BROWSE

ITEM VIEW & DOWNLOAD

Forecasting Korean LNG import price using ARIMAX, VECM, LSTM and hybrid models

Cited 0 times inthomson ciCited 0 times inthomson ci
Title
Forecasting Korean LNG import price using ARIMAX, VECM, LSTM and hybrid models
Author
SEO, SUNG HYUN
Advisor
Woo, Han-Gyun
Keywords
LNG; ARIMAX; VECM; LSTM; ARIMAX-LSTM; VECM-LSTM
Issue Date
2021-02
Publisher
Graduate School of Technology and Innovation Management
Abstract
In this paper, an optimal forecasting model for the South Korean LNG import price was explored by combining an econometric model, a machine learning model, and a hybrid model. The autoregressive moving average model with extrinsic inputs model (ARIMAX) and VECM were the econometric models, and LSTM was the selected machine learning model. ARIMAX-LSTM and VECM-LSTM were used as hybrid models. Various independent variables, such as the Dubai oil price, European gas price, Australian Newcastle coal price, US natural gas price, Japanese liquified natural gas price and system marginal price in Korea were used for forecasting models. As it was proved that granger causality of each independent variables toward South Korean LNG import price is stronger in the order of the Dubai oil price, European gas price, Australian Newcastle coal price, US natural gas price, Japanese liquified natural gas price and SMP, the variables used for forecasting were added one by one in the order of strong granger causality. Optimal lags were derived from VECM analysis for each variable combination and these were used for VECM and LSTM prediction. As a result of forecasting, 6 LSTM models, 4 VECM-LSTM were ranked in the top 10 forecasting models out of the total 90 models. Single econometric models were not included in the list. The best forecasting model was the LSTM with Dubai oil price, European gas price, Australian Newcastle coal price, US natural gas price, and Japanese liquified natural gas price with lag of 6, and its mean absolute percentage error (MAPE) was 3.5209. In addition, because LNG price forecasting is more important when price fluctuation is high, forecasting models were employed for 11 months with high fluctuation among the test periods. Seven hybrid models, one LSTM models, and two ARIMAX models were ranked in the top 10 forecasting models. VECM-LSTM using Dubai oil price with lag of 5 was derived as the best model with a MAPE of 4.9360. As a result of two forecasting analyses for both the whole and high fluctuation periods, we found that LSTM using Dubai oil price, European gas price, Australian Newcastle coal price, US natural gas price, and Japanese liquified natural gas price with a lag of 6 and VECM-LSTM using Dubai oil price, European gas price with a lag of 5 were ranked within the third best for both tests. Of the two models, the VECM-LSTM is in particular considered as the optimal model in that it has both high forecast accuracy and interpretability.
Description
Department of Management Technology and Innovation Management
URI
Go to Link
Appears in Collections:
SBA_Theses_Master
Files in This Item:
There are no files associated with this item.

find_unist can give you direct access to the published full text of this article. (UNISTARs only)

Show full item record

qrcode

  • mendeley

    citeulike

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

MENU