Statistical Theory for Long-memory Property of Economic Time Series and Structural Breaks
| Project/Area Number |
13630027
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Economic statistics
|
|---|
| Research Institution | Hitotsubashi University |
|---|
Principal Investigator |
TANAKA Katsuto Hitotsubashi University, Graduate School of Economics, Professor, 大学院・経済学研究科, 教授 (40126595)
|
|---|
| Project Period (FY) |
2001 – 2002
|
| Project Status |
Completed (Fiscal Year 2002)
|
| Budget Amount *help |
¥1,700,000 (Direct Cost: ¥1,700,000)
Fiscal Year 2002: ¥600,000 (Direct Cost: ¥600,000)
Fiscal Year 2001: ¥1,100,000 (Direct Cost: ¥1,100,000)
|
|---|
| Keywords | deterministic trend / stochastic trend / long-memory / fractional ARIMA / maximum likelihood estimation / wavelet / simulation / 長記憶時系列 / 構造変化 / 共和文 |
|---|
| Research Abstract |
The following are main results obtained in the present research project.
1. Trend, which is the source of nonstationarity of economic time series, can be classified into two categories : one is deterministic trend, and the other stochastic trend. The difference between the two was made clear. The power of statistical tests for determining which trend the actual time series has is quite low. Thus the probability is rather high that we mistakenly conclude the nature of the trend.
2. Stochastic trend may be modeled as fractional ARIMA. In this project, various statistical properties were explored in terms of asymptotic theory.
3. A statistical test was developed if actual time series data is contaminated by noise. At first the test was devised in the time domain, but it was found that the test becomes eventually meaningless as the variation of the noise becomes larger. This is because the signal becomes negligible as the noise becomes larger so that the signal tends to be unidentifiable.
4. Various estimation methods based on wavelets were devised for the long-memory signal plus noise models. The methods were compared with frequency domain methods. The estimation methods employed were OLS and ML methods. It was found that the frequency domain MLE behaves better for stationary cases, but the wavelet-based MLE behaves better for nonstationary case. Since most economic time series exhibits nonstaionarity, this is an advantage of wavelet methods.
|
Report
(3 results)
Research Products
(21 results)
