Industries / Finance and business services / Finance and banking / Wealth management
Portfolios
Portfolio work is choosing and running the mix of investments a client holds: setting the split between stocks, bonds and other assets, matching it to the risk the client can bear, and trading back to it as markets move.
7
Milestones achieved, from 1975 to 2025.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1975
One fee for everything
A wrap account charges the client one yearly fee for advice and trading instead of a commission on each trade. Trade press credits James Lockwood with creating the first wrap-fee account at E.F. Hutton in 1975, the year US stock commissions were deregulated, giving retail investors "an institutional-type of managed account" for a 3% fee; Hutton's consulting unit later became the Smith Barney Consulting Group. Managed accounts now include separately managed accounts, run by a hired manager in the client's own name, and unified managed accounts, which hold several managers' strategies in one account. The Money Management Institute and Cerulli Associates count US managed-account assets at $16.3 trillion at the end of 2025, after growth of 19.1% that year.
Credit
James Lockwood; E.F. Hutton
1986
The 93.6% misreading
Gary Brinson, Randolph Hood and Gilbert Beebower studied the quarterly returns of 91 large US pension funds from 1974 to 1983 and found that the asset mix, the split between stocks, bonds and cash, explained 93.6% of the variation in a fund's quarterly returns over time. The figure was widely quoted as "93.6% of the return"; an unpublished 1998 study by Jennifer and John Nuttall found 49 of 50 citations it checked inaccurate. In 2000 Roger Ibbotson and Paul Kaplan split the question three ways: the mix explained about 90% of a fund's ups and downs over time, about 40% of the differences between funds, and about 100% of the average level of return.
LessonA research figure answers one exact question, and quoting it for another misleads clients.
Credit
Gary Brinson, Randolph Hood and Gilbert Beebower; Roger Ibbotson and Paul Kaplan
2008
Same date, different risk
A target-date fund shifts a saver's mix from stocks toward bonds as a chosen retirement year approaches. Under the US Pension Protection Act of 2006 the Labor Department made such funds an approved default for workers who make no choice, and by 2009 a study cited by the SEC found that 70% of US employers used them as the default. In 2008, funds aimed at people retiring in 2010 lost almost 25% on average, with losses ranging from about 4% to 41%, and the SEC and the Labor Department held a joint hearing on 18 June 2009.
LessonTwo funds with the same target year can hold very different mixes, so the label alone does not tell a saver the risk.
Credit
the US Department of Labor and the SEC, which examined the losses
2011
Risk tools found flawed
Advisers often score a client's appetite for risk with a questionnaire, then map the score to a portfolio. In January 2011 the UK Financial Services Authority (FSA) reported that 9 of the 11 risk-profiling tools it reviewed had weaknesses "which could, in certain circumstances, lead to flawed outputs", and that half of the investment files it rated unsuitable between March 2008 and September 2010 failed because the investments did not match the risk the customer was willing and able to take. A 2015 CFA Institute Research Foundation brief by Joachim Klement found that questionnaires typically explain less than 15% of the differences in how much risk investors hold.
LessonA risk score is only as good as the questions and scoring behind it, and few firms test them.
Credit
the UK Financial Services Authority; Joachim Klement
2022
Stocks and bonds fall together
A 60/40 portfolio, 60% stocks and 40% bonds, is the standard balanced mix, and it relies on bonds rising when stocks fall. In 2022 the S&P 500 lost 18.1% and 20-year US Treasury bonds lost 26.1%, which Larry Swedroe, writing in WealthManagement.com, calls the first year both fell by double digits; InvestmentNews, using the Vanguard Wellington Fund as a stand-in, reported a 14.3% fall, the worst year for 60/40 portfolios since 2008. AQR researchers found that the stock-bond correlation, a measure of whether the two move together, was about +0.18 from 1926 to 1999 and about −0.2 from January 2000 to November 2022.
LessonBonds cushion stocks mainly when news about growth drives markets, and an inflation shock can push both down together.
Credit
AQR (Brixton, Brooks, Hecht, Ilmanen, Maloney and McQuinn), as reported by Larry Swedroe
2022
A fund limits withdrawals
Advisers had placed many wealthy clients in BREIT, Blackstone's $69 billion real estate fund, which does not trade on an exchange and lets investors withdraw only within set limits. In December 2022 BREIT said withdrawal requests had passed its limits of 2% of net asset value a month and 5% a quarter, so it paid out only part. Requests peaked at $5.3 billion in January 2023, and the fund did not meet all requests in full until February 2024, after returning more than $15 billion over 15 months.
LessonA fund that holds hard-to-sell assets can offer only limited exits, and the limit binds when many clients want out at once.
Credit
Blackstone
2025
Ready-made models grow
A model portfolio is a ready-made mix, designed by an asset manager, that advisers copy into client accounts and rebalance. Morningstar counted more than $645 billion in third-party models at the end of March 2025, up 62% since its June 2023 survey and not counting the in-house models of large brokerages; BlackRock took $30.2 billion, more than 80%, of the net money that flowed into models in 2024. When Morningstar compared the 60/40 models of ten leading teams it found "a lot of room for variation": weights in high-yield bonds ranged from 0% to 10%.
Credit
Morningstar Manager Research (Stephen Margaria, Jason Kephart, Nour Al Twal)
Watch for
Two-thirds less portfolio time
Vanguard launched Vanguard Custom Model Portfolios, which advisers can adjust and run through the portfolio software of Vestmark, SS&C Black Diamond and Orion, and says that using models for core portfolio construction "can reduce the amount of time advisors spend on portfolio management by two-thirds". The figure comes from Vanguard's own 2026 research.
Achieved whenA party that sells no models measures adviser hours with and without models across many firms and finds a similar cut.
Claimed by
Vanguard, August 2026
Private markets in one account
BlackRock and Partners Group launched three separately managed accounts (income, balanced and growth) on Morgan Stanley's wealth platform, each holding stakes in seven evergreen private-market funds run by BlackRock, HPS and Partners Group; an evergreen fund has no end date and allows only limited withdrawals. They call it a "first-of-its-kind solution", and say portfolios that added private markets showed "approximately 100 points of incremental return annually at comparable risk levels" on data from March 2015 to March 2025, without saying what kind of points.
Achieved whenAn independent tracker such as Morningstar or Cerulli reports the accounts' assets and live results, and the underlying funds meet all withdrawal requests through a period of heavy redemptions.
Claimed by
BlackRock and Partners Group, January 2026
AI notes on portfolios
Vanguard says its Expert Insights tool, built on generative AI, "turns complex portfolio data into straightforward, client-ready observations and recommendations" for advisers. In April 2026 it was "currently in pilot with select advisors" and due to be "widely available later in 2026".
Achieved whenThe tool is generally available and an outside reviewer checks a sample of its outputs against the portfolio data and publishes an error rate.
Claimed by
Vanguard, April 2026
Open
01
A test bench for rebalancers
A public set of 1,000 invented accounts, each with a target mix, holdings broken into tax lots (the date and price of each purchase) and a schedule of deposits and withdrawals, plus a harness that runs several open-source rebalancing tools on the same accounts over public historical prices. Rebalancing is trading a portfolio back to its target mix. Score each tool on drift from the target, yearly turnover, trading costs and realized gains. The results score software, not investments.
The briefCredit · unclaimed02
Overlap between model portfolios
An open archive of the holdings that model-portfolio providers publish in their quarterly factsheets, with each holding labeled by asset class and region. For every pair of models with the same risk label, such as "moderate" or 60/40, publish an overlap score: the sum, across holdings, of the smaller of the two weights. The scores measure similarity and do not rank what to buy.
The briefCredit · unclaimed
Open 01 · the brief
A test bench for rebalancers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of 1,000 invented accounts, each with a target mix, holdings broken into tax lots (the date and price of each purchase) and a schedule of deposits and withdrawals, plus a harness that runs several open-source rebalancing tools on the same accounts over public historical prices. Rebalancing is trading a portfolio back to its target mix. Score each tool on drift from the target, yearly turnover, trading costs and realized gains. The results score software, not investments.
Why it is still open
The closest public benchmark, by Yin and co-authors (March 2026), ran 15 strategies through five open-source backtest engines on 180 S&P 500 stocks and varied only trading costs, with no client cash flows or tax lots. Cvxportfolio, an open-source optimizer, models holding and trading costs but its page does not mention taxes, and a Hugging Face dataset of 10,000 rebalance rows counts lots per account without lot-level records and does not call itself a benchmark.
Built with what exists
cvxportfolio (GPL-3.0); PyPortfolioOpt (MIT); Double Finance's Oracle, an MIT-licensed tax-aware rebalancer with wash-sale checks; public daily price data; Inkling 2 Small to write and check the harness's tests (see the Note "Checked without being asked").
First step
Write and publish the account generator and file layout for the first 100 invented accounts with three target mixes.
Done looks like
One command that reproduces, for each tool on the same 1,000 accounts, its average drift in percentage points, yearly turnover, costs and realized gains, and a second person who reruns it and matches every figure to within 0.01 percentage points.
Open 02 · the brief
Overlap between model portfolios
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open archive of the holdings that model-portfolio providers publish in their quarterly factsheets, with each holding labeled by asset class and region. For every pair of models with the same risk label, such as "moderate" or 60/40, publish an overlap score: the sum, across holdings, of the smaller of the two weights. The scores measure similarity and do not rank what to buy.
Why it is still open
Morningstar's 2025 comparison covered the 60/40 models of only ten teams, at the level of asset classes, using its own survey and its paid Morningstar Direct database. This research found no open archive at the level of holdings, and factsheet links disappear: two BlackRock model factsheet links found in search returned "404 Not Found" on 19 November 2026.
Built with what exists
Browser Use 1.0 to collect public factsheets (see the Note "Deleting 80% to start over"); pdftotext and Python to pull weights from the PDFs; Gemini Decision 1 to label thousands of holdings by asset class and region in one pass (see the Note "A thousand questions, one pass").
First step
Download the current factsheets of the moderate or 60/40 models of ten providers and label them by hand, as a test set for the automatic labeler.
Done looks like
At least 200 models from at least 20 providers in at least two countries, each with dated holdings; overlap scores for every pair with the same label; and a hand check of 50 random holding labels that agrees at least 95% of the time.