Research Approach
Paradigm — Research Approach
A conceptual overview of how strategies are built, tested, and validated on Paradigm.
1. Philosophy
Durable edges in markets come from systematic research, not from black-box automation. We believe that language models are powerful tools for accelerating research — surfacing patterns, screening a large universe of candidates, and drafting testable hypotheses faster than a single analyst could. But acceleration is not a substitute for judgment.
Every position taken on Paradigm is the result of a human decision. The model advises; the human is accountable. This keeps the research loop grounded, auditable, and improvable over time — qualities that fully autonomous systems routinely sacrifice in exchange for speed.
2. The Research Loop
Every strategy moves through a defined loop before any capital — real or paper — is committed to it:
- 1
Human Pattern Recognition
Domain experts identify emerging sector themes and structural market shifts. Investment theses originate with a human.
- 2
Python Candidate Screening
Systematic quantitative filters screen the watchlist: volume surge detection, price structure analysis, ATR-normalised range, relative strength ranking.
- 3
LLM Hypothesis Generation
Language models synthesise sector research, earnings data, news flow, and technical positioning to generate directional trade hypotheses with explicit rationale.
- 4
Reinforcement Learning Validation
An RL model trained on historical signal outcomes continuously updates signal weights, penalising recency bias and rewarding regime-appropriate signals.
- 5
Human Final Approval
No position is opened without human review of the full hypothesis chain. The system surfaces; the human decides.
3. Why Paper First
Backtesting is a necessary condition for confidence — but it is not sufficient. Overfitting, look-ahead bias, and slippage assumptions can all make a strategy look better on historical data than it will perform in live markets. Paper trading is the bridge between a backtest and a real-money commitment.
On Paradigm, every strategy that clears backtesting enters a live paper portfolio tracked at market prices with realistic execution assumptions. Win rate, expectancy per trade, maximum drawdown, and average hold duration are all monitored continuously. The current live paper portfolio started at $100,000. All performance figures shown on the homepage reflect this paper portfolio — no real capital is at risk.
Only after a meaningful paper-trading sample — enough trades to have statistical significance across varying market conditions — will a strategy be considered for live deployment.
4. Pipeline Architecture & Monitoring
The system runs seven modular service stages in a defined sequence, each with a scoped responsibility and audit trail. All pipeline activity is logged with timestamps and signal chain. No stage can trigger trade execution without a valid watchlist entry and confirmed structure check.
Theme Detector
Scans sector ETF flows, news velocity, and social signal frequency to score emerging themes (0–1 confidence).
Watchlist Builder
Populates the candidate pool from ETF holdings and fallback sector universes. Refresh rate: 3-hour cycle outside market hours.
Structure Checker
Flags candidates with clean base patterns (tight consolidation, controlled volume). Rejects extended or climactic structures.
Breakout Scanner
Screens for volume-confirmed price breakouts above identified resistance levels. Requires directional volume — filters panic-selling spikes.
Trade Executor
Submits orders via Alpaca paper API. Position sizing uses ATR-based stops with predefined risk-per-trade limits.
Risk Manager
Runs every 5 minutes during market hours. Enforces stop-loss execution, maximum drawdown limits, and pyramid eligibility checks.
Health Check
16-point system diagnostic running post-scan. Broadcasts status to owner via WhatsApp notification.
5. Go/No-Go Protocol
Before transitioning from paper to live capital, the system must pass a multi-layer checklist. Only when all checks pass does the system flag LIVE_READY=true. Live deployment requires explicit owner action — it is never automatic.
Technical Checks
- All 7 pipeline stages operational with no FAIL status in the last 7 days.
- Database integrity verified — no orphaned positions, no status inconsistencies.
- Notification pipeline live-tested (WhatsApp delivery confirmed).
- API connectivity stable — Alpaca, Finnhub, data feeds — no degraded responses in 48h.
Strategy Checks
- Minimum 30 completed paper trades with positive expectancy.
- Win rate ≥ 50% on paper capital.
- Maximum drawdown on paper ≤ 8%.
- Average hold time consistent with strategy thesis (2–10 days).
- No single position responsible for > 40% of total return.
Operational Checks
- Owner has reviewed and approved the strategy parameters.
- Emergency stop (kill switch) tested and confirmed functional.
- Position size limits verified against available capital.
- Risk-per-trade hard cap reviewed.
6. Strategy Principles
The current strategies on Paradigm operate at a conceptual level around three principles. Specific parameters, thresholds, and entry criteria are not disclosed.
Sector-rotation momentum
Strategies focus on identifying sectors where capital is actively rotating and price momentum is accelerating. The assumption is that institutional flows create multi-week trends that can be systematically identified and exploited.
Catalyst-driven entry signals
Entries are not made purely on technicals. A confirmable catalyst — whether macro, earnings-driven, or structural — is part of the entry thesis. This reduces the chance of entering on noise.
Conviction and volatility-based position sizing
Position size is a function of two inputs: the researcher's conviction in the thesis, and the recent volatility of the instrument. Higher volatility reduces position size. Higher conviction may increase it, within defined limits.
7. What Paradigm Will Offer
The goal of Paradigm is to bring research-grade infrastructure to individual traders who currently lack access to institutional-quality tooling. The planned platform includes:
- Backtesting infrastructure with walk-forward validation, built in Python and accessible through a clean interface.
- LLM-assisted screening that surfaces candidates and generates annotated hypotheses for human review.
- Human-in-the-loop execution flows — the platform is designed to support the researcher, not to trade autonomously.
- Performance tracking across paper and live portfolios with consistent, auditable metrics.
Paradigm is currently in private research mode. If you want to be among the first to access the platform when it opens, join the waitlist.
This document describes the research approach used internally at Paradigm. It is not financial advice. All performance figures refer to paper-traded simulations. Past simulated performance does not guarantee future results.