# Idea 004: AUDUSD Multi-timeframe trend filter

## Summary

Document a PyP Quant strategy candidate for AUDUSD on 1m candles using a RandomForestClassifier approach focused on multi-timeframe trend filter.

## Market Hypothesis

AUDUSD may show repeatable behavior when multi-timeframe trend filter conditions align with volatility, trend, and recent structure filters. The first implementation should stay conservative and treat this as an educational research candidate, not a production trading recommendation.

## Candidate Features

- Recent return over 3, 6, and 12 bars.
- ATR-normalized candle range and close location value.
- Rolling volatility percentile over 50 and 200 bars.
- Distance from EMA 20, EMA 50, and EMA 200.
- Session flag or market-hours bucket where applicable.
- Prior swing high and swing low distance.

## Label Design

- Predict whether the forward 1m move exceeds an ATR-normalized threshold.
- Use neutral labels when the forward return is too small to justify risk.
- Tune the horizon separately for trend, range, and scalp variants.

## Risk Controls

- Require minimum confidence before emitting UP or DOWN.
- Block signals during abnormal spread, missing candles, or low-liquidity windows.
- Cap exposure per instrument and stop after a daily drawdown limit.
- Validate stop-loss and take-profit levels with PPE simulation before live use.

## Backtest Checklist

- Compare against HOLD and simple moving-average baselines.
- Run walk-forward splits with no future leakage.
- Inspect trade count, profit factor, max drawdown, win rate, and average adverse excursion.
- Review behavior by session, volatility regime, and weekday.

## Implementation Notes

This idea is intentionally Markdown-only. A future template can add `strategy.py`, `quant.config.json`, and a focused README once PPE results justify turning the idea into executable code.

Closes #7