Dollar Cost Averaging vs Lump Sum vs a 200-Day Moving Average: 273 S&P 500 Start Dates

TL;DR: the answer box

• Lump sum usually wins. $120,000 invested on day one beat the same money spread over 12 months in 76.9% of 273 rolling 10-year windows, by a median 5.8%. Against a 120-month plan it won 90.5% of the time.
• 12-month DCA barely changes the decade's worst fall. It was more than 1 point shallower than lump sum's in only 8.8% of windows. Median worst fall: 47.5% for both.
• A 200-day switch halves the fall and costs money. Median worst fall 20.0%, and no window ended below $120,000. Plain lump sum still ended with more in 81.0% of windows.
• Full history, 1993 to 2026: lump sum grew 10.88% a year with a 55.2% worst fall. The switch grew 8.40% a year with a 26.8% worst fall.
• The smaller fall beat random timing. None of 100 random-timing runs with the same time in the market fell as little as the switch.

How I tested lump sum, DCA and the 200-day switch

You have $120,000. Do you put it all into the S&P 500 today, feed it in month by month, or add a rule that steps aside when the market falls? Most people only compare the first two. The third is the one almost nobody tests.

Three plain definitions first.

• Lump sum: all the money goes in on day one.
• Dollar cost averaging (DCA): the same amount goes in every month, whatever the price. I tested 12, 36 and 120 months.
• The 200-day switch: the 200-day moving average is the average closing price of the last 200 trading days. If SPY closes above it, you hold SPY. If it closes below, you hold cash.

If your money arrives with each paycheck, monthly investing is simply how you invest. If you already hold the lump, DCA means keeping part of it in cash for a while. That is the decision this study tests.

Eight rules, one pot of money. The four plain plans above, and the same four with the 200-day switch on top. With the switch, a monthly payment that arrives while SPY is below its average waits in cash.

The windows. Every month from December 1993 to August 2016 is a start date: 273 in all. Each one is judged 10 years later. The windows overlap, so they are not 273 separate tries. They cover about three separate decades.

The market. SPY, daily bars from 1993-01-29 to 2026-08-20 (8,447 bars), with dividends reinvested. The average is drawn on the price without dividends, the line your chart shows.

The fills: Execution rule v2. Every decision is made at a close and filled at the next open. Costs on every fill: $0.005 commission plus $0.01 slippage per share per side. Cash earns 0%, which is a little unfair to DCA and the switch, since both hold cash.

Two measures matter most. Ending wealth is what you have after 10 years, counting any money still waiting in cash. Worst fall (drawdown) is the biggest drop from a high point, waiting cash included.

Is it better to invest a lump sum or dollar cost average?

Lump sum. Usually, and not by a little.

The median gap was 5.8%. In dollars, the median lump sum ended at $269,872. The median 12-month plan ended at $256,038.

Spread it out longer and the gap grows. Lump sum beat a 36-month plan in 76.2% of windows, by a median 16.6%. It beat a 120-month plan in 90.5%, by 51.7%.

Vanguard's 2023 paper Cost averaging: Invest now or temporarily hold your cash? found lump sum ahead about two thirds of the time. On SPY from 1993, with costs and next-open fills, I measured 76.9% against a 12-month plan. Same direction, a bit bigger margin, on one market.

When does dollar cost averaging beat lump sum?

Mostly when the plan started in a bear-market year.

12-month DCA ended ahead in 63 of 273 start months. Of those, 45 started in 2000 to 2002 or 2007 to 2009, the years of the two big bear markets. The other 18 were spread across the rest.

That makes sense. When prices fall right after you start, money that is still waiting buys cheaper. DCA's best start was June 2008: $399,826 against $287,394 for lump sum, 39.1% more. Its worst was March 2009, the bottom: $422,905 against $572,713, 26.2% less.

The heatmap below shows every start year at once. Each cell is the average value after 10 years for starts in that year.

Lump sum led in 17 of 24 start years. 12-month DCA led in 3, lump sum with the switch in 1, and 12-month DCA with the switch in 3.

Look at 1999. A lump sum started that year averaged $100,406 after 10 years, less than the money put in. The 12-month plan with the switch averaged $150,407. Now look at 2009. Lump sum averaged $455,750. The same switch plan averaged $245,473.

I entered the markets in 1998, close to the dot-com peak. So this one is personal. $120,000 put in at the top in March 2000 was worth $114,105 ten years later. A 12-month plan did no better: $110,573. Lump sum with the switch ended at $176,785.

Does dollar cost averaging reduce risk?

For a 12-month plan, barely. In these 10-year SPY windows, 12-month DCA had almost the same worst fall as lump sum. Once the money is all in, you hold the same index through the same crash.

The 12-month plan's worst fall was more than 1 point shallower than lump sum's in only 8.8% of windows. The median difference was 0.01 points. In these decades, the big falls came after the first 12 months, when the DCA investor was already fully in.

Lump sum ended a decade below its $120,000 in 24 of 273 windows. The 12-month plan did so in 22. That is almost no difference.

The 120-month plan looks safer, with a median worst fall of 27.6%. But that is because about half its money sits in cash on average. It is not timing anything. You pay for that cushion: a median $204,102 after 10 years, against $269,872 for lump sum.

So for the 12-month plan, the skeptics are right. It did not protect you from a large fall. It delayed full investment by 12 months, and the big falls in these decades came after those 12 months were over.

Does a 200-day moving average make DCA safer?

Yes on the fall. No on the money.

Think of the switch as an umbrella. It keeps you dry in the storm. You also carry it on a lot of sunny days, and on some of them it blows inside out.

The storm part is real. No plan with the switch ended a decade below the $120,000 put in. The median longest stretch below an old high fell from 57.2 months to 28.6.

The sunny-day part is real too. Plain lump sum ended with more money than lump sum with the switch in 81.0% of windows, by a median 14.7%. With the switch, the 12-month plan beat its plain version in only 21.6% of windows.

Here is how the switch treats each month's payment. If SPY closes above its average on the day the money arrives, the money buys at the next open. If SPY closes below, the money waits.

Per point of drawdown, the switch scored higher in 80.6% of windows. Its median yearly return divided by its worst fall was 0.345, against 0.174 for lump sum. So the switch ended with less money in 81.0% of windows, but it earned more return for each point of pain in 80.6%.

Does the 200-day switch beat buy and hold?

Not on return. On drawdown, by a wide margin.

Here every rule runs as one account over the full history, December 1993 to August 2026.

Holding through grew 10.88% a year and turned $120,000 into $3,521,236. The switch grew 8.40% a year and ended at $1,678,079. What the switch bought was a worst fall of 26.8% instead of 55.2%. In dollars, the biggest drop was $521,879 for lump sum and $265,090 with the switch.

The switch made 112 trips into the market and stayed 55 trading days on average. It paid $11,517 in costs, against $69 for holding.

So a simple moving average did not match buy and hold on return in this test. It bought a smaller fall instead.

The paycheck investor sees the same trade. Put $1,000 in on the first trading day of every month since December 1993, and you paid in $393,000.

The plain plan ended at $3,282,327. With the switch, $1,878,814. Same trade as the lump sum: more money plain, half the fall with the switch.

How did the switch do in 2008 and March 2020?

It lost less in every crisis I measured. It made money in none of them.

With next-open fills and cash at 0%, the switch ended 2008 at -0.35% and March 2020 at -4.95%. Far better than holding through. Still a loss.

2022 is where it helped least: -16.54% against -18.18%. The market kept crossing the average, and every cross costs a little.

Here is what that crossing looks like.

The first exit filled on 28 February 2020. From 19 February to 23 March the switch fell 18.90% against 33.72% for holding through. Then SPY sat on its average, and the switch filled 10 times by 30 June. That back and forth, buying just above the line and selling just below it, is called whipsaw. It is the price of the umbrella.

Is the switch's smaller drawdown just luck?

Not against this test. Here is the obvious objection. Maybe any rule that sits in cash this often would fall this little. Then the switch is just less invested, not smart.

So I tested it directly. I took the switch's own in-market and out-of-market stretches and shuffled them into random order, 100 times. Same time in the market, 74.9%. Same number of trips, same costs and fills. Only the timing is random.

The typical random run fell 52.4%. The shallowest fell 33.7%. None of the 100 fell as little as the switch's 26.8%. On return divided by worst fall, 1 of 100 matched it.

On return alone, 33 of 100 random runs matched or beat the switch's 8.40%. So its return was ordinary for its time in the market. Its drawdown was not.

What about markets that never came back, like the Nikkei?

The Nikkei 225 closed at a record 38,915.87 on 29 December 1989. It first closed above that on 22 February 2024, at 39,098.68. That took 34 years and 2 months, per Nippon.com. So it did come back, very slowly.

I could not test DCA or the switch on the Nikkei here. The engine has no Nikkei price data. For the S&P 500's own long slumps, see S&P 500 drawdowns since 1870.

The honest limits

• One market (SPY), one average (200 days), one 10-year horizon, plus one full-history run. Nothing was tuned. Other settings could give other numbers.
• The 273 windows overlap. They are about three separate decades, not 273 separate tries.
• Cash earns 0%. Interest on cash would help DCA and the switch.
• Taxes are not counted. The switch made 112 trips into the market, and every trip out is a sale.
• Costs per share are charged on dividend-adjusted prices, which slightly overstates them in the early years.
• ES futures are not tested: a monthly payment cannot buy whole contracts.

What this means for you

1. Know the base rate before you wait. In these 273 SPY windows, investing at once beat a 12-month plan 76.9% of the time.
2. If you invest monthly because the money comes with your paycheck, keep going. That is not a timing choice. It is how money arrives.
3. Do not count on a 12-month plan as crash protection. The median decade's worst fall was 47.5% for 12-month DCA and lump sum alike.
4. If a 50% fall would make you sell, weigh the switch. It cut the median worst fall to 20.0%. It also ended with less money than plain lump sum in 81.0% of windows, and it whipsawed through 2022. If you build your own version, test whether its record survives reshuffling before you trust it. That is the job AlgoChef does.
5. Decide which number you care about before the crash, not during it. The 2014 Turtle Traders lesson that rebuilt my own trading was the same one: simple rules, and risk management you will actually follow.

The same idea across several asset classes, not just the S&P 500, is in the tactical asset allocation study.

Reference tables

Table 1. All eight rules, 273 rolling 10-year windows

Worst fall counts money still waiting in cash. Start dates December 1993 to August 2016, each judged 10 years later. Execution rule v2.

Table 2. Full history, one account each, 1993-12-01 to 2026-08-20

SPY with dividends, 32.72 years, next-open fills, costs included, cash at 0%.

Lump sum usually ended with more. The switch's median fall was less than half of lump sum's. Pick the one you can hold through the next crash.

Everything here is research, not investment advice. It is a backtest on historical data with the assumptions stated above. It is not a recommendation to buy or sell anything, and past results do not predict future ones.

I test one strategy everyone repeats, properly, and leave the parts that failed in. StatOasis.com/Overfit