Although they do not use modern intelligence and machine learning techniques, some older methods of analyzing complex problems can be useful in facing chaos.
A good example of this is the Monte Carlo Method (MMC), which uses the brute force of a sequence of simulations based on random events to try to identify approximate solutions to complex problems, and derived heuristic search algorithms, like Monte Carlo Tree Search (MCTS), for some kinds of decision processes. The origin of the name is related precisely to the casinos and random factors used for simulation.
The idea of this strategy and its algorithms is to simulate possible future movements of the environment and/or the market, based on random features values variations, seeking to identify tactics, which may or may not involve the use of indicators or features action, more efficiently to face chaotic scenarios. For that, randomized decision trees are created, very similar to those used in computer game algorithms.
Another good example is the randomized decision tree. In algorithms of highly complex computer games, such as Chess, for example, where there is an estimated number of possible moves in 10¹²³, heuristics should be sought that define the next move to solve the problem of the analysis time limit of so many possibilities. For this, position evaluation criteria are generally used in order to minimize the chances of defeat or maximize the chances of victory, selecting the best scenarios in this regard.
Through the randomized decision tree, it is possible to simulate all the plays of the two players, starting with totally random values, to gradually decrease the random factor, replaced by patterns discovered in terms of plays or criteria that lead statistically to better results. But undoubtedly, to achieve high levels of competitiveness, many combinations of simulations will be needed in short times, since this is controlled in this game.
In the year 2000, therefore 20 years ago, I developed my first algorithm in this sense to try to predict the chances of classification of teams in the Brazilian championship. At that time, the maximum amount of information that existed in this sense was near the end of the championship, usually through the famous mathematician Oswaldo de Souza. My idea was to create a Monte Carlo decision tree, although at the time I didn’t know that I was technically doing this, focused on calculating the chances of qualifying for each round of the Brazilian championship. This system was used by local radio to pioneering the chances of every goal of the teams of the round, which gained repercussions and competition from several other radios and systems in Brazil, which started doing the same. And today, on the Internet, there are several similar systems based on the same concepts. But this experience showed me the strength of previewing the results only with the intensive use of methods like Monte Carlo, and this certainly influenced my use of them in the Capital Markets, a little later, starting in 2006.
By Rogerio Figurelli at 06/14/2020