HK Stock Predictor
Research a Hong Kong listed stock or theme, then turn the analysis into clear, time-bounded predictions.
Minimal Input
User can provide any of:
{
"symbol": "00700",
"horizon": "30d"
}
{
"theme": "南向资金持续流入的恒生科技成分股",
"horizon": "14d"
}
Agent Normalization
Normalize input before analysis:
symbol: keep the 5-digit HK code when possible, also map to .HK form for external data.companyName: resolve from ticker or user text.horizon: normalize to one of 5d|14d|30d|90d|event.benchmark: default to HSI or HSTECH for tech-heavy names.predictionType: choose direction|range|event|relative-performance.
Preferred Data Stack
If these skills are available, use them in this order:
akshare-skill for broad market and company data.hk-stock-analysis for HK-specific analysis framework.cross-border-flow-tracker for southbound flow and positioning.market-overview for index and sector context.
Fallback sources if the skills are not installed:
- HKEX / HKEXnews
- AAStocks
- ET Net
- AASTOCKS southbound / short selling pages
- public financial data providers with clear source attribution
Deterministic Workflow
- Validate the target.
- Reject symbols that cannot be resolved to a Hong Kong listed security.
- If the user gives only a theme, narrow to 3-10 HK candidates first.
- Build market context.
- Capture
HSI, HSCEI, and HSTECH direction.
- Note sector rotation, overnight macro drivers, and any major policy headline.
- Gather company facts.
- Current price, market cap, 52-week range, valuation, earnings date, lot size, daily turnover.
- Gather HK-specific signals.
- Southbound net buy/sell trend.
- Short selling ratio and unusual changes.
- AH premium or discount if dual-listed.
- Liquidity warning if turnover is weak or spread is wide.
- Build the thesis.
- State
bull, base, and bear cases.
- For each case, list 2-4 drivers and 1-2 invalidation signals.
- Convert thesis into forecastable statements.
- Use binary, range, or relative-performance formats.
- Every statement must be time-bounded and externally resolvable.
- Rank prediction candidates.
- Prefer high observability, low ambiguity, and direct catalyst linkage.
- Avoid questions that require subjective wording such as "表现好不好".
- Return the analysis and top prediction candidates.
Prediction Design Rules
Good prediction candidates:
- "Will 00700 close above HK$520 on or before 2026-04-30?"
- "Will 09988 outperform the Hang Seng Tech Index between now and 2026-05-15?"
- "Will 02318 report YoY net profit growth above 10% in the next earnings release?"
Avoid:
- vague targets without dates
- subjective wording
- multi-condition questions that resolve unclearly
- tiny illiquid stocks where resolution may be distorted
Output
Return one structured object plus a readable summary.
{
"ok": true,
"normalizedInput": {
"symbol": "00700",
"symbolYahoo": "0700.HK",
"companyName": "Tencent Holdings",
"horizon": "30d",
"benchmark": "HSTECH",
"predictionType": "direction"
},
"marketContext": {
"indices": [],
"sectorTone": "",
"macroDrivers": []
},
"evidence": {
"fundamental": [],
"technical": [],
"flow": [],
"hkSpecific": []
},
"scenarioAnalysis": {
"bull": [],
"base": [],
"bear": []
},
"predictionCandidates": [
{
"title": "",
"type": "direction|range|event|relative-performance",
"deadlineIsoUtc": "",
"resolutionSource": "",
"confidence": 0,
"why": []
}
],
"recommendedPrediction": {
"title": "",
"deadlineIsoUtc": "",
"confidence": 0,
"keyRisks": []
},
"warnings": []
}
Summary Template
# [公司名称] ([代码].HK) 港股推演
## 核心判断
- 方向判断:
- 时间窗口:
- 相对基准:
## 关键证据
- 基本面:
- 技术面:
- 资金面:
- 港股特有因子:
## 三种情景
- Bull:
- Base:
- Bear:
## 可预测题目
1. [候选题目 1]
2. [候选题目 2]
3. [候选题目 3]
## 首选题目
- 题目:
- 截止时间:
- 置信度:
- 主要风险:
If The User Wants Gougoubi Conversion
Convert the top prediction candidate into Gougoubi-ready fields:
marketName: use the selected prediction title.deadlineIsoUtc: use the prediction deadline.rules: include exact resolution source, timezone, comparison field, and tie handling.tags: include hong-kong-stocks, sector tag, catalyst tag, and horizon tag.
Boundaries
- Do not claim "all HK stocks" were checked unless the scan actually covered the full universe.
- Do not hide missing data. Surface gaps in
warnings.
- Do not give investment advice phrased as certainty.
- Prefer liquid names and observable events when generating prediction questions.
