Imagine being able to simulate the functioning of the human brain, with its intelligence and sensitivity, and apply it to a computer. Well, the ANNs, the artificial neural networks, were born for just this. Today, neural networks are an established reality for various investment funds, especially hedge ones, that are those who want to get more by taking more risks. The great strength of these systems is that they are able to obtain knowledge and value from Big Data, that is, collections of very substantial data from which it is essential to obtain the right information in a few seconds. Neural networks certainly cannot “predict” the future. But they can help analysts understand if there are otherwise subtle patterns or signals that, in the past, might have led to crises or runaway market seasons. Behind AIs there are mathematicians, engineers, and physicists, and in fact, these are the most sought-after professionals by the big hedge funds, Renaissance and Bridgewater in the lead.
The first attempts to emulate the neurons of the human brain were made as early as the 1970s. However, at the time there was neither the computing power nor the amount of data necessary to make the best use of the mathematical models available. Today the ANNs are able to guarantee efficiency in the analysis of huge amounts of data in a very short time. They also possess the great ability to learn, just like our brains, after several attempts. Evolutionary and adaptive learning.
At a financial level, it was initially thought of using them to predict the performance of a stock, or an index, looking for patterns and trends in the past of that same stock. For example, a neural network can be trained to analyze the historical behavior of the SP 500, so that from the trend of the last month, it can extrapolate a probable trend for tomorrow. But this idea couldn’t work. The neural network learned to do what was required of it. The problem, however, was upstream. The price of a stock today is not directly influenced by the price of the same stock in the past, or at least not only.
In the meantime, the use of ANNs has multiplied in various areas: virtual assistants and chatbots, retail, shopping and fashion, security and surveillance, sports analytics and activities, manufacturing and production, livestock and inventory management, warehousing and logistic supply chain, agriculture and farming, autonomous flying, self-driving cars, and autonomous vehicles, healthcare and medical imaging analysis. And, after the first attempts dating back to the end of the last century, the boom in the computational power of computers has given rise to a second spring for machine learning and neural networks applied to finance. Before, it was difficult to integrate everything into an automated system. Today, however, we can read the data more in-depth, especially if we work on the historical series. Thanks to the computing powers achieved, it is now possible to make multiple attempts in a few instants to train the AI, those that in the past took years.
In the stock market, many things have changed in the space of just a couple of years. Immediately after the financial crisis of 2008, hedge funds experienced a strong process of concentration, favored by regulatory and compliance changes that rewarded the consolidation of large players, the megafunds, that are organizations capable of managing tens if not hundreds of billions of dollars. For example, Bridgewater Associates manages over $ 160 billion a year. But with the pandemic and now war, volatility is more difficult to predict and these mega-funds struggle to compete with smaller, more agile rival structures. Realities with high performance and the ability to implement highly innovative trading methods.
75% of the trading volume is nowadays done with high-frequency trading (HFT), or systematic trading, which is automated trading used by large investment banks, hedge funds, and institutional investors. The historical leader in this field is Renaissance Technologies, a fund created in 1982 by a professional mathematician, Jim Simmons, who controls, among others, the Medallion fund, famed for the best record in investing history. It specializes in the application of systematic trading using quantitative models derived from mathematical and statistical analysis to analyze and perform operations, and it was the first to bring High-Frequency Trading to the financial markets.
It is easy to understand that today those who control the most innovative technologies in the AI field can count on an undisputed competitive advantage. In addition to Jim Simmons’ Renaissance fund, there is also the Prediction Company, a company founded in Santa Fe, New Mexico, USA, in March 1991 by J. Doyne Farmer, Norman Packard, and James McGill, among the most important in the application of trading techniques based on particularly advanced mathematical and physical models, including the machine learning.
The reality is that those who trade in a traditional way, for example using charts, lose sight of a lot of fundamental information and get a poor general picture. Thanks to neural networks, effective mathematical models, and Big Data it is possible to grasp what a human eye often does not see. Certainly small nuances on trends and patterns, but also for example a wave of pessimistic comments via Twitter regarding a particular company, or even articles in local economic newspapers that could escape the human analyst but not the machine.
From these examples, it clearly emerges that neural networks are fundamental especially if we think about data management. The problem today is no longer the lack of information needed to understand where to direct the investment. From the extraction of purely quantitative information from Big Data, we have moved on to more qualitative data. For example, from satellite images, we can understand how many people go to the supermarket and predict a store chain’s future revenue, or how many cargo ships are moving along the routes of international trade and predict GDP figures.
There are databases that do just that. They gather quality information and then resell it to hedge funds. This created a florid market of “useful” data. The more we move forward with the digital revolution, the more necessary is the need to have “clean” data available, collected in the sea of hyper-information. And to identify them, the artificial eye is more reliable and takes much less time than the human eye. Thus there is a frantic search for quality data especially if we think of what bad data input could lead to. For example, once, fake news relaunched online about Obama’s death, obviously later denied, caused a sudden collapse in stock prices, because a high-frequency model had misread the information and started selling.