“Machine learning and AI skills are no longer options, and these technologies transform data analytics and prediction paradigms. “That was the last words of financial services recruiter Selby Jennings. [date/DB backlink here] An article covering the rise of Python programming in the industry.
However, the AI's final words are far from here. The changes that computing has brought in the finance industry are no longer recognised. Guido Van Rossum I developed Python in the 1980s. Still, the conversation about AI and machine learning has just begun.
In the beginning of the follow-up series, disruption banking examines the global phenomenon of AI in trading and its impact on the financial services industry.
AI trading is equivalent to a large company
Financial Services' global AI market was expected to reach $22 billion by 2024 by Data Cruncher Gitnux. Yahoo Finance went a step further. It forecasts an increase from an estimated $38.36 billion to $19.03 billion by 2030.
By 2023, Gitnux rated that the use of deep learning techniques in financial modeling had increased by 95% since 2019. Between that year and 2022, machine learning in quantitative finance increased by 58%.
And that's just the tip of the iceberg. According to quantified strategies, the 2003 algorithm or “Argo” trading accounted for about a tenth of US stock orders. According to a survey published in the 2011 Financial Journal, by 2009, algo transactions accounted for 73% of high-frequency transactions.The starting point near zero in the 1990s. In 2023, the amount of orders processed by the algorithmic trading system exceeded 25 billion a day worldwide.
A snapshot of your data should tell you one thing. The future of trading is already here, dominated by AI and machine learning.
Similarly, but not synonymous
They overlap, but AI and machine learning are not exactly the same. Machine learning, a subset of AI, allows systems to learn that they are not being monitored from the data and improve over time, without direct intervention.
In terms of financial transactions, it corresponds to most forecasting models that use historical data to predict future market movements. A more specific example is large-scale language model (LLM)-based applications such as BlackRock's Theme Robots. This is used by bank portfolio managers to blend human insights and big data to make better stock trading decisions.
In fact, LLM is another iteration of machine learning and is relatively new to the game.
LLM for inventory trading
Machine learning has been used in financial transactions for quite some time. The old “traditional” approach involves using numerical data such as price and volume to try to accurately predict where the market will move next. However, the newer version also includes using text-based LLM. Analyze useful information such as news, filing, and revenue calls.
Perhaps the most well-known recent experiment involving LLMS and trading markets was the Social Science Research Network (SSRN) 2023 study. Can ChatGpt predict stock price movements? Returns predictability and large language models.
As a result, SSRN concluded, “Sophisticated return forecasting is a new capability of AI systems, and that these technologies can change the spread of information and decision-making processes in financial markets.
Expose the hype
However, ChatGpt created headlines as a mainstream LLM, but it was far from what was originally used to predict the stock market. And if published research passes, it may not be the best.
Journal of Emerging Technologies A study published last September last year that introduced Google AI's deep learning model Bert and its financial services spinoff Finbert. This was three years ago when Openai's ChatGPT3 was released, shocking the global technological media.
The Journal survey conducted a test in 2024 to supply newspaper headline data to both LLMs to predict the stock market. The accuracy of Bert and Finbert respectively “After a thorough comparison with the benchmark algorithm, we found that there are 86.25% and 83.6% accuracy.
Research author Ritesh Tandon “The findings show that the system may be used in the stock market and provide important information on LLM algorithms for stock data forecasting and sentiment analysis using financial news headlines.”
Burt defeated the GPT
“There is no evidence that GPT models perform much better in predicting correlations compared to BERT models,” agrees another study published in November by Cornell University. “When it comes to optimal strategies, the BERT model is […] It's a better strategy. ”
Simple Science, which worked from Cornell's research and released its findings from its own portfolio management tests on BERT and GPT this May, was further highlighted. Based on predictions from both models, we calculated various allocation strategies across assets such as stocks, bonds, and real estate. The aim was to minimize risk and increase returns.
“As long as it is [overall] The results were promising and there were clear distinctions in the model,” it says. [Federal Reserve] Beige book. The simplicity of Bert's approach allowed us to adapt to a variety of market scenarios. ”
Trad ML is also struggling
However, simple science highlighted that traditional models struggle to navigate the complex landscape of stocks, warning of overly enthusiastic news reports that are nothing more than “gossip.”
“Market data can be not only loud, but also complicated,” he said. “This noise means predicting how the market will behave is a major challenge. Traditional models often struggle to keep up with the chaotic nature of market data.
Beware of bias
We get it – predicting inventory is tough at the best. However, that consideration never allows the LLMS to be removed from the hook.
Another issue highlighted in ChatGpt in Cornell's study was “appearance bias.” This occurs when the training model uses data that was not accessible during the test period. Think of it as something that gives you a past self from today's newspaper a week ago and is astonished when that past self brings amazing foresight.
As a result, backtesting of how the model works in stock market forecasts will give you a better score than it is, not true. guess what? they are.
Training vs. trading reality
The second problem facing LLMS is “overfitting”. This is when the model learns data specific to the training set, but it is not necessarily applicable to future live forecasts. result? After working well in exams run on historical data, delivering in real life situations fails miserably.
Or, as Tech HQ defines, “If a model is too specialized in training data and loses its ability to generalize to an invisible scenario, it causes overfitting. This reduces the overall effect of the model and undermines its purpose.”
At the time of writing, efforts are underway to mitigate problems that overfits may cause. According to MIT Technology ReviewAI benchmarks like ARC-AGI darken the test data, so LLMS cannot fall into overfit traps.
LLMs should be risk assessed
It has not yet been decided whether this kind of thing will become an industry standard, but earlier this year, Cornell University researchers called for a larger risk analysis for LLM.
“Standard benchmarks pinpoint how well a large-scale language model agent works in finance, but rarely say whether it's safe to deploy. Scholars warn that they “overlook vulnerabilities such as illusions of reliability, hallucination facts, old data, and quick manipulation of adversity.”
They add: “Financial LLM agents need to be evaluated first and foremost on their risk profile, not on point user performance.”
Speaking of risk, next week we will look at how AI itself will be used to protect it in finance. Until then, stay safe!
Author: Damien Black
The #DisuptionBanking editorial team takes all precautions in this article to ensure that people or organizations are not being affected or provided financial advice. This article is definitely not financial advice.
#capitalmarkets #machinelearning #ai #financialmarkets #fintech
reference:
Python: Finance's primary language? |Confused Banking
Quantitative Financial Experts Looking back on the Future of the Market Turbulent Year and Discussion Industry at QUANT STRATS 2025 in London | Confusing Banking
AI5: A new powerhouse that redefines the AI era | Confusion Banking
