A rule that holds cash and buys only after the US market has fallen 30% below an all-time high finished the past century with 84% of the money that steady monthly investing produced. Move the trigger to 10% and the gap nearly closes, at 97%. Split the record into 20-year windows instead, and that same 30% rule beat steady investing in 13.8% of them.
So the answer isn't "waiting never works". It's that the answer swings almost entirely on where the trigger sits and on what the cash earns while it waits. Both of those are choices made before any dip arrives.
How the test works
Two investors each put $100 to work on the first trading day of every month. The steady investor buys the market immediately, every month, without exception. The dip investor puts each $100 into one-month Treasury bills and moves the whole accumulated pile into equities on the first day the market closes at or below a set distance from its running all-time high. While the market stays under that line, new contributions go straight in as well. Once it recovers, the waiting resumes.
The price series is the Fama/French daily market return, which is the value-weighted return of every CRSP-listed US firm on the NYSE, AMEX and NASDAQ. Cash earns the daily one-month Treasury bill rate from the same file. The series runs from 1 July 1926 to 30 June 2026 and covers 26,274 trading days. Over that century the market compounded at 10.27% a year and the bill rate at 3.19%.
Three caveats travel with every figure below. There are no fees, taxes or dealing spreads, and the $100 is never adjusted for inflation. Drawdowns are measured against a total-return high rather than a price high, so a trigger fires slightly less often than a headline "market down 30%" would suggest. And the waiting cash earns interest, which is kinder to the dip rule than the way most people actually hold it. Ken French's file also notes that bill rates come from Ibbotson Associates through May 2024 and from an ICE BofA index after that.
What a century of data shows
Run both strategies across the full hundred years and the dip rule loses at every threshold tested. Shallow triggers barely lose, because they fire so often that cash is rarely idle for long. Deep triggers lose badly.
| Trigger depth | Final balance vs steady | Fresh episodes since 1926 | Longest wait between them |
|---|---|---|---|
| 5% | 99.2% | — | — |
| 10% | 97.3% | 31 | 16.9 years |
| 20% | 95.2% | 16 | 16.9 years |
| 30% | 83.9% | 7 | 40.5 years |
| 40% | 86.9% | 4 | 44.8 years |
| 50% | 85.9% | 2 | 78.0 years |
An episode counts once each time the market first breaches that depth after having regained an all-time high. On that basis the market went 30% or more below a high on seven occasions in a hundred years: October 1929, May 1970, July 1974, October 1987, March 2001, October 2008 and March 2020. Between the first two sits a gap of 40.5 years.
The 40% and 50% rows finish slightly ahead of the 30% row, which looks odd until you notice why. A rule that almost never fires spends most of its life holding bills, and bills at least earn something. The 30% rule fires just often enough to buy into several long declines and then ride them further down. This is the same mechanism that makes rebalancing into the 2008 and 2020 crashes uncomfortable in real time.
Twenty-year windows are the fairer test
One hundred-year run is a single path, and single paths flatter whichever strategy happened to suit them. Cutting the record into overlapping 20-year windows, one starting every month, gives 960 of them and a distribution rather than an anecdote.
| Trigger | Windows won by dip rule | Median vs steady | Worst window | Best window | Average cash weight |
|---|---|---|---|---|---|
| 5% | 24.1% | 99.7% | 95.8% | 100.8% | 0.6% |
| 10% | 18.1% | 97.9% | 73.2% | 101.9% | 4.0% |
| 20% | 36.2% | 98.2% | 71.8% | 113.6% | 8.4% |
| 30% | 13.8% | 85.2% | 24.2% | 117.7% | 33.7% |
Two things stand out. The win rate isn't monotonic. A 20% trigger won more windows than either a 10% or a 30% one, because 20% falls are common enough to keep the cash moving and deep enough to be worth the wait. And the payoff is badly skewed. At a 30% trigger the best window gained 17.7% and the worst lost 75.8%.
Horizon matters too. Over 1,080 ten-year windows the 20% trigger won 43.6% of the time. Over 840 thirty-year windows it won 38.7%. The 10% trigger goes the other way, falling from a 24.3% win rate at ten years to 8.7% at thirty, because small repeated drags compound.
Cash drag does most of the damage
The dip rule doesn't lose because it buys at bad prices. It buys at good ones. It loses because of the months and years when it owns nothing.
Across those 960 twenty-year windows, a 30% rule held some uninvested cash on 86.9% of trading days and averaged 33.7% of the portfolio in bills. A 20% rule held cash on 76.8% of days but only 8.4% of the portfolio, because it deploys often enough that the pile never gets large. Even the 5% rule, which fires constantly, sat on cash 42.1% of the time.
That idle share is expensive because of when markets rise. Just under half of all trading days since 1926 — 47.9% of them — occurred with the market within 5% of an all-time high. A rule that only buys after a large fall spends most of its life sitting out the conditions under which the market spends most of its life. That's a different problem from the one covered in the difference between drawdown and standard deviation as risk measures, and arguably a worse one, because you never see it on a statement.
When the dip never comes
The worst 20-year window for a 30% rule began in August 1944. Over the next two decades the market never once closed 30% below an all-time high. The dip investor's $24,000 of contributions sat in Treasury bills for the entire period and grew to $29,530. The steady investor finished with $121,874. That's 24.2%, and it came from a rule that was never wrong about anything. It simply never got a signal.
This isn't only a mid-century curiosity. An investor who started a 30% rule in April 2020 has now waited more than six years without a trigger. Contributions of $7,500 have grown to $8,508 in bills, against $12,884 for steady monthly buying. A 30% rule starting in January 2010 fired exactly once, in March 2020, and ended at 70% of the steady balance.
The waiting gaps are long by any standard. Fresh 20% drawdowns arrived at a median of 4.6 years apart, with a maximum gap of 16.9 years. Fresh 30% drawdowns arrived a median of 12.4 years apart. As of June 2026 it had been 4.1 years since the last 20% breach and 6.3 years since the last 30% one. Anyone reading how long past bear markets have taken to recover in nominal and real terms will recognise that these are not rare events so much as irregular ones.
The case for waiting
The strongest counter-argument is real, and the data supports it. Over a long horizon with a deep threshold, a dip rule can genuinely beat steady investing by a wide margin.
The best 20-year window for a 30% rule started in August 1962. The rule sat in bills through the rest of the decade, deployed in May 1970, deployed again in July 1974, and kept buying through the grind that followed. It finished with $56,875 against the steady investor's $48,324, or 17.7% ahead. It worked because those two decades contained two of the century's seven qualifying drawdowns and the market went almost nowhere in between. A 20% rule over the same window finished 10.3% ahead.
AQR's 2025 study "Hold the Dip" finds an echo of this. Across 196 dip-buying strategies on the S&P 500 from January 1965 to September 2025, the versions defined on a five-year dip length produced a higher average Sharpe ratio of 0.21 and an average alpha of 0.8%, though at a t-statistic of only about 1.0. The authors note the catch themselves: it's unlikely many investors would consider a five-year drawdown a dip.
What that upside costs is the tail. In 86.2% of 20-year windows the 30% rule lost, and at the tenth percentile it finished with 39.6% of the steady investor's balance. The trade is roughly a one-in-seven shot at a 17% gain against a one-in-ten risk of losing 60% of your outcome. Nobody chooses that trade explicitly, but a rule that waits for a deep fall chooses it by construction.
What the wider evidence says
The broader literature lands in the same place, from different angles. AQR's 196 dip strategies produced an average Sharpe ratio 0.04 below simply holding the index over 1965 to 2025, roughly a 16% degradation, with more than 60% of implementations underperforming. On the shorter 1989 to 2025 sample that includes dividends, the shortfall widened to 0.27, a 47% degradation. Only 16 of the 196 alphas cleared conventional significance, which is 8%, about what data-mining alone would produce. Those results are gross of fees and trading costs, and the long sample uses price returns only, both of which flatter the dip rules.
AQR's authors also flag, in a footnote, that whether a regular contributor should deviate from a fixed schedule to wait for dips is the subject of follow-on research they had not yet published. The test above is one answer to that question, on different data.
Asness, Ilmanen and Maloney reached a similar verdict on valuation-based timing in the Journal of Investment Management in 2017. From 1900 to 2015, a value-timing strategy earned a 7.4% excess return against 6.6% for buy-and-hold, but with volatility of 20.0% against 17.5%, so the Sharpe ratios were 0.37 and 0.38. From 1958 to 2015 the timing strategy simply earned less. None of the differences were statistically significant. Goyal and Welch, writing in the Review of Financial Studies in 2008, put it more bluntly still: the standard predictor variables "would not have helped an investor with access only to available information to profitably time the market".
Vanguard's 2023 study of cost averaging approaches the cash question from the other side. Splitting a windfall into three monthly tranches lost to investing it at once in 68% of rolling one-year periods on MSCI World data from 1976 to 2022, and US stocks beat cash 76% of the time over that span. That paper assumes no interest at all on the uninvested portion, which overstates the drag relative to the test here. It also answers a different question from this one, since a windfall is a one-off and monthly contributions are not.
Morningstar's Mind the Gap 2025 measures the cost of mistimed cash flows in the real world rather than in a backtest. The average dollar in US funds and ETFs earned 7.0% a year over the decade to 31 December 2024 against 8.2% for the funds themselves, a 1.2 percentage point gap across more than 25,000 funds. The gap widened to 1.8% a year for the quintile of funds with the most volatile cash flows and narrowed to 0.8% for the most stable. That study measures all timing decisions together rather than dip rules specifically, but it points the same way as our earlier breakdown of the behaviour gap and what bad timing costs.
What would change the conclusion
A higher cash rate would change the arithmetic directly. The test uses realised bill returns, which averaged 3.19% against 10.27% for equities. If bills paid 6% and equities returned 7%, waiting would cost far less and the deep-threshold rules would win more windows.
A market with weaker upward drift would change it too. Everything here rests on a series that rose relentlessly, and it is a single country's series at that. The US was the best-performing large equity market of the twentieth century, which is precisely why it survives in datasets. A market that spent decades flat would reward waiting.
The choice of threshold is the biggest soft spot. Testing five depths on one series and then reporting that 20% did best is exactly the data-mining AQR warns about. The 20% result should be read as a description of what happened, not a level to adopt. Nothing in the data says 20% will keep being the sweet spot.
Behaviour could flip the answer for an individual. A rule that keeps somebody invested who would otherwise sell in a panic can be worth several percentage points a year, which is more than the drag measured here. That argument is about commitment rather than returns, and it belongs alongside the case for writing rules down in advance that runs through the evidence on investment policy statements and pre-commitment.
Finally, the test compares one specific rule against one specific alternative. Critics of this framing would point out that a trend-following rule, which buys strength rather than weakness, sits on the other side of the same trade and has better published evidence behind it. AQR measured an average correlation of −0.14 between dip strategies and the SG Trend index, and a trend alpha of 4.7% against 0.3% for the average dip rule from 2000 to 2025. That's a different argument from the one tested here, and it doesn't rescue the waiting strategy.