Will artificial intelligence end financial crises?

by bold-lichterman

At the risk of disappointing some, the answer is no. That said, artificial intelligence (artificial intelligence – AI) allows certain investors to anticipate areas of stock market turbulence and to guard against them, or even benefit from them. Artificial intelligence will not solve our social problems but will make some investors wiser.

What is artificial intelligence?

The principle is simple in theory: endow software with the same cognitive capacities as humans. In practice, it should be programmed not to solve a given problem, but to teach it, from a large set of examples, to solve a problem by itself. Indeed, it is currently the best way to approach very (even too) complex problems for which there is no direct algorithmic solution.

To do facial recognition for example: the idea is not to mathematically define what a mouth or an eye is in an image, but to teach the software to recognize them by itself, by showing it numerous examples (images of human faces, animal heads, objects, etc.) and giving him the answer each time. He thus learns to recognize the elements that make up a human face. Artificial intelligence (here more precisely machine learning) is similar to learning by mimicry in children.

The two main families of problematic are the regression and the classification of a large number of data. To solve them, various techniques are possible such as decision trees coupled with Gradient Boosting, or SVMs (Vector Machines Support) for the separation of data along linear as well as non-linear boundaries, or even powerful probabilistic tools such as Bayesian networks.

What could it be used for in finance?

Among the warning signs of a major financial crisis, we can distinguish periods of euphoria (sharp rise in asset prices) and periods of panic (sudden fall in these same prices), as was the case at the beginning of the years. 2000 with the Internet bubble. During these two periods, the behaviors of the actors are not only based on quantitative data and objective analyzes, but also on typically human emotions: greed during euphoria and fear during panic (for example during the day of the Brexit announcement, when the European stock markets over-reacted negatively).

The important point to understand and remember is that with artificial intelligence in finance, we add an additional dimension of complexity: no more idealized world where everything is linear, where everything is Gaussian, where algebrists can apply their favorite theorems. Here we are in a complex non-linear world where the data are of various kinds: thus it is necessary to know how to treat both asset prices and macroeconomic indicators such as the level of American employment, the confidence index of German consumers, or the monetary policies of central banks. It is also a world where each correlation between two financial assets is multi-factorial: the world economy will have a very different configuration if a barrel of oil is below 50 dollars than if it is around 100 dollars. You have to know how to manage situations where a multitude of causes can generate the same consequence.

Artificial intelligence makes it possible to design algorithms that learn to detect periods of euphoria and panic in financial markets (we are talking about market regime). By having well defined these market regimes, the investor can modulate his investments and thus benefit from the performance of the financial markets, but without being caught in excess sentiment. This is precisely what allows better management of its risks.

So what are the limits of AI in finance?

In the financial markets, tomorrow or a deadline in 6 months are not very different because the resilience of the economy depends on political decisions, psychology, etc. There are many situations which belong to the realm of absolute uncertainty, in other words that it is not possible – even with consequent cognitive, human or artificial efforts – to derive any useful information from them.

The question of the temporality of predictions is essential. If we take the example of the film “The Big Short», Michael Burry (played by Christian Bale) – who made his fortune by foreseeing the crisis of subprime – would have gone bankrupt if the crisis had started a few months later. We can continue with Keynes who said “The long term is not an interesting horizon. In the long run, we’ll all be dead. Economists do nothing if, in the middle of a storm, all they can say is that once the storm has passed the sea will be calm.“. This is still as relevant as ever. The IA in finance currently makes it possible to determine the arrival of thunderstorms but not yet their precise eruption dates. In short, this is already a revolution for risk management in finance, but it is by no means the ultimate solution.

However, we must qualify the limits of AI in finance because even if we cannot know when the storm will arrive exactly, we know that in the coming months things will “heat up” on the markets. This is more than enough to reallocate your portfolio and manage your risks prudently.

Why is bank infrastructure a technological debt to innovate in AI?

It is interesting to note that innovation in artificial intelligence is the fruit of the efforts of a few major players in the web and technology: by Facebook with chat bots to automatically respond on Messenger, by Google with the AlphaGo program which beats the best human to the game of Go, by IBM with the software Watson, champion of the American game Jeopardy !, and not from the banks.

Let’s take a look at what happened in High Frequency Trading (HFT). In today’s world, virtually all transactions are automated. Whether it’s to minimize the impact of a massive buy or sell or to seize arbitrage opportunities, automatic trading has become the norm. The issue here concerns the speed of access to information (telecommunications technologies) but also to avoid market manipulation. But in this, the IA does not intervene or while in its most rudimentary stages. It is above all an infrastructure issue. And yet, already the banks have failed to take this turn. The leaders in this market are called Virtu, Tower Citadel, etc. Only Goldman Sachs has managed to go the distance. Why? The answer will be the same for AI: technological debt. The computer systems of banks are incapable of integrating this new financial reality.

It is, moreover, an open secret: each quarter the bank staff discuss at loggerheads the hundreds of millions of euros to be transferred to “allowances for exceptional depreciation of fixed assets” to upgrade their information system. . Despite the billions of euros invested over the past 20 years, their systems have suffered extreme obsolescence. SQL bases, Excel spreadsheets, Murex, Sophis, Calypso and others. All of this is incapable of responding on their own to the challenges of deep learning and fast learning. These old dusty IT factories must be replaced by remote high-performance computing (high performance computing coupled with cloud computing), of big data, direct connection to ultra-high-speed stock exchanges, machine learning, etc. not even to mention the digitization expectations of customers.

And in fact, in this game, start-ups and banks start on the same starting line … except that start-ups are more agile and therefore they advance de facto faster.

What is the difference with the current state of mind in banks?

Engineering is highly developed within banks, in particular financial engineering. It is based on powerful mathematical tools such as stochastic calculus or linear algebra. The aim is, for example, to model the future evolution of the prices of complex assets (structured products) or to study the correlation between the variations of the prices of different assets in order to bet up or down on these assets. . It is a real industry where the workers are the mathematicians and the developers.

The popularity of El Karoui’s master’s degree in probability and finance, which regularly made the headlines Financial Times and sometimes even “debates and opinions” pages from the newspaper The world is the proof. We learn to manipulate Markov chains, martingales, and stochastic calculus using Wiener processes and Itô’s lemma. Unfortunately, all these great engineers and scientists are not equipped to evolve in this new El Dorado that represents artificial intelligence. This calls for other skills, at the crossroads of several sciences such as probability, statistics and algorithms as well as macroeconomics and computer science (via big data).

The culture necessary to excel in AI is much closer to the free spirit of start-ups than to “big brother ‘ major banks. The paranoia that has raged since the great post-crisis purges of subprime is also not very conducive to the state of creativity that these talents might have in other work environments.

This is why start-ups innovate more quickly. They intrinsically have a strong Tech and scientific culture. This constitutes their competitive advantage. All the more so as they are participating in another revolution, that of the digitization and democratization of finance. They can emancipate themselves from the big historical players because they have the capacity to speak directly to the customer, by directly providing him with the tools necessary for his empowerment in the financial markets.

jean-christophe-dornstetterJean-Christophe Dornstetter is responsible for artificial intelligence at Marie Quantier, the first technological platform for personal portfolio management.

Polytechnician, holder of a research master’s degree in quantum physics from ENS Ulm and a doctorate in applied physics from X, Jean-Christophe worked at Misys, the world leader in software for banking and finance, before joining Marie Quantier .

Photo credit: Fotolia, royalty-free stock images, vectors and videos