When one reads something about the performance of stock exchanges, one perhaps imagines rowdy places, populated by unscrupulous traders who buy and sell stocks based on their hunches, moving huge sums of money through the chaos as one often sees in movies and pictures accompanying articles about particularly complicated days on the stock exchange. Lo and behold, this is not the case, and not only because today stocks can also be bought and sold with a smartphone: today much of the financial transactions are directly carried out by computer systems. But human beings still matter quite a bit.
In the jargon it is called algotrading or algorithmic trading, and with the evolution of technology and artificial intelligence it is becoming more and more sophisticated: sometimes it is used to help traders figure out what to do at a certain time-whether to buy, sell, or stand still-while at other times it directly makes trades based on indications and criteria provided by the traders themselves. This is not new. These tools have been adopted over the past two decades by large banks and funds because they limit human error and cost: it may be cheaper for a firm to have five algorithm-driven traders than ten without an algorithm.
These algorithms are basically computer tools that process all the information that can influence the financial markets: political news, economic data, catastrophic events, financial statement publications, and many others. They do this much more completely and quickly than a human being, arriving in a short time at investment decisions that are considered optimal given the information available and the instructions originally given.
There is no single type of algorithm. Some are for speculative trading, that is, for executing lots of trades in an attempt to make small systematic gains day after day. Others help investors build and maintain a diversified and balanced portfolio.
Let us take an example. Let's imagine that an investment fund has set a minimum gain target of 5 percent and a maximum loss target of 5 percent as well, and that it has chosen to diversify its investments based on their geographic origin (assumptions: 30 percent in Europe, 30 percent in Asia, and 40 percent in the United States) and their nature (50 percent stocks and 50 percent bonds). The fund may also have set more specific parameters on individual securities, such as the average price over a time frame or the volatility of their prices.
This just described is a simple investment strategy. With the help of the algorithm, traders can execute it much more easily and quickly, without recalibrating all their portfolios for every single move: the algorithm already provides the complete package of things to do for each event they have to respond to. So far, it may therefore seem that the algorithm not only facilitates but replaces the trader's work. It is not - yet - so.
Davide Biocchi, a trader and popularizer, explains that algorithms follow instructions designed and coded by people, and even when they operate autonomously "there is always a human, a trader or a financial engineer, who checks that everything is going according to predictions" and is ready to correct any anomalies. Among traders they call them "little machines," which buy and sell on their own under human supervision.
The superficial narrative that these computer systems control the market, and are the concrete architects of the big rises and falls, is not so well founded. "The system proposes but there must always be a human who decides, in order to protect the customer but more importantly the market itself," says Chiara Frigerio, secretary general of Cetif, a research center atCattolica Università that deals with the organization and innovation of the financial system. And this is for three reasons: one organizational, one legal and one technical.
The organizational issue is trivial: the ultimate responsibility for how an investment goes cannot lie with a computer but with the trader, the head of his or her division, and ultimately with the company itself, which is accountable to clients and shareholders.
Then there is the whole legal issue: companies that use algorithms have an obligation to report that they use them to Consob-the authority that oversees the regular functioning of financial markets-and are subject to the same rules as those that do not. There are no additional obligations or prohibitions on behavior, but the rules unequivocally state that the responsibility for investment and its control lies with individuals. Finally, there is the technical issue, perhaps the most counterintuitive and interesting.
One might be led to think that algorithms are useful to traders especially in high-stress situations, such as in those times of high volatility and uncertainty when no one understands anything: quite the opposite is the case. Frigerio argues that "precisely because the algorithm is unemotional, it cannot handle an environment of very high uncertainty. When there are large and unpredictable swings, the algorithm is unable to decide because it does not find the same dynamics in its experience."
Algorithms do in fact learn from the past, and they run into trouble when they fail to find dynamics they already know. They don't know how to put themselves in people's shoes, they have no capacity for innovation or interpretation of new phenomena: in the case of large swings the most frequent response they suggest is to stay put, which a trader who has to contain losses, for example, cannot do. "In uncertainty, the human being still wins," says Frigerio. Biocchi himself recounts that it is precisely in these moments that traders turn off algorithms because they are inadequate to handle such situations.
In the ordinary course of business, on the other hand, the systems give the optimal solutions, both in terms of appropriateness of choices and speed of trading, and according to Biocchi they also become a great guide for the market: everyone knows that big investors are using them and therefore can observe their moves and understand what the algorithms are suggesting to do at that moment. Conversely, the fact that everyone knows that during the most confusing phases their input is less becomes an additional element of uncertainty.
For the past few years, algorithms have also begun to garner the interest of small investors, albeit still on an experimental basis.
Algorithms drive, for example, what are called "robo advisors," automated advisory systems used in online investment platforms. It means that the algorithm tries to do the job of financial advisors, who through interviews with the investor figure out how much money he or she wants to invest, how much he or she wants to earn, and how much he or she is willing to risk losing: that is, they do what is called the client's "risk profile." The algorithm does the same thing and costs much less than the advisor; the disadvantage is the loss of the human relationship, which for less accustomed investors can be very important.
"Tools like these," Biocchi says, "can be a help and can be wrong just like advisors," so "the more an investor wants to be autonomous, the more they need to prepare and be aware."