As obras disponibilizadas nesta Biblioteca Digital foram publicadas sob expressa autorização dos respectivos autores, em conformidade com a Lei 9610/98.
A consulta aos textos, permitida por seus respectivos autores, é livre, bem como a impressão de trechos ou de um exemplar completo exclusivamente para uso próprio. Não são permitidas a impressão e a reprodução de obras completas com qualquer outra finalidade que não o uso próprio de quem imprime.
A reprodução de pequenos trechos, na forma de citações em trabalhos de terceiros que não o próprio autor do texto consultado,é permitida, na medida justificada para a compreeensão da citação e mediante a informação, junto à citação, do nome do autor do texto original, bem como da fonte da pesquisa.
A violação de direitos autorais é passível de sanções civis e penais.
This dissertation presents the development of a hybrid
system, based in Algorithms Genetics (AG) and Neural
Networks (RN), for the selection of stocks, for the
determination of the percentage to invest in each asset
called weight of the stocks on the portfolio and investmet
portfolio management. The objective multiples (return and
risk) where desired to choose a set of actions of
compaines with profit perspectives to form the investment
portfolio. This choice difficult must to the great number
of possiblities and parameters be considered, as: return,
risk, correlation volatility, among others; reason by
which it is considered as problem NP-Complete.
The research work was developed in 5 main stages: a study
on the investment portfolio area; a study on the models
that use techniques of computacinal intelligence in this
area; the dffinition of a hybrid model Genetic-Neural for
the selection and manages of portfolio for the variant
case in the time; and the study of cases.
The study of the investment portfolio area it involved all
the necessary theory for the construction and investment
The study the techniques of computacional intelligence it
defines the main concepts of Genetic Algorithms and Neural
Networks used in this dissertation.
The hybrid modeling Genetic-Neural for the classic or
stationary case, consisted basically in the use of a
Genetic Algorithm to select the stocks of the portfolio
from a subgroup of assets negotiated in the Stock exchange
of São Paulo - Brazil (BOVESPA). A Neural Network assists
in the management of the portfolio, making forecasts of
the returns of the assets for the next period to
evaluation of the portfolio.
In the asset seletion, two genetic algorithm are shaped:
the first selects 12 amongst 137 assets negotiated in the
São Paulo Stock Exchange, that present greater return
expectation, with lesser risk and that they present low
correlation with the others assets; and the second selects
the assets using the model of Markowitz and the Criterion
of Efficient Frontier.
The forecast of returns of the stocks is a strategy that
it aims at to improve the investment portfolio
performance, typically, they consider only the average
return of the asset. Diferent models of neural networks
had been tested as: Neural Back Propagation, Networks
Bayesianas, Hierarchic Neuro-Fuzzy System and Neural
Networks with Filters of Kalman. The best ones resulted of
forecast had been gotten with the neural network the
weekly returns, as Filters of Kalman. For the stationary
case they had been used as entred of the neural network
the weekly returns, as much of the asset as of the index
of the market, using itself the method of sliding window
to make the forecast a step the front.
The hybrid modeling Genetic-Neural for the variant case in
the time, consisted of the use of 3 models: a AG to make
the choice of the assets of the portfolio; model GARCH to
make the forecasts of the volatility of the assets and the
calculation of the risk of each asset is given by the VAR
(measured of risk that tries to quantify the maximum loss
that portfolio (or asset) can have in a horizon of time
and with a confidence interval); e a RN to make the
forecasts of the returns of the assets for the next period
to evaluation of the portfolio.
In the construction of the portfolio, the Criterion of
Efficient Frontier for the selection of the assets was
used, also amongst the 137 negotiated in the São Paulo
The forecast of the volatility of the assets is a form to
indicate how much it can vary the price of the assets,
measured useful to determine the risk of an asset
represented for the VAR. For this case job model GARCH to
make this forecast.
For the forecast of the returns os the assets they had
been used as inputs of the Neural Networks Back
Propagation the 10 last weekly returns of the assets and
the volatily of the asset, using itself also the method of