Research · Dataset
Buy gold a little every month, or buy it all at once?
Dollar-cost averaging is the most widely repeated piece of retail gold advice: a fixed amount every month, and you never have to guess the bottom. It is also almost never tested against the thing it is supposed to beat. This page runs the comparison properly — every possible start month, the same cash on both sides — and then takes the result apart to find out why it comes out the way it does.
802 monthly observations, 1960-01 to 2026-10 · Source data updated November 03, 2026
The answer, first
Averaged across all 743 windows, the monthly programme ended -30.85 percentage points behind the lump sum (29.8% against 60.6%). The moving-block bootstrap 95% interval on that difference is -55.56 to -5.03 pp, which does not cross zero.
That interval deserves a caveat, because it is the trap this page is built around. Consecutive 60-month windows overlap by 59 months, so the 743 windows are not 743 independent observations — the naive t-test (p = 1.39e-15) is meaningless and the block bootstrap still inherits much of that dependence. Taking every 60th window instead gives 13 genuinely independent comparisons: the gap is -26.37 pp with a p-value of 0.525 and a confidence interval of -72.24 to +26.53 pp. That interval crosses zero. So the honest statement is that DCA is behind on average across the full set of start months, and that on a statistically clean sample the size of the gap is not precisely measured. Both things are true, and only reporting the first one would be misleading.
But that is the surface answer, and taking it at face value would be wrong — as the section on exposure duration below shows, most of that gap is arithmetic, not skill.
Source and method
The comparison holds cash constant. The DCA programme contributes 1/60 of the total every month for 60 months; the lump sum invests the same total on day one and holds it for the same 60 months. Both are marked to market at the 60-th month's average price. No leverage, no withdrawals, no rebalancing, no taxes.
Contributions are modelled at each month's AVERAGE price, which is an approximation of a real fill. No spread, commission, financing or storage cost is included, so the DCA side is if anything flattered. No taxes, no leverage, no withdrawals.
One sample caveat that matters more than usual here: the official gold price was legally fixed at $35/oz from 1960-01 to 1967-12 in this series, and the London free market only opened in 1968. Windows that start inside the peg largely buy a frozen price, so they are counted separately below rather than left to distort the headline.
Where DCA does win — and it is exactly where you'd expect
Split the 743 windows by whether gold rose or fell over the 60-month window, and the picture inverts completely.
In a rising market, buying everything on day one is simply better, and DCA wins only 9.9% of the time. In a falling market, the monthly programme buys into the decline and wins 83.9% of the time.
This is what DCA is actually for. It is not a return enhancer; it is a mechanism that reduces the pain of a falling price by buying more units as the price drops. Whether that trade is worth making depends entirely on your own behaviour and cash flow, not on an edge.
The gap is mostly exposure arithmetic
Here is the part that changes the interpretation. A lump sum is fully invested for all 60 months. A DCA programme's average dollar is only invested for about half that time — the first contribution gets 60 months of work, the last gets one. So part of any DCA-versus-lump gap is simply "the money was in the market for less time", and comparing against a lump sum held for the full 60 months flatters the lump sum.
The fair benchmark is a lump sum held for the DCA programme's average exposure period, which is 30 months. Same money, same expected time in the market.
| Benchmark | Mean result | DCA win share | Bootstrap 95% CI on the gap |
|---|---|---|---|
| Lump sum held 60 months (full exposure) | +60.6% | 27.9% | -48.66 to -15.23 pp |
| Lump sum held 30 months (average DCA exposure) | +25.8% | 59.1% | -18.85 to +21.87 pp |
| DCA programme | +29.8% | — | — |
Once exposure is equalised, the picture changes: DCA ends +3.99 pp ahead of the half-period lump sum on average, and wins 59.1% of windows. The bootstrap interval on that difference (-18.85 to +21.87 pp) straddles zero, so there is no measurable edge either way.
In other words: the headline DCA deficit is largely the price of sitting in cash for the second half of the programme, not a penalty for bad entry timing. Once you charge both programmes the same exposure, they are indistinguishable.
Money-weighted IRR: the metric a saver actually feels
Terminal profit ignores when the cash was paid, which is the single most important fact about a DCA programme. A monthly saver typically spends most of their money near the end of the programme, so most of it is exposed for only a few months. Money-weighted IRR charges the programme for that.
On the metric that matters to someone actually saving, the two are statistically indistinguishable: a mean difference of +0.50 percentage points per year with a p-value of 0.769 and a bootstrap interval of -7.87 to +8.29 pp. DCA is slightly ahead on average (61.1% of windows) but the effect is indistinguishable from noise.
The IRR sample starts in 1968 and excludes the 1960-1967 peg era, where the price was frozen at $35 and any "return" was the peg being removed rather than an investment outcome. The 5-year and 10-year IRA/Roth analogy is worth stating plainly: this page's numbers describe accumulation into a pile, not a periodic-withdrawal retirement plan.
How long do you wait to be in profit?
This is the question DCA is actually sold on, and it has a real answer. A month counts as "underwater" when the portfolio is worth less than the cash contributed so far — the number on the statement is below what you have paid in.
| Time spent underwater | Windows | Share |
|---|---|---|
| never underwater | 48 | 6.5% |
| 1-11 months | 315 | 42.4% |
| 12-23 months | 199 | 26.8% |
| 2-4 years | 164 | 22.1% |
| 4-6 years | 17 | 2.3% |
| 6+ years | 0 | 0.0% |
The median DCA investor waits 12 months before the portfolio is worth more than the cash paid in; the mean is 15.6 months. Only 6.5% of windows never go underwater at all, and 24.4% spend two years or more in the red.
For comparison, the lump sum's own record is harsher. The longest run of months in which gold traded below its prior all-time high was 315 months — from the 1980-01 peak of $675/oz. That episode reached a maximum drawdown of -62.1%, bottoming at $256/oz in 1999-07.
Note the difference between two numbers that are easy to conflate: the price at the end of that streak was $611/oz (-9.48% below the peak), because the market had already recovered most of the way by then. The depth that actually had to be endured was -62.1%.
Stress test: what DCA does in the bad decades
The average hides the tail. Here is what happens when the monthly programme starts at each major gold peak, over the following 60 months.
| Peak | Gold's decline | DCA underwater | DCA worst paper loss | Lump sum worst | End result: DCA vs lump |
|---|---|---|---|---|---|
| 1974-12 | -40.2% over 39m | 24 months | -24.3% | -40.2% | +22.3 pp |
| 1980-01 | -62.1% over 316m | 25 months | -33.3% | -53.3% | +27.6 pp |
| 2008-03 | -21.4% over 18m | 6 months | -12.1% | -21.4% | -32.5 pp |
| 2011-09 | -39.3% over 106m | 44 months | -21.4% | -39.3% | +23.6 pp |
The 1980 peak is the case that decides the argument. Gold fell -62.1% over 316 months and the lump sum's paper loss reached -53.3%. The monthly programme was underwater for 25 months with a worst paper loss of -33.3% — a much smaller hole — and still ended +27.6 pp ahead of the lump sum, because it bought more units at $256/oz.
But note what it did not do: it did not turn the loss into a profit. Over the 60 months from that peak, DCA still ended -25.0%. It made a bad outcome less bad, which is a different claim from making it good.
The 2008 peak is the mirror image and the strongest argument against DCA as a return strategy. Gold fell -21.4% in 18 months, DCA was underwater for only 6 months — and then it gave all of that back, ending -32.5 pp behind the lump sum. The safety during the fall did not translate into a better outcome, because the recovery was the whole prize and the lump sum was fully there for it.
The 2011 peak sits between the two: a -39.3% decline over 106 months, DCA underwater for 44 months, and a +23.6 pp advantage at the 60-month mark.
Does the answer depend on how long you keep the money in?
Yes — and the size of the gap grows sharply with the horizon, though the win rate does not move monotonically.
| Horizon | Windows | DCA win share | Mean gap | Median gap |
|---|---|---|---|---|
| 12 months | 791 | 45.1% | -4.70 pp | -1.15 pp |
| 24 months | 779 | 43.3% | -11.32 pp | -2.38 pp |
| 36 months | 767 | 40.7% | -18.21 pp | -5.25 pp |
| 60 months | 743 | 27.9% | -30.85 pp | -10.87 pp |
| 120 months | 683 | 30.2% | -102.78 pp | -40.52 pp |
At one year the two are nearly even (45.1% DCA win share, mean gap -4.70 pp). At ten years the lump sum is ahead by 102.78 pp on average. Note that the win rate does not fall in step: at ten years DCA still wins 30.2% of windows, slightly more than its 27.9% at five years. What grows with the horizon is the size of the losses in the windows where DCA loses, not the frequency with which it loses.
This is not a finding about DCA being bad — it is the same exposure arithmetic as before, amplified. More time in the market beats less time in the market, and a longer horizon means the difference between "in from month one" and "in from month 60" compounds for longer. A lump sum has an extra 60 months of average exposure at the ten-year horizon, against 30 months at five years.
Decade by decade
The same comparison, grouped by the decade the programme started in. The sign flips constantly, which is the point: there is no regime where DCA reliably wins.
| Start decade | Starts | DCA win share | Mean DCA | Mean lump sum | Mean gap | Median underwater |
|---|---|---|---|---|---|---|
| 1960-1969 | 120 | 0.0% | +37.8% | +52.3% | -14.45 pp | 15.0 months |
| 1970-1979 | 120 | 7.5% | +71.6% | +175.8% | -104.25 pp | 5.0 months |
| 1980-1989 | 120 | 68.3% | -0.7% | -5.6% | +4.97 pp | 28.0 months |
| 1990-1999 | 120 | 62.5% | +0.7% | -3.6% | +4.27 pp | 22.5 months |
| 2000-2009 | 120 | 0.0% | +48.3% | +113.4% | -65.14 pp | 2.0 months |
| 2010-2019 | 120 | 34.2% | +11.3% | +20.9% | -9.52 pp | 11.0 months |
| 2020-2029 | 23 | 0.0% | +80.5% | +116.5% | -36.06 pp | 6.0 months |
DCA won more often in the 1980s (68.3%) and 1990s (62.5%) — the two decades gold went nowhere, so buying slowly beat buying once. It won 0.0% of the time in the 1960s and 0.0% in the 2000s, and never in the 2020s so far. The pattern is not "DCA is good in some eras" — it is "DCA is good when the market is flat and bad when the market trends".
The 1970s row is the largest gap in the table (-104.25 pp) simply because gold rose enormously from a fixed $35 base: a lump sum in 1974 captured a doubling that a monthly programme spread across a rising market.
The peg era, separated out
The official gold price was fixed at $35/oz from 1960-01 to 1967-12 in this series. A DCA programme starting in 1961 would have bought $35 gold for most of its life, then watched the price quadruple in 1974-75. That is not a strategy, it is an artefact of the price being administratively frozen, so it is worth separating from the real comparison.
The free-market-era figure (-34.77 pp) is close to the full-sample headline, so the peg is not doing the work here. In the free-market era, DCA's mean result was 32.9% against a lump sum's 67.7%, and DCA won 32.0% of windows.
Where this leaves the retail recommendation
The commonly repeated version — "buy a fixed amount every month and you will do better than someone who puts it all in at once" — is not supported by 66 years of monthly gold prices. On terminal profit the lump sum wins 67.0% of windows; on money-weighted IRR the two are indistinguishable; and once exposure duration is equalised, so is the terminal-profit comparison.
What survives the test is narrower and more useful:
- DCA reliably reduces the size of a paper loss. Its worst paper loss was -33.3% against the lump sum's -53.3% in the worst episode in the sample.
- DCA does not make bad outcomes good. From the 1980 peak it still ended -25.0% over the 60-month horizon.
- The cost of that insurance is giving up the compounding a lump sum gets from earlier entry — worth 102.78 pp on average over ten years, in the windows where DCA loses.
- A median wait of 12 months before the statement is above the money paid in is a real operational fact for anyone with recurring income, and it is worth checking against your own cash flow.
If the goal is accumulation of a fixed sum over a fixed period, the arithmetic favours investing earlier and holding longer, and the data here shows how much of the popular recommendation is really a behavioural suggestion dressed as a return claim.
What this page is not
It measures monthly average prices, so a "contribution" is modelled as filled at that month's average. A real order fills at some intraday price, and the DCA side would be hurt slightly more than the lump-sum side by anything it pays in spread. No transaction cost, financing, storage or tax is included anywhere, so the DCA result is if anything flattered. These are gold price returns in dollars, with no currency effect for a non-dollar holder.
This is also not a study of periodic-withdrawal retirement plans. It compares two ways of accumulating a fixed sum. A withdrawal strategy has a completely different cash-flow profile and would need its own analysis.
Nothing on this page is investment advice, and past results do not predict future returns. Gold prices are volatile and a monthly programme can be underwater for years.
Current context
Gold was $4,105/oz on average in 2026-10, +1.2% over the trailing twelve months and +131.0% over five years. A DCA programme started today would inherit those numbers — and, given that the last decade is one in which DCA won 0.0% of start months, it would not be a flattering way in.
Frequently asked questions
Does dollar-cost averaging beat investing the same cash at once in gold?
No, not on this sample. Across 743 rolling 60-month windows from 1960-01 to 2026-10, lump sum beat DCA in 67% of windows and DCA won 27.9%, with 38 ties. Mean terminal profit was 60.6% for lump sum against 29.8% for DCA, a gap of -30.85 pp. The gap widens with horizon: mean DCA-minus-lump difference is -4.7 pp at 12 months and -102.78 pp at 120 months.
Is the DCA deficit just an artefact of being in the market for less time?
Partly, but not entirely. The fair comparison is a lump sum held for the average DCA exposure period, which is half the horizon. Against that benchmark DCA is +3.99 pp on average and wins 59.1% of windows, so exposure timing explains part of the gap. But against a full lump sum DCA is still -30.85 pp and wins only 27.9% of windows, with a 95% bootstrap interval of -48.7 to -15.2 percentage points that excludes zero. The remaining gap is the sequence of purchase prices, not just the time in market.
Which market regime did DCA work best in?
Declining markets. DCA won 83.9% of the 180 windows that ended lower, against 9.9% of the 563 windows that ended higher. The mean gap in rising windows is far more negative. That is the well-known behaviour of averaging into a falling price, and it is also exactly the condition under which the average purchase price is lowest.
Does DCA produce a higher return per unit of risk than lump sum?
On an IRR basis the two are close: across 108 sample programmes the mean annualised return was 8.8% for DCA against 8.3% for lump sum, a difference of +0.5 pp. DCA had the higher IRR in 61.1% of samples, but the bootstrap interval runs from -7.9 to 8.3 percentage points and the naive test gives p = 0.769. There is no measurable IRR advantage; the return gap comes from the deployment schedule, not from a better rate of compounding.
How much of this is distorted by the fixed $35/oz gold price?
A real amount, which is why the free-market figures are published separately. The official price was fixed at $35/oz from 1960-01 up to 1967-12 in this series, so 96 of the 743 headline windows start inside that stretch and largely buy a frozen price. Restricting to the 647 free-market windows, the mean DCA-minus-lump gap is -34.77 pp and DCA wins 32% of them, with mean DCA profit 32.9% against 67.7% for lump sum. The conclusion strengthens once the peg is removed.
How reliable is the statistical significance of the DCA deficit?
Less than the headline t-statistic suggests. Consecutive 60-month windows overlap by 59 months, so the naive t-statistic of -8.1 (p = 1.39e-15) is inflated by the overlap. On the non-overlapping subsample of 13 windows the difference is -26.37 pp with t = -0.6 and p = 0.525, and the non-overlapping bootstrap interval runs from -72.2 to 26.5 percentage points, which includes zero. Read the interval, not the point estimate: the deficit is large and consistent in direction, but the overlap-corrected interval is wide.
What does starting a DCA programme now inherit?
The starting conditions, not a special advantage. Gold was 4,105 per ounce in 2026-10, up 1.2% over 12 months and 131% over 60 months. A programme started at that level inherits those prices, and this page's own results show the arithmetic of averaging into a level that has already risen sharply is unfavourable relative to the same cash deployed at once.
Sources
- World Bank Commodity Price Data (The Pink Sheet) — Monthly Prices sheet, gold column, monthly averages in USD per troy ounce. Data updated November 03, 2026.
-
scripts/gold_dca_analysis.py— production script, writessrc/data/research-gold-dca.jsonand both CSVs. -
scripts/verify_gold_dca.py— independent second implementation; recomputes every published statistic from the published CSVs with numpy and a different code path.
Last updated: 2026-11-09