> ## Content Index
> Fetch the complete content index at: https://simplestocktrade.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# The Five Deepest Dips on the Nasdaq 100: half of buy-and-hold's return, a 23% worst loss against 83%
- URL: https://simplestocktrade.com/five-deepest-dips-nasdaq-100/
- Published: 2026-09-14T12:03:16.000Z
- Updated: 2026-09-14T12:03:14.000Z
- Author: Yevhen S
- Tags: Strategy reviews, Mean reversion, Backtesting, Risk, Nasdaq 100

## The short version

| **What it does**           | Buys large US tech shares that have just fallen hard — up to five at a time — and sells each one on the first sign of a bounce. It holds them under five days. |
| -------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **What it made**           | 4.28% a year over 26.69 years, against 7.89% a year for simply buying the index and holding it.                                                                |
| **What it cost**           | At its worst it was 22.99% below its previous high. Holding the index put you 82.99% below.                                                                    |
| **How invested it was**    | About an eighth of the money on an average day. On two days out of three it held nothing at all.                                                               |
| **The test that mattered** | I rebuilt it 200 times with the buy day chosen by a calendar instead of by the rule. Not one of the 200 did as well.                                           |
| **The catch**              | Half the index's return, almost all of the advantage earned in the 2000–2002 crash, and the test pays no interest on the cash it sits in.                      |

Everything below is a backtest of my own rules on past prices. It is a record of what I measured, not advice and not a recommendation to trade anything.

## What this is, and what it isn't

The Five Deepest Dips is a mean-reversion strategy: a bet that a share which has just fallen hard will bounce back rather than keep falling. It watches the Nasdaq 100, using the companies the index really held on each day rather than today's list. When the market looks healthy it buys the five most beaten-up names and sells each on the first sign of recovery.

The number in the table above — 22.99% at its worst against the index's 82.99% — is called drawdown: the deepest fall from a peak in the account to the low that follows, the stretch you would have had to sit through without quitting. So about half the return, and a bit over a quarter of the pain.

Almost all of that advantage came in one stretch. From 2000 to 2002 the strategy made 4.11% a year while the index lost 36.48% a year. By January 2010 $40,000 had become $66,578, against $19,301 for the index investor, who was still below the 2000 peak. Buy-and-hold caught up for good on 29 May 2020.

So it is a defensive thing, not a growth thing. Those are different products and people routinely mix them up.

I wrote this formula myself, but the idea is old and heavily mined: buy weakness inside an uptrend, sell into the first strength. Twenty years of people picking that over makes a hidden secret unlikely, and the numbers below describe a known effect with the usual costs attached.

## The rules

Three gauges do the work, and I use the same names for them throughout.

- the **dip gauge** says where today's close sits inside the last three days' range, on a scale of 0 to 100\. Near 0 means the share closed at the very bottom of it.
- the **bounce gauge** is a fast measure of recent strength, also 0 to 100\. It shoots up as soon as a share stops falling.
- the **trend-strength gauge** measures how firmly a move is trending, in either direction, over 12 days.

In plain terms. The index has to be above its own 200-day average, or it buys nothing that day. Then it wants shares also above theirs, still trending, that have just closed near the bottom of their three-day range. It buys the five that fell hardest, a fifth of the account each, at the next morning's open. It sells on the first sign of a bounce, at a 4% profit, or at a 25% loss.

#### The rules in full, precise enough to rebuild

- ****What it can buy.** The Nasdaq 100 as it really stood each day, including companies later dropped or delisted. Testing on today's list would hand the strategy the survivors and hide the disasters.
- ****When it is allowed to buy at all.** The index must be above its own 200-day average price.
- ****Which shares qualify.** Above its own 200-day average, trend-strength gauge rising, and trading enough volume to be worth touching.
- ****The trigger.** The dip gauge closes below 10.
- ****Choosing between them.** More names usually qualify than there is room for. The five that fell hardest over the previous five days win.
- ****Size.** Five slots, a fifth of the account each. Nothing is ever borrowed.
- ****Buying.** Decided on the close, bought at the next morning's open, commission $0.005 a share. The test adds no slippage (the gap between the price you expect and the one you get) and pays nothing on idle cash.
- ****Selling, whichever comes first.** The bounce gauge closes above 65; a 4% profit; two up days in a row; or a 25% loss.

That last list is the vague part: three of those four can happen on the same day, and the order you check them in changes the record. Two honest people building this from the same description get different answers there.

![Four gates and a ranking step. Every gate has to be open, which is why the account is in cash on two days out of three.](https://storage.ghost.io/c/8a/20/8a2056d3-43f4-497a-abef-84bc9d80e6a7/content/images/2026/09/dipfive-mechanism.png)

## One real trade, start to finish

From 25 to 31 January 2024 a large Nasdaq share closed down five days running, 195.22 to 184.40\. On the close of 31 January the dip gauge read 8.3 — under the threshold of 10 — and the bounce gauge read 1.0, about as washed out as it goes. The market filter was on, the share was above its 200-day average, the trend-strength gauge was rising. It qualified.

Next morning, 1 February, it was bought at the open at 183.99\. It went against me by 2.57% at its worst before it turned. On 5 February the bounce gauge closed at 66.3, above the exit level of 65, so the sell went in for the following morning. On 6 February it sold at 186.86: up 1.56% in four days.

What did not happen matters too: the 4% profit target, at 191.35, was never reached, as it usually isn't. This is built to collect a lot of small moves and leave. Two trades in three made money, the average winner made 2.53% and the average loser lost 3.29%. Frequently right, and modestly so.

![One real trade, checked bar by bar: five red closes, a buy at the next open, and an exit four days later on the bounce gauge rather than on the 4% target, which was never reached.](https://storage.ghost.io/c/8a/20/8a2056d3-43f4-497a-abef-84bc9d80e6a7/content/images/2026/09/dipfive-setup.png)

## What it did over 26 years

The conditions, because a number without them is not a number. Nasdaq 100 with its real historical membership, 3 January 2000 to 11 September 2026\. $40,000 to start, five slots at 20%, $0.005 a share, bought at the next open, no interest on idle cash, 1,115 trades. A backtest, not a live track record.

| Over the full 26.69 years | The strategy | Holding the index |
| ------------------------- | ------------ | ----------------- |
| $40,000 became            | $122,368     | $298,660          |
| Return a year             | 4.28%        | 7.89%             |
| Worst loss from a peak    | 22.99%       | 82.99%            |
| Sharpe ratio              | 0.52         | 0.42              |

Sharpe compares the return against how bumpy the ride was, so a higher number means the same gain arrived more smoothly.

The figure that changes the interpretation most is how much money was actually working: 12.6% on an average day, the rest in cash. It held at least one position on only 32% of days, usually one or two names rather than five. This is a part-time job, not a salary.

| Era       | Strategy | Its worst loss | The index | Its worst loss |
| --------- | -------- | -------------- | --------- | -------------- |
| 2000–2002 | +4.11%   | 23.0%          | −36.48%   | 83.0%          |
| 2003–2007 | +7.40%   | 10.6%          | +15.08%   | 17.4%          |
| 2008–2012 | +0.74%   | 18.3%          | +5.24%    | 49.5%          |
| 2013–2019 | +5.72%   | 9.3%           | +17.98%   | 23.2%          |
| 2020–2026 | +3.28%   | 15.1%          | +19.58%   | 35.6%          |

The first row is the whole case. After it the index wins on return every time.

Cut the history in half rather than by era and neither half collapses: 3.79% a year on a 23.0% worst loss over 2000–2013, 4.78% on 15.1% over 2013–2026\. It was ahead of buy-and-hold on 5,081 of the 6,713 trading days, 76% of them, and still finished behind.

One detail worth keeping: the trigger fired on 2,589 days that had passed every filter, and only 1,115 became trades. Three signals in five never got a slot, because a market-wide dip turns a dozen names oversold on the same morning.

![The strategy was ahead for twenty years and finished behind. The lower panel is the strategy's drawdown; the index's worst was 82.99%.](https://storage.ghost.io/c/8a/20/8a2056d3-43f4-497a-abef-84bc9d80e6a7/content/images/2026/09/dipfive-equity.png)

## What I tried to break

Every run is here, including the ones that beat my own defaults. A review that shows only the flattering tests is an advertisement. "Capital in market" is the average share of the account invested. MAR is the return a year divided by the worst loss, so a higher MAR means more return earned per unit of worst loss. A basis point is one hundredth of one percent of the price.

The full stress battery — 39 rows, every run I made 

| Group                       | Variant                        | CAR% | maxDD% | MAR  | Trades | Win% | Capital in market % |
| --------------------------- | ------------------------------ | ---- | ------ | ---- | ------ | ---- | ------------------- |
| Baseline                    | as written                     | 4.28 | 22.99  | 0.19 | 1115   | 65.8 | 12.6                |
| Dip-gauge threshold         | below 4                        | 1.22 | 8.90   | 0.14 | 179    | 68.7 | 2.0                 |
| Dip-gauge threshold         | below 7                        | 2.46 | 21.09  | 0.12 | 557    | 66.1 | 6.4                 |
| Dip-gauge threshold         | below 10 (default)             | 4.28 | 22.99  | 0.19 | 1115   | 65.8 | 12.6                |
| Dip-gauge threshold         | below 13                       | 6.72 | 22.47  | 0.30 | 1735   | 64.2 | 19.6                |
| Dip-gauge threshold         | below 15                       | 8.71 | 20.53  | 0.42 | 2149   | 64.9 | 24.0                |
| Profit target               | 2%                             | 4.03 | 19.23  | 0.21 | 1133   | 69.1 | 10.8                |
| Profit target               | 3%                             | 4.03 | 21.18  | 0.19 | 1117   | 66.8 | 12.1                |
| Profit target               | 4% (default)                   | 4.28 | 22.99  | 0.19 | 1115   | 65.8 | 12.6                |
| Profit target               | 6%                             | 4.21 | 22.15  | 0.19 | 1114   | 65.3 | 13.1                |
| Profit target               | 8%                             | 4.22 | 22.11  | 0.19 | 1113   | 65.0 | 13.2                |
| Stop loss                   | 15%                            | 3.41 | 25.46  | 0.13 | 1116   | 65.6 | 12.4                |
| Stop loss                   | 20%                            | 3.73 | 25.46  | 0.15 | 1116   | 65.7 | 12.5                |
| Stop loss                   | 25% (default)                  | 4.28 | 22.99  | 0.19 | 1115   | 65.8 | 12.6                |
| Stop loss                   | 30%                            | 4.05 | 20.35  | 0.20 | 1113   | 65.9 | 12.6                |
| Stop loss                   | 40%                            | 4.69 | 18.29  | 0.26 | 1113   | 65.9 | 12.7                |
| Trend-strength gauge period | 8                              | 5.87 | 23.01  | 0.26 | 1322   | 66.3 | 14.8                |
| Trend-strength gauge period | 10                             | 5.07 | 24.31  | 0.21 | 1212   | 65.6 | 13.6                |
| Trend-strength gauge period | 12 (default)                   | 4.28 | 22.99  | 0.19 | 1115   | 65.8 | 12.6                |
| Trend-strength gauge period | 16                             | 3.95 | 27.95  | 0.14 | 1035   | 67.1 | 11.5                |
| Trend-strength gauge period | 20                             | 3.69 | 21.93  | 0.17 | 926    | 67.9 | 10.2                |
| Bounce-gauge exit level     | 55                             | 4.01 | 22.28  | 0.18 | 1138   | 64.6 | 10.9                |
| Bounce-gauge exit level     | 60                             | 4.02 | 22.28  | 0.18 | 1126   | 65.1 | 11.8                |
| Bounce-gauge exit level     | 65 (default)                   | 4.28 | 22.99  | 0.19 | 1115   | 65.8 | 12.6                |
| Bounce-gauge exit level     | 70                             | 4.32 | 23.88  | 0.18 | 1107   | 66.0 | 13.2                |
| Bounce-gauge exit level     | 75                             | 4.13 | 24.64  | 0.17 | 1094   | 66.5 | 14.0                |
| Slippage each way           | 0 bp (default)                 | 4.28 | 22.99  | 0.19 | 1115   | 65.8 | 12.6                |
| Slippage each way           | 1 bp                           | 4.13 | 23.01  | 0.18 | 1115   | 65.6 | 12.6                |
| Slippage each way           | 3 bp                           | 3.82 | 23.04  | 0.17 | 1115   | 64.8 | 12.6                |
| Slippage each way           | 5 bp                           | 3.51 | 23.07  | 0.15 | 1115   | 64.2 | 12.6                |
| Remove a filter             | all filters on (default)       | 4.28 | 22.99  | 0.19 | 1115   | 65.8 | 12.6                |
| Remove a filter             | no rising trend-strength gauge | 8.27 | 23.01  | 0.36 | 1987   | 65.4 | 22.3                |
| Remove a filter             | no 200-day on the stock        | 5.21 | 23.77  | 0.22 | 2284   | 63.3 | 27.1                |
| Remove a filter             | no index filter                | 6.34 | 22.99  | 0.28 | 1331   | 66.7 | 14.6                |
| Ranking rule                | biggest 5-day fall (default)   | 4.28 | 22.99  | 0.19 | 1115   | 65.8 | 12.6                |
| Ranking rule                | most oversold by the dip gauge | 4.42 | 29.28  | 0.15 | 1124   | 66.3 | 12.4                |
| Ranking rule                | lowest bounce gauge            | 3.95 | 29.04  | 0.14 | 1123   | 66.2 | 12.5                |
| Fill assumption             | next open (default)            | 4.28 | 22.99  | 0.19 | 1115   | 65.8 | 12.6                |
| Fill assumption             | that day's close               | 3.67 | 18.93  | 0.19 | 1120   | 63.6 | 12.6                |

**The placebo test — the run that could have killed it.** A placebo here is the same strategy with the buy day chosen by something that cannot know anything: a fixed calendar. I built 200 of them. Each keeps every filter, every exit, the five slots and the costs. Each buys every 147th day that had already passed all the filters, starting from a different point for every share and every run.

Why every 147th? Because a placebo that trades a different number of times is not a fair test. The obvious spacing is every 74th day, which is how often the real trigger fires. But that version makes 2,182 trades against the real 1,115, because real signals arrive in clumps and a calendar does not. At 147 days apart the placebos traded 1,084 to 1,118 times.

| Across the 200 placebos | Worst  | Middle | Best  |
| ----------------------- | ------ | ------ | ----- |
| Return a year           | −0.89% | 1.17%  | 3.36% |
| Worst loss              | 8.8%   | 16.4%  | 35.9% |

The real rule made 4.28% a year on a 22.99% worst loss. **Not one of the 200 matched it.** That is the strongest result here, and the one I expected to fail. A great many published bounce strategies turn out to be "buy anything that passed a trend filter, sell on the first strength" with a decorative trigger bolted on.

The placebos say something uncomfortable too. At 22.99%, the real strategy's worst loss is deeper than the typical placebo's 16.4%. The real trigger buys several names on the same bad morning; the calendar versions spread their entries out. The entry rule buys concentration along with its edge.

![Two hundred versions with the buy day chosen by the calendar instead of by the dip gauge. The real rule sits outside the whole distribution.](https://storage.ghost.io/c/8a/20/8a2056d3-43f4-497a-abef-84bc9d80e6a7/content/images/2026/09/dipfive-controls.png)

**The trigger level is a dial, not a discovery.** Loosen it and both the return and the money at work rise together, all the way to the loosest setting I tested.

| Dip gauge must close below | Return a year | Worst loss | Money at work | MAR  |
| -------------------------- | ------------- | ---------- | ------------- | ---- |
| 4 — strictest tested       | 1.22%         | 8.90%      | 2.0%          | 0.14 |
| 10 — the default           | 4.28%         | 22.99%     | 12.6%         | 0.19 |
| 15 — loosest tested        | 8.71%         | 20.53%     | 24.0%         | 0.42 |

So the default is not the best value even in its own neighbourhood. On the headline numbers the loosest setting simply won, with more return *and* less drawdown. But it did not find better trades. It just put more money in the market for longer. The threshold decides how exposed you are; it does not make the signal better.

#### The arithmetic behind that

For every 1% of the account actually working, the strictest setting earned about 0.61% a year, the default 0.34% and the loosest 0.36%. The money the loose settings put to work earns no more per unit than the default does. And the strict setting that ratio flatters made 1.22% a year in absolute terms, which nobody would trade.

**Every filter I removed made more money.**

| What I removed                  | Return a year | Worst loss | Money at work |
| ------------------------------- | ------------- | ---------- | ------------- |
| Nothing — the default           | 4.28%         | 22.99%     | 12.6%         |
| The rising trend-strength gauge | 8.27%         | 23.01%     | 22.3%         |
| The index's 200-day average     | 6.34%         | 22.99%     | 14.6%         |
| The share's own 200-day average | 5.21%         | 23.77%     | 27.1%         |

Each removal bought more exposure along with the extra return, and not one cost as much as a single point of worst loss. Filters are supposed to buy safety with return. Here they bought very little.

The index filter also failed on the one occasion it was meant to matter. The worst loss in the whole test, 23%, happened in two weeks: a peak on 31 March 2000, a trough on 14 April 2000\. The filter was on the whole way through, because the Nasdaq was still above its 200-day average on 31 March. A 200-day average cannot react to a two-week collapse.

## What survives

**The trigger carries real information.** Not one of the 200 calendar placebos beat the real rule's 4.28% a year on a 22.99% worst loss. The best managed 3.36%, the middle one 1.17%, and their own worst losses ran 8.8% to 35.9%. The dip trigger is not decoration.

**It survived being cut in half.** 3.79% a year on a 23.0% worst loss over the first thirteen years, 4.78% on 15.1% over the second. Different markets, no collapse in either.

#### Two more things that held up

****Both sell rules sit on flat ground.** Move either across its whole tested range and almost nothing happens: the profit target from 2% to 8% spans 4.03% to 4.28% a year on worst losses of 19.23% to 22.99%, the bounce exit from 55 to 75 spans 4.01% to 4.32% on 22.28% to 24.64%. That is robustness, and it also means 4% and 65 carry no edge of their own.

****Costs are survivable, because of what it trades.** Every extra basis point lost on each fill costs about 0.154 of a point of annual return: 4.28% at zero, 3.51% at five, worst loss unchanged at 23.07%. Gentle only because these are the most liquid shares in the US market.

## What breaks

**The cash assumption is worth more than the strategy.** The test pays nothing on the 87% of the account sitting idle. Pay it something, on exactly the same trades, and the picture moves.

| Interest paid on the idle cash | Return a year |
| ------------------------------ | ------------- |
| None, as the test was run      | 4.28%         |
| A flat 2%                      | 6.11%         |
| A flat 4%                      | 7.98%         |

The worst loss barely moves: 22.91% at a flat 4%, against 22.99% as it stands. And 7.98% beats buy-and-hold's 7.89%. Two caveats. It is arithmetic applied to the existing curve, not a fresh run, and a flat 4% was not on offer across 2000–2026, where there were years near zero. Even so, it moved the result by more than any rule change I tested.

**The defence rests on one crash.** 2000–2002 earned the whole advantage: +4.11% a year against the index's −36.48%. Every era since went the other way, the index making 17.98% against 5.72% in 2013–2019 and 19.58% against 3.28% in 2020–2026\. The only thing it ever protected against was a slow, multi-year bear market.

**When you buy matters more than it should.** Buy at that day's close instead of the next morning's open and the return drops from 4.28% to 3.67%, the winners from 65.8% to 63.6%, on a worst loss of 18.93% rather than 22.99%. The overnight gap between deciding and buying is part of the result, not a rounding error.

**The stop loss is almost never what gets you out.** Moving it from 25% to 40% changed the trade count by two out of 1,115, and improved both numbers at once: 4.69% a year on an 18.29% worst loss. Tightening it was worse on both. 15% gave 3.41% on a 25.46% worst loss.

**Running it bigger does not scale the way it looks.** With only 12.6% of the money at work and nothing borrowed, three or four times the size looks available. That multiplies the 22.99% worst loss by the same amount, and the trigger fills several slots on the same bad morning, exactly when borrowed money is least forgiving.

## Verdict

This suits someone who already owns the index and wants a mostly-idle second holding beside it, in the market only during quiet stretches. They need somewhere sensible to park the cash, because here the interest on that cash moved the result more than the rules did. And they have to care more about the 82.99% fall they did not sit through than about the gap between 4.28% a year and 7.89%.

It does not suit anyone looking for growth: 4.28% a year against the index's 7.89% over 26.69 years, beaten on return in every era since 2002\. Nor anyone who wants the settings to mean something. The thresholds sit on plateaus, every filter removed made more money, the stop barely fires. What earns its keep is a trigger that beat all 200 placebos, a sell rule that leaves quickly, and a broker that pays interest on cash.

**If you want to run it yourself.** The formula is on my Patreon in a version that needs no data vendor: it reads the list of index members from a plain text file, so it runs on any daily database. The post that comes with it measures something this article leaves out — how much of a backtest's result is flattery when you test on today's index members instead of the ones that were really there at the time: [patreon.com/cw/SimpleStockTrade](https://www.patreon.com/cw/SimpleStockTrade?ref=simplestocktrade.com).

---

## Sources — every number traceable

Every figure in the text comes from one of the runs below, and each one records the conditions it was run under.

#### What each number came from

****The baseline run.** The full trade list, all 1,115 of them, and the account value for every one of the 6,713 trading days along with how many of the five slots were filled that day. The era numbers, the two halves and the idle-cash overlay are all worked out from that daily curve.

****The benchmark.** Buy and hold the same index over the same window, from the same starting money.

****The stress battery.** Thirty runs, one for each row of the table above, plus the benchmark.

****The placebo spacing.** Before the 200 placebos were run, five calibration runs at spacings of 74, 130, 150, 170 and 190 days produced 2,182, 1,245, 1,086, 947 and 843 trades. 147 is the spacing that lands on the real strategy's 1,115.

****The 200 placebos.** Two hundred separate runs, each a complete backtest with the buy day moved onto the calendar.

****How often the trigger fires.** It was checked on 190,270 share-days and fired on 2,589 of them.

****The illustrated trade.** Daily prices for that share around February 2024, with the gauges recomputed to confirm the entry and exit days.

**Cross-check.** The figures in this article were computed from the daily equity curve by an independent script, not read off the backtesting software's own report. The two agree: the software reports 4.28% a year, a worst drawdown of 22.99%, 12.93% of capital in the market, 1,115 trades, 739 winners (66.28%) and an average holding time of 4.78 bars. The script computed 4.28%, 22.99%, 12.6% and the same trade count from the curve alone. Every run in this article — the baseline, its 29 variants, the benchmark and all 200 placebos — came out of the same software, so there is no second version of the rules that could have drifted from the first.