University of Twente Student Theses
One-step ahead forecasting IRSS using a hybrid approach: Combining time series models with machine learning models
Broekman, Dave (2024) One-step ahead forecasting IRSS using a hybrid approach: Combining time series models with machine learning models.
|
PDF
10MB |
| Abstract: | This study researches the forecasting accuracy of Interest Rate Swap Spreads (IRSS) us- ing both econometric models and machine learning techniques, with a focus on integrating these approaches in a hybrid model to enhance predictive accuracy. Given the important role of IRSS for financial institutions, accurate forecasting is essential for effective risk man- agement. Our research evaluates the performance of the Hull-White Two-Factor (HW2F) model, the Long Short-Term Memory (LSTM) networks, and introduces an integrative hy- brid model combining the strengths of these two individual models. Through an elaborate literature review, we have identified the important predictors in- fluencing IRSS including financial predictors - such as the zero-coupon bond yield, the Trea- sury yield and the TED Spread - and macroeconomic predictors, such as the gross domestic product, the inflation rate and the unemployment rate. Our methodology involved data preprocessing strategies - including stationarity tests, normalisation and outlier adjustment - to ensure the quality of our dataset. The HW2F model demonstrated high out-of-sample forecasting accuracy with a Root Mean Squared Error (RMSE) of 2.298, a Mean Absolute Error (MAE) of 1.550, and a Mean Absolute Percentage Error (MAPE) of 3.118%. It economically outperformed a naive model for forecasting IRSS. The LSTM model showed it is prone to overfitting risks and strug- gled in high volatility market conditions, with out-of-sample RMSE of 3.708, MAE of 2.571, and MAPE of 5.041%. The hybrid model, designed to leverage the HW2F’s stability and the LSTM’s pattern recognition capabilities using residual correction, showed promise in reducing forecast volatility. However, it did not consistently outperform the HW2F model in out-of-sample IRSS forecasting, with out-of-sample RMSE of 2.680, MAE of 2.131, and MAPE of 4.123%. Our findings highlight the importance of balancing model complexity and stability in financial forecasting. Additionally, we recommend future research to research further model enhancements and improve the model interpretability, for example by Explainable Artificial Intelligence (XAI) techniques, ensuring that advanced models can provide transparent in- sights for financial institutions. This study highlights the continued relevance of traditional econometric models and explores the potential of integrating machine learning for improved financial forecasting. |
| Item Type: | Essay (Master) |
| Clients: | EY, Amsterdam, The Netherlands |
| Faculty: | BMS: Behavioural, Management and Social Sciences |
| Subject: | 83 economics |
| Programme: | Industrial Engineering and Management MSc (60029) |
| Link to this item: | https://purl.utwente.nl/essays/102962 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page