You pick 60% stocks and 40% bonds
That split is a decision about risk. It is the line every later step is measured against.
Reports · The map · October 2026
How often should you rebalance?
What the research says, and where AI helps. A 60/40 portfolio does not stay 60/40. Watch it drift, test a rebalancing band on your own numbers, and leave with a written policy an AI assistant can follow.
Watch a portfolio driftRebalancing is about risk control. Multi-asset portfolios drift away from the allocation you chose, and rebalancing brings the risk back to where you intended it.
The research favours a rule that checks every day and trades only when an asset drifts too far. Few people can keep that up. An AI assistant can do the watching, as long as code does the arithmetic and you place the order.
The walkthrough
Left alone, a 60/40 portfolio slowly turns into something else.
One simulated portfolio, 30 years of monthly returns, one step at a time.
Stocks rise 25%, bonds go nowhere
Without buying or selling anything, 60/40 becomes roughly 65/35. You now hold a riskier portfolio than the one you chose.
Left alone, the mix edges towards 80/20
Stocks usually outperform bonds over the long run, so an untouched portfolio keeps drifting the same way. This path ends near 80% stocks.
Check often, act only outside the band
Vanguard's 2024 rule for its target-date funds: monitor daily, trade only when an asset drifts 2 percentage points from target. The shaded strip is that band.
Trade back to the edge, not to the target
When the band is breached, Vanguard trades back to 1.75 points from target rather than all the way. Smaller trades, and the lowest transaction cost per rebalance in its research.
Each tick on the chart is a rebalance in this simulation.
target
left alone
rebalanced inside the band
Simulated monthly returns, illustrative only: stocks average 0.65% a month with 4.2% volatility, bonds 0.25% with 1.2%. One fixed random path, checked monthly. Not a forecast and not historical data.
The bench
Now try it on your own numbers.
Enter what you hold, the split you want and the band you will tolerate. The calculation runs in your browser and tells you what an assistant should say this month.
Your portfolio
Vanguard uses 2 points at institutional cost. Paying retail fees, a wider band may make sense.
What the assistant should tell you
Amounts are in any one currency. Two assets only, no fees, spreads or tax. When a sale is needed, what it costs you depends on your broker and the rules where you live, and belongs in your written policy. Nothing you type leaves this page.
Finding 1 · Risk, not return
Rebalancing usually costs a little return.
Some writing online sells rebalancing as a way to buy low and sell high. The logic is mean reversion: sell what ran ahead, buy what lagged, and collect a bonus. Between stocks and bonds it rarely works, because stocks are expected to keep outperforming for the extra risk they carry.
Michael Kitces puts it this way: when two assets have different expected returns, rebalancing usually trims long-run returns slightly, in exchange for keeping risk where you set it. AQR reached the same conclusion in 2017: rebalancing does not harvest a return premium; it keeps your risk level stable.
The return gain, in studies that do find one
0.5%
a year at most, and heavily dependent on which assets are involved
Upper bound as summarised by Kitces. Treat any tool that promises a "rebalancing bonus" accordingly.
Finding 2 · How often
The best rule needs someone watching every day.
Vanguard's classic study found no optimal frequency. Monthly, quarterly or annual rebalancing gave risk-adjusted returns that were not meaningfully different, while trades and costs rose with frequency.
Its 2024 research gave a sharper answer, but one that only works with daily monitoring, which is not practical for most individual investors. Nobody opens a brokerage app every morning to compute drift across five funds. Software can, and it can stay quiet for months.
Two Vanguard answers, 14 years apart
The classic study (US data 1926–2009)
- Check
- Once or twice a year
- Act when
- Drift reaches 5 points
- Finding
- No optimal frequency
The rebalancing edge (December 2024)
- Check
- Daily
- Act when
- Drift reaches 2 points
- Trade back to
- 1.75 points from target
- Benefit
- 5–21 basis points a year against calendar rebalancing
Same simulated path, 2-point band: where should a rebalance stop?
0
rebalances in 30 years
0
average trade, points of the portfolio
0
total traded, points of the portfolio
Computed in your browser on the simulated path from the walkthrough, checked monthly rather than daily. Stopping at the edge means more, much smaller trades, and less trading overall. Illustrative only.
Finding 3 · New money first
Contributions do the work, until the portfolio outgrows them.
If you still add money, send each contribution to whatever is underweight. Nothing is sold, so there is no commission or spread on a sale and, depending on where you live, no tax to account for. Dividends and interest can do the same job.
The method has a ceiling. As the portfolio grows, contributions shrink next to it. A portfolio of 100,000 at 75/25 needs 25,000 of new bonds to get back to 60/40. At 500 a month, that is more than 4 years.
How long new money alone takes to fix a 75/25 drift back to 60/40
50
contributions, all into bonds
year 1year 10
Bonds needed = 0.25 × portfolio, because the stock holding must become 60% of a larger total. Past a few years, a sale is the realistic fix.
Finding 4 · The set-up
The agent reads and proposes. You approve and act.
Brokers give agents very different levels of access. For rebalancing, read-only access plus a written proposal is enough. A monthly rebalance is never urgent, so there is no reason to let software press the button.
The simplest version needs no connection at all: almost every broker lets you export your holdings as a CSV file, and a monthly export is all a rebalancing check needs.
Five steps, in order. Select one
- 1
Read
Holdings from a CSV export or read-only access
- 2
Calculate
Drift and trades in code, against your policy
- 3
Explain
The model writes the proposal in plain language
- 4
Approve
You check it and decide
- 5
Act
You invest the contribution or place the order
IG
Connects its platform to agents over MCP on a read-only basis.
Robinhood
Lets an agent trade, but only inside a separate account of its own.
Interactive Brokers
Keeps the agent from reaching the order book at all.
Finding 5 · The maths
Never let the model do the arithmetic.
A language model predicts text. Asked for a drift percentage, it can produce a confident number that looks plausible and is wrong. Both ChatGPT and Claude can run code: have the weights, drift and trade sizes computed in code, with the calculation shown, and use the model for reading a messy export and explaining the result.
The same goes for instrument names. Many indices are tracked by several funds, some reinvesting dividends and some paying them out, in different currencies and at different costs, each with its own identifier. Name every holding by its exact identifier, so the assistant cannot confuse two siblings.
The same question, two ways of answering it
Asked in text: what is my drift?
"Your stocks are at about 64%,
so you are roughly 4 points over."
(an illustrative text-only answer: a guess, no working shown)
Asked to compute it in code
The right-hand answer is computed live from the numbers in the bench above.
The experiment
Write your policy, then hand it over.
Without a written policy, the assistant improvises, and improvising is the opposite of a rebalancing rule. Set the lines below, copy the result, and paste it at the top of a conversation or into your assistant's project instructions, with your latest holdings export.
Target, band, destination and contribution come from the bench above.
An agent should check more often, not trade more often.
Most months, a well-set-up assistant has nothing to report beyond where this month's money goes. That is the rule working.
For agents and sceptics
Every number on this page, and where it came from.
Every figure traces to a published source, or is computed in your browser from assumptions stated beside it. Purple, which publishes Hello Purple, builds technology that connects AI agents to trading accounts.
- Yorben De Reuse, "How often should you rebalance? What the research says, and where AI helps", Hello Purple Reports, October 2026. The argument, the worked examples and the set-up on this page come from this piece.
- Vanguard, Best practices for portfolio rebalancing (AAII summary). US data 1926–2009, no optimal frequency, annual or semi-annual monitoring with 5% thresholds.https://www.aaii.com/journal/article/best-practices-for-portfolio-rebalancing
- Vanguard, The rebalancing edge, December 2024. Daily monitoring, 200 basis point threshold, 175 basis point destination, 5–21 basis points a year against calendar rebalancing.https://corporate.vanguard.com/content/corporatesite/us/en/corp/articles/balancing-act-enhancing-target-date-fund-efficiency.html
- Vanguard, Rational rebalancing, October 2022. Threshold rules require daily monitoring, impractical for many investors.https://corporate.vanguard.com/content/dam/corp/research/pdf/rational_rebalancing_analytical_approach_to_multiasset_portfolio_rebalancing.pdf
- Kitces, rebalancing usually reduces long-term returns but manages risk; return gains typically at most 0.5% a year.https://www.kitces.com/blog/how-rebalancing-usually-reduces-long-term-returns-but-is-good-risk-management-anyway/
- AQR, Portfolio rebalancing: common misconceptions, 2017.https://www.aqr.com/-/media/AQR/Documents/Whitepapers/AQR_Portfolio-Rebalancing_Common-Misconceptions.pdf
- Broker access for AI agents: IG read-only over MCP, Robinhood in a separate agent account, Interactive Brokers away from the order book.https://www.stockbrokers.com/guides/ai-agent-brokershttps://dev.to/barissozen/brokers-are-racing-to-give-ai-agents-a-trading-seat-nobody-is-racing-to-give-them-settlement-2kch
- Simulation on this page: one fixed pseudo-random path of 360 monthly returns. Stocks mean 0.65%, volatility 4.2%; bonds mean 0.25%, volatility 1.2%; no correlation, no costs. Bench and ceiling figures are arithmetic on the values you enter.