Discovering macroeconomic phases using statistical methods.

• Main Author: Wong, Jian Cheng.
• Other Authors: Lian Heng
• Format: Final Year Project
• Language: English
• Published: 2009
• Subjects: DRNTU::Science::Mathematics::Statistics
• Online Access: http://hdl.handle.net/10356/14733
• Institution: Nanyang Technological University

Economies and financial markets are complex system which can exist in different macroeconomic phases. To build predictive models, we need to know how many such phases there are, and what their statistical properties are. In this report, a recursive segmentation scheme based on the Jensen-Shannon divergence is employed to extract statistical signatures of different market phases from the Dow Jones Industrial Average half-hourly time series from January 1997 to March 2007. Each stationary segment of the segmented time series is described by a different Gaussian model. We find a total of 102 segments, and these segments can be grouped into five clusters by hierarchical clustering. There is a low volatility phase that can be associated with expansion, a high volatility phase within which economic contraction occurs, a moderate volatility correction phase, an extremely high volatility crash phase, and also an extremely-low-volatility phase whose macroeconomic nature is not yet understood. The market is predominantly found in the low and high volatility phases. These are interrupted by the correction phase. Moreover, the crash phase is mostly found within the high volatility phase. Transition from low to high and high to low occurs over a period of roughly one year.