Comparison of double auction bidding strategies for automated trading agents

• Název práce v češtině: Comparison of double auction bidding strategies for automated trading agents
• Název v anglickém jazyce: Comparison of double auction bidding strategies for automated trading agents
• Klíčová slova: Algoritmické obchodování, Strategie nabízení, ZIP, Zero-Intelligence Plus, GDX, Adaptivní Agresivita, AA, Obchodování agentů, Autonomní obchodování, Ojovo pravidlo
• Klíčová slova anglicky: Algorithmic trading, Bidding strategy, ZIP, Zero-Intelligence Plus, GDX, Adaptive Aggressiveness, AA, Agent trading, Autonomous trading, Oja's rule
• Akademický rok vypsání: 2013/2014
• Typ práce: diplomová práce
• Jazyk práce: angličtina
• Ústav: Institut ekonomických studií (23-IES)
• Vedoucí / školitel: Ing. Aleš Maršál, Ph.D.
• Řešitel: Mgr. Daniel Vach - zadáno vedoucím/školitelem
• Datum přihlášení: 19.06.2014
• Datum zadání: 19.06.2014
• Datum a čas obhajoby: 23.09.2015 00:00
• Místo konání obhajoby: IES
• Datum odevzdání elektronické podoby: 31.07.2015
• Datum proběhlé obhajoby: 23.09.2015
• Oponenti: Martin Burda, Ph.D.

Předběžná náplň práce

Motivation:
Algorithmic trading is of great importance in markets today. It is a phenomenon which deserves a lot of attention. Trading algorithms for continuous double-auction are studied and developed vastly in literature.

There have been three main strategies examined thoroughly in literature: namely ZIP (zero intelligence plus), the GD (Gjerstad-Dickhaut) class and AA (adaptive aggressiveness). ZIP and AA make use of Widrow-Hoff adaptation and GD is based on belief functions. Luca and Cliff (2011) compares these three strategies against each other and against humans in artificial market. They come to a conclusion that algorithmic strategies outperform human traders and that AA performs better than the rest of algorithmic strategies.

I am going to challenge the algo-to-algo results of De Luca and Cliff (2011) by repeating their experiment, but with different parameters of the artificial world. Moreover, I will try to test these three families of algos on historical data obtained from the real market.

Last but not least, I will try to develop my own algorithmic strategy. I will test it on ex post real market data, in the artificial world and I will show how these algorithmic strategies perform on generated stochastic process simulating the ex ante real market data.

Hypotheses:
1. AA trade strategy does not outperform the other trading strategies in different artificial world setting as significantly as in De Luca and Cliff (2011).
2. There is a difference in how trade strategies perform based on the number of rival strategies included in experiment.
3. There is a difference in how trade strategies perform based on the number of agents following each rule.
4. It is possible to rank these algorithmic strategies with sufficient significance and this order is robust to changes in the experimental setup.


Methodology:
I am going to test ZIP60, GDX, AA and newly developed trading strategy against each other in competitive environment similar to one used by De Luca and Cliff (2011). I will use OpEx (Open Exchange Experimental Laborator). Compared to them, I will change reasonably the artificial world setting in number of participating strategies and in number of agents of each strategy in one experiment. Based on this I will produce sensitive analysis of outcomes on different number of strategies and agents included. Then I will use real order book data from one day as an input for decision making of strategies and I will compare their successfulness in this experiment. Then I will compare these results with artificial world results. I will also try to model artificial orderbooks with the same generic properties as my real order book simulated by Hawkes process similarly to Hewlett (2006). I will apply all four strategies on these orderbooks and see whether there are changes to the pattern of success. This step should make the results more robust to the randomness.

Expected Contribution:
This thesis should discover whether the results of De Luca and Cliff (2011) are robust to some changes in artificial world setting. I will also show whether these results are valid for example of historical data as well. I will challenge the rank order of trading algos introduced in De Luca and Cliff (2011). I will possibly show that it is not a consistent result or at least I will provide some more evidence to this topic. I will introduce completely new algorithmic trading strategy and show whether it performs worse, comparable or better than some known algos.

Outline:
1) Introduction
2) Literature Review
3) Definition of Trade Agents
4) Definition of Artificial World
5) Results in artificial world
6) Results with historical prices
7) Results with market simulating stochastic function
8) Comparison and Discussion
9) Conclusion


Core Bibiliography:
1) Wellman, Michael P. (2011) "Trading Agents", Synthesis Lectures on Artificial Intelligence and Machine Learning, Volume.5, Issue.3, pp.1, ISSN: 19394608
2) D. Cliff (1998) "Genetic optimization of adaptive trading agents for double-auction markets", Proc. Comput. Intell. Financial Engineering (CIFEr), pp.252 -258
3) D. Cliff (2006) ZIP60: Further explorations in the evolutionary design of trader agents and online auction market mechanisms.
4) M. De Luca and D. Cliff (2011) Human-agent auction interactions: adaptive-aggressive agents dominate. In Proc. ICAART, ISBN: 978-1-57735-513-7
5) D. Cliff and J. Bruten (1997) Minimal intelligence agents for bargaining behaviors in market-based environments. Technical report, HPL-97-91, HP Labs
6) Vytelingum, P. (2006) The structure and behaviour of the Continuous Double Auction. PhD Thesis. University of Southampton.
7) Anonymous: OpEx: An Open Exchange Experimental Economics Laboratory in a Box. Technical Report in preparation.
8) Preist, C., Van Tol, M. 1998. Adaptive agents in a persistent shout double auction. In Proceedings of the First International Conference on Information and Computation Economies. ACM Press,11-18.
9) Tesauro, G., Das, R. 2001. High-performance bidding agents for the continuous double auction. In Proceedings of the Third ACM Conference on Electronic Commerce, 206-209.
10) Foucault, T., O. Kadan, E. Kandel. 2005. Limit order book as a market for liquidity. Review of Financial Studies, 18(4), pp. 1171–1217.
11) Hewlett, P. 2006. Clustering of order arrivals, price impact and trade path optimization. Workshop on Financial Modeling with Jump processes, Ecole Polytechnique, 6–8 September 2006.