Artificial intelligence (AI) has taken over the financial industry. For many sectors, it is no longer a luxury to give you the edge over your competitor, but rather a necessity for your company to survive. However, one sector it is yet to take over is investing.
Saying this, the clock is ticking and it is only a matter of time before AI finds its place in the investment market according to Dr. Anna Becker, co-founder and CEO of EndoTech.io, the alternative investment platform. With over 20 years of experience in the trading market, Dr. Becker spoke to The Fintech Times about why AI does not yet work in investing, as far as outperforming human trading decisions.
Why AI doesn’t work for investing…..yet
Researchers at Bloomberg and Johns Hopkins recently announced the creation of BloombergGPT. It is the world’s first large language model AI system trained specifically on financial data and information. Bloomberg’s project is an exciting first step toward developing LLMs for the financial world.
This technology, the same type of tech behind ChatGPT, has the potential to greatly improve market analysis and investment decisions. But it is still not enough to deliver on the elusive goal of beating the market. For that, we need a new type of AI—as the current approach of neural networks and deep learning, even advanced ones like LLMS, is clearly not working.
Why AI hasn’t worked yet in financial trading
AI has beat players in chess, learned to drive vehicles, and composed beautiful literary works. But AI systems that execute market trades and investments have yet to outperform humans, at least on a consistent and reliable long-term basis. For example, since its launch in 2017, the AIEQ has delivered a 40 percent return to its investors. Compare this to the 72 per cent gain on the S&P 500 Total Return Index, and it shows there is still a way to go. Although since 2023 began, AIEQ is ahead of the S&P 500 on returns.
While AI can help human traders and financial advisors obtain information about markets and related topics, like tax code changes, quickly, it has not really outdone them on trading. This is because deep learning, at the heart of most AI approaches today, is not suited to financial markets. Successful deep-learning based models rely on millions of data points for training. They are tasked with creating a specific output, such as writing an essay, or identifying tumors on MRI images.
Note that this is quite different from how humans perform tasks; as humans cannot store millions of data points. They instead base their decisions and actions on fewer inputs. This is why in many use-cases, deep learning AI is more accurate and faster than humans.
But when it comes to financial trading, there is not enough data available for accurate deep learning; the data points are in the thousands, not the millions. And, there is not a clear understanding of the goal because there is a lack of consensus on what actually drives financial markets and what rules, if any, they follow. For example, some say market behavior is based on efficiency; others believe markets follow economic fundamentals, and still others think the movements are completely random.
Echoing the noise, rather than creating direction
In addition, unlike creating articles or images, when it comes to investing, the AI must not just imitate the world, it must make decisions about the future with lots of changing parameters. It must give a yes-or-no answer and play a zero-sum game; this is a much different task than producing an essay.
At most, in reality today’s AI-based trading signals give us little information beyond market sentiment, or the general feeling or attitude of the crowd toward a certain asset. This is not adequate for making the best decision; this does not tell us to go with or against market sentiment; to buy or sell. It’s simply a more accurate way to track how the herd is moving.
A new approach to AI
The key to harnessing AI for trading is to use it to know whether to go with market sentiment or against it. To do that, we need a system that runs like the intuition of a human trader, that takes into account lots of different factors and decides what is the most advantageous move in a given situation. This means we need to move away from deep-learning models and focus on systems based on causal networks.
Causal networks analyse decision-making, the reasons behind decisions, and the success of specific decisions in certain specific situations. They can answer what-if questions. For investing, these types of systems can consider dozens of different elements, including support and resistance levels and candlestick charts, taking them all into account in order to figure out what to do in order to have the highest probability for the desired outcome.
These systems use the latest AI techniques, which connect together a large web of elements; give the correct weight to each element; and help find the causal relationship between them. Advances like self-attention,which allow inputs to interact with each other, play a key role in this last step.
Current advancements to a 50-year-old concept
Although causal networks have been around since the 1970s, they have experienced advancements in recent years. They have fresh potential due to new forms of interconnected analysis, as well as more accurate and comprehensive elements. Each element can contain within it machine-learning. Other types of technology, including pattern recognition similar to what a computer might do with computer vision in analysing handwriting, help the system better understand and consider each element.
How causal networks will advance trading and our understanding of the market
In addition to thinking like a trader, AI systems need to be able to instantly and automatically execute trades and other moves. If signals are simply released to investors to act on, it will be too late. Such a powerful automatic tool, which can also be tailored to the level of risk that investors want to take, will turn financial trading from a guessing game into a math-based pursuit. Returns will more reliably match the level of risk taken, potentially pushing Sharpe ratios higher than we have ever seen.
Because data scientists understand the inner-workings of causal networks, the deployment of these systems will likely deepen our understanding of how financial markets work, helping shed light on the age-old debates about what really moves markets. This type of AI is not a black box, a challenge that faces many large language models, including ChatGPT, fueling fears about the unpredictable and dangerous behaviour of machines.
The fintech sector must recognise that we don’t fully understand what drives market movements. Therefore, we cannot crack trading with deep learning, as we have done for writing and self-driving cars. Rather, we need to build a system that works like human intuition, weighing the likely consequences of different actions based on our knowledge and experience. But this system will be more accurate and faster than the human brain, outperforming even the most talented financial traders and giving us new insights on the workings of markets.
