How should we regulate AI in trading?

The emerging Artificial Intelligence (AI) boom has already seen a host of applications unveiled as supposedly advanced trading tools.

Just this week we’ve seen the launch of Meta Trade Bot (unrelated to the Facebook-owning company), a cloud-hosted auto-trading tool designed to “eliminate human emotion in trading, enhance trading efficiency, and aiming to deliver consistent profits”, according to reports, while Lion Group Holding announced their ‘Phoenix’ platform, said to provide users with “an intelligent trading service experience deeply supported by AI algorithms.”

The hype is fuelled by the hope – and expectation – that in the years to come predictive analytics will revolutionise trading to the same extent that computers and algorithms did in the 80s and 90s. Except that it won’t, because while computers represented what Peter Thiel would term a ‘0 to 1’ innovation when it comes to trading, artificial intelligence doesn’t. This is simply because past market behaviour gives very little – if any – steer with regards to what is likely to happen in the future.

I’m also firmly of the view that human emotion is critical to effective trading. After all, markets are driven by emotion as much as anything else, and professional trading insight evolves through the emotional training undergone while trading live markets over sustained periods of time, and riding the emotional rollercoaster that comes with it.  

Proponents of AI trading tools will disagree of course. So how should regulators approach the issue?

The U.S. Securities and Exchange Commission (SEC)’s recent proposals for market structure reforms include specific attention paid to the integration of predictive analytics and AI into trading practices. Here's an overview of what the proposals say:

1. Enhanced Transparency: The reforms stress the importance of transparency in AI-driven trading. Under the new rules, firms engaged in AI-based trading would be required to disclose their trading strategies and algorithms to both the SEC and the public. This heightened transparency is designed to reduce information asymmetry in the market.

2. Risk Mitigation Measures: The SEC's reforms include provisions to manage risks associated with AI in trading. Firms that employ AI algorithms must establish robust risk management mechanisms to ensure that the speed and automation of AI do not compromise market stability.

3. Fair Access to Data: The proposals aim to level the playing field by advocating for equal access to market data. This would prevent certain market participants, often high-frequency trading firms, from exploiting information advantages gained through superior AI technology.

4. Market Data Fees: In addition to transparency and risk mitigation, the SEC is considering changes to market data fees. Lowering these fees could potentially reduce the cost barriers for smaller market players, thereby promoting a more competitive environment.

I think the reforms are largely well intentioned, and clearly aim to promote investor protection, fairness and market stability. However, I do still have some concerns. 

Firstly, the reforms will need to strike the right balance between fostering innovation and imposing necessary regulations. Excessive regulation will hinder progress, while insufficient oversight may invite market abuses.

Secondly, compliance will become costly. Implementing and enforcing these reforms will be a substantial challenge. Regulators will need advanced tools and expertise to effectively monitor AI algorithms. Such costs will also accrue to market participants and their internal compliance teams, who will face difficulties in complying with new disclosure and risk management requirements. This might force smaller firms out of the market.

Finally, it’s important to remember that the pace of technological change in finance is relentless. Complex rules can never keep up and are quickly outgrown. As such, we should focus on principles-based regulation. Any regulatory framework must be adaptable to evolving AI technologies, not just addressing current challenges but also anticipating future developments.

I remain somewhat ambivalent about the application of AI to trading, and don’t think it will change market making as much as some people think, but we still need to make sure it works in favour of retail investors and professionals alike.