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:

CODEBLOCK0

CODEBLOCK1

Agent Normalization

Normalize input before analysis:

• - symbol: keep the 5-digit HK code when possible, also map to .HK form for external data.
• INLINECODE2: resolve from ticker or user text.
• INLINECODE3: normalize to one of 5d|14d|30d|90d|event.
• INLINECODE5: default to HSI or HSTECH for tech-heavy names.
• INLINECODE8: choose direction|range|event|relative-performance.

Preferred Data Stack

If these skills are available, use them in this order:

1. 1. akshare-skill for broad market and company data.
2. INLINECODE11 for HK-specific analysis framework.
3. INLINECODE12 for southbound flow and positioning.
4. INLINECODE13 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

1. 1. 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.

1. 2. Build market context.

- Capture HSI, HSCEI, and HSTECH direction. - Note sector rotation, overnight macro drivers, and any major policy headline.

1. 3. Gather company facts.

- Current price, market cap, 52-week range, valuation, earnings date, lot size, daily turnover.

1. 4. 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.

1. 5. Build the thesis.

- State bull, base, and bear cases. - For each case, list 2-4 drivers and 1-2 invalidation signals.

1. 6. Convert thesis into forecastable statements.

- Use binary, range, or relative-performance formats. - Every statement must be time-bounded and externally resolvable.

1. 7. 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".

1. 8. 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.

CODEBLOCK2

Summary Template

CODEBLOCK3

If The User Wants Gougoubi Conversion

Convert the top prediction candidate into Gougoubi-ready fields:

• - marketName: use the selected prediction title.
• INLINECODE21: use the prediction deadline.
• INLINECODE22: include exact resolution source, timezone, comparison field, and tie handling.
• INLINECODE23: 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.