View : 821 Download: 0

Full metadata record

DC Field Value Language
dc.contributor.advisor신동완-
dc.contributor.author김미현-
dc.creator김미현-
dc.date.accessioned2021-01-25T16:30:24Z-
dc.date.available2021-01-25T16:30:24Z-
dc.date.issued2021-
dc.identifier.otherOAK-000000173381-
dc.identifier.urihttp://dcollection.ewha.ac.kr/common/orgView/000000173381en_US
dc.identifier.urihttps://dspace.ewha.ac.kr/handle/2015.oak/256125-
dc.description.abstractThis study considers a combination VECM(Vector Error Correction Model) and LSTM(Long Short-Term Memory) for multivariate time series to increase forecast performance. VECM is a traditional linear method of multivariate time series prediction, used when there are some cointegration relationships between nonstationary time series. LSTM is a nonparametric nonlinear deep learning method that reflects short-term and long-term relationships of time series variables. The linear relation among nonstationary variables is first explained by VECM and the unexplained nonlinear relation in the VECM is handled by the deep learning method of LSTM. The combined model of VECM and LSTM has better performance than the single model of VECM.;본 연구는 공적분 관계가 있는 다변량 시계열에 대해 VECM(Vector Error Correction Model)과 LSTM(Long Short-Term Memory)을 결합하여 예측력을 높였다. VECM은 다변량 시계열 예측의 전통 선형적 방법으로, 비정상 시계열 간 공적분 관계가 있을 때 사용한다. LSTM은 시계열 변수의 단·장기적 관계를 반영하는 비모수 비선형 딥러닝 방법이다. 모형 결합 방법은 VECM으로 자료를 설명하고 남은 잔차를 딥러닝으로 예측하여 결합하는 것이다. 해당 결합방법은 VECM 단일 모형보다 성능이 높다. 동시에 VECM의 설명력을 그대로 가져갈 수 있다는 점에서 해석이 불가능하다는 딥러닝 단일모델의 단점 또한 보완된다.-
dc.description.tableofcontentsI. Introduction 1 II. Related Works 3 III. Method 4 A. Model Description 4 1. VECM 4 2. LSTM 5 B. Normalization: min-max normalization & widow moving scaling 11 C. Combining Models 13 IV. Data 15 A. Exchange rate & composite stock price index 15 B. Vegetable price data 17 V. Model Fitting Information 20 A. Exchange rate & composite stock price index 20 1. VECM 20 2. LSTM 21 B. Vegetable price data 21 1. VECM 21 2. LSTM 22 VI. Forecast results 23 A. Exchange rate & composite stock price index 23 B. Vegetable price data 24 VII. Conclusion. 25 Bibliography 26 국문초록 27-
dc.formatapplication/pdf-
dc.format.extent943316 bytes-
dc.languageeng-
dc.publisher이화여자대학교 대학원-
dc.subject.ddc500-
dc.titleA Hybrid of VECM and LSTM for Forecasting Multivariate Time Series-
dc.typeMaster's Thesis-
dc.creator.othernameKim, Meehyun-
dc.format.pageiv, 27 p.-
dc.identifier.thesisdegreeMaster-
dc.identifier.major대학원 통계학과-
dc.date.awarded2021. 2-
Appears in Collections:
일반대학원 > 통계학과 > Theses_Master
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

BROWSE