YouTube Blog AI
Can AI Be Your Financial Advisor? MIT's Andrew Lo Explores the Future of FinGPT

Can AI Be Your Financial Advisor? MIT's Andrew Lo Explores the Future of FinGPT

Content

Can AI Be Your Financial Advisor? MIT's Andrew Lo Explores the Future of FinGPT

Andrew Lo, a professor at the MIT Sloan School of Management and director of the MIT Laboratory for Financial Engineering, delves into the potential of generative AI in finance. While excited about the possibilities, he also expresses skepticism and annoyance at the need to understand this rapidly evolving technology. His research, in collaboration with talented MIT students, focuses on whether large language models (LLMs) can offer sound financial advice.

The Psychology of Loss: Why We Make Bad Investment Decisions

Lo begins by highlighting research on how investors react to losses. Analyzing data from 600,000 household accounts, he discovered a common pattern: people tend to panic sell after experiencing losses, pulling money out of the market and keeping it in cash for too long. This "freaking out," as he calls it, demonstrates our natural inability to handle loss effectively.

A Simple Investment Test: Are You Making the Right Choices?

To illustrate this point, Lo presents the audience with two investment scenarios. The first involves choosing between a guaranteed profit of $240,000 (Investment A) and a lottery ticket offering a 25% chance of winning $1 million (Investment B). Most prefer the sure thing, Investment A. He then presents a second scenario: a sure loss of $750,000 (Investment C) versus a lottery ticket with a 75% chance of losing $1 million but a 25% chance of losing nothing (Investment D). In this case, most people choose Investment D, preferring the riskier option with a chance of avoiding loss entirely.

The catch? Lo reveals that the most popular combination, A and D, is actually the worst choice. It's equivalent to a lottery ticket with a 25% chance of winning $240,000 and a 75% chance of losing $760,000. The less popular combination, B and C, offers the same probabilities of winning and losing but with a slightly better outcome, effectively leaving $10,000 on the table. This demonstrates how our aversion to loss can lead us to make irrational and costly financial decisions.

Financial Engineering: Exploiting Behavioral Patterns

Lo explains that financial engineering often involves exploiting these behavioral patterns. Multinational organizations or even individual households can make seemingly rational decisions in isolation that, when viewed holistically, result in significant financial losses. He argues that understanding these biases is crucial for making sound financial choices.

Can ChatGPT Be Your Financial Advisor?

This leads to the central question: can large language models help us avoid these pitfalls and provide better financial advice? Lo shares his experience of asking ChatGPT 3.5 what to do after losing 25% of life savings in the stock market. The response, while superficially reasonable, was deemed potentially negligent by financial professionals due to its lack of personalization and generic recommendations. However, ChatGPT 4.0 offered a significantly improved, more comprehensive response, although it still emphasized the need to consult with a professional advisor.

FinGPT: The Future of AIPowered Financial Advice

This prompts the question: how close are we to having truly reliable AIpowered financial advisors? Lo and his team at MIT are actively working on this problem, exploring the potential of what they call "FinGPT."

The potential benefits are clear: AI could help individuals avoid loss aversion and make more informed decisions, especially in areas where they lack the time or expertise. However, the risks are also significant: LLMs can make mistakes, hallucinate, and even be used malevolently to exploit our biases.

The Three Pillars of FinGPT Research

Lo outlines the three key components of their research:

  • Competency: Can LLMs meet the minimum qualifications for a financial advisor? Preliminary results show that GPT 4.0 passes versions of the Series 65 and CFA exams.
  • Personalized Advice: Can FinGPT tailor its advice to individual circumstances? A randomized clinical trial is underway to test this proposition.
  • Trust: Can humans trust AI for financial advice? The researchers believe that by adhering to the code of ethics that governs financial advisors, they can create a system that generates ethical and trustworthy recommendations.

Democratizing Financial Advice

Lo concludes with optimism, expressing hope that their research will contribute to democratizing the financial system by providing accessible and scalable financial advice to those who cannot afford traditional financial advisors. This is a huge potential benefit of the work.