Clustering a Large Number of Time Series with Vector Autoregression Model

• Activity Number: 402
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Chiu Tzu-En*+ and Jeng Shuen-Lin
• Companies: National Cheng Kung University and National Cheng Kung University
• Keywords: clustering ; penalized regression ; similarity measure ; time series ; vector autoregression model

Abstract

This research explores the clustering methods on a large number of return of stocks through the application of the vector autoregression models. When the amount of time series in consideration is large, one effective way to resolve the challenge of the high dimension of the data is through the compact modeling. Then the clustering will be based on certain similarity measures in lower dimension on the parameter space of the model. We will discuss the use of the vector autoregression (VAR) model in two approaches. The first one is to utilize the penalized regression technique (e.g., LASSO and SCAD) in time series model fitting. The second one is to implement the VAR model on certain pre-specified subgroups such as categories of industries and/or random subgroups of stocks. A simulation study will provide the comparisons for the two approaches. The further discussion will focus on the findings of the similar patterns in the stocks from USA and Taiwan markets.