Multi Composite Scoring Based on Neural Network Discoveries

Introduction

When neural network research starts feeding back into discretionary and rule-based trading, the results can be surprisingly elegant. This strategy distils insights from machine learning experimentation into a composite mean reversion system that layers RSI extremes with high-quality candlestick reversal patterns on AUDCAD 15-minute charts. The approach is directionally neutral — going long at oversold bounces confirmed by Bullish Pin Bars, and short at overbought rejections confirmed by Evening Star formations. It is designed to perform best in range-bound or choppy market conditions where price oscillates between defined extremes rather than trending persistently in one direction.

The Australian and Canadian dollars are both commodity-linked currencies, and in early 2026 commodity markets have been under pressure as global trade uncertainty mounts. With the US Federal Reserve holding rates steady and the Reserve Bank of Australia navigating a delicate inflation path, AUDCAD has been trading in a well-defined range — precisely the environment where mean reversion strategies carry their strongest edge. Traders watching the pair have noted that intraday swings are reliably snapping back to the mean after emotional overextension, creating repeatable setups for rules-based systems like this one.

The Anatomy of the Trade

The Logic: What Inefficiency Are We Exploiting?

Two commodity currencies from similar economic cycles tend to track each other closely over the medium term. When short-term sentiment pushes AUDCAD to statistical extremes, there is a strong structural pull back toward equilibrium. The RSI quantifies that extension objectively, flagging moments when momentum has overshot. But RSI alone is notoriously unreliable as a standalone trigger — it can stay pinned at extremes during momentum-driven moves. The candlestick confirmation layer addresses this directly: we don’t trade every RSI extreme, only those where price itself provides a visual signal of shifting supply and demand. This confluence of oscillator and price structure is where the genuine edge lives.

The Bullish Pin Bar for long entries represents buyers aggressively rejecting lower prices within a single candle — wicking deeply into oversold territory before closing near the high. The Evening Star for short entries is a three-candle reversal pattern signalling that a rally has exhausted itself and sellers have taken control. Together with an RSI cross through an extreme threshold, these patterns represent the highest-probability confluence available on a 15-minute chart without introducing lag from additional indicators.

Setup Requirements

• Primary Indicator: RSI with default settings (14 periods), using 30 and 70 as the oversold and overbought thresholds
• Long Confirmation: Bullish Pin Bar candlestick pattern completing at or after the RSI cross above 30
• Short Confirmation: Evening Star three-candle formation completing at or after the RSI cross below 70
• Risk Management: ATR (14 periods) for dynamic, volatility-adjusted stop placement
• Primary Symbol: AUDCAD — a correlated commodity-currency pair with tight spreads and reliable mean-reversion behaviour, not recently used in the rotation
• Timeframe: 15-minute — optimal balance between signal frequency and noise, giving RSI readings statistical significance without excessive lag
• Adaptability: The same logic applies well to other correlated commodity-currency pairs such as AUDNZD or CADJPY during ranging market phases

Entry Rules

All conditions must align before entering. A partial setup — RSI cross without the candlestick pattern, or the pattern without the RSI trigger — is not a valid trade.

• Long Entry: RSI crosses above 30 and a Bullish Pin Bar candle completes on the 15-minute chart
• Short Entry: RSI crosses below 70 and an Evening Star formation completes on the 15-minute chart

Enter at the close of the confirmation candle. For the Evening Star this means waiting for the third candle to fully form before placing the order. Anticipating the pattern risks entering on an incomplete signal that may not materialise.

Exit Rules

• Stop Loss: 1.5 × ATR from the entry price, placed below the Pin Bar wick for longs and above the Evening Star high for shorts
• Take Profit: Minimum 2:1 reward-to-risk ratio — at least twice the distance of the stop loss from entry
• Secondary Exit: RSI reaches the opposite extreme (above 70 for longs, below 30 for shorts) or divergence between price and RSI is detected, signalling momentum exhaustion ahead of the target

The stop loss is non-negotiable. AUDCAD is a low-volatility pair but it can trend aggressively during commodity shocks or central bank surprises. An ATR-based stop ensures you are not swept out by ordinary intraday noise while still protecting against a genuine directional breakout that invalidates the mean reversion premise.

Risk Management

• Risk Per Trade: Maximum 1–2% of account equity per position
• Risk-to-Reward Ratio: Minimum 2:1, targeting 3:1 when the RSI is deeply extended (below 25 or above 75) and the candlestick pattern is textbook quality
• Position Sizing: Calculate units as (Account Risk $) ÷ (1.5 × ATR in price). For example, with a $10,000 account risking 1% ($100) and an ATR of 0.0008: $100 ÷ 0.0012 = ~83,333 units (0.83 standard lots)
• Maximum Concurrent Positions: 2 trades maximum to limit correlated exposure across commodity-currency pairs

SYMBOL:       AUDCAD

TIMEFRAME:    15m

LONG ENTRY:  RSI crosses above 30
              AND Bullish Pin Bar pattern completes

SHORT ENTRY: RSI crosses below 70
              AND Evening Star formation completes

STOP LOSS:   1.5 × ATR from entry

TAKE PROFIT: 2:1 minimum reward-to-risk
              // Or RSI reaches opposite extreme

RISK:        1-2% per trade, max 2 concurrent positions

Copy the above rules into your Arconomy Strategy Designer. See Working with Strategy Notes for guidance on adding a Strategy Note to your rule set.

Common Pitfalls

Mean reversion strategies on correlated currency pairs carry specific failure modes that differ from high-volatility assets. Understanding them before going live is as important as knowing the entry rules.

Ignoring Extended Trending Conditions

AUDCAD can trend persistently for weeks when commodity cycles diverge — for instance when iron ore prices move independently from oil. During such phases, RSI will signal oversold or overbought repeatedly while price continues in the same direction. If price is making consecutive higher highs or lower lows over several hours, pause all mean reversion entries regardless of the RSI reading.

Trading Through High-Impact Macro Events

The Australian and Canadian dollars react sharply to central bank rate decisions (RBA and BoC), commodity price shocks (oil and iron ore), and Chinese economic data releases. A single CPI surprise can push AUDCAD through ATR-based stops in seconds. Check the economic calendar before each session. If a Tier 1 event is scheduled within four hours, skip the setup entirely — the risk/reward calculus changes fundamentally around news.

Overtrading and Skipping the Candlestick Filter

After a string of clean setups, the temptation is to trade every RSI extreme without waiting for the Pin Bar or Evening Star to complete. Resist this entirely. The candlestick confirmation is not an optional bonus — it is the mechanism that separates genuine reversals from momentum traps. Removing it transforms a selective edge into random noise.

Curve-Fitting RSI Thresholds

Default RSI parameters (14 periods, 30/70 thresholds) exist because they capture genuine statistical extremes across a wide range of market conditions and instruments. It is tempting to adjust thresholds to 25/75 or optimise the period based on recent backtest performance. Over-fitting to historical data produces strategies that fall apart in live conditions. Stick with defaults unless you have a compelling structural reason — and test any change across at least 12 months of data before adopting it.

Revenge Trading After Drawdowns

Mean reversion strategies on low-volatility pairs like AUDCAD experience long stretches of small wins followed by occasional outsized losses when a trending move catches open positions. After two consecutive stops, step back and reassess whether the market has shifted from range-bound to trending. Doubling position size after a losing trade to recover losses is the single fastest path to a blown account. Discipline in drawdown is what separates systematic traders from gamblers.

Build Strategy using Arconomy

The Multi Composite Scoring Based on Neural Network Discoveries strategy can be assembled inside the Arconomy Strategy Designer using the following rule configuration — no coding required.

Backtest Considerations

Run this strategy across a minimum of six months of AUDCAD 15-minute data, ensuring the sample includes both clearly ranging periods and at least one sustained trending phase. Testing only on a range-bound window will inflate results dramatically. If possible, include the commodity-shock volatility periods of 2024–2025 as a stress test against the mean reversion assumption — the strategy should show reduced trade frequency and wider drawdowns during those windows, which is expected and healthy behaviour.

Key metrics to watch: profit factor should exceed 1.3 on an honest backtest with realistic spreads; maximum drawdown should stay below 15% of peak equity; and trade distribution should show the strategy sitting on the sidelines during trending phases rather than churning losing trades. A lopsided distribution of losses clustered in a short window likely indicates a trending period where the filter failed. Consult the Arconomy Backtesting Guide for methodology and metric benchmarks.

AUDCAD typically trades with spreads of 1.5–3 pips during London and New York sessions, widening to 4–6 pips during the Asian session when liquidity thins. Build in at least 2 pips of round-trip spread cost plus 0.5 pips slippage on stop orders in your backtest assumptions. On 15-minute charts with ATR-based stops, these costs are manageable, but they meaningfully impact strategies that trade frequently on small R multiples. Prefer setups where RSI is deeply extended (sub-25 or above 75) for larger expected moves that justify the transaction cost.

Credits

This strategy was shared by Artcheezy on r/algotrading.