【计算机科学】【2013.01】人工神经网络预测金融时间序列的性能分析

本文为沙迦美国大学（作者：Assia Lasfer）的硕士论文，共115页。

对希望降低投资风险的投资者来说，预测股票价格至关重要。预测是基于股票价格有规律变动的思想。到目前为止，人们都知道发达市场、新兴市场和前沿市场具有不同的一般特征。

本研究使用实验设计（DOE）来研究人工神经网络（ANN）设计参数的意义和行为，以及它们对发达、新兴和前沿市场预测变化性能的影响。在本研究中，每种分类用两种市场指数表示。研究数据基于摩根士丹利国家指数（MSCI），包括阿联酋、约旦、埃及、土耳其、日本和英国的指数。

设计了两个实验，其中的5个神经网络设计参数在两个层次之间变化。第一个模型是4因子全因子分解模型，包括网络类型参数、隐层神经元数量、输出传递函数类型以及Levenberg-Marquardt（LM）算法的学习率。第二个模型是一个5因子分数因子分解模型，包含了所有前四个参数加上隐藏层sigmoid函数的形状。结果表明，对于特定的金融市场，DOE是识别最重要的人工神经网络设计参数的有效工具。此外，研究结果表明，在所有受试市场中，有时仅在同一类别的市场中存在一些普遍显著和普遍不显著的因素。然而，根据市场分类，人工神经网络设计参数的影响似乎没有任何差异；所有主要影响和相互作用似乎在所有测试市场中表现相似。

Forecasting stock prices is of criticalimportance for investors who wish to reduce investment risks. Forecasting isbased on the idea that stock prices move in patterns. So far, it is understoodthat developed, emerging, and frontier markets have different generalcharacteristics. Subsequently, this research uses design of experiments (DOE)to study the significance and behavior of artificial neural networks‟ (ANN)design parameters and their effect on the performance of predicting movement ofdeveloped, emerging, and frontier markets. In this study, each classificationis represented by two market indices. The data is based on Morgan StanleyCountry Index (MSCI), and includes the indices of UAE, Jordan, Egypt, Turkey,Japan, and UK. Two designed experiments are conducted where 5 neural networkdesign parameters are varied between two levels. The first model is a 4 factorfull factorial, which includes the parameters of type of network, number ofhidden layer neurons, type of output transfer function, and the learning rateof Levenberg-Marquardt (LM) algorithm. The second model, a 5 factor fractionalfactorial, includes all previous four parameters plus the shape of hidden layersigmoid function. The results show that, for a specific financial market, DOEis a useful tool in identifying the most significant ANN design parameters.Furthermore, the results show that there exist some commonly significant andcommonly insignificant factors among all tested markets, and sometimes amongmarkets of the same classification only. However, there does not seem to be anydifferences in ANN design parameters‟ effect based on market classification;all main effects and interactions that appear to be significant behavesimilarly through all tested markets.

1 引言
2 文献回顾
3 人工神经网络
4 实验设计DoE
5 研究方法
6 结果
7 结论

更多精彩文章请关注公众号：