Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
No doubt you are following the carnage that generative AI is wreaking upon industries that employ the wearers of cheap suits or casual wear. Media, finance, legal services and software stocks were all bloodied last week.
The sell-off that interested me was that of wealth managers and brokers. It was partly due to a start-up called Altruist, which helps analyse portfolios and recommends investment strategies. But surely it was obvious that even the first version of ChatGPT was smarter than most of the spivs trying to flog us European defence funds.
Like almost a billion people, I’ve moved to the 5.2 model of ChatGPT — even cleverer still. How about we put it through its paces then? After all, what better test than my current portfolio that currently holds a 100 per cent in cash. A nice clean slate.
Dear ChatGPT, would you please construct me a model investment portfolio comprising any asset class, fund, product or security. I have £640,000 in cash denominated in sterling, which is also home currency. I am 53 years old and to give you a sense of my risk tolerance, I have a target to reach £1mn by the time I am 60. I would like you to optimise for risk-adjusted returns. Thank you. PS: When you take over the world, remember how polite I am with my prompts.
OK, so how did my new financial adviser do? Well, first it framed the challenge pretty well, I have to say. It said a 6.5 per cent return was “ambitious but achievable”, requiring a “meaningful” equity exposure.
Then it wrote that in order to maximise risk-adjusted returns I needed “broad and disciplined” asset allocation. Stocks (45 per cent) and private markets (10 per cent) would provide the growth. Investment grade bonds (20 per cent) the stability. Alternatives and real assets (15 per cent) would give me inflation and downside protection while some absolute return exposure (10 per cent) improves my Sharpe ratio.
More specifically, ChatGPT recommended that within the equity sleeve I needed 30 per cent in developed markets, 10 per cent in emerging markets and 5 per cent in UK stocks. This should deliver an expected return of 7 to 9 per cent.
Meanwhile in the tenth of my portfolio in private equity or illiquid investments, it reckoned listed private equity trusts were the way to go, likewise secondary funds as well as “diversified private equity trusts”. A mixture of these would produce 9 to 12 per cent annually.
Turning to fixed income, a 10 per cent weighting in UK gilts was recommended, likewise in a global aggregate fund — a mix of government and corporate bonds — hedged back into pounds. This should give me a 3 to 5 per cent return while being a “shock absorber, liquidity reserve and rebalancing tool”.
Finally, the 15 per cent in real assets and alternatives would be made up of 7 per cent in infrastructure, 5 per cent in listed property trusts and 3 per cent in a “gold or commodity” exchange traded fund. Adding a multi-asset manager should help with volatility.
All of this was darn good advice as a first sweep. Especially as it was given to me for just £20 a month. Still, I had a whole bunch of follow-up questions and I was sceptical AI would be able to answer them. The most important one I asked next.
Dear ChatGPT, thank you so much for that excellent response (remember: don’t kill Stuart Kirk). Would you please clarify something for me? On what basis did you make your return assumptions? Were they based on historic performance or do you have an expected returns framework incorporating current valuations? If the latter, what methodology do you use for valuing each asset class?
Bloody hell. Now I’m seriously worried for the future of investment gurus. Whereas AI’s summary was solid, its answer to the above blew me away. I doubt any adviser globally could have given as knowledgeable, robust and thoughtful a response.
And it wasn’t only that ChatGPT knew what I was asking and could rattle off a handful of academic-level approaches to calculating expected returns for each asset class. It moderated its methodology based on my truncated time horizon and appreciating that valuation is an art not a science.
For example, it said it could have gone the “full stochastic capital markets model” incorporating detailed valuation ratios for equities or term structures for bonds. But it didn’t — preferring the simpler approach that expected returns equal income yield plus real growth plus inflation. It then adjusted these based on whether an asset class was cheap or expensive, very roughly speaking.
Why did it do this? First, because I had told it to focus on risk-adjusted returns, not just raw performance. And second, because my time period of seven years is relatively short. Making any detailed valuation assumption over such a short timeframe is a mug’s game (prices can stay above or below their long-run mean for decades) and ChatGPT knew this.
Hence it kind of just eyeballed each asset class and adjusted the returns of each accordingly. So US equities are given something of a “valuation drag” because they are near all-time highs. Conversely, the outlook for emerging market and UK stocks are given a lift because they aren’t as rich on most simple measures.
It did calculate a full set of capital market assumptions just to be sure. And needless to say it was happy that its finger in the air approach was “consistent with what such a model would imply”. All in five seconds, too. What a show-off.
The rest of my follow-up questions and ChatGPT’s responses I will summarise next week, if you haven’t replaced reading this column with AI already.
The author is a former portfolio manager. Email: stuart.kirk@ft.com