Dollar-cost averaging vs lump sum — what the evidence says

The question a windfall forces

An inheritance arrives, a house sells, a bonus lands, stock options vest. You now hold cash you intend to invest for the long term, and one decision sits in front of you: put it all in the market today (a lump sum), or split it into equal pieces and invest them on a schedule (dollar-cost averaging, or DCA). Both paths end fully invested. The only difference is how long part of the money waits in cash — which means the question is really how much time out of the market are you choosing, and what does that time cost or save you?

Most write-ups answer with one backtest window from one start date, which lets the author pick the story. The honest way to answer is to roll the comparison across every start year the data allows and report the distribution: how often each side wins, by how much in the middle, and what the tails look like.

What the rolling-window backtest shows

We ran the comparison on the annual S&P 500 total-return series (dividends reinvested) with undeployed DCA cash credited the 3-month T-bill — data from Aswath Damodaran's NYU Stern dataset, 1928–2025, asOf 2026-01. Each row rolls a $100,000 deployment across every start year with a full window, evaluated at the end of the deployment period (after that, both strategies hold the same portfolio and compound identically). These are nominal, pre-tax, no-fee figures; the calculator lets you toggle each of those layers.

Spread length	Start years (n)	Lump-sum win rate	Median Lump − DCA	p5 / p95 of the spread
2 years	97	67.0%	+$5,873	−$13,181 / +$19,328
3 years	96	74.0%	+$10,066	−$21,598 / +$37,272
5 years	94	73.4%	+$22,570	−$25,700 / +$66,572
10 years	89	80.9%	+$66,416	−$28,285 / +$208,282

Read the middle column first: investing at once came out ahead in roughly two out of three start years for a two-year spread (67.0%) and about three out of four for the default three-year spread (74.0%), rising toward 81% as the spread stretches to a decade. The longer you drip, the longer the average dollar sits out of the market, and the more often that costs you. But read the last column too — the p5 end of the spread is negative in every row. In the worst twentieth of start years, DCA finished five figures ahead on a $100,000 deployment. Both facts are true at once; a projection that shows only one of them is marketing.

These figures cross-check against the published research. Vanguard's 2012 paper "Dollar-cost averaging just means taking risk later" (Shtekhman, Tasopoulos & Wimmer) found lump-sum investing ahead of a 12-month DCA roughly two-thirds of the time across the US, UK, and Australia on 1926–2011 rolling windows. Vanguard's February 2023 update, "Cost averaging: Invest now or temporarily hold your cash?", put the one-year figure at 68% on MSCI World data for 1976–2022. Our annual-granularity backtest lands in the same band from an independent dataset, which is what you would hope: this is a robust historical regularity, not an artifact of one study's choices.

Why the odds favor investing at once

There is no mystery here. Equities have historically carried a positive expected return over cash — that premium is the entire reason to own them. A DCA schedule holds part of the capital in cash for months or years, so on average it forfeits part of that premium. Vanguard's 2012 title says it precisely: dollar-cost averaging just means taking the risk later. It is not a return-enhancement technique; it is a schedule for delaying equity exposure, and delay has an expected cost whenever the asset you are delaying into has a positive expected excess return.

The magnitudes are not symmetric either. In our three-year run, when the lump sum won it won by an average of about 17% of the deployed capital; when DCA won, its average margin was about 13%. So the lump sum wins more often and by somewhat more when it wins — in the historical record, on this dataset. That is the projection's center of gravity.

What DCA actually buys: downside insurance

If DCA loses most of the time, why does anyone use it? Because the start years where it wins are exactly the ones people fear. Here are the crash-adjacent three-year windows from the same backtest, $100,000 deployed, nominal and pre-tax:

Window	Lump sum	DCA (3 tranches)	Outcome
1929–1931	$38,562	$47,505	DCA ahead by $8,943
1973–1975	$86,990	$117,937	DCA ahead by $30,947
2000–2002	$62,572	$73,691	DCA ahead by $11,119
2008–2010	$91,751	$118,327	DCA ahead by $26,576
2020–2022	$124,268	$104,076	Lump sum ahead by $20,192
2022–2024	$129,024	$141,320	DCA ahead by $12,295

Starting into 1929, 1973, 2000, 2008, or 2022, the drip schedule softened the damage — sometimes by a quarter of the capital. Notice two honest details, though. First, DCA softens; it does not rescue. The worst DCA start year in the whole sample (1929) still ended at $47,505 on $100,000 — better than the lump sum's $38,562, but a deep loss either way. Second, 2020 is the counterexample people forget: the crash came and went inside the calendar year, the market finished 2020 up 18.0% on a total-return basis, and the lump sum beat the drip by $20,192 despite starting into a pandemic. A crash you can see coming after the fact is not a crash you could schedule around in advance.

The insurance premium, priced

The cleanest way to think about DCA is as an insurance contract you write against your own start date. In the three-year run, the premium — the median amount you give up for choosing DCA — was about $10,066 per $100,000, roughly 10% of capital. What the premium bought: in 15 of 96 start years, DCA finished more than 10% of capital ahead, and it cut the depth of the worst outcomes. Whether that trade reads as expensive or cheap depends on your circumstances — the size of the windfall relative to your net worth, how close you are to needing the money, and honestly, how you behave when a new investment is immediately down 30%. The projection prices the premium; it cannot price your sleep.

One behavioral point belongs in any honest treatment: the numbers above assume you hold to the end of the window. A lump sum abandoned in the middle of a drawdown — sold after the fall, re-entered after the recovery — does worse than either line in the table. If a drip schedule is what keeps a plan intact through its first bear market, its value shows up in a column no backtest prints. That is an argument about you, not about markets, and only you have the data.

DCA-ing money you already have is market timing

Here is the sharp edge of the topic. The phrase "dollar-cost averaging" gets used for two different things, and only one of them is a strategy choice at all:

• Investing salary as it arrives — a 401(k) contribution every payday — is not DCA in any decision-relevant sense. The money did not exist earlier; each dollar is deployed as early as it can be. There is no lump-sum alternative to weigh.
• Holding a lump you already have in cash and dripping it in is an active bet. Spreading $100,000 over three years is an implicit forecast that near-term equity returns will be lower than cash — otherwise the delay has negative expected value. That forecast is market timing, whatever it is called.

A useful consistency test: if the same $100,000 were already invested, would you sell it all today just so you could drip it back in over three years? Almost no one says yes — yet holding a new lump in cash and DCA-ing it is the same portfolio position with the same expected cost. If the answer to one is obviously no, the other deserves the same scrutiny. None of this makes drip schedules irrational; it makes them a priced choice — the insurance framing above — rather than a free prudence upgrade.

What flips a naive comparison

Many DCA-vs-lump-sum comparisons floating around are built on shortcuts that quietly move the answer. Four to check before trusting any chart, each of them a toggle in our calculator:

• Price-only returns. Leaving out dividends understates the cost of sitting in cash, because the waiting money forfeits the dividend too. The backtest defaults to total return; switch dividends off to see the distortion.
• Idle cash earning 0%. This one cuts the other way — it makes DCA look worse than it is. Credit the waiting cash the T-bill and the three-year lump-sum win rate drops from 80.2% to 74.0%. An honest comparison credits the cash; run it with the credit off to see how much of DCA's bad press comes from this shortcut.
• Average purchase price. "DCA lowers your average cost per share" is arithmetic, not evidence — it compares share counts, not outcomes. The decision-relevant quantities are terminal value and money-weighted IRR, which is what the calculator reports.
• Nominal, pre-tax, pre-fee dollars. Inflation, account type, and fee drag change levels for both strategies; the real-dollar view and the taxable-account view keep the comparison honest about what you would actually keep.

Run it on your own numbers

Every figure in this guide is reproducible in the DCA vs Lump Sum calculator — same data, same engine, every input in the URL so you can share a scenario. Useful starting points:

• Default: $100k over 3 years, IRA, total return, T-bill credit — the headline 74.0% win-rate run.
• Short spread: 2 years — the two-in-three (67.0%) case closest to the published 12-month studies.
• Long spread: 10 years — watch the median cost of waiting grow to two-thirds of the capital.
• House-sale case: $500k, taxable, 25 bps fee, real dollars — a windfall with every honesty layer on.

Limitations worth keeping in view

This is a historical backtest, not a forecast. The sample is one country's equity market over 98 years — a market that happened to compound well; start years overlap, so the 96 windows are not 96 independent draws. Granularity is annual (tranches invest at the start of each year), which slightly coarsens results versus monthly schedules. The taxable layer applies long-term capital-gains tax at liquidation but does not yet model annual dividend tax. And nothing here weighs your liquidity needs, other assets, or risk capacity. The full method, sources, and every assumption are on the methodology page; all outputs are estimates presented as ranges because that is the only honest shape for them.