Stock Exchange Forcasting with the help of Neural Networks

Ph.D. student Alexandra Cristea University of Electro-Communication, Tokyo, Information Systems, Department of Artificial Intelligence

Keywords and Dictionary

NN= Neural Networks
DSP= Digital Signal Processing
NP complete= Nondeterministic- Polynomial problems, problems of extreme difficulty.
securities= Secure goods, with stable or slowly increasing value (price) on the market.

Starting point and goals

With this study I intend to achieve several goals:

Research a different application-field of NN, as I have already seen and implemented NN resolving tools in my past study, to be applied for classical problems, such as the N queens problem, the N-color map, but also reality-based problems, such as the DSP problem of computation of filter coeficients with NN. With this opportunity, I've developed a more general tool, that can stay as a basis to several other NN aplications and may also be used for the present study.
Connect the study in the NN field with my study of economics, by applying NN to a typical problem of today's world economy.
Increase by the choice of stock exchange the data complexity of the input problem given to a NN based solving tool.
Stock exchange, as market occurencies that do not follow a given function or equation, can be most probably approached with a better solving chance by approximative methods, such as NN methods.
The range and complexity of the problem can be varied by allowing a gradual increase of external and internal interferencies to a basic model, therefore, the chosen problem offers a wide possibility of research.
The considered goal ( Forcasting of Stock Exchange processes) is of great actuality and interest, as any problem connected to large amounts of money in our economy-driven world. Computers are and can interfere more, by modeling, forcasting, guidelining and acting as a permanent tool at hand for human activity.
As far as I know, such studies have already been made. Stock Exchange modeling has been tried with several tools, such as function-decomposition [Biblography, a.o. This problem has been, I've heard, also approached with NN solutions. I've been told that there are people making a living out of that, so they can provide an average salary only by playing the real-world stock-exchange game. These solutions are unknown to me at present, so there is therefore no comparision and/or optimisation I could make. I will approach the problem as a new one, and try to solve it.
The program and project may also be developed to have a parallel stock-exchange simultation game, in order to check the program's and the human skill as well to find it's way in the economical labyrinth of prices, buys and sells.

Possible tools and ways to the solution

temporal sequence processing An article of Lipo Wang , called "A general design for temporal sequence processing using any arbitrary associative NN" [Biblography brought me to the idea that, as stock exchange prices are having a time development, and they increase or decrease as a function of time, they may be regarded as temporal sequencies, and be applied the usual temporal sequencies tools. The article considered studies graphic patterns, with a relatively short lenght of the considered training sequencies. The present study will refer to a less complex input pattern ( as prices are uni-dimensional and not bidimensional, as graphic input), but a definitely longer training sequence. The exact and optimal lenght will be considered later during the study.
filter coefficients computation The problem offers some similarities to the filter coeficient computation that was studied by me before, as already mentioned, in the respect of being uni-dimensional forcasting of a certain number of real values. The amplitudes of the frequencies can be equivalated with the price amplitude of stocks on the market. This second model can be considered even if the patterns (sequencies) that are searched are not of a constant length, as the length itself, even if fix for the filtering problem, has little influence of the algorithm itself.
game approach Together with the solutioning mechanism there can be implemented a problem creating mechanism, that allows the premanent testing of the tool characteristics and performances, and may also allow the connection to the user by assuring a better understanding of the given problem through game-interaction. This kind of help tool is useful during a further optimization process of the NN algorithm.

Problem determination and delimitation

The studied problem has to be set in a certain frame, to clarify the already set purposes and aims. There are several problems in economical forecasting, such as the International Journal of Forecasting, the official publication of the International Institute of Forecasters, North-Holland, Amsterdam, treates: New Products Forecasting; Financial Forecasting; Economic Analysis; Production Forecasting;Technological Forecasting; Legal and Political Aspects; Implementation Research; Judgemental/Psychological Aspects of Forecasting; Impact of Uncertainty on Decision Making; Seasonal Adjustments; Marketing Forecasting; Time Series Analysis; Time Series Forecasting; Organizational Aspects of Forecasting; Economic and Econometric Forecasting; Evaluation of Forecasting Methods and Approaches; Forecasting Applications in Business, Government and the Military. The field is too broad and I won't try to cover it. I will select only a few aspects that seem quite general for economical forecasting problems, and try to analise them. Of course, similar methods can be applied for the most of these problems, given their common features, so therefore, generality is important. But also specificity is important, in order to definitely point to a typical problem and genuinly resolv it, instead of beeing lost somewhere in the field of infinite parameter functions that lead to NP problems.

Stock Exchange Mechanism: A Simple View

The mechanism of the buying-selling procedure can be simply illustrated by looking at the following example [see Biblographie: A person X decides to invest in securities. Therefore, it askes a broker from a Stock Exchange Company to search for the best investment possible for a long-term period. The optimisation and goal is to obtain the good with the best relation between bid and offer. On the other hand, a different person, Y, wants to have quickly some liquidities, in order to purchase some expensive goods. Therefore, it decides to sell some shares of a company, and contacts in the following it's broker and encharges him with the job. Both brokers access the electronic market data system and aquire data about the quotes on the market. The broker of person Y is concerned with one single good that that person has to sell, but the broker of person X looks for different goods and tries to optimise the buy price/sell price relationship. Thereafter, both brokers forward their data to the customers. If person X and Y decide to buy respectively sell the same good G (here:shares) in a certain amount A, at the current market price, the respective brokers either transmit the orders directly to the trading post, through a system called the "Super Dot" system, or they have to follow the hierarchical order of sending their orders to the floor of the Stock Exchange (in this example considered the same), wherefrom they are sent to each brokerage firm's ownclerks and given to the firm 's brokers, who take it to the trading post of the Stock Exchange Floor. At the post, the specialist in that company assures fare and orderly transactions. The floor brokers on the trading post try to get the best price for their customers, and the brokers representing persons X and Y agree finally on a price. The transaction is entered electronically, and persons X and Y are informed. Through the computer, the transaction is reported within minutes and appears on tickertapes across the country and around the world. The transaction takes place electronically, crediting X's brokerage firm and debiting Y's. X settles it's account within 3 business days, by paying for the amount A and a comission to his broker. Y settles it's account within 3 business days, by collecting the money from the broker minus the comission. The Stock Exchange is, in short terms, an auction market, where stocks are bought and sold at prices determined by the bids and offers of investors. These are represented on the trading floor by floor professionals, who use their skill, judgment and experience in order to obtain the best possible prices for their customers. The whole process is supported and made easier by the extensive usage of advanced tehnology [see Biblographie.

Biblography

Okamoto,T., Ueda,Y., Kunishige, M.: "The Distributed Multimedia Learning Environment Employing Gaming/Simulation Method with Expert System in the world of Macro Economics" Computer and Artificial Intelligence, Vol. 14, 1995, No.4, 395-415

Wang,L.: "A General Design For Temporal Sequence Processing Using Any Arbitrary Associative Neural Network" , Artificial Intelligence - Sowing the seeds to the future, Armidale New Wales, Australia, AI 21-25 Nov. 1994, Proceedings of the 7th Australian Joint Conference on Artificial Intelligence, World Scientific, Singapore

NYSE Net: About the NYSE, Capt. About the Exchange, HTTP adress http://www.nyse.com/public/about/market/flowchrt.htlm

<- back to my homepage