Neñer, Julián M. (2021) Redes neuronales para la búsqueda de estrategias óptimas en problemas de econofísica / Neural networks for finding optimal strategies in econophysics problem. Maestría en Ciencias Físicas, Universidad Nacional de Cuyo, Instituto Balseiro.
![[img]](/style/images/fileicons/application_pdf.png)
| PDF (Tesis) Español 21Mb |
Resumen en español
En esta tesis se desarrollaron técnicas computacionales para estudiar dos problemas dentro del área de la econofísica: modelos multi-agente de distribución de riqueza, y predicción de series temporales en mercados nancieros. Se estudiaron modelos cinéticos de intercambio de riqueza basados en sistemas de agentes. En particular, se estudiaron los modelos conocidos como Yard-Sale y Merger- Spinoff, cuyos resultados a nivel macroscópico han sido de especial interés debido a sus analogías con casos empíricos. Se hace especialmente evidente la acumulación de riqueza en pocos individuos, una de las características centrales en sistemas económicos capitalistas. Se analizó la aparición de leyes de potencias en las distribuciones de riqueza, a partir de las cuales se determinaron los correspondientes índices de Pareto e índices de Gini, entre otras características típicas de sistemas económicos. Se hizo especial hincapié en realizar un análisis de las dinámicas microscópicas individuales de los agentes. La implementación de algoritmos eficientes programados en GPU utilizando CUDA C/C++ hizo posible la obtención de buena estadística en estos modelos de carácter estocástico, permitiendo visualizar la aparición de ergodicidad y fenómenos invariantes de escala. El modelo Yard-Sale resultó ser de especial complejidad a nivel microscópico. Se encontraron estrategias óptimas que maximizan la riqueza individual de cada agente, realizando su entrenamiento mediante un algoritmo genético. La adición de diferentes niveles de racionalidad, dada por la cantidad de información que poseen de su entorno, mostró resultados prometedores tanto a nivel macroscópico como microscópico. Por otro lado se estudió la implementación de redes neuronales para la predicción de la tendencia del par EUR/USD. Se investigaron diferentes arquitecturas y métodos de entrenamiento que llevaron al desarrollo de un algoritmo propio. Este resultó ser incluso más eficiente que las librerías mas comúnmente utilizadas, tanto en tiempo como en utilización de memoria.
Resumen en inglés
In this work computational techniques were developed to study two main problems within the area of econophysics: multi-agent models for wealth distribution, and time series prediction of nancial markets. Kinetic wealth exchange models based on systems of agents were studied. In particular, the studied models are known as Yard-Sale and Merger-Spinoff, whose results at the macroscopic level have been of special interest due to analogies with empirical data. It is made specially clear how the wealth is acculumated by very few individuals, one of the main characteristics of capitalist economic systems. The emergence of power laws in the wealth distributions was analized, from which the corresponding Pareto and Gini indices were determined, among other typical characteristics of economic systems. Special emphasis was made in analyzing the individual microscopic dynamics of the agents. The implementation of effcient algorithms in GPU using CUDA C/C++ was of utter importance to obtain enough sampling in these models of stochastic nature, allowing for the visualization of new previously unobserved characteristics such as ergodicity and scale invariant phenomena. The Yard-Sale model turned out to be of special complexity at the microscopic level. Optimal strategies were found that maximize the individual wealth of each agent, by performing their training through a genetic algorithm. The addition of different levels of rationality given by the amount of available information from their environment showed promising results, both at the macroscopic and microscopic level. On the other hand, the implementation of neural networks for the prediction of the tendency of the pair EUR/USD was studied. Different architectures and training methods were explored, which led to the development of our own algorithm. The algorithm proved to be more effcient than the best known libraries usually implemented in this context, both in time and memory usage.
| Tipo de objeto: | Tesis (Maestría en Ciencias Físicas) |
|---|---|
| Palabras Clave: | Neural networks; Redes neuronales; [Econophysics; Econofísica; Wealth distributions; Distribución de riqueza; Forex; CUDA] |
| Referencias: | [1] Mantegna, R. N., Stanley, H. E. An Introduction to Econophysics: Correlations and Complexity in Finance. Cambridge University Press, 1999. 1 [2] Slanina, F. Essentials of Econophysics Modelling. Oxford University Press, 2014. 1 [3] Stanley, H. E. Anomalous uctuations in the dynamics of complex systems: from dna and physiology to econophysics. Physica A: Statistical Mechanics and its Applications, 224, 302{321, 1996. 1 [4] Schinckus, C. 1996{2016: Two decades of econophysics: Between methodological diversication and conceptual coherence. Eur. Phys. J. Special Topics, 225, 3299{3311, 2016. 1 [5] Reisman, G. Capitalism: A Complete and Integrated Understanding of the Nature and Value of Human Economic Life. Jameson Books, 1997. 2 [6] Kaul, I. Human Development Report 1992. Oxford University Press, 1992. 2 [7] Pareto, V. Cours d'Economie Politique. F. Rouge, Lausanne, 1897. 2 [8] Pianegonda, S., Iglesias, J. R. Inequalities of wealth distribution in a conservative economy. Physica A: Statistical Mechanics and its Applications, 342, 193{199, 2004. 3 [9] Souma, W. Universal structure of the personal income distribution. Fractals: Complex Geometry, Patterns, and Scaling in Nature and Society, 9, 463{470, 2001. 3 [10] Dragulescu, A. A., Yakovenko, V. M. Statistical mechanics of money, income, and wealth: A short survey. Modeling of Complex Systems: Seventh Granada Lectures, pag. 180{183, 2003. 3 [11] Dragulescu, A. A., Yakovenko, V. M. Exponential and power-law probability distributions of wealth and income in the united kingdom and the united states. Physica A: Statistical Mechanics and its Applications, 299, 213{221, 2001. 3 [12] Nirei, M., Souma, W. A two factor model of income distribution dynamics. Review of Income and Wealth, 53, 440{459, 2007. 3 [13] Cardoso, B. F., Goncalves, S., Iglesias, J. R. Wealth distribution models with regulations: Dynamics and equilibria. Physica A: Statistical Mechanics and its Applications, 2019. 3, 8, 66 [14] Banerjee, A., Yakovenko, V. M. Universal patterns of inequality. New Journal of Physics, 12, 2010. 4 [15] Li, S. Changes in the distribution of wealth in china 1995-2002. World Institute for Development Economic Research (UNU-WIDER), Working Papers, 2007. 4 [16] Chatterjee, A., Chakrabarti, B. K., Manna, S. S. Pareto law in a kinetic model of market with random saving propensity. Physica A: Statistical Mechanics and its Applications, 335, 155{163, 2004. 4, 16 [17] Chakraborti, A., Chakrabarti, B. Statistical mechanics of money: how saving propensity affects its distribution. Eur. Phys. J. B, 17, 167{170, 2000. [18] Caon, G., Goncalves, S., Iglesias, J. The unfair consequences of equal opportunities: Comparing exchange models of wealth distribution. Eur. Phys. J. Special Topics, 143, 69{74, 2007. [19] Chakraborti, A., Toke, I. M., Patriarca, M., Abergel, F. Econophysics: Empirical facts and agent-based models. Quantitative Finance, 11, 991{1012, 2011. 4, 16 [20] Patriarca, M., Heinsalu, E., Chakraborti, A. Basic kinetic wealth-exchange models: common features and open problems. Physics of Condensed Matter, 73, 145{153, 2010. 4, 8 [21] Hayes, B. Follow the money. American Scientist, 90, 400{405, 2002. 5, 8 [22] Boghosian, B. M., Devitt-Lee, A., Wang, H. The growth of oligarchy in a yardsale model of asset exchange - a logistic equation for wealth condensation. 1st International Conference on Complex Information Systems, 2016. 8 [23] Chorro, C. A simple probabilistic approach of the yard-sale model. Statistics and Probability Letters, 112, 35{40, 2016. 8 [24] Sinha, S. Stochastic maps, wealth distribution in random asset exchange models and the marginal utility of relative wealth. Physica Scripta, T106, 59{64, 2003. 8 [25] Risau-Gusman, S., Laguna, M. F., Iglesias, J. R. Wealth distribution in a network with correlations between links and success. Econophysics of Wealth Distributions: Econophys-Kolkata I, págs. 149{158, 2005. 9 [26] Iglesias, J. R., Risau-Gusman, S., Laguna, M. F. Inequalities of wealth distribution in a society with social classes, 2006. 5, 9 [27] Plerou, V., et al. Econophysics: nancial time series from a statistical physics point of view. Physica A, 279, 443{456, 2000. 5 [28] Immonen, E. Simple agent-based dynamical system models for ecient nancial markets: Theory and examples. Journal of Mathematical Economics, 2016. 5 [29] Vargas, M. R., de Lima, B. S. L. P., Evsukoff, A. Deep learning for stock market prediction from nancial news articles. CIVEMSA, 2017. 5 [30] Pahwa, N. S., et al. Stock prediction using machine learning a review paper. International Journal of Computer Applications, 163, 36{43, 2017. 5 [31] Scafetta, N., Picozzi, S., West, B. J. Pareto's law: a model of human sharing and creativity, 2002. 9 [32] Shorrocks, A., Davies, J., Lluberas, R. Global Wealth Report 2019. Credit Suisse Research Institute, 2019. 9 [33] Matsumoto, M., Nishimura, T. Mersenne twister: A 623-dimensionally equidistributed uniform pseudo-random number generator. ACM Transactions on Modeling and Computer Simulation, 8, 3{30, 1998. 9 [34] Patriarca, M., Chakraborti, A., Germano, G. In uence of saving propensity on the power-law tail of the wealth distribution. Physica A: Statistical Mechanics and its Applications, 369, 723{736, 2006. 15, 16 [35] Patriarca, M., Chakraborti, A., Kaski, K., Germano, G. Kinetic theory models for the distribution of wealth: power law from overlap of exponentials. New Economic Windows, 2, 93{110, 2005. 16 [36] Mount, J. Ergodic theory for interested computer scientists, 2012. 25 [37] Gorban, A. N., Smirnova, E. V., Tyukina, T. A. Correlations, risk and crisis: From physiology to nance. Physica A: Statistical Mechanics and its Applications, 389, 3193{3217, 2010. 37 [38] Sosa, I. Procesos de intercambio de energía en sistemas de osciladores mecánicos acoplados. Tesina de grado del Instituto Balseiro, 2019. 37 [39] Csáji, B. C. Approximation with articial neural networks, 2001. 42 [40] Neñer, J., Laguna, M. F. En búsqueda de estrategias óptimas en modelos de sistemas capitalistas. Tesina de grado del Instituto Balseiro, 2019. 52 [41] www.metatrader4.com. 70 [42] Rumelhart, D. E., Hinton, G. E., Williams, R. J. Learning representations by back-propagating errors. Nature, 323, 533{536, 1986. 75 [43] Ruder, S. An overview of gradient descent optimization algorithms. ArXiv, 2016. 77 [44] Kingma, D. P., Ba, J. Adam: A method for stochastic optimization. ArXiv, 2014. 77 [45] Srivastava, N., et al. Dropout: A simple way to prevent neural networks from overtting. Journal of Machine Learning Research, págs.. 1929{1958, 2014. 78 [46] Glorot, X., Bengio, Y. Understanding the diculty of training deep feedforward neural networks. JMLR Workshop and Conference Proceedings, págs. 249{256, 2010. 78 [47] McInemey, M., Dhawan, A. P. Use of genetic algorithms with back propagation in training of feed-forward neural networks. IEEE, 1993. 79 [48] Gill, J., Singh, S. Training back propagation neural networks with genetic algorithm for weather forecasting. IEEE, 2010. 79 [49] Tealab, A. Time series forecasting using articial neural networks methodologies: A systematic review. Future Computing and Informatics Journal, 3, 334{340, 2018. 85 [50] Amidi, A., Amidi, S. Recurrent neural networks cheatsheet. Stanford University, 2018. [51] Bayer, J. Learning sequence representations. Tesis de la Universidad Tecnica de Munich, 2015. 85, 86, 87 [52] Hochreiter, S. Long short-term memory. Neural Computation, 8, 1735{1780, 1997. 86 [53] Cho, K., et al. Learning phrase representations using rnn encoder{decoder for statistical machine translation. Conference on Empirical Methods in Natural Lan- guage Processing, 2014. 88 [54] Karpathy, A. The unreasonable effectiveness of recurrent neural networks. Andrej Karpathy blog, 2015. 90 [55] Neñer, J., Laguna, M. F. Optimal risk in wealth exchange models: Agent dynamics from a microscopic perspective. Physica A, 566, 2020. 101 |
| Materias: | Física |
| Divisiones: | Gcia. de área de Investigación y aplicaciones no nucleares > Gcia. de Física > Sistemas complejos y altas energías > Física estadística interdisciplinaria |
| Código ID: | 946 |
| Depositado Por: | Marisa G. Velazco Aldao |
| Depositado En: | 21 Jul 2021 13:53 |
| Última Modificación: | 21 Jul 2021 13:53 |
Personal del repositorio solamente: página de control del documento
