How a seven-agent trading swarm works.

Seven agents look for different evidence

The system does not ask one model for a single prediction. It combines seven signal agents: SMA crossover, RSI, momentum, Bollinger Bands, volume and on-balance volume, MACD, and headline sentiment.

Each agent returns an action and confidence. The result is a swarm consensus rather than a single indicator presented as certainty.

• SMA Crossover compares 20-day and 50-day moving averages.
• RSI checks overbought, oversold, and divergence conditions.
• Momentum measures a 10-day price rate of change.
• Bollinger Bands evaluate price relative to two-sigma bands.
• Volume and OBV look for confirmation or divergence.
• MACD evaluates the MACD and signal-line relationship.
• Sentiment scores Yahoo Finance RSS headlines.

Weights adapt after results

The README says every agent begins with a weight of 1.0. An agent that was right has its weight multiplied by 1.1, capped at 2.0. An agent that was wrong has its weight multiplied by 0.9, floored at 0.5.

This is a simple feedback mechanism. It lets the system favor signals that have recently been more reliable while preventing any weight from growing without limit or disappearing completely.

Backtests are evidence, not guarantees

The public README reports one-year walk-forward tests for several tickers. Walk-forward testing is designed so each historical decision uses information available at that time.

Historical tests can still overfit, miss trading costs, or fail under different market conditions. The repository supports simulation and optional paper trading. None of this establishes future profitability.

The correct way to describe it

Trading Swarm is best described as a public multi-agent market-signal experiment. It demonstrates signal combination, adaptive weights, scheduling, alerts, and a dashboard.

It should not be described as financial advice, a guaranteed trading system, or proof of investment performance.