FREE NEWSLETTER
I find it rather annoying to waste my time. Unfortunately, it seems I have to tag that handle onto my threshold rebalancing strategy over the last ten year period. A quick recap: I follow a once and done rebalance strategy anytime my equity holdings drift 15% from their normal allocation. The idea is to capture upside during a rebound.
After a recent post about my experience doing a rebalance during the April ’25 “liberation day” tariff announcement, a comment from a fellow HD reader mentioned she used a 10% threshold. I speculated about which strategy would have generated better alpha but pointed out I was too lazy to try the comparison.
The idea stayed in my mind unresolved until I had the simple epiphany to ask my friendly AI Claude to have a stab at crunching the numbers for a side by side comparison of the two rebalance points over the last ten years. I duly received the results which proved interesting. Essentially they are neck and neck — 108.2% return vs 108.8% in favour of the 10% trigger point.
The fly in the ointment occurred when Claude pointed out that a 60:40 portfolio with an annual rebalance beat both, with a ten year return of 111%. It seems I’ve done a bit of faffing about for no good reason over the last ten years — all a bit of a bummer if you ask me!
I have to admit, it’s a bit of a black box situation. I truly have no idea of the maths methodology Claude used but I’m taking a leap of faith and assuming the results have some basis in reality. If nothing else, it might show the power the average investor now has to run complex calculations using AI on their investment portfolio.
Only five years ago I would have needed to pay someone for this type of analysis. And if I’m truthful, it would have been just as much of a black box result coming from an advisor. Maybe this is a genuine use case for AI for the average Joe — the ability to somewhat level the financial landscape. If I choose to, I could pepper an AI with questions about its methodology for hours; I’d be far too uncomfortable doing the same with an advisor.
I asked Claude to provide a short summary of its method and I’ve tacked it on at the end of the post. . Although it disregarded dividend reinvestment and used a flat bond scenario, which would skew the returns downwards, I assume that would translate reasonably equally across the projections.
So where will I go from here? I’m going to stick with my 15% threshold rebalancing, I’m too individualistic and stubborn: data informs my choices but doesn’t dictate my actions. The last ten years might not have provided any alpha if Claude is to be believed, but I’m going with my gut feeling that the period in question has been nearly one continuous bull market, and that’s probably doing a lot of the work in these results.
A sideways or sawtooth market is a different beast entirely; I feel that’s the environment where a threshold strategy should earn its keep, selling the spikes and buying the troughs. Maybe someday, if I’m feeling a bit bored, I’ll try to craft an AI prompt to test that theory… just not today.
(Methodology: Claude used real S&P 500 monthly closing prices from March 2016 to March 2026, sourced from public historical data. It modelled a simple 60/40 portfolio — 60% equity (S&P 500) and 40% bonds — with bonds assumed flat throughout. For the threshold strategies, a rebalance trigger fires whenever the equity position moves 10% or 15% from its last reset point in either direction, at which point the portfolio is reset once to 60/40. The annual rebalance version resets to 60/40 every March. No transaction costs, taxes, or dividends were included in the model. **Please note this is an AI generated paragraph.**)
I ran some rebalancing scenarios using AI a few months ago – another rabbit hole!
I initially used ChatGPT but this proved to be very frustrating. I switched to Gemini and found this more to my liking. During an episode of one of the podcasts I listen to regularly, the hosts indicated they achieved easier/better results for financial analysis using Gemini, and that inspired that change.
I ended up running 20 or more scenarios, changing the prompts and guide rails. I won’t go down that entire rabbit hole here. But here is a summary:
I started with a hypothetical $1m 50:50 portfolio, S&P 500 for stocks and 20-year treasuries for bonds as baseline for all scenarios. I have several reasons for picking the 20-year treasury, the main reason being I was going to run a 20 year back test and for academic purposes felt that a constant rate of return would reduce one of the variables. By coincidence the coupon interest rate was also very similar to today 20 years ago. The scenarios I ran assumed the portfolio is within an IRA, do not address taxes, RMD’s or other withdrawals. Also note that the duration of any back test and start date can skew results, so take this all with a pinch of salt.
Scenario 1: I ran a simple annual rebalance back test through end of 2025 which resulted in growth to $4.24m
Scenario 2: I then compared to 100% S&P 500 portfolio, to see what I was “leaving on the table” with a 50:50, which grew to $7.5m
The subsequent process of refinements I sought to close the gap between 1 and 2. I had AI advise the number of “trades” required to maintain the strategy as I didn’t want a rebalancing process that required a lot of maintenance………………………….the refinement process took a couple hours, but I can only imagine the months it would have taken an analyst back in the day.
Scenario 17 rules for rebalancing / guard rails:
Net result of all these gyrations, portfolio grows to $6.88m in the 20 year back test, with bond portion reaching $1m. This is much closer to Scenario 2 the 100% S&P 500 portfolio. This scenario would have required on average 3.15 trades per year with a peak of 6 in 2014, 2016, 2019.
Scenario 18 is the same as 17 except for the stock portion I used a 4 fund portfolio (equally split between large cap blend (S&P 500), large cap value, small cap blend and small cap value, which matches equity positions in my portfolio) and this portfolio grows to $10.96m in the 20 year back test with bond portion reaching $1m. This is significantly larger than Scenario 2 the 100% S&P 500 portfolio. This scenario would have required a lot more maintenance, on average 6.5 trades per year with the peak year in 2022 requiring 20 trades.
For scenarios 17 and 18 the increase in frequency of trades happens when the 70:30 tether is met after 7 years or so. The four fund portfolio has more trades because there are 4 funds vs 1 in the S&P 500 version, but in reality the maintenance required is very similar performing trades on a quarterly basis.
This version of back testing shows that having a rebalancing strategy could potentially be effective matching or beating 100% S&P 500 performance with a 50:50 portfolio over a 20 year period. The added potential benefit for those who are about to retire or in early retirement and concerned about sequence of return risk, the severity of drawdowns during the first 6-7 years in these back tests were significantly mitigated by the bond holding.
For what it’s worth! Adding to the conversation.
That’s a great illustration of the power AI has put at our fingertips, and one has to wonder where it will all end. The potential seems vast, perhaps only limited by our imagination and ability to craft the right prompts.
What stands out most is that both Scenario 17 and 18 achieved this while maintaining a protective bond allocation, the kind of cushion that could make a real difference for anyone navigating the early years of retirement. The fact that 17 nearly matched a 100% S&P 500 portfolio on those terms alone is impressive, and 18 actually surpassing it is remarkable.
Mind you, we’ve still got a long way to go. Just today I read about an AI being used to analyse drone footage of the Giant’s Causeway here in Northern Ireland, and it produced wildly inaccurate visitor numbers because it couldn’t distinguish between a top-down view of the basalt columns and the people walking among them.
Then again, maybe the data was fine. Maybe it was just an unusually large group of very stoned visitors that day. 😉