A Novel Hybrid Decision Making Framework for FX Hedging Strategy Selection Using Machine Learning and Fuzzy TOPSIS - Master’s Thesis

Aug 5, 2026

Foreign exchange markets combine high trading volume with strong volatility, creating substantial FX risk for daily users. Because hedging choices depend on multiple market and firm characteristics, treasury dealers may face inconsistent decisions under time pressure. This thesis proposes a rule-based, data-driven framework that integrates machine learning signals with fuzzy TOPSIS to select FX hedging strategies across criteria and scenarios. LSTM and XGBoost are trained from Jan 2020 to Jul 2025, then assessed via backtesting and comparisons against random strategy baselines.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages102

Reading Time4 h 17 min

AccessFree to download

Abstract

Acknowledgements

Symbols and abbreviations

1 Introduction

2 Literature Review

2.1 Foreign Exchange Hedging Solutions

2.2 Exchange Rate Forecasting

2.3 Multiple Criteria Decision Making

2.4 Hybrid Multiple Criteria Decision Making Models

2.5 Research Gap

3 Theoretical Background

A NOVEL HYBRID DECISION MAKING FRAMEWORK FOR FX HEDGING STRATEGY SELECTION USING MACHINE LEARNING AND FUZZY TOPSIS Lappeenranta–Lahti University of Technology LUT Master’s Program in Business Analytics 2025 Ariba Rehan Siddiqui Examiner(s): Professor Jan Stoklasa Assistant Professor Mahinda Mailagaha Kumbure ABSTRACT Lappeenranta–Lahti University of Technology LUTLUT Business School Business Analytics Ariba Rehan Siddiqui A novel hybrid decision making framework for FX hedging strategy selection using machine learning and fuzzy TOPSIS Master’s thesis 2025 94 pages, 26 figures, 11 tables and 1 appendix Examiner(s): Professor Jan Stoklasa and Assistant Professor Mahinda Mailagaha Kumbure Keywords: machine learning, treasury hedging, FX forecasts, fuzzy TOPSIS, hybrid MCDM, FX hedging strategy FX market is one of the most highly traded and volatile markets in the world with a huge daily trading volume. Corporations, who deal with FX markets on a daily basis, face significant FX risks due to volatile and unpredictable nature of the market. Hence, hedging FX exposures is of paramount importance to avoid losses when the markets turn in unfavourable directions. Previous research has shown that hedging decisions are a multiple criteria problem where several market and firm related characteristics should be considered together. Treasury dealers in a bank, who make these hedging decisions on behalf of a customer, can often make inconsistent decisions when multiple criteria have to be evaluated in a fast-paced and volatile dealing room environment. Hence this study builds a rule-based data-driven decision making framework, using machine learning and fuzzy TOPSIS models, to recommend best hedging strategy under multiple criteria and scenarios. The study trained LSTM and XGBoost with data from January 2020 till July 2025 and generated market signals for the period August 2025 till September 2025. The signals were fed into fuzzy TOPSIS model where a rule-based framework was created for strategy selection. Strategies selected for each case were then back tested and compared with returns from randomly selected strategies and other alternatives. The results showed significance performance of the decision making framework in the 7 day horizon, especially in bearish markets. Performance declined in the longer horizons as forecast accuracy of the machine learning model also declined in longer tenors. Overall, results revealed that our decision making framework can be used to protect the downside risk in short horizon while longer tenors need improvement. ACKNOWLEDGEMENTS First and foremost, I would like to thank my thesis supervisor, Professor Jan Stoklasa, for his invaluable support, insightful guidance and constructive feedback throughout the journey. Thank you for providing me with the guidance and direction that shaped the depth and breadth of this thesis. I would like to thank all the treasury dealers, and my ex-colleagues, who participated in this thesis and provided me with their valuable suggestions and evaluations. Thank you for the time and effort that significantly helped my results. Lastly, and most importantly, I would like to thank my family, especially my husband and my daughter, for their unwavering support during the entire Master’s program. Thank you for being the best source of motivation, emotional support and strength, especially during moments of doubt. This thesis would not have been possible without you. Lappeenranta, 30th November 2025 Ariba Rehan Siddiqui SYMBOLS AND ABBREVIATIONS Abbreviations FX EUR USD MCDM TOPSIS AHP ANP ML LSTM XGBoost RNN ANN SVR ARIMA VAR BERT MSE RMSE MAE MAPE EMA RSI Foreign Exchange Euro US Dollars Multiple Criteria Decision Making Technique for Order Preference by Similarity to Ideal Solution Analytical Hierarchy Process Analytical Network Process Machine Learning Long Short-Term Memory

FAQs

What problem does the thesis address in FX hedging decisions?

It addresses inconsistent hedging strategy decisions caused by the need to evaluate multiple criteria in fast, volatile trading environments.

Which models and method are combined in the proposed framework?

The framework combines machine learning models (LSTM and XGBoost) that generate market signals with a fuzzy TOPSIS-based rule framework for strategy selection.

How are the selected strategies evaluated and what do the results show?

Strategies are back tested for the selected cases and compared with returns from randomly selected strategies and other alternatives; performance is strongest in the 7-day horizon, especially during bearish markets, while it declines in longer horizons due to reduced forecast accuracy.

Aug 5, 2026

Upload by Genevieve