Abstract
Technology and Finance are a match made in heaven. In the vast and intricate world of Finance, few areas have evolved technologically as much as FX trading over the past few decades. From being mostly voice based in the early 1990s to the lightning-fast
algorithmic executions of the present, the transformation of FX markets shaped by technological advancements has been both fascinating and intriguing. The latest edition of the technological disruption in FX will supposedly be led by Generative AI, a buzzword
that garners apprehension as much as it raises interest among the financial markets community in general and the FX world in particular.
Gen AI has been in the spotlight with the emergence of ChatGPT1, a conversational AI platform developed by OpenAI. It took the whole world by storm since its launch in late 2022. It receives 10 million queries daily, a testimony to how it has captured people’s
imagination by helping them write eulogies, draft quizzes and even compose rap songs! It crossed 100 million users in a record two months, with the first one million users clocked within five days of release. It has mesmerized the novice and the connoisseur
alike with its ability to generate impressive textual responses to human prompts.
This unprecedented curiosity in Gen AI and it’s apparent prowess has ushered in a new area of possibilities and FX is certainly no exception. Gen AI has the potential to redefine the way institutions operate within the FX markets. Armed with the ability
to learn from vast amounts of historical data and generate new content, Gen AI appears to be a perfect panacea for many predicaments the FX industry has been facing over the years. Will Gen AI indeed fulfill its promise or fall out as another one of those
technological gimmicks? Only time will tell.
What is Generative AI?
Generative AI is a subset of artificial intelligence that has the ability to create new content, such as text, images, videos, computer code, music etc. It does so by leveraging a ‘Large Language Model’ (LLM) which is statistical machine learning (ML) model
designed to identify patterns in the human language and use it to predict next words in a sentence that are contextually relevant and coherent. The LLMs are complex mathematical algorithms trained on large datasets that learn about the patterns in the language.
One of the popular Gen AI models GPT-3 contains one hundred seventy-five billion parameters2 and it is trained on five hundred billion tokens3. To state it simply, Gen AI LLM models can “learn” from existing data and generate new content that is similar to
the original data.
Power of Gen AI in FX
The potential use cases for the application of Gen AI in FX are large. By analyzing vast amounts of data, Generative AI can identify patterns and trends that would be difficult if not impossible for humans to spot. This information can be used to generate
new trading strategies, identify profitable opportunities and manage risk more effectively.
1) Trading signals
Gen AI models can be trained on historical data to predict market movements, indicate potential arbitrage opportunities and allow traders to optimize their trading strategies and portfolios. AI/ML can assist traders in making informed trading decisions by
presenting trading options along with their potential outcomes which are based on historical data but tailored for current market conditions. Many Fintech firms and PTFs like XTX markets, Jump trading, Virtu, LMax, IG Markets etc. are leveraging AI to identify
lucrative trading opportunities and manage risk.
On the execution side, Gen AI models can predict order flow and market impact. AI-driven execution Algos can help prevent slippage and minimize the costs of execution. This will eventually lead to increased order flow and more business for the sell side.
2) Hidden patterns
Gen AI can be leveraged to analyze large volumes of data and identify patterns that can be utilized to understand customer behaviour and generate business insights to tailor customized solutions. Gen AI can analyze the client data and segment clients based
on their trading behaviour and preferences, risk profiles etc. The trading institution can create offerings for each specific client segment keeping in mind their investment goals, risk tolerance and investment outlook.
Advanced pattern recognition can also be utilized to indicate fraudulent activities or market manipulation activities like front running, wash trading etc., making it a positive use case for regulatory and supervisory functions.
Moreover, Gen AI models can also analyze social media and news feeds to gauge market sentiment and track the latest market trends.
3) Synthetic data
Generative AI can be used to create data resembling the characteristics of real-world transactional and market data. It can help synthesize datasets that are related to but cannot be mapped back to the original data. Gen AI could be particularly useful to
generate quotes for currencies whose pricing is not readily available, for ex. the emerging and frontier currencies, which are not as liquid as the freely traded currencies. Another important use could be to generate data in outlier scenarios involving sudden
and extreme market movements like the Japanese Yen flash crash of 2019 or the Turkish Lira crisis of 2018 to name a few. These events are rare but critical to analyze to design robust risk management strategies.
This synthetic data thus generated can be immensely helpful in many ways. It can be used to train and test advanced ML models and strategies for FX trading and execution. Unlike real data, anonymized synthetic data could be shared internally or externally
for research or academic purposes without risking regulatory breaches.
4) Automated Trading
Gen AI can serve as the foundation for building fully automated trading systems. The LLM models could be trained to execute trades based on predefined rules and market patterns. Machine-based trading can circumvent human emotions and cognitive bias,
thus avoiding impulsive decisions and potentially improving trading outcomes. A couple of areas where Gen AI could be especially useful are automated hedging and FX options pricing.
A high level of automation would ensure human traders can focus on other areas like developing complex trading strategies, analyzing market trends or conducting advanced quantitative analysis to fine-tune the strategies.
5) Risk management
Gen AI can help traders respond quickly to changing market conditions, compare it with past market data and identify potential risks or outliers. This can help traders to anticipate and mitigate potential losses. Moreover, Gen AI can identify correlations
and diversification opportunities among different currency pairs. Another use case could be for Gen AI to help back-test trading strategies and improve them based on patterns identified in historical market data.
6) Operational efficiency
Gen AI can automate tasks that are time-consuming and resource intensive like creating reports, analyzing market trends, data analysis, identifying investment opportunities, trade execution, liquidity analysis, client servicing etc., thereby reducing the
manual workload and enhancing overall productivity.
Challenges
Generative AI models need a substantial volume of high-quality data to train. This data can be difficult to obtain, especially for FX markets which by nature are highly fragmented and siloed. Moreover, the data used to train the AI models could have an inherent
bias which may perpetuate and cascade down to the generated outputs as well, resulting in biased decisions or signals. Data security and privacy concern is another critical data related issue that can’t be overlooked.
Gen AI models are complex and dynamic with billions of parameters at play, constantly changing as they learn from the data fed into the model. It may get difficult for traders to trace back the rationale behind the trading signals generated by the AI models.
This “black box” behaviour of AI models may raise red flags for regulators as there is less transparency on how these models work.
Gen AI models are trained on historical data and may not be able to align with current market conditions which are dynamic and ever-changing. The performance may diminish significantly or even become inaccurate in volatile market conditions.
LLMs can hallucinate facts, or create unrealistic, misleading information which is absent in the original training data. This could lead to severe financial and reputational loss primarily when these models are used for trading or decision-making purposes.
Rigorous validation, testing procedures and human oversight are essential to mitigate the impact of hallucination in generative AI models.
In his recent speech, SEC chair Gary Gensler4 pointed out, “AI may heighten financial fragility as it could promote herding with individual actors making similar decisions because they are getting the same signal from a base model or data aggregator.
This could encourage monocultures. It also could exacerbate the inherent network interconnectedness of the global financial system. Thus, AI may play a central role in the after-action reports of a future financial crisis”
Thus, Gen AI could create systemic risk in the markets as too many firms rely on similarly constructed AI models. This common foundational AI platform across multiple organizations may promote ‘herding’ as a similar trading signal is generated from the underlying
model.
Conclusion
Generative AI is poised to revolutionize the larger capital markets landscape and FX markets are no exception. This cutting-edge technology has the potential to impact FX trading activities across front to back by bringing forth a multitude of benefits along
the entire FX value chain.
From automating trade execution and algo trading strategies to enhancing risk management and providing real-time insights, generative AI could prove to be a game changer. The ability of Gen AI to analyze vast amounts of data, identify patterns, simulate
scenarios and generate trade signals opens up new opportunities for traders to capitalize on market sentiment, predict volatility and adapt to ever-changing market conditions swiftly. Gen AI’s adaptive learning capabilities ensure that trading strategies evolve
with the market dynamics, enhancing performance over time.
However, it’s important to acknowledge that although Gen AI has immense transformative potential, it’s far from being a magic bullet. The technology is still relatively new, and it's not without its share of problems. Unavailability of accurate and high-quality
training data, complex AI models and their “black-box” behaviour, biased results, potential systemic risks and ethical concerns are some of the burning issues that need to be addressed strategically before it becomes mainstream.
Keeping these challenges in mind, human expertise and oversight in running these AI systems and interpreting the generated insights is indispensable. As technology continues to evolve, integration of Gen AI in FX markets likely will become more sophisticated,
offering the participants new tools to navigate the complex and ever-changing FX landscape.
References
- Statistics on chatGPT -
https://www.demandsage.com/chatgpt-statistics/
- “Parameters” refer to the learned weights and biases in the neural network model. These weights are updated during the training process to optimize the model’s ability to generate coherent and contextually relevant text.
- “Tokens” are the individual units of text that a language model processes. A token can be as short as a single character or as long as a word.
- SEC Chair Gary Gensler’s speech -
SEC.gov | “Isaac Newton to AI” Remarks before the National Press Club
- What are Large Language Models? - by Catherine Breslin (substack.com)