Key takeaways
- Between 9 October 2007 and 9 March 2009, a monthly-rebalanced 100% US equity portfolio fell 54.6% peak to trough, an 80/20 fell 43.9% and a 60/40 fell 31.4% — total return, marked daily.
- 2022 collapsed the gap between those allocations. The 60/40 fell 22.3% against 25.4% for all-equity, because the bond side fell 20.6% at the same time.
- The wait is usually the binding constraint, not the fall. The all-equity portfolio needed 4.4 years to regain its October 2007 peak; the 60/40 needed 2.8 years; a 40/60 needed 1.8 years.
- Self-reported tolerance carries a lot of variation nobody can explain. In Claudia Sahm's Federal Reserve study of 12,000 people, the persistent part of risk tolerance her covariates could not account for had a standard deviation of 0.72, against 0.41 for the part they could. The transitory residual, which she attributes to survey response error and short-lived preference shocks, was 1.55.
- The UK regulator reviewed 11 risk-profiling tools in March 2011 and found 9 of them had features that could produce an output not reflecting the risk a customer was willing or able to take.
Tolerance is a feeling, capacity is arithmetic, and only one of them can be calculated
A questionnaire told you that you are a "moderate" investor, and you want to know what that means and whether it's even the right question. The short version: it's half of the right question.
Risk tolerance is how much volatility you can stomach — a fact about your psychology. Risk capacity is how large a fall your plan can absorb before the plan itself breaks — a fact about your balance sheet, your timetable and what the money is for. A 34-year-old with a stable salary, no debt and a 30-year horizon has enormous capacity and may still be unable to sleep through a 20% fall. A 63-year-old planning to buy a house next spring may be perfectly calm and have almost no capacity at all.
Most questionnaires blend the two into a single word. The US Securities and Exchange Commission's own investor-education page defines risk tolerance as "your ability and willingness to lose some or all of your original investment in exchange for potentially greater returns" — ability and willingness in one sentence, one label, one score. Those are two different measurements with two different units.
Regulators separate the two questions even where questionnaires don't
The clearest official statement of the distinction comes from the UK. In March 2011 the Financial Services Authority — now the Financial Conduct Authority — published finalised guidance titled Assessing suitability: Establishing the risk a customer is willing and able to take. Willing and able. Two verbs, deliberately.
The guidance defines capacity for loss precisely: "the customer's ability to absorb falls in the value of their investment. If any loss of capital would have a materially detrimental effect on their standard of living, this should be taken into account." Its central finding was that firms were measuring one and skipping the other — "although most advisers and investment managers consider a customer's attitude to risk when assessing suitability, many fail to take appropriate account of their capacity for loss."
The regulator also examined the tools themselves. Of 11 risk-profiling tools reviewed, 9 "had features that meant that there was a high probability that, under certain circumstances, the output might not accurately reflect the risk that a customer is willing or able to take." Among the practices flagged as poor: a firm whose method was to have the customer pick a number from 1 to 10, where what each number meant was left entirely to interpretation. And of the investment files the FSA assessed as unsuitable between March 2008 and September 2010, half failed specifically on the risk the customer was willing and able to take.
US rules point the same way without naming the split. FINRA's suitability rule requires a firm to understand a customer's investment profile, which it enumerates as "age, other investments, financial situation and needs, tax status, investment objectives, investment experience, investment time horizon, liquidity needs, risk tolerance." Risk tolerance is one item on a list of nine. Time horizon, liquidity needs and financial situation — all capacity — are three more. A questionnaire that returns "moderate" has compressed nine inputs into one adjective.
How much of a risk-tolerance score is signal
There's a serious empirical answer to this, and it isn't encouraging. Claudia Sahm, then an economist at the Federal Reserve Board, studied the Health and Retirement Study's panel of hypothetical income gambles — the same people asked the same risk question repeatedly across a decade. The sample was 12,000 respondents between 1992 and 2002.
Two results matter here. First, the stable part is real: persistent differences between individuals account for 73% of the systematic variation in measured risk tolerance. People genuinely do differ, and the difference is durable. Second, the systematic part is small next to the noise. The estimated standard deviation of the survey response error was 1.55, against 0.72 for the persistent individual effect — response error more than twice the size of the thing being measured. Sahm's decomposition puts systematic within-person change at 11% of the total variance in the measure, and systematic between-person differences at 45%. A single questionnaire, taken once, is a draw from a distribution with a large error term.
The measured score also moves with the weather. In the same study, a ten-point rise in the University of Michigan consumer sentiment index was associated with a 9% increase in risk tolerance. Average measured risk tolerance rose 36% between October 1992 and February 2000, then fell 15% between May 2002 and February 2003. The effect faded fast — sentiment six months earlier had a weaker association than current sentiment, and sentiment a year earlier had none. That's a description of a measurement that partly reflects the market you happen to be sitting in when you fill in the form.
A better question, and one with an arithmetic answer
The reframing that gets further than "how do you feel about risk" is this: what is the largest peak-to-trough fall you could live through without selling, without stopping contributions, and without changing what the money is for?
That question has a testable answer, because peak-to-trough falls for standard allocations are a matter of record.
What each allocation actually lost in 2008, 2020 and 2022
| Allocation (equity/bonds) | 2007-09 crisis | Feb-Mar 2020 | 2022 | Worst fall since 1962 |
|---|---|---|---|---|
| 100 / 0 | -54.6% | -34.2% | -25.4% | -54.6% (2007-09) |
| 80 / 20 | -43.9% | -26.5% | -23.8% | -43.9% (2007-09) |
| 60 / 40 | -31.4% | -18.5% | -22.3% | -31.8% (1973-74, tied with 2007-09) |
| 40 / 60 | -18.2% | -11.0% | -21.0% | -22.1% (1973-74) |
| 0 / 100 | -14.0% | -6.0% | -20.6% | -27.1% (2020-23) |
Method, so you can check it or disagree with it. The equity leg is the CRSP value-weighted US market total return, daily, from Kenneth French's data library. The bond leg is a constant-maturity 10-year US Treasury total return built from the Federal Reserve's daily 10-year yield series on FRED. Portfolios are rebalanced monthly and marked every trading day, nominal, before tax and costs. That bond construction reproduces Aswath Damodaran's published annual 10-year Treasury returns with a mean absolute error of 1.0 percentage point a year across 1963-2025 — 20.5% against his 20.1% in 2008, and -16.4% against his -17.8% in 2022.
Two honest caveats about the bond leg. A 10-year Treasury has longer duration than a typical aggregate bond fund, so the 2022 losses in the lower rows run deeper than a shorter-duration fund would have shown. And it carries no credit risk, so 2008 is kinder than a portfolio holding corporate bonds would have been. We used the 10-year anyway because the Fed's daily 7-year yield series starts only in July 1969, and the 10-year is what reaches back to 1962.
A third caveat is about precision. The 60/40's two worst episodes, 1973-74 and 2007-09, land within half a point of each other. Which one is deeper depends on details as small as the rebalancing date, so they're best read as tied.
Two of our other pieces measure these same events differently, and the gaps are worth naming. Our piece on rebalancing through a crash puts a 60/40's 2007-09 loss at 22.4%, not 31.4%. That one uses Robert Shiller's monthly data, in real terms, on a portfolio left to drift; this one marks daily and rebalances monthly. Rebalancing through the fall accounts for about six of those nine points, and daily marking for about three. Our volatility piece puts the same bond bear market at 37.3% real from April 2020, against 27.1% nominal from August 2020 here. Consumer prices rose about 18% between those dates, and only a daily series catches the yield low of 4 August 2020.
The 2020 column repays a pause for a different reason. That entire fall happened between 19 February and 23 March — 23 trading days. The chart with this piece plots the last column, the worst fall since 1962, and what it shows is a slope rather than a cliff: across the equity-heavy mixes, each 20 percentage points of equity weight bought roughly 11 points of extra maximum loss.
2022 is the year that breaks the intuition
Read down the 2022 column and the diversification argument thins out. The 60/40 fell 22.3% while the all-equity portfolio fell 25.4%. Holding 40% in bonds saved about 3 percentage points. In the two earlier crises the same 40% saved 23 points and 16 points. The reason is in the bottom row: bonds fell 20.6% over the same stretch.
That bottom row makes a sharper point still. The all-bond portfolio's worst fall in 64 years wasn't 2008 or 2020. It was the 27.1% decline from 4 August 2020 to 19 October 2023 — deeper than anything a 40/60 portfolio went through in the same period.
That comparison is a hostage to one construction choice, so here's the test. A par 10-year Treasury has a modified duration near 9.0 at a 2% yield; a 7-year one, closer to a broad aggregate bond fund, near 6.5. Rebuild the table on a 7-year leg and the all-bond worst fall since 1962 shrinks to about 19%, while the 40/60's stays near 21%. The ordering flips. Against the 40/60's own 2020-23 fall the all-bond lead survives, but by less than a point. So the sharp version of the claim doesn't hold on a shorter bond leg. Only the mild version does: bonds alone have had drawdowns in the range of a conservative balanced portfolio. On a 7-year leg 2022 softens too, with 40% in bonds saving about 5 points rather than 3. "Bonds are the safe bit" is a claim about correlation with equities, not about the absence of drawdown, and correlation isn't a constant.
The wait matters more than the fall
Capacity isn't really about the depth of the hole. It's about whether you need the money before the hole closes.
Measured from peak to the day a new high was reached, the all-equity portfolio needed 4.4 years to recover from its October 2007 peak, the 80/20 needed 3.3 years and the 60/40 needed 2.8 years. From the 2022 peak the ordering inverted: all-equity recovered in 2.0 years, the 60/40 in 2.2 and the 40/60 in 2.6, while the all-bond portfolio had still not regained its 2020 high by May 2026 — nearly six years. Anyone whose plan required a specific sum on a specific date inside those windows had a capacity problem, whatever their questionnaire said. Our separate piece on what recovery looks like once inflation is taken out runs the same arithmetic in inflation-adjusted terms, where the waits get considerably longer.
Frequency belongs in the same calculation. Since 1962, the all-equity portfolio spent 29% of all trading days at least 10% below its previous high, and 14% of days at least 20% below, across 13 separate falls of 20% or more. For the 60/40 the figures are 13% of days, 2% of days, and 6 such falls — roughly one a decade. Being underwater isn't the exception. For an equity-heavy portfolio it is close to a third of the time.
The strongest counter-argument: a person who sells has zero tolerance, whatever their capacity
There's a real case against everything above, and it deserves its best statement. Capacity is a theoretical quantity. It describes what your finances could survive, not what you'll do. Someone with a 30-year horizon, a secure income and no need for the money has, on paper, the capacity to hold 100% equities through a 55% fall. If that person sells at the bottom, their realised outcome is a 55% loss crystallised and a recovery missed. Their effective tolerance was zero, and the capacity calculation was worse than useless — it talked them into a position they couldn't hold. On this reading the psychological question dominates, because behaviour, not arithmetic, determines what actually happens.
The evidence gives that argument partial support. Morningstar's Mind the Gap 2025 study estimates that the average dollar in US funds and ETFs earned 7.0% a year over the ten years to 31 December 2024, against 8.2% for the funds themselves — a 1.2 percentage point annual shortfall it attributes to the timing and size of investors' purchases and sales. And the shortfall tracks volatility. In the most volatile quintile of funds, investors captured 3.4% a year against the funds' 5.4%; in the least volatile quintile, 9.6% against 10.0%. The more the holding moved, the less of its return investors kept. That pattern rests on the same dollar-weighted method as the headline figure, and the method is disputed. Fulkerson, Jordan, Riley and Yan, in the Financial Analysts Journal in 2026, put the cost of poor timing at 0.10% a year rather than 1.2.
Two things temper it. First, in that same study, allocation funds — the all-in-one balanced portfolios — showed a gap of 0.1 percentage points, 6.3% against 6.5%. Whatever destroys returns, it isn't a broad multi-asset portfolio held through a cycle. Second, that dispute is a live one, and we have refereed it separately in our piece on what bad timing really costs.
Actual behaviour in a crash is also less dramatic than the folklore. Vanguard's How America Invests study covers more than five million retail households from 2015 through the first half of 2020. Of the volatile first half of 2020 it reports that only 22% of households traded and that fewer than 1% abandoned equities completely. Of those who did trade, 62% moved money into equities rather than out. Panic selling is real, but it wasn't the majority response even in the fastest crash on record.
Where that leaves the argument: tolerance and capacity are not rivals, and the binding one is whichever is lower. Capacity sets a ceiling you cannot exceed without risking the plan. Tolerance sets a ceiling you cannot exceed without risking yourself. A drawdown number — 20%, 35%, 50% — is one of the few quantities both can be measured against.
What this evidence cannot tell you
Historical drawdowns aren't a bound on future ones. The deepest fall in the table is the 54.6% of 2007-09, but that is an artefact of a sample starting in 1962, which is when the Federal Reserve's daily 10-year yield series begins. On the same daily equity data, the US market fell 84.1% between 3 September 1929 and 8 July 1932. Any figure in that table is a lower bound on what is possible, not an upper one.
The data is also entirely US, nominal, before tax and costs, and assumes an allocation held to target through the fall — which is the behaviour this whole piece is questioning. Every figure is a total return, so it assumes dividends and coupons were reinvested at the worst moments as well as the best.
On the questionnaire side the limits are sharper. Risk-profiling questionnaires vary enormously in quality, from psychometrically constructed instruments to compliance box-ticking. The FCA's finding about 9 of 11 tools concerned a specific set of UK tools in 2011, not a universal law. Sahm's measure is a hypothetical income gamble, not a market question, and how well it maps onto what someone does when their account is down 30% is an inference rather than a measurement. Self-reported tolerance recorded in a calm market is, on that same evidence, partly a reading of the calm market.
What would change the conclusion
If bonds and equities kept falling together. The entire capacity argument for holding less equity rests on the other asset behaving differently. In 2022 it didn't, and the 60/40 lost within 3 points of the all-equity portfolio. A second episode like that would mean equity weight is a weak lever for controlling drawdown, and the useful question becomes total portfolio volatility rather than the stock/bond split.
If someone built a questionnaire that predicted behaviour. The case for the capacity reframing is strong partly because the tolerance measurement is weak. A validated instrument with demonstrated predictive power over actual selling under stress would shift the balance back, and the psychometric literature is actively trying to build one.
If your money has a date on it. Everything here assumes an open-ended horizon. A fixed liability at a fixed date — school fees, a house deposit, a retirement start — turns capacity from a judgement into a constraint, and the recovery column matters more than the drawdown column.
The figure that does the work here isn't a label. It's the largest peak-to-trough fall someone would sit through without changing anything, held as a percentage and as the cash sum it stands for, because those two don't land the same way. LedgerTouch shows the current distance from your own high-water mark, which is that number updating itself. Whether the one you chose was honest is something only a bad year answers.