At the recent FT Weekend Festival in London, an experiment pitted a seasoned former portfolio manager against the artificial intelligence chatbot Claude to determine which could construct a superior investment portfolio. The exercise highlighted both the advances and limitations of AI in the realm of personal investing.
The session, titled “Can an AI chatbot make a better portfolio than you?”, featured the author revisiting his personal pension portfolio, previously analyzed by the AI platform ChatGPT earlier this year. Unlike in February, when the pension fund was held entirely in cash, the current portfolio was heavily weighted in equities, reflecting shifts in the market as well as changes in the author’s investment stance.
To prepare Claude for the challenge, the author worked closely with investment expert Tom Ursell to program the chatbot with a “skill” instructing it to construct a portfolio using traditional investment principles. This included applying capital market assumptions, modern portfolio theory, strategic asset allocation, diversification, and rebalancing rules.
Claude’s recommended portfolio differed significantly from the author’s existing holdings. While the author’s portfolio carried an 82 percent allocation to equities, Claude proposed a reduced equity exposure of 63 percent, citing concerns about excessive concentration and risk. In particular, the AI model sharply lowered the author’s UK equity weight from 30 percent to 10 percent, describing the original allocation as a “single-country bet wearing a diversification costume.” European and Asian stock allocations were similarly trimmed, with Latin America receiving a modest 3 percent allocation.
On fixed income, Claude criticized the author’s reliance on a single UK gilt, pointing out the lack of credit exposure and the absence of a maturity ladder. The AI suggested increasing fixed-income holdings from 18 percent to 25 percent by introducing a global aggregate bond ETF, providing a broader mix of government and corporate bonds.
In terms of alternatives, Claude recommended allocating 10 percent to infrastructure assets such as roads, ports, and railways, a departure from the author’s current portfolio, which held no exposure in this area. This contrasted slightly with ChatGPT’s earlier suggestion of a 15 percent allocation to real assets in general.
Overall, Claude estimated the proposed portfolio would generate an annual return of approximately 6.3 percent, slightly below the 6.6 percent target necessary to reach the author’s goal of £1 million by age 60. Festival attendees noted the relatively low equity weight, which Claude justified by the author’s relatively short investment horizon. When asked to design a portfolio with a 20-year horizon, Claude responded by raising the equity proportion to 78 percent.
The discussion also touched on more sophisticated strategies, such as the inclusion of hedging instruments. However, Claude acknowledged that many of the derivatives it suggested for risk management, including long-dated out-of-the-money puts, were largely inaccessible to typical UK retail investors.
Audience members drew comparisons between the author’s portfolio performance and that of Norway’s sovereign wealth fund, noting an identical annualized return of 14 percent since the author’s investment column began nearly four years ago.
The exercise underscored both the growing capabilities of AI tools in asset management and the enduring need for human judgment, particularly in tailoring portfolios to individual circumstances and practical constraints.