Science has not yet managed to develop a method to predict the future. Tomorrow belongs to tomorrow. However, studies in the area of ​​Anticipatory Engineering, which use the support of so-called Artificial Intelligence (AI), are already able to trace trends and outline, with a good margin of success, scenarios that can be implemented in the future. In his doctoral thesis, defended at the Faculty of Electrical and Computer Engineering (FEEC) at Unicamp, computer scientist Carlos Renato Belo Azevedo used the resources of Anticipatory Engineering to project the behavior of the financial market. By seeking to identify decisions to buy and sell shares that kept a fictitious investor's future options open, the tool increased his accumulated wealth by around 60%, even without having been explicitly programmed to maximize long-term financial gains. The work, which was supervised by professor Fernando José Von Zuben, received the 2015 Capes Thesis Prize.

According to Azevedo, Anticipatory Engineering seeks to integrate all available knowledge on a given topic, with the purpose of pointing out possible future scenarios. Thus, in the case of the financial market, it is important that the computational model is fed with different types of information, such as fluctuations in securities on the stock exchange, the movement of the main companies in the segment and even the investor's profile, which can be more or less accustomed to taking risks. “Put simply, we look for information in the past to outline future scenarios, which in turn will guide present decisions”, explains the author of the thesis.

The principle, according to the researcher, is valid both for meeting the interests of a person and a corporation. “When we imagine these future scenarios and the consequences they can bring to the 'author of the thought', what is expected is that decisions in the present will be more robust and more tolerant to uncertainty”, he adds. Often, Azevedo notes, the mathematical models used in this type of prediction exercise seek to reduce the complexity of the world.

Thus, the tradition is to supply the program only with information considered relevant to solving a specific problem. “In my work, I tried to go beyond this concept. Obviously, the system includes mathematical models and computer simulations. However, I also explored other areas of knowledge at work. I read several articles on psychology, sociology, biology, behavioral economics, etc. In other words, I adopted a transdisciplinary path, in order to make the approach more comprehensive. It is important to draw on the most varied areas of knowledge to create intelligent systems that help us make the best decisions”, he ponders.

One precaution that the computer scientist sought to take in relation to the system he developed was to give it the ability to “learn” from its own actions and to preserve flexibility in relation to the options that will present itself in the future. Complicated to understand? Azevedo breaks down this issue. “The idea is not to commit rigidly, during the decision-making process, to one type of preference. Therefore, if the outlined scenario undergoes a major transformation, due to changes in the economic situation, for example, there will always be the chance to change this decision to adapt it to the new moment. This way, the investor has the alternative of being more aggressive in favorable situations and more conservative in adverse situations”, he details.

The researcher says that he validated the tool in the context of the financial market, taking into account real data from the London and Hong Kong stock exchanges and also the Dow Jones Index of the New York Stock Exchange. Azevedo also worked with simulated markets. “Both approaches were important because they allowed the system to analyze both markets whose roles present great variability, and more stable markets, in which scenarios are easier to capture.”

The author of the thesis says that it was possible to verify, based on this procedure, that in markets where levels of predictability are very low, the attempt to extrapolate information to identify trends is more of a hindrance than a help. “In this case, the system is normally affected by a lot of 'noise'”, he says. As for the markets in which trends are more predictable, Azevedo continues, it was possible to notice that certain subsets of investment securities, when combined, present a pattern of behavior that allows for increased long-term predictability.

But if one of the principles of the methodology adopted by the author of the thesis is to learn from the past to outline the future, to what extent is it recommended to regress in order to glimpse what is to come? The computer scientist states that this definition will depend on the behavior of the analyzed market. If it changes quickly, it is recommended to give more weight to the most recent data. “If the market doesn’t vary that much, it’s worth going back a little further and taking advantage of older information. It is necessary to calibrate the model to find the optimal point. In both situations, the important thing is to make the system flexible, in order to make decisions equally flexible”, he reaffirms.

Although it has been validated within the financial market, the system developed by Azevedo can be applied to other segments. Taking advantage of an example provided by the report, regarding the uncertainties generated by the process of global climate change, the author of the thesis understands that the tool could in fact be useful, for example, in making decisions regarding the future of an agribusiness sector. “It is known that there are conflicting positions on the topic. Which scenario outlined by scientists will come true? Or will none of them be confirmed? Although it is not yet possible to predict the future, through Anticipatory Engineering we could use the available data, even if they are incompatible with each other, to draw up a cautious action plan to minimize any possible impact on food production”, he assures.

For Azevedo, the fact that his doctoral thesis was awarded the Capes Prize (Engineering Area IV) represents an important recognition of his work. “At the same time, the award increases my responsibility to continue producing high-quality research. It is important to note that research like this is not carried out by just one person. I was fortunate to have the guidance of Professor Von Zuben and the collaboration and encouragement of several professors and postgraduate colleagues. I must admit that I was surprised by the award, but very pleased to realize that work with a multidisciplinary approach was included in the area of ​​engineering.”

The message that the thesis tries to convey, adds Azevedo, is that it is important to continue carrying out research related to human intelligence. “We don’t know for sure what intelligence is, although there are several definitions of it. By studying this topic holistically, with an open mind, I believe I have made a small contribution in this regard. One area of ​​knowledge is not capable of handling, on its own, a topic as complex as anticipatory behavior. I bring my view on the subject. Others will certainly come, in order to enrich our understanding of it”, points out the computer scientist, who received a scholarship granted by the São Paulo State Research Support Foundation (Fapesp) and the National Council for Scientific and Technological Development (CNPq).

Publication

Tese: Faculty of Electrical and Computer Engineering (FEEC)

Author: Carlos Renato Belo Azevedo

Advisor: Fernando José Von Zuben

Each: Faculty of Electrical and Computer Engineering (FEEC)

Financing: Fapesp and CNPq