Forex forecasting with Support vector regression and Long short-term memory recurrent neural network

• Název práce v češtině: Předpověď Forexu SVM regresí a LSTM neuronovou sítí
• Název v anglickém jazyce: Forex forecasting with Support vector regression and Long short-term memory recurrent neural network
• Klíčová slova: Forex, předpověď, strojové učení, SVM regrese, rekurentní SVM regrese, LSTM neuronová síť
• Klíčová slova anglicky: Forex, forecasting, machine learning, support vector regression, recurrent support vector regression, long short-term memory recurrent neural network
• Akademický rok vypsání: 2018/2019
• Typ práce: bakalářská práce
• Jazyk práce: angličtina
• Ústav: Institut ekonomických studií (23-IES)
• Vedoucí / školitel: Mgr. Jan Šíla, M.Sc.
• Řešitel: skrytý - zadáno vedoucím/školitelem
• Datum přihlášení: 10.05.2019
• Datum zadání: 28.05.2019
• Datum a čas obhajoby: 09.06.2021 09:00
• Datum odevzdání elektronické podoby: 02.05.2021
• Datum proběhlé obhajoby: 09.06.2021
• Oponenti: prof. PhDr. Ladislav Krištoufek, Ph.D.

Seznam odborné literatury

Alaa F. Sheta, Sara Elsir M. Ahmed and Hossam Faris, “A Comparison between Regression, Artificial Neural Networks and Support Vector Machines for Predicting Stock Market Index” International Journal of Advanced Research in Artificial Intelligence (IJARAI), 4(7), 2015.
Baasher, Areej & Fakhr, Mohamed. (2011). Forex trend classification using machine learning techniques. Proceedings of the 11th WSEAS International Conference on Applied Computer Science. 41-47.
Beneki, Christina & Yarmohammadi, Masoud. (2014). Forecasting exchange rates: An optimal approach. Journal of Systems Science and Complexity. 27. 21-28. 10.1007/s11424-014-3304-5.
Mark, Nelson C. (1995): “Exchange rates and fundamentals: Evidence on longhorizon predictability.” The American Economic Review pp. 201–218.
Rodríguez,P.N.&S.Sosvilla-Rivero(2006): “Using machine learning algorithms to find patterns in stock prices.” FEDEA Working Paper No. 2006-12
Yu, Lean & Wang, Shouyang & Huang, Wei & Lai, Kin Keung. (2019). ARE FOREIGN EXCHANGE RATES PREDICTABLE? A SURVEY FROM ARTIFICIAL NEURAL NETWORKS PERSPECTIVE.

Předběžná náplň práce v anglickém jazyce

Research question and motivation:

The predictivity of exchange rates was researched ever since gold standard was abandoned in favor of floating exchange rates. In fact, Mussa (1979) first recognized, that changes in exchange rates cannot be predicted for spot values. Later Mark (1995) empirically demonstrated, that exchange rates can be predicted for longer time intervals. However, for short-term volatilities his predictions were not good.

Nowadays, Forex is the most liquid market on the world with daily trading volume bigger than 5 trillion USD. Governments, central banks, and other financial institutions use Forex for investing 24 hours per day, 119 consecutive hours in a week. The importance of Forex for international trade is undeniable. Trading made on Forex is estimated to be mainly invested into speculations on currency floats and consequent profits. A recent boom of quantitative trading and algorithmic trading strategies disrupted traditional approaches. It created a space for an adaptability of prediction which may outperform current algorithms.

Supervised machine learning algorithms like neural networks and support vector machines regression can find patterns in large data samples and adapt their prediction on recent changes. Beneki and Yarmohammadi (2014) showed that neural network can compete in accuracy of GBP/EUR prediction with vector singular spectrum analysis and recurrent singular spectrum analysis. This thesis focuses on answering a question: Which one of selected methods is the best for main traded fiat currency pairs and USD/BTC pair trade prediction?

My work demonstrates a comparison of time series regression, support vector machine regression, and artificial neural network with random walk.

Contribution

My aim is to compare methods of prediction for time series data, that were not yet compared by any study. Sheta, Ahmed and Faris (2015) used very similar comparison for S&P 500 index but used panel data with 27 explaining variables.
Results of this work could be used as a basis for further academic research or business decision making process.

Methodology

For running the regression and the development of machine learning methods are used 1M time series data obtained from credible source in specified time interval. Then the results of machine learning are validated in order to confirm an accuracy of their setup for reliable prediction. After all these steps, final findings are compared with real data. This comparison is based on MAE and RMSE evaluation criteria. From obtained results is decided which method produces the most accurate prediction.

Outline

1. Introduction
2. Literature review
3. Methodology
4. Dataset
5. Results
6. Conclusion