Témata prací (Výběr práce)Témata prací (Výběr práce)(verze: 336)
Detail práce
   Přihlásit přes CAS
The LSTM approach for Value at Risk prediction
Název práce v češtině: LSTM metoda pro předpověď Value at Risk
Název v anglickém jazyce: The LSTM approach for Value at Risk prediction
Klíčová slova: Value at risk, VaR, Rekurentní neuronové síte, Value at risk forecasting, Visegrádská skupina, LSTM, Joint Supervision, V4
Klíčová slova anglicky: Value at risk, VaR, Recurrent Neural Networks, Value at risk forecasting, Visegrád Four, LSTM, Joint Supervision, V4
Akademický rok vypsání: 2018/2019
Typ práce: bakalářská práce
Jazyk práce: angličtina
Ústav: Institut ekonomických studií (23-IES)
Vedoucí / školitel: Mgr. Marek Hauzr
Řešitel: skrytý - zadáno vedoucím/školitelem
Datum přihlášení: 24.05.2019
Datum zadání: 26.05.2019
Datum a čas obhajoby: 08.09.2021 09:00
Místo konání obhajoby: Opletalova - Opletalova 26, O105, Opletalova - místn. č. 105
Datum odevzdání elektronické podoby:27.07.2021
Datum proběhlé obhajoby: 08.09.2021
Oponenti: RNDr. Michal Červinka, Ph.D.
 
 
 
Kontrola URKUND:
Seznam odborné literatury
Abad, P., et al. (2013). A comprehensive review of Value at Risk methodologies. The Spanish Review of Financial Economics, 12(1), 15-32. doi: 10.1016/j.srfe.2013.06.001
Christoffersen P.F. (1998). Evaluating Interval Forecasts. International Economic Review, 39(4), 841-862. doi: 10.2307/2527341.
Donaldson, R. G. and Kamstra, M. (1996) Forecast combining with neural networks. Journal of Forecasting, 15, 49-61. doi:10.1002/(SICI)1099-131X(199601)15:1<49::AID-FOR604>3.0.CO;2-2
Kupiec P.H. (1995). Techniques for Verifying the Accuracy of Risk Measurement Models. The Journal of Derivatives, 3(2), 73-84. doi: 10.3905/jod.1995.407942
Olah, C. (2015, August 27). Understanding LSTM Networks. Retrieved from: https://colah.github.io/posts/2015-08-Understanding-LSTMs/
So, M., Yu, P. (2006). Empirical analysis of GARCH models in value at risk estimation. Journal of International Financial Markets, Institutions & Money, 16, 180–197.
Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., Salakhutdinov, R. (2014). Dropout: A Simple Way to Prevent Neural Networks from Overfitting. Journal of Machine Learning Research, 15, 1929–1958. Retrieved from: http://jmlr.org/papers/volume15/srivastava14a/srivastava14a.pdf
Taylor, J. W. (2000): “A Quantile Regression Neural Network Approach to Estimating the Conditional Density of Multiperiod Returns.” Journal of Forecasting 19: pp. 299–311
Widrow, B., Rumelhart, D. E., and Lehr, M. A. (1997). Neural networks: Applications in industry, business and science. Communications of the ACM, 37(3).
White, H. (1988). Economic Prediction using Neural Networks: The Case of IBM Daily Stock Returns. Proceedings of the IEEE International Conference on Neural Networks, 2(2), 451 - 458.
Enke, D., Thawornwong, S. (2005). The use of data mining and neural networks for forecasting stock market returns. Expert Systems with Applications, 29, 927-940.
Pham D.T., Liu X. (1995) Financial Prediction Using Neural Networks. In: Neural Networks for Identification, Prediction and Control. Springer, London
Předběžná náplň práce v anglickém jazyce
Research question and motivation
Value at Risk (VaR) is a comprehensive tail-related market risk quantitative measure. In recent years it has become the most widely used technique for measuring future expected risk in both financial and commercial institutions. Unfortunately, due to certain properties of financial time series data (i.e. heteroscedasticity, heavy-tailedness, skewness, etc.), VaR estimation appeared to be a tough statistical proposition and none of the traditional methods has achieved convincing results.
Recurrent Neural Networks have confirmed to be one of the most efficient instruments to process time series data. While the traditional neural networks assume, that inputs are pairwise independent, Recurrent NNs apply the same task on every element of sequential data with output dependent on previous operations. This characteristic and capability to overcome limitations of the traditional forecasting techniques make RNNs a promising alternative approach for risk management purposes.
This paper aims to investigate the efficiency of Recurrent Neural Networks for predicting Value at Risk and their ability to overperform traditional statistical models.
For comparison EVT-POT model and GARCH(p, q) have been chosen. According to several empirical papers, the approach based on the EVT could be considered as the most accurate for estimating VaR. (Abad, P., et al., 2013) Among of the GARCH family, the GARCH(p, q) model overperforms others in terms of VaR estimation. (So and Yu,2006)
Contribution
Despite the vast amount of theoretical and applied research on the neural networks approach in finance (White, 1988; Enke & Thawornwong, 2005; Pham & Liu, 1995), there is little in terms of risk management. This study advances the existing literature by providing empirical evidence of the adequacy of RNN use in modeling and forecasting Value at Risk in emerging stock markets.

Methodology
Recurrent Neural Network will be applied to the forecasting of 1%, 2.5% and 5% Value at Risk of PX, BUX, WIG20, and SAX indexes. The model will be trained on historical daily market data. The evaluation of RNN forecast will be done by comparing with those by GARCH(p, q) and EVT-based dynamic approach using tests of Kupiec (1995) and Christoffersen. (1998)

Outline
1. Abstract
2. Introduction and summary
3. Literature review
3.1. On Recurrent Neural Networks
3.2. On traditional VaR estimation models
4. Methodology
4.1. GARCH(p,q) model
4.2. EVT-POT model
4.3. RNNs
5. Data
5.1. Description
5.2. Preprocessing
6. Results
6.1. GARCH(p,q) model
6.2. EVT-POT model
6.3. RNNs
6.4. Comparison
7. Conclusion
 
Univerzita Karlova | Informační systém UK