Analytics — where the edge lives
更新日: 2026-08-14
Analytics — where the edge lives
Results say “how much”; Analytics says “where and why”. The same equity/drawdown curves sit on top; breakdowns follow.
Entry-price calibration — the headline tool
On probability markets the entry price is the bet. Calibration sorts trades into entry-price buckets (4/6/8) and answers per bucket: is there an advantage after fees?
| Column | Meaning |
|---|---|
| Range | the entry-price bucket |
| Trades | n < 30 is honestly marked “few trades”, no verdict |
| Edge (realised) · 95% CI | average net P&L per share in pp; the reference point is break-even including fees; intervals corrected for the number of buckets |
| Verdict | above break-even (the edge lives here) · indistinguishable from zero · below break-even |
Above the table — the “Where the edge lives” strip. The classic discovery: all the profit sits in one or two buckets while expensive entries steadily lose → add an entry-price filter and rerun (see the teaching example).
Expanding a bucket row: “Edge when held to resolution” (had you held every trade to expiry) and “Exit-rule contribution” — a negative contribution means your takes and stops take away money versus simply holding.
The other charts
| Chart | How to read |
|---|---|
| P&L distribution | many small wins + a contained left tail = healthy; one giant spike — don’t trust the total |
| MAE vs MFE scatter | a cluster of huge-MAE winners = “sat-through” trades, the stop isn’t working |
| Hold time by exit reason | the strategy’s character: take-profit scalping vs holding to expiry |
| P&L by hour (UTC) / weekday | all profit inside 2–3 hours? — restrict the trading window |
| Performance by regime | regimes from spot: volatile (>1.8× median vol), trending_up/down (±1% over 20h), else ranging; a trend strategy must earn in trending |
| Win rate by hour | green ≥55% · amber 45–55% · red <45%; ignore hours with single-digit n |