For its part, the financial field has long had an interest in applying algorithms to understanding how markets work. This is a natural fit both because there is so much financial data to work with and because there is such a strong incentive to get it right. If you can time the market even a little better than everyone else, you can effectively print money.
This is why decision-makers at financial institutions like JPMorgan Chase are leveraging their own AI solutions or pursuing partnerships with fintech startups.
But what is finance AI? How are these old and new techniques being leveraged to pursue that all-important alpha, and what does that actually look like under the hood?
Today, we’re going to get into the nuts and bolts of finance AI. By the end, you’ll have a much clearer understanding of what AI is and how it’s changing the finance game.
What Is Artificial Intelligence?
For the most part, the actual technology that underlies most AI applications isn’t new. Logistic regressions, neural networks, and everything else that makes modern automation use cases possible is decades old. But with the recent release of ChatGPT, millions of people from all walks of life had access to a powerful machine learning system for the first time.
Suddenly, advancements in the field were real in a way they’d never been before, and many began asking themselves questions like, “What is artificial intelligence?”
Unfortunately, this question is famously difficult to answer, predominantly because the concept of “intelligence” is so slippery. Is a calculator really intelligent where simple arithmetic is concerned, or is it just performing a certain kind of calculation extremely quickly?
Is there even a meaningful difference?
We’re not going to be able to resolve these profound issues today, but we can adopt a definition from a renowned textbook on the subject, Artificial Intelligence: A Modern Approach, by Stuart Russell and Peter Norvig. As they describe it:
“The field of artificial intelligence, or AI, is concerned with not just understanding but also building intelligent entities — machines that can compute how to act effectively and safely in a wide variety of novel situations.“
To put more meat on the bones of this definition — and to better understand what “building intelligent entities” means — we’re going to spend the next few sections walking through common approaches to artificial and some ways in which it is used.
If you were to ask a non-specialist for an example of AI, they’d probably say something like “large language models,” or perhaps “Apple’s Siri.”
These are both examples of machine learning (ML), and while it’s not the same thing as AI, it’s probably AI’s most successful subfield.
By and large, machine learning algorithms come in one of two primary flavors: supervised learning and unsupervised learning. (Some taxonomies also include “reinforcement learning,” but that’s a whole separate approach to artificial intelligence that is beyond the scope of this treatment.)
Supervised learning is trained on structured data containing “features” and “labels.” A standard example is training a linear regression to predict the price of houses. Things like “number of rooms” and “proximity to neighbors” are features of a house, while its price is the label.
What the model does is learn a mathematical relationship between features and labels. It arrives at a representation of how features influence labels, and (hopefully) it will then be able to predict the price of a house it has never seen before based on that house’s features.
Unsupervised learning, by contrast, requires no such labels. It is able to find patterns in unstructured data by looking for things like which words a set of documents has in common.
Unsupervised learning is used in myriad places. A common use case is “clustering,” which is processing an unstructured data set to find similarities among different data points. You could feed an unsupervised machine learning algorithm a bunch of financial data on different companies, for example, and it would generate a set of categories for those companies.
By itself, this likely wouldn’t be enough for a full-fledged trading strategy, but it could be an important input into one.
Deep learning is among the most powerful approaches to building AI applications. Deep learning almost always refers to the creation of a large neural network trained to perform some task. Because they are such a flexible, general-purpose architecture, neural networks have been used in speech recognition, autonomous vehicles, and forecasting profits, among many other things.
But not just any neural network counts as “deep learning.” The “deep” refers specifically to really big neural networks. Like “big data,” there is no agreed-upon threshold at which a neural network becomes “deep,” but today’s networks can have billions of internal parameters and take weeks of continuous training on machines running in the cloud. It’s not likely to be something you’re doing on a laptop.
Implementing a deep learning system is a time-consuming and extremely involved technical undertaking, but at the end of the day, it’s only as good as the data you’re feeding it. If you’re building cutting-edge financial AI applications, be sure you’re also using exemplary financial data from a trusted source.
How Is AI Used in Finance?
AI technologies are a natural fit for decision-makers looking for an edge when forecasting market trends or crafting trading strategies.
Let’s spend some time discussing various ways in which AI is being used to affect a digital transformation across the financial services industry.
Because it’s one of the most important AI use cases for the finance industry, anyone interested in finance AI needs to be familiar with fraud detection. Entities involved in underwriting loans, issuing credit cards, or dealing with money in any way are bound by a panoply of different regulations, including anti-money laundering and know-your-customer laws. Fraud detection is an important part of remaining in compliance.
Let’s walk through an example. When someone is seeking a credit card or a loan, the lender will look at the borrower’s credit history, credit scores, etc., as part of their approach to risk management.
Once the borrower has been approved for a card, various fraud detection algorithms will then monitor their transactions looking for aberrant patterns. You might’ve had a credit card declined when you try to make a purchase out of state, for example, or when you’re buying something highly unusual.
What’s happening under the hood is that the card issuer has a set of automation tools that flag and reject potentially fraudulent card activity. This might be a neural network, an isolation forest, or an exotic graph algorithm, but either way, it’s identifying something that doesn’t seem right and stopping the transaction from going through.
Natural Language Processing
Natural language processing (NLP) is a branch of machine learning aimed squarely at working with language. To put it very simply, it is doing computer stuff with words after you’ve tricked them into being math.
Unlike an artificial language like Python or C++, natural language is messy and open to interpretation, which represents unique challenges where implementing AI applications is concerned.
Nevertheless, natural language processing is found in many use cases. Far and away, the most well-known is generative AI like ChatGPT, which has proved astonishingly good at producing everything from poetry to math proofs.
NLP also powers many chatbot apps which, when fine-tuned correctly, can do a lot to improve the customer experience and increase customer satisfaction by, for example, answering questions effectively. These chatbots are being used by more and more banks in a bid to make their services easier to use.
Another way that firms are leveraging natural language processing is by using it to look for valuable trading signals. You might use deep learning to search over vast amounts of financial news about a company to try and predict a good time to buy or sell it, for example. If there’s suddenly a deluge of negative press about JPMorgan Chase, there might be a good short-selling opportunity and vice versa for positive sentiments.
But there are so many tweets, blog posts, press releases, and related text to get through that it’s effectively impossible for a human to do, and that’s when machine learning engineers build automation tools to do it for them.
Among the most sophisticated uses of AI technologies in the finance industry is algorithmic trading, a term that refers to the use of bots to execute real-time trades.
Though the actual mathematics can be incredibly complex, the basic idea is simple. By using technical indicators, technical analysis, or some other source of insight, it’s possible to program an agent that can automatically execute a trade when some condition obtains.
To use a very basic example, you could set up a simple bot that enters a buy order when a particular Fibonacci retracement occurs. To use a slightly more complex example, you might try to build a high-frequency trading strategy based on a hypothesis about market microstructures.
FAQs About Finance AI
How Is Data Science Different From AI?
At a high level, data science refers to the use of statistics and other kinds of analysis in decision-making, such as when charts are generated to help understand how customer retention is changing over time.
More and more organizations consider themselves “data-driven,” but data science is not AI. Though the term “machine learning engineer” is often used interchangeably with “data scientist,” day-to-day data science involves little engineering. A data scientist might use very simple ML models, like logistic regressions, but they’re not going to be building neural networks or creating algorithmic trading agents.
How Is Blockchain Used in Finance?
This article has been about finance AI, but we’d be remiss if we didn’t at least mention one of the other white-hot technologies that’s exciting the finance industry: blockchain.
A blockchain is a distributed ledger containing some kind of information, usually transactions. In most blockchains, transactions are tracked in blocks, which are chained together over time, hence the name.
There’s a lot that could be said about how blockchain is used in finance, but briefly, we can distinguish a few major applications.
First, the blockchain is being explored as a data storage layer. Just as bitcoin transactions can be stored by the Bitcoin blockchain, a blockchain with high-enough throughput could be used to track all sorts of transactions, storing them in a public database that anyone can inspect.
Then, there’s the possibility of using blockchain data as part of a trading strategy. There is more and more activity happening across blockchain projects, and that’s especially true if you broaden your focus to look at smart contracts, patterns in unspent transaction outputs (UTXOs), etc. All of this is a rich source of information for anyone wanting to execute profitable trades on the basis of data others may have overlooked.
What Are Some of the Benefits of AI in Finance?
The benefits of using AI in finance are largely the same as those found in other industries. There’s only so much human time and attention to devote to answering customer questions, doing research into the fundamentals of different companies, and rooting out fraud. To the extent we can use automation to handle these and related tasks, we free up those humans to focus on places where AI is less useful, like thinking creatively about the long-term trends of financial markets.
What Is the Future of AI in Finance?
It’s difficult to say with any certainty what the future of AI in finance will be. But as AI gets better at handling open-ended, multi-step tasks, generating compelling textual output, and executing sophisticated trades, we predict that it will gradually take on a more and more significant role in the finance industry.
Making the Most of Finance AI
Finance has long been an early adopter of new technologies, like computing and the internet, and that’s no different with artificial intelligence. From natural language processing to algorithmic trading, there are many ways in which AI is impacting the financial services industry.
But you have to remember that AI isn’t magic. A machine learning algorithm that is fed garbage data will produce garbage results, so finding a quality source of data is absolutely crucial.
If you’ve begun to think about trading forex or you just want basic financial news, Tiingo has you covered. We’re obsessed with offering the cleanest data available anywhere, and our company is built from the ground up with an authentic focus on our customers.
Sign up for a Tiingo account to see how we can give you the information you need to power your trading strategies.