A SVM Approach in Forecasting the Moving Direction of Chinese Stock Indices.

摘要

Support vector machine (SVM) has been shown to be a reliable tool in prediction and classification using a convex objective function with constraints integrated in by Lagrange Multipliers and characterized by the involvement of kernel functions and the sparsity of the solution. In this paper, we build a series of SVM models based on the macroeconomic fundamental indicators to investigate the predictability of financial movement direction by forecasting the daily movement trend of main indices in Chinese Shanghai and Shenzhen stock markets: SSE 50 Index, SSE 180 Index, SSE Composite Index, SZSE 100 Index, SZSE Composite index and CSI 300 Index. To the best of our knowledge, this is the first application of SVM techniques to Chinese stock indices. After data cleaning and transformation, only a subset of potential input candidates is delivered to the next step through feature selection. Then two parameters of SVM are selected and tuned based on the selected subsets of features for SSE indices. To evaluate the forecasting ability of SVM, we compare its performance with Neural Networks. The experimental results show that SVM with carefully selected features performs comparably to or better than SVM with the whole variable set. And in accordance with previous studies, SVM outperforms Neural Network in all indices. However, both models have low specificity values. The prediction accuracy of SVM for SSE 50 Index, SSE 180 Index, SSE Composite Index, SZSE 100 Index, SZSE Composite index and CSI 300 Index is 61.06%, 60.47%, 61.65%, 59.74%, 61.58% and 61.85%, respectively. These results support the claim that as an emerging market Chinese stock market is semi-strong form inefficient.