Loading...

An Online Portfolio Selection Algorithm Using Recurrent Neural Networks and Controlling the Risk of Tradings with Value at Risk Method

Karimi, Nima | 2021

406 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 53922 (01)
  4. University: Sharif University of Technology
  5. Department: Industrial Engineering
  6. Advisor(s): Khedmati, Majid
  7. Abstract:
  8. Nowadays, capital markets play a key role in the economies of countries. Hence, this market is expanding more and more every day. In such circumstances, traditional analysis methods such as fundamental analysis and technical analysis have lost their position due to low speed and accuracy. In recent years, automated trading systems have been proposed as a solution to these problems. The online portfolio selection, which sequentially allocates capital among a set of assets aiming to maximize the final return of investment in the long run, is the core problem in algorithmic trading. In this research, we present an online portfolio selection algorithm based on pattern matching principle. Pattern matching algorithms consist of two steps of sample selection or in other words finding time windows similar to the current time window and stock portfolio optimization based on the results of the first step. In the algorithms presented in this research, in the first step, due to the nature of the time series of stock market data, one of the powerful tools in this field, namely recurrent neural networks, has been used. In this regard, two well-known and successful types of recursive neural networks, LSTM and GRU neural networks have been used to predict the data. In the second step, which is the optimization step, the log optimal function is used, and finally this has led to the presentation of two algorithms, LSTMLOG and GRULOG. In addition, in these algorithms, in order to control the risk of transactions and avoid high-risk transactions, the value at risk method has been used. In order to evaluate the results of the proposed algorithms and compare the results with other algorithms in this field, five real data sets in different financial market conditions in terms of ascending and descending have been used and the presented algorithms have been implemented on them. The evaluation results show that the algorithms presented in this study are in a favorable position in terms of return and risk in comparison with the algorithms in the online portfolio selection literature. Finally, to further evaluate the results and the effect of risk control criteria on the results, sensitivity analysis on the risk of the algorithm was performed and the results were recorded
  9. Keywords:
  10. Online Portfolio Selection ; Pattern Matching ; Recurrent Neural Networks ; Risk ; Algorithmic Trading ; Value at Risk ; Risk Control

 Digital Object List

 Bookmark

No TOC