• Search Research Projects
  • Search Researchers
  • How to Use
  1. Back to project page

2018 Fiscal Year Final Research Report

Multivariate time series modeling with sparse regularization and its applications

Research Project

  • PDF
Project/Area Number 16K00067
Research Category

Grant-in-Aid for Scientific Research (C)

Allocation TypeMulti-year Fund
Section一般
Research Field Statistical science
Research InstitutionThe Institute of Statistical Mathematics

Principal Investigator

Kawasaki Yoshinori  統計数理研究所, モデリング研究系, 教授 (70249910)

Project Period (FY) 2016-04-01 – 2019-03-31
Keywordsスパース正則化 / 円滑閾値型推定方程式 / ボラティリティ / 経験類似度 / トピックモデル / 多変量自己回帰モデル / 対数死亡率 / マスク効果
Outline of Final Research Achievements

We promoted the smooth-threshold estimation equations (STEE) to develop a prediction model with high accuracy even in high dimensional time series analysis. First, we worked with variable selection problem in volatility forecasting. We focused on empirical similarity-based models which turned out to produce better forecasting. We also compared topic score series which were extracted news text data using a dynamic topic model. Some topic score series are found to help forecasting. We also applied sparse regularization to vector autoregressive models, especially to the residual vector series from Lee-Carter model for log-mortality. Finally we proposed a variable selection method with which we can salvage true causal variables masked by other variables with strong marginal correlation.

Free Research Field

統計科学

Academic Significance and Societal Importance of the Research Achievements

IoTの推進により,学術・社会の両面でさまざまなセンサーデータが取得可能になっており,その多くは時間と共に観測される時系列データで,往々にして多変量である.従来の多変量時系列モデルは,比較的少数の変数間の相互共分散を通じてリード・ラグ関係を抽出するものであったが,ラグが深くなると高次元では推定が破綻する.本研究で試みたスパース推定との組合せは,今後の大容量の時系列解析につながる成果である.

URL: 

Published: 2020-03-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi