TL;DR
- Buying the dip mostly works on SPY. 35 of 37 rankable entries beat drift, the +0.158% that buying any day's open and holding four days paid.
- The ranking is fragile. Rebuilt four other defensible ways, the median entry moves 13 places. Williams %R ranks anywhere from 2nd to 29th.
- One exit fits almost nobody. Only 5 of 37 entries had their largest edge over drift at a four-day hold. 31 of 37 peak at five days or longer.
- 16 of 40 held up across three decades. 11 decayed to under half their early edge, 10 changed sign in at least one period.
- RSI(2) is decaying, not dead. Its edge fell from +0.549% to +0.198% to +0.096% while it kept firing.
- Regime filters mostly fool you. 34 entries had a filter that raised their average trade over the whole window. 10 had one that helped in every period.
- Of the markets tested, it is strongest on stock indices. 96% of rankable entries beat drift on the Nasdaq 100. On gold, 11%.
Every trader has a favourite dip. The RSI one. The three-down-days one. The close below the lower band. So put all forty on the same data, same exit, same account, and ask one more question: is the edge still there?
I spent years hunting for the best indicator. That hunt was the wrong question. The right one is less exciting: is the edge still there?
How we tested
Forty published buy-the-dip entries, from eight families: oscillators, price position, volatility, momentum, patterns, volume and a few more. Each one is a single rule that says "the market just dipped, buy here".
The testing environment:
- Market: SPY daily, 1993-02-02 to 2026-08-20. 8,445 bars, 33.5 years. Price return, not total return.
- Entry: the rule is read on the close, the trade buys the next open.
- Exit: the open four bars after the entry bar. Holds of 0, 1, 2, 3, 5, 10 and 20 bars were run too, 2,880 backtested variants in all.
- Sizing: $35,000, full-account shares, no compounding, one position at a time.
- Costs: none. Frictionless, and I come back to that.
This is the study in 36 Ways to Buy the Dip, Second Edition, and every number below comes from it.
One idea carries the whole article, so here it is up front.
Over this SPY window, a long position made money just for showing up. Buy SPY at any day's open, sell four days later, and on average you made +0.158%. That is drift. An entry that earns +0.15% a trade has added nothing over drift. It just turned up. So every result here is edge over drift: the entry's average trade minus what entering on any day and holding as long paid.
Does buying the dip work?
Mostly, yes.
Three of the 40 fire fewer than 50 times at the four-day hold, so they're flagged and never ranked. Of the other 37, 35 beat drift. The two that don't are Gap Down at -0.178% and Lower Low + Close > Open at -0.038%.
The top of the list, at the fixed four-day hold:
| Rank | Entry | Edge over drift |
|---|---|---|
| 1 | TSI | +0.812% |
| 1 | TRIX | +0.812% |
| 3 | Percent Below SMA | +0.772% |
| 4 | Large Range Down Day | +0.612% |
| 5 | DV2 | +0.522% |
So the question was never whether these entries work. Most of them do. The question is which ones you'd actually trade, and a ranked list is a surprisingly poor answer.
Why the ranking is not the answer
A ranking is a photograph. It shows you one moment of one set of decisions, frozen. You need the film.
The ranking above rests on four choices: which column to rank on, which hold, which window, which market. Each has an alternative that's easy to defend. So I rebuilt the ranking four more times, changing one choice each time, with no new backtest, just re-reading the same results.
They agree with the headline ranking at a Spearman correlation between 0.40 and 0.70. Not unrelated. Not the same either.
The median entry moves 13 places. Williams %R ranks 2nd on one list and 29th on another. It is a real entry with a real edge, and where it sits in a table is mostly a statement about how the table was built.
Only 3 entries move five places or fewer.
There's a second trap in the ranking column itself. Let profit factor pick each entry's best hold and 22 of 39 entries reach for the longest hold on offer, 20 bars. Average trade percent does it for 31 of 39. Neither column is broken. Both include drift, and drift grows the longer you hold, so given a free choice they rank holding periods instead of entries.
One exit doesn't fit 40 entries
The ranking above holds every entry for the same four days. Here is where each entry's edge actually peaks:
5 of 37. That's how many entries had their largest edge at a four-day hold. 31 of 37 peak at five days or longer, and 15 at ten.
These are mean reversion entries, usually traded as short-horizon ideas. Most of them had their largest measured edge at a longer hold than that.
One warning, because it matters. Those peaks were chosen after reading the results, over the same data. They tell you where to look. They don't tell you where to exit.
Does the edge hold up?
This is the test that changes the answer.
The ranking is an average over 33.5 years. So cut the window into three periods with the same number of bars, measure every entry in each one against that period's own drift, and see what order the results came in.
| Verdict | Entries | What it means |
|---|---|---|
| Stable | 9 | positive in all three periods, the latest within half to 1.5 times the earliest |
| Strengthening | 7 | positive in all three, the latest more than 1.5 times the earliest |
| Decaying | 11 | positive in all three, but the latest under half the earliest |
| Unstable | 10 | the edge changed sign in at least one period |
| Thin sample | 3 | under 20 trades in a period, not judgeable |
Sixteen of the 40 held up. That's the honest headline of this whole study.
Now go back to the ranking. Of its top 10, 6 are Decaying, still positive but worth under half what they were, and 4 held up. The ranking isn't wrong about them. Every bar is a real edge, correctly measured over the whole window. It's just silent about whether the thing it measured is still happening.
What decay looks like: RSI(2)
RSI(2) below 10 is the most famous dip entry there is, and the clearest case.
Read the trade counts first. 158, 145, 137. The entry hasn't stopped firing. What changed is what each signal is worth: +0.549% over drift in the first period, +0.096% in the third. Less than a fifth.
Over the whole window it still looks fine. 440 trades, profit factor 1.61, a Strong verdict. That's the danger. Nothing in those whole-window numbers shows the decay.
Not everything is fading. PPO ran +0.008%, +0.172%, +0.398%. Drop it after testing on data ending in 2004, when it was barely above drift, and you'd have missed the two periods where it paid most.
Should you add a regime filter?
Every entry can be limited to a regime: volatility rising or falling, the long trend up or down. It's an easy way to improve an entry, and an easy way to curve-fit one.
34 of the 40 entries have a filter that raises their average trade over the whole window. Only 10 have one that also raises it in each of the three periods.
The gap is the trap. Forty entries times four filters is 160 chances to look good once. Pick the best of four for each entry and 34 of the 40 turned up a convincing one.
And the 10 survivors still cost something. For 6 of them the filter lowered return to drawdown. DV2's best filter adds to its average trade and drops its Ret/DD from 6.84 to 2.69. Z-Score of IBS keeps 87 of its 440 trades: its average trade rises +0.362% and its Ret/DD falls from 7.49 to 4.17. That isn't the same strategy improved. It's a much smaller one.
Are your entries secretly the same trade?
Two entries can sound unrelated and still fire on the same days.
Across all 780 pairs of entries, the average overlap in firing days is 0.170. Most of them really are different. But 19 pairs overlap more than 0.50, firing together on most of the days either one fires. DPO and PPO, two entries with different names and different formulas, overlap at 0.71.
Run two of those together and on most of their signal days you're taking the same trade twice. That's less diversification than two names suggest.
Does it work on futures and other markets?
On the E-mini S&P 500, mostly yes. Over 4,947 daily bars from 2007-01-03, 26 of the 37 rankable entries carry the same verdict on the contract as on the ETF. 3 come out better and 8 worse, and 2 of those 8 were already Decaying on SPY.
On other markets, it depends on the market.
96% of rankable entries beat drift on the Nasdaq 100, 84% on the Dow, 86% on the Russell 2000 ETF. Then 50% on the 10-year note, 28% on the euro and 11% on gold.
The fair objection is the calendar: the equity markets have shorter histories. So every market was also cut to one shared window starting 2002-04-05. Gold rises to 32% and still fails. The equity markets stay at 84% to 96%. The window explains some of it and nowhere near all of it.
Three example strategies, built from the survivors
An entry isn't a strategy. But you can build a starting point from the ones that held up, with mechanical rules stated up front, so you can change a threshold and rerun the pick yourself.
The pool: entries that are Stable or Strengthening, with at least 150 trades and 50 in every period, and an edge of at least 0.15% in their weakest period. 8 of 40 qualify. Three example strategies come out of them:
| Rule | Hold | SPY trades | Profit factor | Win % | Max DD | Edge over drift | |
|---|---|---|---|---|---|---|---|
| A, single entry | DV2(3, 250) < 0.05 | 4 bars | 267 | 2.21 | 65.54% | 12.28% | +0.52% |
| B, the pair | Range Position or Casey C% | 5 bars | 627 | 1.73 | 61.24% | 11.31% | +0.31% |
| C, with a filter | New 10-bar low above SMA(200), ATR rising | 10 bars | 190 | 2.04 | 64.74% | 10.33% | +0.48% |
All three also earn a Strong whole-window verdict on the futures contract: on ES, profit factors of 1.81, 1.53 and 2.09.
They are deliberately incomplete. No stop, no position size, no portfolio rules. Those are your decisions, and the study doesn't make them for you.
The honest limits
- Frictionless. No commission, slippage or spread were modeled, and the study does not test cost sensitivity. Price in your own costs before you trust any edge here.
- Hypothetical. These are backtests on history. Nothing here guarantees a future result.
- Not a buy-and-hold replacement on money. Holding SPY over the same window turned $35,000 into a $568,230.67 profit, with a worst fall of 56.47%. A dip entry is out of the market most of the time. Its case is the edge per trade and a smaller worst fall, like Example A's 12.28%.
- One market for the main test. The stability test is SPY only. The futures and cross-market checks are there so you can see how far it carries.
- The best holds were chosen after the fact. The ranking uses a fixed hold for exactly that reason.
What this means for you
- Measure against drift, not against zero. If your entry's average trade barely clears what entering on any day and holding as long paid, it adds little over drift.
- Don't treat the top of a ranking as the answer. Check whether an entry holds its place when you change the reasonable choices behind it.
- Cut your history into three and measure again. RSI(2)'s whole-window numbers looked fine while its edge shrank, so look at each period before you trust the average.
- Assume a filter is noise until it works in every period. Then count the trades it costs before you call it an improvement.
- Check overlap before combining entries. Two names can be one trade.
- Know your market. Among the markets tested, they did best on stock indices. On gold, assume nothing.
On the stock indices tested, buying the dip works. Knowing which dip still pays is the edge.
The full study: 36 Ways to Buy the Dip, Second Edition
Everything above is a summary. The book is the whole thing.
- All 40 entries, one page each. The rule, how it works, a real dated trade drawn on its chart, the results at eight holding periods, a verdict, and the three worst trades each one took.
- The evidence chapters. The ranking and why it moves, the three-period stability test, the regime filter test, the overlap between every pair, the same 40 on ES futures and on seven other markets.
- A playbook for beginners. The three example strategies built in full, with exact rules.
- Ready-to-run code in EasyLanguage and Python for all 40 entries and the three example strategies, plus an AI companion file so an AI assistant can write any entry in your own platform's language.
If you're going to trade a dip entry, read the page for it first. See what's inside 36 Ways to Buy the Dip.
If you'd rather start with the free research, every new study goes out in the Overfit newsletter first. StatOasis.com/Overfit.
Every figure above is a frictionless historical backtest on SPY daily data, with futures and cross-market checks, from 36 Ways to Buy the Dip, Second Edition. Hypothetical or simulated performance results have certain inherent limitations. Unlike an actual performance record, simulated results do not represent actual trading, and no representation is made that any account will or is likely to achieve profits or losses similar to those shown. Trading involves risk of loss. This is research, not investment advice.
