|
| 1 | +# Data Source and Factor Roadmap |
| 2 | + |
| 3 | +[English](data_factor_roadmap.md) | [简体中文](data_factor_roadmap.zh-CN.md) |
| 4 | + |
| 5 | +## Current Direction |
| 6 | + |
| 7 | +`QuantAdvisorResearch` should remain the final recommendation repository instead |
| 8 | +of becoming a full multi-factor trading platform. |
| 9 | + |
| 10 | +Two different research paths should stay separated: |
| 11 | + |
| 12 | +- Backtestable/executable path: price, technical, momentum, volatility, snapshot, |
| 13 | + and strategy repositories that may eventually connect to broker platforms. |
| 14 | +- Event/policy/news/AI-shadow path: less stable evidence that should only produce |
| 15 | + non-personalized recommendation reports and review artifacts. |
| 16 | + |
| 17 | +For now, this repository consumes only: |
| 18 | + |
| 19 | +- `PoliticalEventTrackingResearch` for source events and watchlists; |
| 20 | +- `ResearchSignalContextPipelines` for medium-horizon theme context and |
| 21 | + long-horizon AI shadow context. |
| 22 | + |
| 23 | +`UsEquitySnapshotPipelines`, `UsEquityStrategies`, `CryptoSnapshotPipelines`, and |
| 24 | +`CryptoStrategies` remain independent reference material until there is enough |
| 25 | +live evidence to justify a separate integration. |
| 26 | + |
| 27 | +## Current Inputs |
| 28 | + |
| 29 | +### PoliticalEventTrackingResearch |
| 30 | + |
| 31 | +Owns the event evidence layer: |
| 32 | + |
| 33 | +- official or semi-structured records; |
| 34 | +- RSS/Atom feeds from durable public sources; |
| 35 | +- alias-based ticker extraction; |
| 36 | +- event study tooling for later review. |
| 37 | + |
| 38 | +Stable default sources should be official records, issuer releases, regulatory |
| 39 | +feeds, and other replayable primary sources. X, Truth Social, Longbridge login |
| 40 | +sessions, and community content are excluded from the stable default pipeline |
| 41 | +until they have reliable interfaces, clear permission boundaries, and saved |
| 42 | +point-in-time artifacts. |
| 43 | + |
| 44 | +### ResearchSignalContextPipelines |
| 45 | + |
| 46 | +Owns reusable signal context: |
| 47 | + |
| 48 | +- medium-horizon theme momentum (`2-12 weeks`); |
| 49 | +- long-horizon AI shadow artifacts (`1-3 years`); |
| 50 | +- static theme taxonomy and symbol exposures; |
| 51 | +- saved `latest_signal.json` and `signal_history/*.json`; |
| 52 | +- replay based on saved artifacts only. |
| 53 | + |
| 54 | +This repository can provide background regime, theme, and risk context, but it |
| 55 | +must not directly generate orders, target weights, or account actions. |
| 56 | + |
| 57 | +### QuantAdvisorResearch |
| 58 | + |
| 59 | +Owns final non-personalized model recommendations: |
| 60 | + |
| 61 | +- inputs: event CSV, watchlist CSV, saved AI shadow JSON, optional theme momentum; |
| 62 | +- outputs: JSON, Markdown, HTML, RSS, and optional Telegram summary; |
| 63 | +- contract blocks orders, target weights, target share quantities, broker routing, |
| 64 | + account information, and suitability claims. |
| 65 | + |
| 66 | +## Factors to Add Later |
| 67 | + |
| 68 | +Priority order: |
| 69 | + |
| 70 | +1. Primary policy and disclosure sources: SEC, issuer IR, White House, Federal |
| 71 | + Register, Congress, DoD/DOE/CHIPS, Treasury, USAspending, or SAM.gov. |
| 72 | +2. Verified official social media: government, issuer, and executive accounts |
| 73 | + only when replayable and clearly attributable. |
| 74 | +3. Financial media leads: low-confidence discovery only, never high-confidence |
| 75 | + recommendation evidence without primary-source confirmation. |
| 76 | +4. Market confirmation: relative returns, abnormal volume, trend state, |
| 77 | + drawdown, volatility, and sector-relative moves. |
| 78 | +5. Fundamentals and valuation: market cap, revenue growth, margins, leverage, |
| 79 | + earnings dates, and valuation bands. |
| 80 | +6. Macro/risk regime: VIX, rates, dollar, credit spreads, oil, yield curve, and |
| 81 | + sector beta. |
| 82 | + |
| 83 | +## Low-Risk Implementation Order |
| 84 | + |
| 85 | +1. Keep public output focused on final recommendations. Preserve |
| 86 | + `theme_first_candidates[]` as JSON/Markdown explanation and audit material, |
| 87 | + not as a public buy list. |
| 88 | +2. Improve stable real sources in `PoliticalEventTrackingResearch`: RSS, official |
| 89 | + releases, SEC/EDGAR, company IR, policy/procurement sources, alias maps, and |
| 90 | + source registry coverage. |
| 91 | +3. Add optional market confirmation CSVs while keeping report generation working |
| 92 | + when the data is absent. |
| 93 | +4. Add event review inputs for 1/5/20/60 trading-day follow-up. |
| 94 | +5. Add fundamentals/valuation snapshots for risk explanation, not execution. |
| 95 | +6. Only then consider read-only references from existing snapshot repositories. |
| 96 | + |
| 97 | +## Anti-Overfitting Rules |
| 98 | + |
| 99 | +Long-lived advisory research should not chase only the current AI trade. |
| 100 | + |
| 101 | +Use static, versioned taxonomy files in `ResearchSignalContextPipelines`: |
| 102 | + |
| 103 | +```text |
| 104 | +config/theme_taxonomy.csv |
| 105 | +config/symbol_theme_exposure.csv |
| 106 | +``` |
| 107 | + |
| 108 | +Rules: |
| 109 | + |
| 110 | +1. Fix theme membership first, then observe future behavior. |
| 111 | +2. AI may output theme bias and shadow context, but not position sizes. |
| 112 | +3. Advisor may use theme bias and theme momentum as explanation inputs for final |
| 113 | + recommendations; theme candidates remain audit material by default. |
| 114 | +4. Every taxonomy, universe, and scoring-rule change must be versioned. |
| 115 | +5. Do not change weights just because MU, INTC, DELL, or any other name is |
| 116 | + currently popular. |
0 commit comments