guides /
Hit-rate methodology
How are hit rates calculated?
How a firing is scored on daily bars, and which JSON endpoint returns one symbol's leaderboard.
Published · Updated
Every rank-panel rule is scored the same way: run it over about two years of daily bars, record each firing, and measure the log return from the next session's open (or that session's close when no open is printed) to the close 5 trading days, 21 trading days (1 month), and 63 trading days (3 months) later. A short return is the negation of that log return, so a positive number is a decline. The composite score is 0.6 times average log return plus 0.4 times (win rate minus 50), with longer horizons weighted more. Nothing here is a forecast.
Those forward returns use a dividend-adjusted close (total return). The next-bar open is scaled into that series, and the exit is the total-return close at the horizon. Both legs are total-return adjusted. Published price levels — moving averages, prior highs and lows, Fibonacci levels, pivots, anchored VWAP, volume profile, and DeMark levels — use split-adjusted prices that are not adjusted for dividends.
Win rate is the share of non-overlapping fills whose side-adjusted log return was positive. Each horizon has its own sample size: n_5d, count (1 month), and n_3mo. A re-fire inside that horizon's hold is skipped. base_up_5d, base_up_1mo, and base_up_3mo are the symbol's unconditional up-rates. base_win_* is the up-rate for a long and the down-rate for a short. edge_* is the win rate minus base_win_*. anecdote_5d, anecdote_1mo, and anecdote_3mo are true when that horizon's n is under 10. The rate is still returned and is not a signal. anecdote is the 1-month flag and matches anecdote_1mo. On a roughly two-year tape, non-overlapping 3-month holds are structurally small-sample. The headline is walk-forward: the rule is chosen on the first 40% of the tape and the 1-month hit rate is the later 60% only. The base rate is the up-rate for a long rule, or the down-rate for a short rule, over all overlapping sessions in the held-out later-60% window. The rule's trades are thinned so none overlap. Conviction means that held-out 1-month sample beats the base rate at 5 percent with at least 10 trades.
One symbol's leaderboard is GET /api/rank?symbol=SPY (period defaults to 2y). The rule catalog, including stable slugs, is /api/meta. The same scoring write-up is /methodology. The HTML worked example is /t/spy.
Live endpoints
- SPY rank JSON https://lordofta.com/api/rank?symbol=SPY
- Rule catalog JSON https://lordofta.com/api/meta
- Methodology https://lordofta.com/methodology
- Hit rates on SPY https://lordofta.com/t/spy
Plain text: /guides/hit-rates.md.
Descriptive research on daily bars. Not a forecast and not trade advice.