Loading component...
Artificial Intelligence and its Impact on Financial Markets and Financial Stability
September 6, 2024
Loading component...
IMF Communications Department
MEDIA RELATIONS
PRESS OFFICER:
Phone: +1 202 623-7100Email: MEDIA@IMF.org
September 6, 2024
PRESS OFFICER:
Phone: +1 202 623-7100Email: MEDIA@IMF.org
As prepared for delivery
Advancements in Artificial Intelligence (AI) continue to unfold at a rapid pace. In the coming years, these new technologies enabling computers and machines to simulate human learning, comprehension, and problem solving will become further intertwined with our day-to-day lives. Certainly the financial sector is no exception. There, these technologies—in particular the new and dramatic advances in Generative-AI—are poised to impact financial markets. Today, I will discuss some of these recent and potentially far-reaching developments, as well as their potential impact on financial stability.
AI, as it should be broadly understood, has already been impacting financial markets for many years. This is a part of the economy that has been leveraging data and sophisticated analytical methods for decades to improve efficiency and enhance returns for investors, and in many ways, Generative AI is just the latest stop on this journey.
One might observe the impact of AI on financial markets in 3 areas.
First, efficiency. In the financial sector, as in many other industries, AI—and in particular Generative AI—is being used to enhance productivity by speeding up and automating many current tasks.
From our observation and outreach, we see finance as an industry that is particularly ready to take advantage of these advances, as the efficient processing of data is already central to most activities in finance. Therefore, from back-office operations to customer-facing interfaces, and from research to building analytical models, we expect this to take off rapidly.
Second, evolutionary improvements. AI, in the form of machine-learning and neural networks, has been used by cutting-edge investment firms for at least ten years, and in our understanding now plays a significant role in the automated and high-speed trading that dominates many of the most liquid and deepest markets in the world.
What is new is that the large language models that form the backbone of the most impressive generative AI systems are now enabling investors to process very large amounts of unstructured, often text-based, data to enhance their already powerful analytical tools.
This has implications across areas that can benefit from deeper pools of information, including enhancing the power of broad and extensive forecasts, but also for the ability of quantitative investors to quickly analyze complex documents such as bond indentures or corporate earnings releases, thereby improving price discovery across asset classes.
In addition, through efficiency gains stemming from the use of AI-assisted coding, data gathering, and investment analysis, Gen-AI is likely to lower barriers to entry for quantitative investors into less liquid asset classes, such as emerging markets and corporate debt. This should lead to an improvement of market liquidity in these asset classes, but could also create some financial stability challenges, which I will discuss shortly.
Third, revolutionary transformation. While the evolutionary changes are well underway, the much larger jump from AI-generated model inputs to very sophisticated autonomous AI-driven financial agents still seems far off.
Many market observers and academics have been envisioning scenarios and producing papers involving autonomous AIs that generate and execute trades without human oversight, but market participants are not at all comfortable with this idea yet.
At present, market participants are not looking for AI-generated strategies that rely on “black box” analyses that result in unexplainable trading patterns. Moreover, most participants view having human oversight as an essential part of any AI-based strategy for regulatory, risk-management, liability, and ethical reasons.
Still the area is evolving very quickly, and we need to keep our eyes open.
Given all of these changes, what are the implications for financial stability?
We must first acknowledge that AI could be good news from a stability perspective. For financial institutions, AI can bring new opportunities and benefits such as productivity enhancements, cost savings, improved regulatory compliance or RegTech, and more tailored offers to clients.
In financial markets, technology has done a tremendous job in improving price discovery, deepening markets, and often dampening volatility in times of stress. And AI is likely to continue these trends as well.
However, we have also seen some limited negative impact of quantitative trading in some sudden market dislocations, and there are fears that these risks could rise with the use of AI.
We also have to be continuously on the lookout for how AI could exacerbate traditional financial stability channels such as interconnectedness, liquidity, and leverage. Fortunately, regulators are well aware of these issues and, following the Global Financial Crisis, put in place the necessary tools and enacted the appropriate regulations to deal with these questions. Hence resilience of the financial system has increased dramatically.
Are existing supervisory frameworks and tools sufficient?
The level of AI readiness varies substantially across countries, as do the existing frameworks needed to handle these issues. It is important for policymakers to measure the degree of preparedness to be able to identify areas for improvement and assess the need of new tools.
That said, I would highlight a few critical areas to be aware of:
From a structural point of view, markets continue to move faster, and we need to make sure that they are ready to deal with the even greater speeds that could come with AI.
The August 5th selloff in Japanese and US equity markets is a very instructive example here. While it is not clear to what extent sophisticated AI models played a role here, the turmoil was reportedly amplified by sophisticated hedge funds all acting at once and in the same direction when algorithms spotted clear downward trends and volatility spiked. Thus, one can imagine an even more dramatic episode when AI models are more widely used.
As AI increases the ability of markets to move quickly and react to new information, the speed and size of price moves may exceed what was previously envisioned. Lenders may need to reevaluate the amount of leverage they are willing to provide. In general, we need to think about issues like margining requirements, circuit breakers, and the resilience of central counterparties in light of a potentially rapidly changing world.
From a monitoring point of view, the rise of AI means that regulators will need the tools to track developments in these changing markets, and, very importantly, the entities acting in them.
AI could mean a continuing rise in the importance of nonbanks, particularly broker-dealers, trading firms, hedge funds, and related entities who are well placed to take advantage of this new reality without the burden of intrusive supervision. We could wake up to a new reality of them playing a critical role in markets without necessarily a good understanding of who they are, how they are funded, and what they are doing.
Here, I would draw the analogy with the flash crash in the US equity market in May 2010 and the flash rally in the US treasury market in October 2014. After these events, new circuit breakers were introduced to safeguard market functioning, and we have a much better understanding of how this market operates thanks to new reporting and disclosure requirements.
In addition, the provision of critical AI services is currently concentrated in a handful of third parties, and regulators need to pay special attention to the interdependencies that this is creating, and what this might mean for market functioning in the event of a significant outage.
Finally, we are all aware of the danger of cyber-attacks and market manipulation. There have been the notable high-profile cases in the news involving “deep fakes”, but this could just be the tip of the iceberg. It is crucial that regulators are able to fight fire with fire, and that they invest in supervisory technology (so called sup-tech) that can use AI to process information and spot fraud and other potential troubles.
Already this is being effectively used in the important areas of money laundering and counter-terrorism, but as the bad guys get more sophisticated, the good guys need to do so as well.
Let me finish by noting activities of the IMF in this area. The Fund plays a pivotal role in shaping global financial sector policies and collaborates closely with international organizations and standard-setting bodies as new potential risks arise. In addition, we continuously monitor developments, and one of the chapters in our forthcoming Global Financial Stability Report examines how AI is likely to transform capital markets, as well as impact financial stability more broadly. The Fund also actively helps its member countries to build and strengthen capacities to manage financial stability, and we are incorporating work on AI into this assistance. Such a collaborative approach will be needed to deal effectively with all the challenges and opportunities AI will bring in the years to come.
