Nvidia (NVDA): three strategies, 2019–2024.
Holding Nvidia over these years returned 3,875%, a Sharpe ratio of 1.64. At their default settings none of the three rules beat that Sharpe. Tuned on the first 70% of the period and scored on the rest, one of the three held up.
Against holding it
On years it never saw
For each rule, every combination in its grid was run on the period up to 2023-03-15, the best was picked, and only that one was scored on what came after. How walk-forward works.
| Picked | Sharpe, tuned | Sharpe, after | Verdict | |
|---|---|---|---|---|
| Momentum | lookback 10 · MA 200 | 0.58 | 2.56 | held up |
| MACD | 16 / 34 / 9 | 0.23 | −0.48 | failed |
| Bollinger | 10 days · 2σ | 0.60 | −0.72 | failed |
Skill or luck
Three more readings of the same runs. A permutation p-value under 0.05 means the timing beat random timing; a probability of overfitting near 0.5 or above means picking by backtest did no better than picking at random; the rolling column slides the split through the period. What each one asks.
| Permutation p | Probability of overfitting | Across rolling windows | |
|---|---|---|---|
| Momentum | 0.391 | 0.14 | depends on the window |
| MACD | 0.178 | 0.59 | depends on the window |
| Bollinger | 0.014 | 0.44 | negative in every window |
Hypothetical results on past data (2019-01-02 to 2024-12-31), from a script in the repository that re-runs every figure. Not a recommendation to buy, sell or trade Nvidia or any other security.
Run it yourself
Each opens in the workspace on Nvidia over the same dates, straight after sign-up if you are new. Change the ticker or the dates and the numbers change with them, which is the point.
Other US stocks: Apple · Microsoft · Amazon · Alphabet · Meta Platforms · Tesla · JPMorgan Chase · S&P 500 ETF (SPY) · Nasdaq-100 ETF (QQQ) · AMD · Netflix · Berkshire Hathaway · Visa · Bank of America · Walmart · Coca-Cola · Johnson & Johnson · ExxonMobil · Disney · Dow Jones ETF (DIA) · Russell 2000 ETF (IWM). All of them.