Key takeaways
- Across 59 overlapping 40-year US windows from 1928 to 2025, a 75-to-36% glide path ended at 3.13 times contributions. A fixed 55.5% weight ended at 3.52.
- The declining path averaged 55.5% equities but only 46.8% pound-weighted, a gap of 8.7 points, because its boldest years came when the pot was smallest.
- What the shape bought was a calmer ending: a median worst real fall of 5.6% in the final ten years, against 11.3% for the fixed weight.
- Neither dynamic rule paid for itself. A volatility budget finished 19.5% below the fixed control at the median, and its final decade was no smoother.
- Vanguard's own simulation puts a conservative glide path at 11.4 times ending salary and an aggressive one at 13.3, with first-percentile outcomes of 5.7 and 5.5.
Bond allocation by age moved the median outcome by about 11%
You've been told to hold more bonds as you get older. The question underneath that advice is rarely tested: does the shape of the path matter, or only the average amount of equity you hold along it?
Run the arithmetic and the answer is mostly the second one. On US annual returns from 1928 to 2025, a linear glide path that starts at 75% equities and finishes at 36% ended a 40-year working life with a median 3.13 times what was paid in. A fixed weight of 55.5%, which is exactly that glide path's own average, ended at 3.52. The shape cost 11.1% of the median outcome, and it lost in 38 of the 59 windows tested.
That is not nothing. It is also much smaller than the gap between either of those and a portfolio held at 100% equities, which finished at 5.36 times contributions. Bond allocation by age is a second-order decision. How much equity you hold on average is the first-order one.
How the test was built, and what it deliberately holds constant
The return series is Aswath Damodaran's NYU Stern dataset of annual returns on the S&P 500 with dividends and on 10-year US Treasuries, 1928 to 2025. Deflating both by the Bureau of Labor Statistics CPI-U, December to December from a December 1927 base, gives real returns of 6.78% a year for the S&P 500 and 1.46% for Treasuries over the full 98 years.
The saver contributes one inflation-adjusted unit at the start of each of 40 years, from age 25 to 65, and rebalances to target once a year. There are 59 overlapping 40-year windows in the data, the first running 1928 to 1967 and the last 1986 to 2025. Terminal wealth is expressed in real terms as a multiple of the 40 units contributed, so windows with wildly different inflation are comparable.
The control matters more than anything else here. Comparing a 75-to-36% glide path against a fixed 60/40 tells you almost nothing, because the glide path holds less equity on average and less equity means less money over 40 years. So the comparison is against a fixed 55.5%, the glide path's own time-weighted average. Any difference that survives is attributable to when the equity was held, not how much.
Allocation and outcome, 59 overlapping 40-year windows, 1928 to 2025
| Path | Equity at 25 / at 64 | Average weight | Pound-weighted | Median terminal multiple | Median worst real fall, final decade |
|---|---|---|---|---|---|
| Fixed 60% | 60 / 60 | 60.0% | 60.0% | 3.73 | 13.2% |
| Fixed 55.5% (control) | 56 / 56 | 55.5% | 55.5% | 3.52 | 11.3% |
| Linear, 100 minus age | 75 / 36 | 55.5% | 46.8% | 3.13 | 5.6% |
| Dynamic, funded status | 75 / 36 | 51.3% | 39.3% | 2.82 | 6.8% |
| Dynamic, volatility budget | 75 / 42 | 54.5% | 49.5% | 2.83 | 11.1% |
| Fixed 100% equities | 100 / 100 | 100.0% | 100.0% | 5.36 | 22.7% |
The chart plots the median column. The rest of this piece is about the two columns either side of it.
The mechanism is pound-weighted exposure, not average exposure
Here's the part that explains the whole result. The linear path spends its life at an average equity weight of 55.5%, but the money inside it experiences only 46.8%. That 8.7 point gap is the entire story.
A 40-year contribution schedule puts most of the money in the pot late. In year three you own perhaps three units of savings, and the fund's 73% equity weight is applied to those three units. Late in the run the balance is many times larger, and by then the weight is down near 38%. Averaging the weights over time gives every year equal billing. Averaging over the money does not, and the money is what compounds.
This is why a declining path is a lower-risk portfolio than its own headline average suggests. It also explains why the fixed control wins on the median without holding more equity on paper. Nothing about market timing is involved. It is the arithmetic of where the balance sits when the weight is applied.
What the declining path bought: a final decade that fell half as far
A glide path is not sold as a way to accumulate more. It is sold as a way to stop the last few years wrecking the first thirty, and on that measure it delivered.
Measuring the deepest peak-to-trough real fall in the pot over the final ten years of each window, the linear path's median was 5.6%. The fixed 55.5% control's median was 11.3%, and the all-equity portfolio's was 22.7%. The worst single decade in the sample was a 27.9% fall for the linear path against 32.4% for the control and 47.6% for all-equity. The glide path was the smoother of the two in 48 of the 59 windows.
So the trade was roughly this: give up about 11% of the median result to halve the size of the fall you sit through in the years when you have the most to lose and the least time to recover. Whether that's worth paying isn't an arithmetic question, and this piece doesn't answer it. What the arithmetic does say is that the price is real and the protection is real, and both are smaller than the marketing on either side implies.
Dispersion moved less than you'd expect. The 10th-to-90th percentile range for the control ran 1.90 to 4.76 times contributions. For the linear path it ran 1.96 to 4.41. The floor came up a little and the ceiling came down more.
Neither dynamic rule paid for itself on this data
"Dynamic" covers two different families, and both were tested against the same control.
The first responds to funded status. It starts from the linear path, then adds 15 points of equity when the pot is below 90% of a 4% real trajectory and takes 15 points off when it is above 110%. Its median came in at 2.82, which is 19.7% below the fixed control. The reason is not subtle: history beat a 4% real assumption most of the time, so the rule spent most of its life de-risking. Its pound-weighted equity exposure was 39.3%, the lowest of any path tested. It did not so much change the shape as quietly lower the level.
The second responds to realised volatility, which is how real schemes tend to define a glide path. Nest's 2013 investment approach sets its risk budgets as volatility levels: a long-term average of 10 to 12 per cent in the Growth phase, and 7 per cent in the Foundation phase. The rule tested here is my own construction in that spirit. It sets the equity weight so the mix's volatility, measured on up to 20 prior years of returns, hits a budget of 12% that tapers to 7% over the final ten years. Median terminal wealth was 2.83, or 19.5% below the control. Its average weight was 54.5%, close to the control, so this one really is a shape effect.
And it bought nothing. The volatility-budget path's median worst final-decade fall was 11.1% against the control's 11.3%, and it was the smoother of the two in only 22 of 59 windows. It also produced the worst single outcome in the entire study, 1.41 times contributions over 1942 to 1981. Realised volatility spikes after a crash, so a volatility budget cuts equity exactly when the previous drawdown has already happened and forward returns have historically been good. Rules that look backwards at risk tend to sell what has already fallen.
Vanguard's own numbers point the same way, and Vanguard still runs a glide path
This is not a fringe finding, and the industry does not hide it. Vanguard's March 2019 paper on its target-date funds simulates the same 40-year contribution problem across three glide paths. The median investor on Vanguard's own path retired with 12.4 times ending salary. On a more aggressive path the figure was 13.3 times, about 7% more. On a more conservative one it was 11.4 times, about 8% less.
The interesting row is the one underneath. At the 1st percentile the conservative and the Vanguard paths both produced 5.7 times ending salary, and the aggressive one produced 5.5. The bad case barely moved. Almost all of the difference between the glide paths showed up in the middle and the top of the distribution, not the bottom, which is the part a glide path is meant to protect.
Vanguard states the implication plainly: "If maximization of wealth is the primary goal, then a higher equity allocation would be an appropriate strategy." They run the glide path anyway, because they are optimising something else. Since 2016 the path has been set by a utility-based model, one in which, as the 2019 paper's own footnote puts it, "negative outcomes (i.e., financial losses) are penalized more significantly than positive outcomes of equal magnitude or dollar value." The path itself was constructed in 2003 and, as of a January 2021 committee memo, had been adjusted five times in 17 years.
Estrada found the same thing in 19 countries, and the defenders have a real answer
The strongest published version of this critique is Javier Estrada's 2014 paper in the Journal of Portfolio Management. He ran the Dimson-Marsh-Staunton dataset of 19 countries and two regions from 1900 to 2009, over 71 overlapping 40-year lifetimes, with a $1,000 real annual contribution for a cumulative $40,000.
His US result: the five declining lifecycle strategies produced an average median terminal wealth of $100.6 thousand, while a strategy held fully in stocks for the whole 40 years produced roughly 88% more. On mean terminal wealth the gap was wider still, $110.5 thousand against $223.5 thousand. Across the world market, contrarian strategies that ran the glide path backwards gave 18% higher terminal wealth in the worst case and 24% higher across the bottom decile.
Estrada also states the counter-case fairly, and it is the one that matters. As he writes, "the goal of these funds is not to maximize the accumulated savings at retirement, but rather to balance risk and return." A saver who quits during a 40% drawdown at 58 does not get the median outcome in anybody's backtest. Nest's 2013 investment approach reaches the same place from the opposite direction. It runs a low-risk Foundation phase for the first five years because younger savers said they may stop saving after a fall, and Nest's own judgement is that "the impact on final outcomes is negligible for most members, but the impact of stopping saving could be much more harmful."
Neither of those is a claim about compounding. They are claims about people, and the backtest has nothing to say about them.
Nest's growth exposure rises for thirty years before it falls
The linear path this piece tested is a textbook object. Bond allocation by age in a real default fund does not look like it. Nest keeps over 95% of its members in its Retirement Date Funds, more than 13 million people, and its published allocation for December 2025 has a shape that a straight line cannot describe.
Grouped by Nest's own labels, the Higher Growth bucket sits at 50% for the Starter fund and the 2071 fund, the two held by the youngest savers. It climbs to 70% for the 2036 to 2041 funds, held by people roughly 10 to 15 years from retirement. Only then does it fall, to 30% for the 2025 and 2026 funds. Listed equity follows the same arc: 37.8% for the Starter and 2071 funds, 53.0% for the 2036 fund, 22.6% for the 2026 fund.
That is the opposite of the 100 minus age rule at the young end, and it is deliberate. The Foundation phase lasts five years, the Growth phase roughly 30, and the Consolidation phase begins about ten years before the retirement date. On the evidence above, the first of those three decisions costs the least and is defended on behavioural grounds rather than financial ones.
What this test cannot tell you
One country, one asset pair. The equity leg is the S&P 500 and the bond leg is 10-year US Treasuries, so nothing here speaks to gilts, to sterling, or to a portfolio holding property, credit and infrastructure the way Nest's does. Estrada's 19-country sample is the check on that, and it points the same way, but it stops in 2009.
The 59 windows overlap heavily, so they are not 59 independent experiments. Every window from 1942 onwards contains the same 1970s inflation, and every window ending after 1990 contains a large part of a 40-year bond bull market that cannot repeat from current yields. A backtest is a description of one path history took, not a forecast.
The model also charges nothing. No fund fees, no trading costs, no tax, and a contribution that rises exactly with inflation rather than with a real career. Real earnings usually rise faster than prices in mid-career, which pushes even more of the money into the late, low-equity years and would widen the pound-weighted gap rather than narrow it. On equity allocation and crash depth, and on long-run asset class returns, the same dataset supports separate pieces with their own limits.
What would change the conclusion
Three things would, and only one of them is about markets.
The first is the return gap itself. Equities beat Treasuries by 5.32 percentage points a year in real terms over 1928 to 2025. Shrink that to a point or two and the cost of de-risking early collapses, and the shape argument stops mattering in either direction. Every result above is a lever on that single spread.
The second is the scoreboard. Median terminal wealth is one measure, and it is the one that flatters holding equities. Score the same six paths on the probability of clearing a specific spending target, which is what Vanguard's model does, and a path that gives up upside to compress the bottom of the distribution can win on its own terms while losing on this page's.
The third is behaviour, and it is the one no return series can settle. If the 5.6% median final-decade fall rather than the 11.3% one is the difference between staying invested and cashing out at 60, then the 11% of median wealth the glide path gave up was the cheapest thing in the portfolio. Nothing in 98 years of US returns tells you which of those two people you are.