Comparing Robo-Advisor Performance: Why the Numbers Do Not Line Up
Last updated August 2026
Short answer
Published robo-advisor returns are mostly not comparable. Six variables decide the number and providers control all of them: the stock and bond split, how much sits outside the US, the exact period, whether fees are deducted, whether tax effects are counted, and how much cash is held. The differences between those choices are larger than any difference in skill, so a ranking built on them mostly measures who held more stock while stocks were rising. The comparison worth running is your own account against a benchmark matching your own allocation. Walnut is informational and is not an investment adviser.
We do not publish a table of provider returns here, because we do not have audited cross-provider data and a table assembled from each firm's own marketing would be worse than nothing. What is genuinely useful is knowing why those tables disagree with each other, and how to run the one comparison that applies to you.
Six variables, any one of which decides the answer
1. The stock and bond split
The dominant driver of return, and it differs between providers even for the same stated risk level. Two portfolios both called moderate can sit ten or more percentage points apart in equity exposure, which will swamp every other difference on this list.
2. How much sits outside the US
The second largest factor in recent history. Providers differ substantially in international weighting, and whichever way markets happen to lead, that choice alone reorders the rankings.
3. The measurement period
Start and end dates decide the answer. A comparison beginning just before a drawdown tells a different story from one beginning just after, and providers naturally publish periods that flatter them.
4. Whether fees are deducted
Some figures are gross of the management fee, some net, and the difference is roughly the whole margin people are trying to detect between providers.
5. Whether tax effects are counted
Harvesting shows up on a tax return rather than in a return figure, so a provider that adds real value there can look identical to one that does not, and a claimed after-tax benefit rests on assumptions about a bracket that is not yours.
6. Cash holdings and sweep rates
Some portfolios hold a meaningful cash allocation, and what that cash earns varies by provider. It is small in a rising market and less small when rates are high.
The first is the one to internalise. Risk labels are not standardised across the industry, so two portfolios both described as moderate can hold substantially different amounts of stock. Comparing them and attributing the gap to the provider is measuring the label rather than the management.
What a study has to do before its table means anything
| Requirement | Why |
|---|---|
| Open real accounts | Funded accounts at each provider rather than modelled portfolios, because published models are not what clients get |
| Normalise the allocation | Group by actual equity exposure rather than by the provider's risk label, which is not standardised |
| Report net of fees | Every fee, management and underlying funds, on the same basis |
| State the exact period | And show more than one, because a single window is a choice with an outcome attached |
| Separate tax from return | Harvesting benefits reported apart from pre-tax performance, with the bracket assumed stated |
The first row is the one that separates real research from content. Published model portfolios are not what clients receive: models have no cash drag, no deposit timing and no client-specific tax position, so anyone serious opens and funds actual accounts and tracks them for years. When you meet a comparison table, read the methodology before the numbers, and if there is no methodology, the numbers are marketing.
Get a recommendation for your situation
Walnut is the AI that knows your portfolio: ask anything in plain English, research any fund, and get an honest second opinion. On the broker you already use, read-only, and you approve every trade. Walnut is not a registered investment adviser.
How to benchmark your own account
| Step | What to do |
|---|---|
| Find your actual allocation | The current stock, bond, US and international split. It is in your account, not in the marketing |
| Build the matching benchmark | A blend of broad index funds in those same proportions, not the S&P 500 |
| Use time-weighted return | It removes the effect of your deposits, which is what you want when judging the manager |
| Use money-weighted for yourself | This one includes your timing, which is what you want when judging your own decisions |
| Look at three years or more | Anything shorter is mostly noise, and the fee difference needs time to become visible |
Rows three and four are the pair most people conflate. Time-weighted return strips out your deposits, which is the fair way to judge the manager. Money-weighted includes them, which is the honest way to judge yourself. Providers report the first, so a portfolio can have performed perfectly well while your own outcome, driven by when you added money, looks quite different.
If your accounts are spread across providers, doing this by hand is the tedious part, and reading the whole picture at once is the shortcut.
What actually differs between providers, once returns are set aside
Strip out the allocation effect and the remaining differences are real but unglamorous: total cost including the underlying funds, whether tax-loss harvesting is offered and at what account size, which account types are supported, how much cash sits idle and what it earns, and whether human access is included. Those persist across market conditions, unlike a return figure, and they are knowable in advance rather than in hindsight.
Related: do robo-advisors beat the market, and robo-advisor fees explained.
FAQ
Which robo-advisor has the best returns?
There is no reliable answer, because published figures are not comparable. Each provider reports a different allocation over a different period, some net of fees and some not, and the allocation differences alone are larger than any skill difference. Rankings built from those numbers mostly measure who held more stock in a rising market.
Why are robo-advisor returns so hard to compare?
Six variables have to match before a comparison means anything: the stock and bond split, the international weighting, the exact period, whether fees are deducted, whether tax effects are counted, and how much cash is held. Providers control all six when they publish, and rarely align on any of them.
Is there an independent robo-advisor performance study?
Serious attempts exist and they work by opening and funding real accounts at each provider, then tracking them over years, because published model portfolios are not what clients actually receive. Read the methodology before the table: how they normalised allocation and whether returns are net of fees decides what the numbers mean.
How do I compare my robo-advisor's performance properly?
Find your actual allocation, build a benchmark from broad index funds in the same proportions, and compare against that rather than the S&P 500. Use time-weighted return to judge the manager and money-weighted to judge your own timing, and look at three years or more before concluding anything.
What is the difference between time-weighted and money-weighted return?
Time-weighted strips out the effect of deposits and withdrawals, so it measures the portfolio rather than your behaviour. Money-weighted includes them, so it measures what you actually earned. Providers usually report time-weighted, which is the fair way to judge them and not the number that reflects your outcome.
Does a higher return mean a better robo-advisor?
Usually it means a higher stock allocation, which is a risk choice rather than a quality signal, and it will reverse in a decline. Comparing two providers without matching equity exposure first tells you almost nothing about either.
Should I switch robo-advisors because of performance?
Rarely, and not on a short record. Differences between providers at the same allocation are small, while switching a taxable account realises gains and resets nothing in your favour. Cost, account types supported and whether harvesting applies to you are more durable reasons than a return figure.
Where do published robo returns come from?
Usually the provider's own model portfolios rather than aggregated client accounts, which is a meaningful distinction: a model has no cash drag, no deposit timing and no client-specific tax situation. Treat any figure without a stated methodology as marketing.
Related articles
Walnut is informational and is not an investment adviser, and nothing here is investment advice. No provider return figures are published on this page by design. Past performance does not indicate future results.