ResearchMarketsNot sure
No, not as a rule. We tested 34 dip-buying rules on the S&P 500 since 1950, SPY since 1993 and QQQ since 1999. Only 5 earned more per year than holding the index, by 0.8 percentage points at best, and none passed the platform's multiple-testing gate. The rule we picked before looking was to buy SPY after a 10% fall from its high and hold for a year. It earned 8.8% a year against 10.7% for holding. With a pile of cash, investing it all at once beat spreading it over 12 months in 70.4% of 5-year periods. Waiting for a 20% dip meant no dip at all within 5 years in 30.7% of start months. And dips kept falling. Buyers at the first 10% drop in April 2000 and in November 2007 watched the index fall another 42.7% and 51.9%.
1. The idea
"Buying after the market drops beats buying and holding." The logic sounds airtight. A market 10% below its high is 10% cheaper, so money put to work then should earn more than money put to work at a record. People run the idea three ways. Some sit on a pile of cash until the market falls. Some save every month but invest only on red days. Others trade in and out on signals like a run of down days or a VIX spike. We tested all three against what they give up. That means investing the money at once, investing a fixed amount every month (dollar-cost averaging), and the T-bill interest earned while waiting.
2. What the evidence says
On the platform, Lyra Data Engine collects this layer from news, filings and macro releases. For this post we gathered the sources by hand, read each one, and checked every link.
- Lump sum usually beats spreading it out. Vanguard's Cost averaging, invest now or temporarily hold your cash? (2023) finds that lump-sum investing beat cost averaging 68% of the time over 12-month horizons in global stocks from 1976–2022.
- Even perfect dip-buying struggled against monthly investing. Nick Maggiulli's Even God Couldn't Beat Dollar-Cost Averaging (Of Dollars And Data, 2019) saves cash and invests it only at the exact bottom between all-time highs, over 40-year windows that start in 1920–1979. That version "underperforms DCA over 70% of the time". Missing each bottom by two months loses 97% of the time.
- Waiting costs more than perfect timing earns. Schwab's Does Market Timing Work? follows $2,000 a year for 20 years (2005–2024). Investing at each year's lowest close ended at $186,077. Investing on the first day of each year ended at $170,555, only $15,522 behind. Monthly investing ended at $166,591.
- Dip rules do not beat holding. AQR's Hold the Dip (2025) builds 196 buy-the-dip variants on the S&P 500 from 1965 to September 30, 2025. On average their Sharpe ratio was 0.04 below holding the index, about a 16% degradation, and over 60% of them did worse.
- Returns after dips are not special. Javier Estrada's Buy the Dip? Not So Fast (IESE, 2026) uses US data from 1871–2025. One year after a negative month, stocks returned 9.1%, against 11.0% after the average month.
Every source points the same way. Three of them come from fund firms that benefit when investors stay invested, which is a reason to test the claim rather than take it on trust. The tests below are ours.
3. Data
| Series | Source | Range used | Use |
|---|---|---|---|
| S&P 500 index close | Yahoo Finance | 1950-01-03 to 2026-11-17 | dip signals |
| US stock market total return | Ken French Data Library | 1950-01-03 to 2026-09-30, 19,308 days | returns in the tests since 1950 |
| SPY and QQQ adjusted closes | Yahoo Finance | 1993-02-01 and 1999-03-11 to 2026-11-16 | returns with dividends |
| SPY and QQQ adjusted closes, cross-check | Alpaca Market Data | 2016-01-04 to 2026-11-16 | checking the above |
| Nasdaq-100 index close | Yahoo Finance | from 1985 | dip signals for QQQ |
| VIX close | FRED | from 1990 | VIX signals |
| 3-month T-bill rate | FRED | from 1954 | interest on cash while waiting |
Cleaning and checks.
- Free daily S&P 500 data with dividends does not reach back to 1950. So the tests since 1950 read signals from the S&P 500 price index, which is what investors watch. They earn the total return of the whole US market from the French library, which is close to what an index fund pays with dividends. On 19,308 common days the two daily return series correlate at 0.9863. The market's total return compounded at 11.5% a year against 8.3% for the S&P 500 price index. The gap is mostly dividends.
- Cash earns the 3-month T-bill rate from 1954 and the French library's T-bill return before that.
- Yahoo and Alpaca agree on SPY and QQQ. On 2,734 common days the daily-return correlation is 0.9961 for SPY, with 44 days more than 10 bp apart, mostly around dividends. For QQQ it is 1.0, with 2 such days.
- "The high" is the highest close since 1950, or since 1985 for the Nasdaq-100. Every signal uses only closes up to that day, and the trade happens at that day's close.
- The analysis stops at 2026-11-17. The French library ends at 2026-09-30.
4. Method
We fixed the rules and the trial count before running anything.
Dip triggers. A close at least 5%, 10% or 20% below the high. Three or five down closes in a row. A VIX close at or above 30.
Trading rules. Sit in T-bills. On a trigger day, buy the index at the close and hold it for 63 or 252 trading days, about three months or a year. Each new trigger extends the hold. Costs are 5 bp per switch. Six triggers times two holds makes 12 rules per market. VIX data starts in 1990, so the market since 1950 has 10. That is 34 trials in all. The one we graded was named in advance. It buys SPY 10% below the high and holds for 252 days. The other 33 count against it.
The test is the gap to buy-and-hold. The idea claims to beat holding, so the platform grades the daily return of the rule minus the daily return of the index, not the rule's own return.
Two cash-flow tests, all reported.
- One pile of cash. On the first trading day of every month since 1950, an investor has $1 and five years. One choice invests it all at once. One invests a twelfth each month for 12 months. Three hold T-bills until the S&P 500 closes 5%, 10% or 20% below its highest close since the start, then invest it all. If the dip never comes, the money stays in T-bills.
- Saving every month. $1 arrives on the first trading day of every month. The monthly investor buys at once. The dip saver keeps it in T-bills and invests everything waiting on any day the S&P 500 is at least 5%, 10% or 20% below its high. For reference we add a hindsight saver that invests only at the exact lowest close of each dip of 10% or 20%. Nobody can trade that one.
Did returns after dips beat other days? We take the first close past −5%, −10% and −20% in each fall and compare the next 30 and 365 days with random days. That makes 18 event tests across three depths, two horizons and three markets.
Grading uses Lyra's own code. The pipeline imports the platform's check modules unchanged.
- Past cases. The gap series is cut into 30-day blocks. A block is a win if the rule beat holding. The idea agrees when more than half the blocks win and the average block is positive.
- Sharpe gate. The deflated Sharpe ratio must clear 0.95 against a bar that rises with the trial count (34 here) and with the spread of the trials' results. The probability of backtest overfitting (PBO) must be under 0.5.
- CPCV. History is cut into 10 blocks, and every pair takes a turn as the test set, up to 45 paths. Days within 395 of a test block are purged, which covers the 30-day horizon plus the one-year hold, and the platform embargoes more. More than 50% of paths must be positive, and the 5th-percentile path Sharpe must be above −0.5.
- Event placebo. Each event test draws 2,000 sets of random days and asks how often they beat the dip days. It must pass at p ≤ 0.0028, which is 0.05 split across the 18 tests.
- Sealed holdout. The year 2025-10-01 to 2026-09-30 stays out of every test above.
- Grade. A single hand-picked ticker can reach Good Idea at most. Good Idea needs the past cases to agree. Awesome Idea also needs the gate, the holdout and CPCV.
5. Results
Dips are common and records are rare
Since 1950 the S&P 500 has fallen 10% from a record 26 times, about 3.4 times a decade. It has fallen 20% 11 times. It closed at a record on only 8.0% of days and sat at least 10% below one on 35.2% of days. A dip buyer is rarely short of dips. The question is whether money that waits for one ends up ahead of money that never waited.
Returns after a dip were no better than on other days
The average year after a 5% fall returned 12.3%, the same as any year (12.6%). After a 10% fall it returned 8.4%, which is lower. Only falls of 20% or more were followed by more, 20.0% on average across 11 cases. Two of those 11 were followed by a year of heavy losses, and the placebo puts the gap at p = 0.061, short of the 0.0028 bar. The 30-day view shows the clearest bounce. After a 5% fall the next 30 days returned 2.1% against 1.0% on all days (p = 0.01). That also falls short once 18 tests are counted, and a month is too short for a buy-and-hold investor to cash in.
The dips that kept falling
Buying at −10% meant a median further fall of 9.9%. In 46.2% of cases the index fell at least another 10%, and in 23.1% at least another 20%. On 2000-04-14 the first −10% close was 1,356.56. The index then fell another 42.7% and did not close above that price again until 2006-10-12, 6.5 years later. On 2007-11-26 the buy price was 1,407.22. The index fell another 51.9% to 2009-03-09 and took 4.3 years to come back. The Nasdaq-100 was worse. Its first −10% close of the 2000 crash came on 2000-04-03. Another 80.3% fall followed, and the index took 14.4 years to get back to that price.
One pile of cash, invested now, spread out, or held for a dip
Investing at once beat 12 monthly installments in 70.4% of the 861 start months, close to Vanguard's 68%. Spreading it out cost a median 3.9%. Over 10 years the share was 69.2%.
Waiting for a dip did no better. Waiting for a 5% pullback beat investing at once in 55.6% of starts, by a median 1.1%. That is a coin flip with small stakes. A tenth of starts lost 11.5% or more and a tenth gained 5.8% or more. Waiting for 10% won 46.1% of the time, with a median gap of −1.7% and a worst tenth of −36.0% or less. Waiting for 20% won 43.3% of the time, with a median gap of −5.4%.
The dip often came at a higher price than the day the wait began. When the 10% dip came, the price was above the start 54.4% of the time. The market rose first, then fell 10% from the new high.
The median wait was 9.9 months for a 10% dip and 25.9 months for a 20% one. In 30.7% of start months the 20% dip never came within 5 years. Those savers stayed in T-bills and ended a median 44.7% behind the investor who bought on day one.
On QQQ the same test runs from 1999-03-11. Investing at once beat monthly installments 72.9% of the time. Waiting for 10% won 49.8% of the time. Waiting for 20% won 35.9% of the time and never came within 5 years in 24.2% of starts.
Saving every month, the dip saver's cash sits idle for years
The cost of waiting shows up as time in cash. With a 10% rule the average deposit waited 9.1 months before it was invested, and 25.7% of deposits waited more than a year. With a 20% rule the average wait was 31.2 months, and 62.3% of deposits waited over a year. The S&P 500 went 6.8 years without a 10% dip from 1991 to 1997, and 12.3 years without a 20% dip from 1988 to 2001. By March 2001 the 20% saver had 148 months of deposits sitting in T-bills.
Every rule a real investor could follow lost to monthly investing more often than it won. The 10% saver came out ahead in 36.0% of 10-year windows and 34.8% of 20-year windows. Over the whole stretch from 1950 to September 2026 it ended 4.0% behind the monthly investor. The 5% saver ended 1.1% behind and the 20% saver 18.6% behind.
The hindsight saver did beat monthly investing in our data, in 71.9% of 10-year windows for 10% dips. That is the opposite of Maggiulli's result, and the definitions explain it. We let hindsight pick only the bottoms of dips of 10% or more in daily data. His version buys at the bottom between every pair of all-time highs over 40-year windows. Both need to know where the bottom is. The same hindsight saver on QQQ ended 16.5% behind monthly investing. After the 2002 bottom the Nasdaq-100 did not set a new record until 2015, and the cash waited for the next bottom.
Trading rules, 34 variants and none clearly better than holding
Of the 34 rules, 25 earned less than holding and 4 tied because they were invested almost all the time. The other 5 earned more, by 0.8 percentage points a year at most. The graded rule on SPY was invested 76.3% of the time and earned 8.8% a year against 10.7%, with a Sharpe ratio of 0.57 against 0.63. On the S&P 500 since 1951 the same rule earned 10.4% against 11.3%. Waiting for the VIX to pass 30 cost 4.1 points a year on SPY.
Deep-dip rules have one real effect. They shrink the worst loss. Holding for a year after a 20% fall cut the S&P 500's maximum drawdown from −54.6% to −47.1%, and 15 of the 34 rules had a shallower worst loss than holding. On the S&P 500 and SPY the 20% rules paid for it with 2.9 and 4.8 points a year, because they sat in cash through much of each rally. On QQQ the same rule came out 0.8 points ahead, one of the 5 winners, inside a test that barely separates the rules from holding, as section 6 explains.
6. Validation
All of this comes from the platform's check code, run on the main period up to 2025-09-30.
| Test | SPY, graded | S&P 500 since 1951 | QQQ | Bar |
|---|---|---|---|---|
| Past cases, 30-day blocks won | 37 of 385, mean −0.16% | 108 of 909, mean −0.08% | 2 of 311, mean −0.04% | more than half, mean above 0 |
| Blocks where the rule differed from holding | 37 won of 97 (38.1%) | 108 of 255 (42.4%) | 2 of 8 | |
| Sharpe of the gap, annualised | −0.29 | −0.18 | −0.2 | about 0.5 at trial 34 |
| Deflated Sharpe | 0.0 | 0.0 | 0.001 | 0.95 |
| PBO across the market's rules | 0.071 | 0.2 | 0.786 | below 0.5 |
| CPCV paths positive, p5 | 17.1%, −0.98 (35 paths) | 31.8%, −0.63 (44 paths) | 0.0%, −0.86 (9 paths) | over 50%, above −0.5 |
| Grade | Not sure | Not sure | Not sure |
Ties. When the rule is fully invested it is buy-and-hold, and the block is a tie that cannot count as a win. That is most blocks, 288 of 385 on SPY. Counting only the 97 blocks where the rule was partly in cash, it beat holding in just 38.1% of them. QQQ is almost all ties. The Nasdaq-100 spent 2000 to 2015 below its 2000 high, so a rule that buys 10% below the high was invested 97.6% of the time and the test has little to grade.
Multiple testing. We tried 34 rules and graded one, so the bar is set at trial 34. The deflated Sharpe is 0.0 against 0.95. The graded rule's gap to holding is negative, so it fails before any deflation. The low PBO on SPY (0.071) only says the rankings were stable, and the rankings are of rules that lose.
Placebo. We kept the rule's exact time in the market and shifted its timing by a random offset of at least a year, 1,000 times. On SPY the real rule's Sharpe of 0.56 sits at the median of the shifted copies (0.57), with p = 0.564, and below holding at 0.63. Buying after dips did no better than buying at random times for the same total time in the market. The S&P 500 since 1951 gives p = 0.429 and QQQ gives p = 0.707.
Event placebo. None of the 18 event tests passed. The closest were the 30-day bounce after a 5% fall on the S&P 500 (p = 0.01) and the year after a 20% fall (p = 0.061). Seven of the 18 had fewer than the 10 separate falls the platform needs, all on SPY or QQQ, and were not graded.
Other splits. Walk-forward on SPY, with 30 windows and a 30-day purge, had a median Sharpe of −1.27 and 36.7% of windows positive. Purged 5-fold left 3 usable folds on SPY, with a median of −0.43.
Holdout. The platform keeps the sealed year shut when the gate fails. We opened it for this post. The graded SPY rule won 2 of 12 blocks and ended 6.4% behind holding. The April 2025 fall kept it invested until 2026-04-28, and it sat in cash for the last five months of the holdout.
7. Verdict and limits
The grade is Not sure. Past cases do not back the idea on any of the three markets.
- Buying after a fall did not beat buying and holding. The graded rule lost 1.9 points a year on SPY, 25 of 34 variants lost, and the 5 winners won by less than a point.
- Returns after dips were no better than returns on other days. Falls of 20% or more may be an exception, but they come about 1.4 times a decade and are too few to test.
- Waiting has a known cost and an unknown payoff. Investing a pile of cash at once beat spreading it out about 70% of the time, and a 20% dip failed to show up within 5 years about 30% of the time.
- What the rules did do is cut the worst drawdown, at the price of return. That is a risk preference, not an edge.
Limits.
- Before 1993 the returns are the whole US market from the French library, not the S&P 500. The two correlate at 0.9863 a day, and the signals come from the S&P 500 itself.
- Trading at the close of the signal day is a little generous. Real orders fill near the close, not at it.
- We did not model taxes. Taxes would hurt the trading rules more than holding, since the rules sell.
- The QQQ test barely separates the rule from holding, for the reason given in section 6.
- We tested common retail definitions of a dip. Rules that combine dips with valuation or trend are separate ideas and would count as new trials.
- Our hindsight saver differs from Maggiulli's because the definitions differ. We report both, and neither can be traded.
FAQ
Does buying the dip actually work? Not as a rule that beats holding. Across 34 dip rules on the S&P 500, SPY and QQQ, 25 earned less than buy-and-hold, and none passed a multiple-testing check. A dip is a fine day to invest money you already meant to invest. Waiting for one with money you could invest today has cost more than it saved.
Is buying the dip better than dollar-cost averaging? Mostly not. A saver who invested only when the S&P 500 was 10% below its high beat a plain monthly investor in 36.0% of 10-year windows since 1950, and in 34.8% of 20-year windows.
Should I invest a lump sum now or wait for a dip? History favours investing now. Since 1950 a lump sum beat waiting for a 10% dip in more than half of 5-year periods, and beat 12 monthly installments 70.4% of the time. Waiting for a 20% dip lost a median 5.4%, and in 30.7% of start months the dip never came within 5 years.
How often does the S&P 500 drop 10% or 20%? Since 1950 the index has fallen 10% from a record close 26 times, about 3.4 times a decade. It has fallen 20% 11 times, about 1.4 times a decade. It closed at a record on 8.0% of trading days.
Sources
- Vanguard, Cost averaging, invest now or temporarily hold your cash? (2023)
- Of Dollars And Data, Even God Couldn't Beat Dollar-Cost Averaging (2019)
- Charles Schwab, Does Market Timing Work?
- AQR, Hold the Dip (Alternative Thinking, 2025)
- Javier Estrada, Buy the Dip? Not So Fast (IESE, 2026)
- Kenneth R. French Data Library
- FRED, CBOE Volatility Index (VIX)
- FRED, 3-Month Treasury Bill Secondary Market Rate
- Alpaca Market Data, historical stock bars
Related stocks
SPYQQQ
