About Q Signals
Q Signals is a free, open research tool designed to help individual investors and students of the market explore stock signals, understand model-based scoring, and learn how algorithmic analysis can approach stock evaluation. It is built and operated by an independent developer based in Massachusetts and is offered to the public for research and educational use only.
What Q Signals Does
Q Signals runs a 23-module analysis pipeline against any ticker in its coverage universe of 150+ stocks across 15 market sectors. Each module independently scores the stock on a scale from -1.0 (strongly bearish) to +1.0 (strongly bullish), then a two-gate decision system combines those scores into a final BUY, HOLD, or SELL recommendation alongside price targets and risk parameters.
The 23 modules span five categories:
- Technical analysis — Fibonacci retracement and extension levels, Elliott Wave pattern detection, swing strategy overlays incorporating RSI, Bollinger Bands, and SMA crossovers, moving average alignment, and 52-week high/low proximity scoring.
- Market microstructure — Volume profile and on-balance accumulation/distribution, multi-timeframe trend alignment across daily and weekly bars, and ATR-based volatility regime classification (calm, elevated, extreme).
- Sentiment and news — News headline NLP sentiment scoring via natural language processing, social media mention velocity and directional bias from Reddit and StockTwits, Federal Reserve minutes hawkishness/dovishness scoring, and VIX-based market fear regime classification.
- Fundamental and institutional signals — SEC insider transaction net direction, institutional holdings change signals from 13F filings, Wall Street analyst consensus and price target upside, Congressional STOCK Act trade disclosures, and macroeconomic indicators including yield curve spread, employment trend, and CPI proxy signals.
- Peer and sector context — Competitor relative-strength scoring, sector rotation momentum ranking across 15 sectors, cross-asset correlation regime shift detection, and supply chain stress scoring based on SEC filing relationships.
How the Signal Decision Works
Q Signals uses a two-gate decision system. Both gates must pass simultaneously for a BUY or SELL to be issued:
- Composite score threshold — The weighted composite score must exceed +0.10 to issue a BUY, or fall below -0.10 to issue a SELL. A score between those bounds produces a HOLD regardless of other factors.
- Vote consensus gate — At least 60% of contributing modules must independently agree on the signal direction. A high composite score driven by just a few dominant modules while others disagree will be held back as HOLD. This gate prevents weak consensus from triggering signals.
This conservative design is intentional. In the context of research tools, false signals are more harmful to decision quality than missed opportunities. The tool is tuned to require meaningful, broad-based agreement before flagging an actionable signal.
Reinforcement Learning Agent
Q Signals includes a Deep Q-Network (DQN) reinforcement learning agent built with PyTorch. The agent architecture uses a 10-feature input (the 10 offline-computable module scores), two 128-neuron hidden layers with ReLU activation, and a 3-output layer representing Q-values for BUY, HOLD, and SELL actions — 18,307 trainable parameters total.
The DQN uses experience replay (10,000-step buffer, 32-sample minibatches), a soft-target network with Polyak averaging (tau=0.01), Huber loss for gradient stability, and epsilon-greedy exploration starting at 15% and decaying toward 1%. Weights are serialized as base64-encoded PyTorch state dictionaries and persisted to Supabase after every session, allowing the agent to accumulate learning across all users and sessions over time.
The RL agent contributes its learned signal direction as one of the 23 module inputs — it does not override the two-gate system but earns influence through its historical accuracy like any other module.
Price Targets and Hold Duration
Every BUY or SELL signal includes three price targets: conservative (T1), moderate (T2), and aggressive (T3), each computed from the stock's Average True Range scaled by horizon-specific multipliers. Stop-loss distance also derives from ATR. Hold duration is computed per-stock from the ATR-to-price ratio — higher-volatility stocks get shorter suggested hold windows, which scales naturally with market conditions rather than using fixed calendar windows.
Three investment horizons are supported: Short-Term (days to 2 weeks), Mid-Term (2 weeks to 3 months), and Long-Term (3 months to 1 year+). Each uses different ATR multipliers and evaluation windows.
Adaptive Module Weighting
As signal outcomes accumulate in Supabase, each module's historical hit rate is tracked. Modules that consistently contribute to correct predictions receive an effective weight boost through the formula: effective_weight = base_weight × (0.5 + hit_rate). Modules that underperform lose influence proportionally. This keeps the weighting system self-tuning as more real-world outcome data accumulates.
Coverage Universe
The default universe covers 150+ tickers across Technology, Semiconductors, Banking, Automotive, Industrials/Defense, Mining, Energy, Healthcare, Consumer, Real Estate, Utilities, Telecom, Indices/ETFs, and MegaCap sectors. Users can add custom tickers for research purposes.
Who Built This
Q Signals is built and operated by an individual developer based in Massachusetts. It is offered free to the public as a research and learning tool. It is not affiliated with any broker, financial institution, fund, or investment firm. The developer holds no investment advisory registration and does not provide personalized financial advice.
Q Signals is for educational and research purposes only. Nothing on this site constitutes investment advice or a recommendation to buy, sell, or hold any security. All trading and investing involves substantial risk of loss.