模型预测和基于“分解-集成”框架下预测这两种方法相结合，进行非线性集成，得到的最终预测结果也有所提升。为了得到更精准的预测数据，体现模型的优势，为后续制定策略提供强有力支撑数据。在训练过程中，通过向量加权平均优化算法（INFO）对模型的权值和偏置进行优化，旨在提高预测精度。实证结果表明，引入技术指标的VMD分解下的INFO优化的KELM组合预测模型（VMD-INFO-KELM），表现出更好的预测效果。

最后，基于最优组合的预测结果，设计出统计套利交易策略。实证交易回测表明，在本研究价差预测方法下，结合均值回复思想的套利策略在获得了更高的收益，提出开仓条件中加上价差处于在均值的三倍标准差之外的限制条件套利策略最有可借鉴价值。从投资角度来看，量化投资项目的可行性可以借助计算机进行观察，便于及时调整经营策略，实现资产价值最大化。结果分析作为有借鉴价值的参考材料，有助于投资者的最终决策。

In recent years, the influence of computer technology on the financial market has been deepening. In the financial market, data has gradually become rich, and investors also use computer technology to make quantitative investment decision analysis, aiming at improving investment efficiency. It is worth noting that this has reduced the investment risk to some extent. Based on the forecasting method of introducing technical indicators, this study constructs the VMD-INFO-KELM model and puts forward the cross-variety statistical arbitrage strategy in the agricultural futures market.

First of all, the E-G cointegration test and dynamic time warping (DTW) technology are combined to select the best statistical arbitrage combination from 8 kinds of actively traded agricultural futures as optional trading objects. The technical indicators with high relevance and importance were selected by feature selection and input into the KELM, RF and LSTM prediction models optimized by the vector weighted average optimization algorithm (INFO) respectively, and it was verified that the technical indicators played a role in improving the prediction accuracy. The backtest results of the statistical arbitrage strategy developed showed that, The INFO-KELM model with technical index is better than other models in the backtest results.

Secondly, based on the "decomposition-integration" framework, the variational modal decomposition (VMD) and sample entropy (SE) methods are used to decompose and reconstruct a new series of the spread series of the closing price of arbitrage varieties. VMD is to decompose the highly volatile original data into relatively stable, periodic and logically interpretable components. Then, three machine learning methods, RF, KELM and LSTM, are used to integrate the sub-sequences respectively. Empirical analysis shows that the prediction effect under VMD decomposition is better than that of a single machine learning model. Then, the technical indicators after feature screening are combined with machine learning model prediction and prediction based on "decomposition-integration" framework for nonlinear integration, and the final prediction result is also improved. In order to get more accurate forecast data, reflect the advantages of the model, and provide strong supporting data for subsequent strategy formulation. In the training process, the weight and bias of the model are optimized by the vector weighted average optimization algorithm (INFO) to improve the prediction accuracy. The empirical results show that the information-optimized KELM combined forecasting model (VMD-INFO-KELM) with VMD decomposition of technical indicators shows better forecasting effect.

Finally, based on the prediction results of the optimal combination, a statistical arbitrage trading strategy is designed. Empirical transaction backtesting shows that the arbitrage strategy combined with the idea of mean recovery has achieved higher returns under the spread forecasting method in this study, and it is most valuable to put forward the restrictive arbitrage strategy that the spread is beyond three standard deviations of the mean value in the opening conditions. From the investment point of view, the feasibility of quantitative investment projects can be observed with the help of computers, which is convenient for timely adjustment of business strategies and maximization of asset value. As a valuable reference material, the result analysis is helpful to investors' final decision.