Loading...

An Online Portfolio Selection Algorithm Using Pattern-matching Principle

Azin, Pejman | 2018

658 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 51413 (01)
  4. University: Sharif University of Technology
  5. Department: Industrial Engineering
  6. Advisor(s): Khedmati, Majid
  7. Abstract:
  8. According to the rise of turnover and pace of trading, accelerating of analysis and making decision is unavoidable. Humans are unable to analyze big data quickly without behavioral biases so, using machines to analyze big data seems critical. Hence, financial markets tend to apply algorithmic trading in which some techniques like data mining and machine learning are notable. OLPS 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. This article presents an online portfolio selection algorithm. The online portfolio selection sequentially selects a portfolio over a set of assets in order to maximize the expected return, during a long period. Accordingly, it is made a decision on optimal portfolio through algorithms following the pattern matching principle in each period and the optimal portfolio is upgraded at the beginning of each period. The proposed method has two steps: i) sample selection ii) portfolio optimization. First, in the sample selection, clustering algorithms including k-means, k-medoids, spectral clustering and hierarchical clustering are applied to discover time windows similar to the recent time window. Second, in portfolio optimization, after similar time windows found and the market behavior based on the following day are predicted by considering similar days, the log optimum function is used along with the transaction cost. Hence, four algorithms including KMNLOG, KMDLOG, SPCLOG and HRCLOG are provided. The presented algorithms are executed over the portfolio comprises 15 most active stocks of NASDAQ, in the 2-year period between January 4, 2016 and December 29, 2017, and their performance has been evaluated. The results shows that the provided algorithms in this article have more appropriate performance in terms of efficiency compared to the algorithms proposed in the literature. Finally, algorithms are ranked by TOPSIS method along with 4 attributes of the return, risk, the prediction accuracy and the running time of algorithms. Accordingly, the HRCLOG, KMNLOG, KMDLOG and SPCLOG algorithms achieved the first to fourth place, respectively
  9. Keywords:
  10. Data Mining ; Clustering ; Pattern Matching ; Portfolio Selection Problem ; Online Portfolio Selection ; Algorithmic Trading

 Digital Object List

 Bookmark

No TOC