Deep Learning for New Tanker Shipbuilding Price Forecasting
- 주제(키워드) Shipbuilding Price Forecasting , Tanker Newbuilding Prices
- 주제(DDC) 658
- 발행기관 아주대학교 일반대학원
- 지도교수 장병윤
- 발행년도 2026
- 학위수여년월 2026. 2
- 학위명 석사
- 학과 및 전공 일반대학원 경영학과
- 실제URI http://www.dcollection.net/handler/ajou/000000035971
- 본문언어 영어
- 저작권 아주대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
Accurate forecasting of shipbuilding prices is critical for shipyards, shipowners, and financial stakeholders, as these prices are highly sensitive to fluctuations in freight markets, steel costs, macroeconomic conditions, and broader fleet dynamics. Traditional time-series approaches such as ARIMA and SARIMA have long been employed in maritime economics due to their transparency and effectiveness in modelling structured price movements. However, their predominantly linear formulation may limit their ability to fully capture nonlinear and short-term dynamics present in volatile shipbuilding markets. Recent advances in data-driven modelling have led to increasing interest in deep learning techniques, particularly Convolutional Neural Networks (CNN) and hybrid CNN-LSTM architectures, which are capable of learning complex temporal patterns from historical price data. Accordingly, this study evaluates and compares the forecasting performance of ARIMA, SARIMA, CNN, CNN-LSTM, and a hybrid SARIMA+CNN-LSTM framework using historical newbuilding price data for major tanker segments. Forecast accuracy is assessed using RMSE, MAE, and MAPE based on out-of-sample forecasts. The empirical findings indicate that no single model consistently outperforms others across all tanker categories. Linear models such as ARIMA and SARIMA remain highly competitive, particularly for tanker segments exhibiting relatively stable and trend-dominated price behaviour. In contrast, deep learning models, including CNN and CNN-LSTM, demonstrate improved performance for segments characterised by stronger nonlinear dynamics and short-term variability. The hybrid SARIMA+CNN-LSTM approach provides accuracy gains in selected cases but does not consistently outperform the best standalone models. Overall, the results highlight the importance of model selection tailored to tanker-specific price dynamics, rather than reliance on a universally superior forecasting framework. These findings contribute to the shipbuilding economics literature by offering a nuanced evaluation of statistical, deep learning, and hybrid approaches for price forecasting in an increasingly volatile maritime environment.
more목차
1. Introduction 1
2. Literature Review 3
3. Methodology 8
3.1. Research Design 8
3.2. VLCC – Very Large Crude Carrier (~320,000 DWT): 10
3.3. Suezmax (~150,000 DWT): 11
3.4. Aframax (~110,000–120,000 DWT): 11
3.5. LR2 Product Tanker (~105,000 DWT): 11
3.6. LR1 Product Tanker (~70,000–80,000 DWT): 11
3.7. MR Tanker – Medium Range: 12
3.8. Scaling for neural network models 14
3.9. Forecasting Model 15
3.10. ARIMA Model 17
3.11. SARIMA Model 20
3.12. CNN-Style Weekly Forecasting Model (MLP with Sliding Window) 21
3.13. CNN-LSTM–Style Weekly Forecasting Model (Deep Nonlinear Autoregression) 24
3.14. Hybrid SARIMA + CNN-LSTM–Style Model 27
4. RESULTS AND DISCUSSION 30
4.1. VLCC – Very Large Crude Carrier (~320,000 DWT) 33
4.2. Suezmax (~150,000 DWT) 37
4.3. Aframax (~110,000–120,000 DWT) 41
4.4. LR2 Product Tanker (~105,000 DWT) 45
4.5. LR1 Product Tanker (~70,000–80,000 DWT) 49
4.6. MR Tanker – Medium Range 52
4.7. Discussion of Findings 57
5. CONCLUSION 60
REFERENCE 62
APPENDIX 1 67
APPENDIX 2 71
APPENDIX 3 75
