Vinaora Nivo Slider 3.xVinaora Nivo Slider 3.xVinaora Nivo Slider 3.x
Welcome to the ORCS Lab
Welcome to the Laboratory of Operational Research and Complex Systems. Feel free to explore the website and get to know our projects and the people behind the algorithms.
ORCS Lab - Optimisation, Data Science and Complex Systems
Research in the ORCS Lab focuses mainly on the development and application of optimisation methods to complex systems; on data science and statistical modeling; and on the applications of computational intelligence to large-scale power and energy systems.

Latest News

We're happy to invite everyone to Ciniro's presentation of his dissertation,

Construção de um Robô Investidor baseado em Redes Neurais Artificiais e Preditores Econométricos.

on February 20th 2017, 9:30a.m., at EEUFMG's Seminar Room III. Besides Ciniro's advisor, ORCS Lab's Prof. Eduardo Carrano, the dissertation committee will be composed by:

    • Prof. Felipe Campelo (ORCS Lab, Dept. Electrical Engineering, UFMG)
    • Prof. Adriano César Machado Pereira (Dept. Computer Science, UFMG)
    • Prof. Rodrigo Tomás Nogueira Cardoso (Dept. Physics and Mathematics, CEFET-MG)

Click below to read the abstract [UPDATE: and for a picture.]

Abstract: The forecast of stock market trends is a difficult task, as traded goods tend to suffer the most diverse influences. Factors observed on a global scale, such as international politics and exchange flow, are as determinant for variations in the value of an asset as local factors, such as the culture of a given country or the financial indices of a given industry. This perceived complexity, however, when viewed through techniques of pattern detection and time series analysis, allows the implementation of several studies that seek to mathematically model the possible trends in specific financial series. This field of study known as Quantitative Finance is interdisciplinary and involves areas such as statistics, economics and computing. In this context, this paper presents an automated negotiation strategy (robot investor) that combines predictions made by neural networks and econometric predictors through intelligent selectors and different decision committees that are progressively optimized in time to generate profitable scenarios. A methodology was developed for the selection of stocks with high liquidity history and, combined with that, a negotiation algorithm that has risk management was optimized temporarily. All the data of the series that served as the basis for the research are real, but the negotiation process was done with simulated capital. In the experiments the robot investor used a history of 3 years to generate the investment strategies applied to ten shares of several brazilian companies and to the Bovespa index simultaneously. In trading sessions that happened between 2012 and 2013, the results showed statistical significance for both the success rate and for financial return against all benchmarks, even with brokerage costs. Considering the number of trading sessions in the period and the number of assets involved, the working time of the robot in unknown environment refers to a time series of 22 years. The comparisons of several metrics were made against the best predictors of each class when applied individually, to the mean of classes, to the Buy and Hold market technique, and to the evolution of BOVESPA index. The contributions of this study add to other studies in the field which, for being considered recent in the literature, propose a combination of techniques that model the volatility of returns and that, to find optimal parameterizations in the methods, take as their main mechanism classic techniques of artificial intelligence and optimization.