Lab 4 Bot Checking…
Last bar:
Bot cockpit · why the bot fired (or didn't) on the last tick
no snapshot yet
Williams %R(14) Daily — entry signal
−100 (oversold) %R = — 0 (overbought)
Gates checked this tick
·Waiting for first cron tick…
About AUD_v1 — Lab 4 Williams %R(14) Daily
single-position scalper, walk-forward verified, K-fold robust

AUD_v1 trades AUDUSD using a single boolean entry signal: Williams %R(14) computed on Daily-resampled bars. When Williams crosses up through −50 (close moves to upper half of recent 14-day range) a LONG opens; when it crosses down through −50, a SHORT opens. Each position has a fixed take-profit, stop-loss, 1-week timeout, and Friday-21:00-UTC weekend force-close. No grid, no martingale, no compound-tricks — just a clean single-position bot whose only job is "trade in the direction of the daily-autocorrelation regime."

Strategy spec (live)

  • Pair: AUDUSD (Exness Cent #160001992)
  • Indicator: Williams %R period 14 on Daily-resampled bars
  • Entry: cross at −50, both directions (long on up-cross, short on down-cross)
  • Take-profit: 150 pips (was 200 — E46 walk-forward zoom found 150 better)
  • Stop-loss: 30 pips
  • Timeout: 168 H1 bars (1 week)
  • Weekend close: Friday ≥ 21:00 UTC
  • Risk per trade: 5% of broker equity (was 10% — reduced after K-fold revealed wider DD bands)
  • Magic number: 777401
  • Tier choice: Tier C of E52 K-fold ranking — most regime-robust

Backtest performance (22-yr AUDUSD H1)

  • Out-of-sample Calmar: 28.34 (vs deployed EMA's 4.3 = 6.6× lift)
  • Out-of-sample CAGR: +641% (4-year IS vs 11-year OOS at the 2014-12-31 split)
  • Out-of-sample max drawdown: 22.6%
  • Out-of-sample profit factor: 4.22
  • ~32 trades/year, ~57% win rate (most non-TP closes are also profitable timeouts)
  • K-fold across 6 regimes (2004-2026): all profitable, max DD bounded at 24% in every regime
  • K-fold Calmar CV: 0.54 (3-4× more regime-stable than aggressive tiers)

Why Williams %R, not EMA?

  • Tested 161 indicators across 17 phases (MA crosses, oscillators, channels, trend strength, statistical, Ehlers cycle, regime detectors, fractals, candle patterns, multi-timeframe, adaptive MAs)
  • None beat Williams %R Daily on out-of-sample Calmar
  • L1 logistic regression and Lasso decode confirmed Williams entries are not filterable ex-ante — the edge is irreducible
  • The "edge" is positive autocorrelation in AUDUSD daily returns, likely China/iron-ore commodity-cycle driven (multi-pair test confirmed AUDUSD is structurally 10-30× stronger than EUR/GBP/JPY/CAD/CHF on this pattern)

Audit trail

  • Strategy file: /home/ubuntu/live_trade/lab4_code/lab4_trader.py
  • State persistence: ~/live_trade/lab4/AUD_v1/state.json
  • Immutable event log: ~/live_trade/lab4/AUD_v1/events.csv
  • Cron log: ~/live_trade/lab4/AUD_v1/cron.log
  • Pre-Williams backup: lab4_trader.py.pre_williams.bak (2026-05-07 14:30 UTC swap)
  • Email alerts on entries, exits, errors via Gmail SMTP
Account balance
In trade
Direction
Total trades
Win rate
Cumulative P&L
Cumulative P&L curve
Account equity from $100 seed (10,000 USC), updated per closed trade
TP hits
+150 pips each
SL hits
−30 pips each
Timeouts
closed at week timeout
Weekend closes
closed Fri 21:00 UTC
Closed trades
most recent first
EntryExitDir Entry $Exit $ LotsReason P&LBars
Loading…
Recent events
last 30 cron-bar events from events.csv (immutable audit trail)
Time BKKEventIn pos?TicketLast bar
Loading…