signals / Stoch %K/%D cross (overbought)
Stoch %K/%D cross (overbought) is a short-side study. Fires when Stochastic %K crosses below %D while in the overbought zone (above 80).
FAQ
What is Stoch %K/%D cross (overbought)?
Stoch %K/%D cross (overbought) is a short-side study. Fires when Stochastic %K crosses below %D while in the overbought zone (above 80).
Which names are firing it?
Among mega-caps in the last 5 daily bars: COST, AMZN, AVGO, TSLA, NVDA, MSFT, AMD.
What is the sample?
Hit rates use the same rank engine as /t/spy on a fixed sample (SPY, SPX, QQQ, IWM, AAPL, MSFT, NVDA, TSLA). 8 of those names have a measurable history for this rule. Windows are 5 days, 1 month and 3 months on daily bars. Last bar in sample: 2026-10-12. Method: /methodology. Caveats: /help.
What does Stoch %K/%D cross (overbought) measure?
Fires when Stochastic %K crosses below %D and %K is in the overbought zone on this bar or the prior bar (above 80). Underlying math: see How it works on this page. Catalog line: Fires when Stochastic %K crosses below %D while in the overbought zone (above 80).
What does a firing mean?
A firing is the daily bar where this rule’s condition becomes true (a cross or a flip). It is a study event, not a recommendation to buy or sell. Forward windows on this site are close-to-close with no costs. Study side is short-side (SHORT). A positive forward return is a decline in the close after the print (side-adjusted) — a scoring convention, not a recommendation to sell short. Method: /methodology.
Why can the sample be small?
Hit rates need enough dated prints with a completed forward window. A row with fewer than 10 firings is anecdote, not evidence — the same caveat as /help. Slow rules (especially 50/200 SMA crosses) often have single-digit counts on a two-year tape. We do not invent a hit rate when the count is missing. Names in this sample with a measurable history include AAPL · IWM · TSLA · NVDA.
How it works
Slow stochastic on daily bars: lookback 14, %K smoothed 3, %D = 3-period SMA of %K. Raw %K is 100 × (close − 14-bar low) / (14-bar high − 14-bar low). The oversold / overbought zone uses the current or prior %K (a cross still counts if %K has just left 20 / 80).
Fires when Stochastic %K crosses below %D and %K is in the overbought zone on this bar or the prior bar (above 80).
Definition comes from the rank-rule catalog and the indicator defaults in this repo. Slug stoch-kd-cross-overbought is listed on /api/meta.
How to read it
Study side is short-side (labeled SHORT on firing chips). A positive forward return on this page is a decline in the close after the print (side-adjusted). That is how the event is scored — not a recommendation to sell short. A %K/%D cross inside an extreme zone is a classic oscillator setup. Because the zone test allows the prior bar, a print can occur on the bar %K leaves 20 or 80. That is a definition detail, not extra edge. Use the count column: many stochastic crosses on a two-year tape are still a small sample once you want a 3-month window.
Scoring on this site is shared across every rule (5-day / 1-month / 3-month close-to-close windows, side-adjusted). See How it is scored here below. Names in the tracked sample with a measurable history include AAPL · IWM · TSLA · NVDA. Those figures are the same rows as the hit-rate table below — not a second ranking. Historical hit rates, not a prediction / not trade advice.
When it fails
- Choppy ranges produce many %K/%D crosses; most are not durable turns.
- A strong trend can print oversold or overbought crosses that fail immediately.
- The prior-bar zone allowance means a print can occur after %K has already left 20 / 80.
- Past ≠ future. A historical hit rate is not a prediction and not investment advice.
- No costs modeled. Forward returns are close-to-close with no commissions, slippage or borrow.
- Small samples. A row with fewer than 10 firings is anecdote, not evidence.
- Overlap bias. Clustered prints can credit the same later move more than once.
Same caveats as the rest of the site: /methodology · /help.
How it is scored here
On this site every rule is scored the same way: run it over about two years of daily bars, record each firing, and measure what the name did 5 days, 1 month and 3 months later. Sell signals are measured as shorts, so a positive return is a profitable short (side-adjusted). The composite score is 0.6 × average return + 0.4 × (win rate − 50), with longer horizons weighted more. Nothing here is a forecast.
Full catalog: /signals. Method: /methodology. Slug stoch-kd-cross-overbought is stable and listed on /api/meta.
Same family: Stoch %K/%D cross (oversold)
Firing now (mega-caps, last 5 bars)
| Symbol | Last date | Win 1mo | Count |
|---|---|---|---|
| COST | 2026-10-09Win 57.1% · n=21FIRED | 57.1%n=21 | 21 |
| AMZN | 2026-10-12Win 40.7% · n=27STUDY | 40.7%n=27 | 27 |
| AVGO | 2026-10-08Win 40.0% · n=25STUDY | 40.0%n=25 | 25 |
| TSLA | 2026-10-08Win 41.7% · n=24STUDY | 41.7%n=24 | 24 |
| NVDA | 2026-10-07Win 36.7% · n=30STUDY | 36.7%n=30 | 30 |
| MSFT | 2026-10-08Win 29.6% · n=27STUDY | 29.6%n=27 | 27 |
| AMD | 2026-10-08Win 33.3% · n=24STUDY | 33.3%n=24 | 24 |
Hit rates on tracked names
Same rank engine as /t/spy, /t/nvda, and /t/spx, restricted to this rule.
| Symbol | Win 5d | Win 1mo | Win 3mo | Count | Last bar |
|---|---|---|---|---|---|
| AAPL | 42.9% | 53.1%n=32 | 34.5% | 32 | 2026-10-12 |
| TSLA | 44.0% | 41.7%n=24 | 54.5% | 24 | 2026-10-12 |
| IWM | 50.0% | 44.1%n=34 | 28.1% | 34 | 2026-10-12 |
| MSFT | 37.0% | 29.6%n=27 | 39.1% | 27 | 2026-10-12 |
| NVDA | 50.0% | 36.7%n=30 | 30.8% | 30 | 2026-10-12 |
| SPX | 57.9% | 35.1%n=37 | 21.2% | 37 | 2026-10-12 |
| QQQ | 41.0% | 31.6%n=38 | 27.0% | 38 | 2026-10-12 |
| SPY | 55.6% | 25.7%n=35 | 21.9% | 35 | 2026-10-12 |
Honest caveats
- Past ≠ future. Everything here is a historical measurement on daily bars, not a prediction and not investment advice.
- No costs modeled. Forward returns are close-to-close with no commissions, slippage or borrow.
- Overlap bias. Signals that fire in clusters get credit for the same move more than once.
- Small samples. A row with fewer than 10 firings is anecdote, not evidence.
Full method: help manual. Interactive chart: terminal.