Two Algorithms with Logarithmic Regret for Online Portfolio Selection
In online portfolio selection, an online investor needs to distribute her wealth iteratively and hopes to maximize her final wealth. To model the behavior of the prices of assets in a financial market, we consider two measures of the price relative vectors: quadratic variability and deviation. There exist algorithms which achieve good performance in terms of these two measures. However, the theoretical guarantees depend on an additional parameter, which may not be available before the investor chooses her strategies. In this paper, the performances of the algorithms are tested using real stock market data to understand the influence of this additional parameter.
KeywordsOnline portfolio selection Regret Quadratic variability Deviation
- 1.Agarwal, A., Hazan, E., Kale, S., Schapire, R.E.: Algorithms for portfolio management based on the Newton method. In: ICML, pp. 9–16 (2006)Google Scholar
- 2.Chiang, C.K., Yang, T., Lee, C.J., Mahdavi, M., Lu, C.J., Jin, R., Zhu, S.: Online optimization with gradual variations. J. Mach. Learn. Res. Proceedings Track 23, 6.1–6.20 (2012)Google Scholar
- 4.Das, P., Banerjee, A.: Meta optimization and its application to portfolio selection. In: Proceedings of International Conference on Knowledge Discovery and Data Mining (2011)Google Scholar
- 6.Helmbold, D.P., Schapire, R.E., Singer, Y., Warmuth, M.K.: On-line portfolio selection using multiplicative updates. In: ICML, pp. 243–251 (1996)Google Scholar
- 8.Zinkevich, M.: Online convex programming and generalized infinitesimal gradient ascent. In: ICML, pp. 928–936 (2003)Google Scholar