Good to know · Case study
Buying the dip in NVDA: one year against buy and hold and chance
Question
A common idea: don't buy at the high, buy after a drop — and only once the price turns up again, so you don't catch a falling knife. I put this rule on NVDA in the rulio Lab and computed one year back. The question: does it return more than simply holding the stock, and more than the same number of buys on random days?
Rule
- Step 1: −10% from the new high (within 30 d) — the high moves up while the price rises
- Step 2: +3% from the running low — the low moves down while the price falls
- Then: Buy · TP +3.4% (3ר) · SL −4.5% (4ר) · no time limit
- Once — the backtest still counts every trigger in the period, like the Lab
Result
| With the rule, after costs | +20.6% |
|---|---|
| Buy and hold | +15.2% |
| Chance: as many buys on random daysmedian, 8 in 10 versions between −19.0% and +11.7% | −5.0% |
The rule is ahead of 95% of the random versions. Largest drop of the account with the rule: −4.7%.
Costs, period, number of cases
- Costs: 0.1% per buy and per sell
- Period: 24 October 2025 to 23 October 2026
- Cases: 12 trades, 11 closed, 1 still open at the end
- Computed on 24 October 2026 with the rulio Lab
Other stocks
The same rule, unchanged, over the same period on other stocks:
| Stock | Trades | Rule | Buy and hold | ahead of chance |
|---|---|---|---|---|
| AMD | 27 | +29.9% | +115.0% | 92% |
| INTC | 29 | +8.9% | +112.6% | 80% |
| AVGO | 14 | −17.8% | −8.0% | 19% |
| AMZN | 5 | −7.8% | +12.8% | 33% |
| MSFT | 6 | −4.9% | +0.6% | 49% |
| BIDU | 15 | −24.4% | −36.5% | 15% |
How to read this
In NVDA the rule was ahead of buy and hold this year and ahead of almost all random versions. It rarely risked much: target and stop sit close to the entry, and the largest drop of the account stayed below 5%.
On other stocks the same rule does not hold up. In AMD and Intel it stayed far behind holding; in Broadcom, Amazon, Microsoft and Baidu it lost. Target and stop are cut to NVDA's daily range — on another stock the same percentages are too tight or too wide.
What the figures do not say
- One period, one stock, one rule — chosen in hindsight. Try enough rules and one will always come out ahead. That it fails on other stocks suggests luck played a part here too.
- 11 closed trades are few; one or two trades more or less shift the result noticeably.
- It was a rising year for NVDA. A falling year would look different.
- Target +3.4% and stop −4.5% are fixed percentages in the rule: three and four times NVDA's daily range when the rule was built. The backtest uses these percentages for the whole year.
- Costs are a flat rate; gaps and partial fills can cost more in real trading.
- Not investment advice. The calculation shows what was — not what will be.
Who writes this
I run rulio. The Lab used for these figures is part of the app — so I have an interest in people trying it. The figures come from a stored run and were not copied by hand.
Sources
- Prices: Alpaca Markets (hourly or daily bars), 24 October 2025 to 23 October 2026.
- Chance: the same number of buys on random days of the same stock, computed the same way, with a fixed seed.
- Assumptions: hourly bars, 0.1% costs per buy and per sell, results compounded; a trade still open at the end is not counted.
Check it yourself
The Lab opens with exactly this rule and this period. Check the figures there and change the rule.
More good to know
- GuideWhat does a short stock trade cost? Four routes compared from a euro accountCapital.com, Trading 212, Alpaca and Interactive Brokers from a euro account: deposit, trades, withdrawal — the whole bill for €300 and €3,000, line by line.
- Case studyTight stop, expensive stop: what 4,893 trades say about the stop-lossA stop-loss just below the entry feels safe. Measured over four years it cost more than it protected — and what to size it by instead.

Open the app