Frequently Asked Questions
Common questions about how Poros Alpha works
General
What is Poros Alpha?
A quantitative investment platform. Factor-based research, regime-aware portfolio construction, strategy validation, and live trade execution, accessible through a conversational AI agent with 125 specialized tools.
How does it work?
Ask a question in plain language. The agent selects tools, evaluates data against current market conditions, and returns structured analysis with full reasoning visible. Complex queries delegate to specialist subagents (market, risk, fundamental, sentiment, general). Trade proposals go through risk validation and compliance checks before execution.
Is Poros free?
Yes. Free to use with supported AI models. Bring-your-own-key is supported for all providers.
What AI models are available?
Multiple providers: Anthropic, OpenAI, DeepSeek, Google, and Groq. Models can be switched per conversation. Bring-your-own-key is supported for any provider.
Trading & Portfolio
Which brokers are supported?
Alpaca, Interactive Brokers, and Trading 212. Each connects via OAuth or API key. Paper portfolios are available without a broker connection.
Can Poros execute trades automatically?
Three autonomy tiers tied to strategy lifecycle stage. Sandbox (research strategies): analysis only, no execution. Paper (candidate strategies): simulated fills, requires user approval. Live (approved strategies): executes within guardrails: blocked/allowed tickers, max position size, mandatory stop-loss. Every trade proposal includes Almgren-Chriss impact estimates and compliance validation regardless of tier.
How does portfolio construction work?
Five optimizers available: Hierarchical Risk Parity (primary), mean-variance, equal-weight, min-variance, and risk parity. HRP uses Ledoit-Wolf covariance shrinkage and Ward hierarchical clustering. Constraints enforced via clip-without-redistribute: position caps, sector caps, cash buffer, minimum position floor. Signal conviction scores optionally blend into target weights.
What risk controls exist?
Three independent circuit breakers: drawdown halt (portfolio-level max drawdown from peak), CVaR breach (95% expected shortfall), and correlation spike (average pairwise correlation). Breakers scale position sizes proportionally or halt execution entirely. Drift-triggered rebalancing fires when any position deviates beyond threshold.
Research & Data
What data sources are used?
Six external providers: Alpaca for real-time and historical quotes, Tiingo for OHLCV history and fundamentals, SEC EDGAR for regulatory filings, FRED for economic indicators, financial news sentiment, and arXiv for academic research. Prices are cached in Redis (5-min TTL) and stored in TimescaleDB for backtests and reduced API load.
What factor models are available?
Five cross-sectional factors: momentum (12-month minus 1-month return), value (P/E + P/B composite), quality (financial health metrics), low volatility, and carry (dividend yield). Scored with sector-neutral z-score normalization, winsorized at 3 sigma. Minimum 80% coverage required per factor. Factor weights tilt with the detected market regime.
How does strategy validation work?
Three layers. Layer 1 (viability): Sharpe ratio above 1.0, max drawdown below 30%, minimum 30 trades, 2+ years history, ADV participation below 10%. Layer 2 (walk-forward): 5-fold anchored analysis with 6-month OOS windows and 14-day embargo, all folds must show positive Sharpe. Layer 3 (deflated Sharpe): DSR above 0.95, correcting for multiple comparisons using Bailey & Lopez de Prado method. All three layers must pass before a strategy can go live.
How does regime detection work?
A 3-state Gaussian Hidden Markov Model trained on six features: SPY daily return, 21-day realized volatility, VIX proxy (VIXY z-score), credit spread (HYG minus LQD), yield curve slope (TLT/SHY ratio), and SPY-credit rolling correlation. Classifies into risk-on, risk-off, or crisis, with automatic recovery detection. Fallback to rule-based volatility regime if model unavailable.
Security & Privacy
Is portfolio data secure?
All connections use TLS encryption. Broker credentials are encrypted at rest. Authentication uses JWT with refresh token rotation. User data is fully isolated.
Is data shared with third parties?
Conversation queries are processed by the selected AI provider. No analytics trackers or advertising. The privacy policy details data handling.
Can data be deleted?
Full data rights supported: access, correction, deletion, and portability under GDPR and CCPA.