基于L1-LSSVR模型的金融状况指数的构建
【摘要】 金融状况指数(FCI)综合多个金融变量的信息,相比单个指标而言,能更直观、全面地反映金融市场的整体运行状况。本文结合最小二乘函数和L1范数的优秀性能,采用L1范数最小二乘支持向量回归机(L1-LSSVR)模型构建金融状况指数。通过人工数据和实际数据验证模型的特征选择功能,发现L1-LSSVR模型不但具有良好的回归效果和特征选择功能,而且运算速度快。最后进一步研究了金融状况指数与通货膨胀之间的关系,并且建立出通货膨胀率的预测模型。
【关键词】金融状况指数,最小二乘函数,L1范数,L1-LSSVR模型,通货膨胀预测
The construction of financial condition index
based on L1-LSSVR model
【Abstract】 The Financial Condition Index (FCI) combines the information of multiple financial variables to reflect the overall operational status of financial markets more intuitively and comprehensively than individual indicators.Based on the best performance of least squares function and L1 norm,this paper constructs the financial condition index by L1 norm least squares support vector regression (L1-LSSVR)model.Through the artificial data and the actual data to verify the model's feature selection function,we found that the L1-LSSVR model not only has good regression effect and feature selection function,but also has high computing speed.Finally,the relationship between the financial condition index and the inflation is further studied.And establish the prediction model of the inflation rate.
【Key Words】 Financial condition index,Least squares function,L1-norm,L1-LSSVR,Inflation forecast
表目录
表3-1 SVR与L1-LSSVR模型对人工数据的特征选择结果 15
表3-2 UCI数据集 15
表3-3 SVR与L1-LSSVR模型对UCI数据集的特征选择和回归结果 15
表4-1 变量选取 17
表4-2 L1-LSSVR的特征选择结果 19
表4-3 KMO和Bartlett的检验 20
表4-4 潜在金融因子所包含的信息 20
表5-1 通货膨胀率与一阶差分后的金融状况指数的相关分析 23
表5-2 FCI与通货膨胀的Granger因果检验 24
表5-3 单位根检验 24
表5-4 确定最优滞后期的信息准则比较表