FREE NEWSLETTER
Go to main Forum page »
There’s a debate ongoing in the UK at the moment around a cash-only tax-advantaged account, and if the benefit should be reduced from a yearly £20,000 deposit allowance to £4,000. This is with the aim of making people favor equity-based, tax-advantaged accounts to enhance returns. Very UK specific, but it got me thinking once again about the general idea of holding cash as a defensive asset in your portfolio for sequence of returns (SOR) risk when in retirement.
Truthfully, the maths was beyond me for a quick article, but I thought this was maybe something I could test with one of the shiny new LLMs that are popping up everywhere. I chose Google’s Gemini, and after politely introducing myself, I posed a question:
“Can you quantify the difference on a standard 60:40 $1,000,000 portfolio of holding a five-year cash buffer during a three-year 20% market drawdown, and project out to two years after the recovery at average market returns?” Here’s the LLM’s condensed summary:
“For retirees with a $1,000,000 60:40 portfolio, a $200,000 (5-year) cash buffer significantly mitigates ‘sequence of returns risk.’ Our analysis shows that during a hypothetical 3-year, 20% market drawdown, the buffered portfolio avoids forced selling of depressed assets. After the drawdown and two subsequent years of average returns, the portfolio with the cash buffer would be worth $916,649. The unbuffered portfolio, forced to sell assets at a loss, would stand at $886,906. This $29,743 difference highlights the cash buffer’s crucial role in preserving capital and enhancing long-term portfolio resilience, especially when regular withdrawals are necessary. It demonstrates how avoiding sales during downturns allows the invested portion to recover on a larger base, ensuring greater financial stability.”
This seems a reasonable outcome to me and highlights a possible benefit of holding cash if you are of the opinion there’s going to be a large prolonged drawdown. Then again I’ve no clue about the calculations the AI used to come to these conclusions.
But what has captured my imagination, and is, I think, something worthy of serious debate going forward as these AI systems develop in complexity in the coming years: Could this be the start of a democratization of financial advice for each individual, with the ability to perform very powerful financial simulations on portfolios that only a few years ago would have been nearly impossible? There are deep implications for this going forward.
I can think of a few areas of concern. When I was a young teenager, I loved programming but always had to be careful of the “garbage in, garbage out” mantra. This would definitely apply here. Other things that come to mind include fundamentally trusting the outputs and issues with privacy of the data used in the simulations, maybe even people over-relying on their AI models. But it offers intriguing possibilities for personal portfolio management in the coming decade and I think very new and unique dangers to navigate.
The simple fact that this morning I could, within minutes of watching a news article, come up with a powerful simulation using an AI is thought-provoking and, in some ways, an alarming demonstration of technological advancement. Let’s hope we can successfully embrace its benefits.
In this case it seems a reasonable answer was provided. It’s early days yet for AI and there is still an unacceptably higher error rate. I’ve used it off and on and have noticed an improvement in the quality of the responses. I think part of it is my improvement in the question posed and part of it is the AI tool improvement. I only use Perplexity. That said, I don’t plan on engaging in a relationship with a Perplexity bot. 🙂
The CEO of Salesforce, Marc Benioff, wrote in the Financial Times today that “humans must remain at the heart of the AI story.” His words rekindled my hope for a future where technology serves humanity—not the other way around. I find myself both awed by and wary of AI’s responses. It delivers data with dazzling speed and precision, yet often with the tone of an all-knowing overlord. I have to consciously remind myself: humans are driven by emotion, by intuition, by the ineffable messiness of being alive. AI, on the other hand, is driven by what I’ve come to call “ee motion”—endlessly efficient motion. It never sleeps, never doubts, never pauses to wonder why. That’s its strength—and its danger. Because in a world increasingly shaped by algorithms, the human pause might be the most radical act of all.
HAL was one step ahead of IBM—literally, one letter at a time—but he was ultimately defeated by Dave’s silence and determination to pull the plug. That quiet act of resistance reminds us: the power belongs not to the loudest intelligence, but to the most resolutely human.
Isn’t the size of withdrawals a key variable? What if you don’t need to touch the portfolio?
AI may go the route of Goggle search. Ask a question and get many questionable results, some paid for by “advertisers”. It is a mistake to think that META, Google, Microsoft and Apple, etc. are doing any of this altruistically.
Slightly tongue-in-cheek, maybe we could then ask the AI to filter out the junk from its reply.
I’m not sure AI has told you something that’s a new idea, if we know it’s a bear market of course lower Std Dev improves returns.
The question that’s interesting is how long do you maintain a cash moat and if you do have a cash moat, how does it reduce the long term purchasing power of ones portfolio.
The question posed to the AI was really just a narrative tool to develop the idea around the future of financial AI development and subsequent positives and dangers. But, I do take your point.
I thought you were asking an interesting question, but I think we need to push the AI further if we want truly useful answers.
I tried a few versions of what I asked above and the AI kitchen sinked me.
Maybe break it into separate queries and get the AI to combine the results?
That seems consistent with “regular” advice around “low volatility” buffers and SORR. I suspect it hasn’t had to scrape too many sources to get there.
The challenge is what sources is it scraping and weighting highest – if you asked for strategies to maximise returns is it going to weight unduly crypto prophets and Robinhood meme stonkers or is it going to have some ballast in the form of traditional diversification advice?
I think we’re pretty much at an inflection point re personal use of AI. Do we lift up our skirt and start sharing personal data to get it to advise us better (at the risk we become mined and targeted by others) or do we try to keep it at arm’s length as much as possible?
Because to truly utilise its capabilities in personal finance we probably need to get quite specific as everyone has individual circumstances which differ.
I am no authority on AI for sure. But I have thoughts.
Privacy is key, of course, with our financial data. I hope and intend all my queries to be private, and for that purpose use only AI that keeps my AI chatter private, provides encryption end-to-end, and an assurance that my data is not made available for training. While free AI is cheap, doesn’t it make sense that using free AI generates information to train the models?
I have a residual fear, despite promises my AI back and forth is on machine only, how am I to be assured of privacy? To maximize my chances, I elected to pay a monthly fee.
YMMV but be safe out there!