基于斐波那契数列采样的BP神经网络金融时间序列短期趋势预测

Financial Time Series Prediction of Short-term Trend Using Neural Networks Base on Fibonacci Series Sampling

金融市场趋势预测研究一直都是热点题目,针对金融时间序列非线性特点用神经网络建立预测模型早已是研究热点。本文在吸取前人利用神经网络预测的经验下,提出利用自然界神奇数列斐波那契数列对金融时间序列采样。实验证明对金融时间序列短期趋势预测斐波那契数列采样的优越性,为了去除金融时间序列存在的大量噪音,本文利用指数平均数平滑原始价格数据,避免了利用移动平均平滑造成的滞后问题。本文所用的金融时间序列数据是欧元/美元的1小时收盘价,实验结果表明外推一步预测方向命中正确率可以达到75%以上,同时做了多步外推预测,证明5步外推趋势正确命中依然可以达到60%以上。

Predicting the trend of financial market is always an attractive research branch,especially using of Neutral Network to model nonlinear financial time series. We propose a new sampling method using Fibonacci series based on the previous researcher＇s experience of prediction with Neutral Network. Our experiments show the advantage of using Fibonacci series for sampling. In order to eliminate massive noises in financial time series,we use exponential average to preprocess raw data,which can avoid the lagging problem brought by moving average. In this paper,1 hour close price of EUR /USD is used. Our results show the accuracy of prediction for direction of extrapolation by one step can be improved to 75%. Furthermore,we modified our model to predict direction of extrapolation by multiple steps. The results still show the accuracy is above 60%.