基于小波降噪的美国月度新房开工率的时间序列分析

Time Series Analysis of United States Monthly Housing Starts Based on Wavelet Denoising

基于有明显季节性规律性的时间序列数据与信号数据表现相似的特征,本文运用信号处理中的小波方法对原始时间序列数据进行降噪,提升模型的预测效果.针对选取的不同类型小波,通过更改阈值对原始数据进行降噪;以美国新建私有房屋开工率1990年1月至2014年5月的数据为样本,运用降噪后的数据进行模型拟合与预测,并与传统的纯时间序列分析建模结果相比较.结果表明对数据进行小波变换降噪的预处理之后效果更好:减弱了因局部高频变化数据导致的模型不稳定性;使数据更加平滑;提升了预测值与真实值的接近程度.

According to the obvious similar characteristics between seasonal time series data pattern and periodic signal data,this paper comes up with applying wavelet transformation method in signal process to denoise serial data and raises accuracy of model prediction.We use different threshold values,based on different types of wavelet,to remove noise in original series.Take the monthly housing starts data in United States from Jan 1990 to May 2014 as data sample.After that,we build model based on those denoised data and then compare it with the model approached by traditional time series method.The final outcome shows that the model based on denoised data has a better performance：reduce the instability of model caused by local high frequency change;make the data smoother;improve the degree of closeness between prediction and real value.