Source: research/002-ai-chain-full-market-leaders/research-002-method-plan.md
Last modified: 2026-06-04 15:45:37 · 3,811 bytes

Mini-Cypress Research Series #002 Method Plan

Objective

#002 expands #001 from P1 holdings to the full-market A-share AI-chain leader set, then adds the first four follow-up research directions from #001 section 9:

1. stock-specific pre-breakout anchor, using the trading day before the first 60D breakout;

2. expanded historical cycle samples, beyond the 2020-2021 new-energy sample;

3. speed comparison between earnings realization and PE expansion;

4. top-signal matching 1-3 months before historical peaks.

The fifth item, V_R regime-conditional study, is excluded from #002 and reserved for a later v1.6 risk study.

First-stage Universe

The universe is rule-based, not P1-holding based:

- Base data: Task #5 all-market technical prefilter and strict weekly candidate cache through 2026-06-03.

- Industry filter: SWS L2/L3 AI-chain scope, using the 2026 SWS snapshot.

- Included SWS L2: 半导体, 通信设备, 元件, 消费电子, 光学光电子, 电子化学品Ⅱ, 计算机设备, 软件开发, IT服务Ⅱ, 自动化设备.

- Explicit exclusions at this stage: 电池, 光伏设备, 风电设备, 电网设备. These may be studied as separate power-electronics/new-energy adjacencies, but they should not dilute the AI-chain report.

Leader Tiers

Two leader tiers are used so the report can keep both price leadership and fundamental quality visible:

- strict_leader: AI-chain stock with at least one Task #5 strict weekly candidate row since 2025-06-30. This means technical trend/RS, basic earnings quality, liquidity, and non-ST gates passed.

- technical_leader_fundamental_caveat: AI-chain stock that did not pass the strict fundamental gate but has leader-like price behavior since 2025-06-30: tech_score >= 3.15, rs_pct_batch >= 0.95, trend_count >= 7, and 20D traded value >= RMB 80m.

The second tier is included because AI-chain cycles often include turnaround, loss-to-profit, or heavy-capex names where strict trailing fundamentals lag the price cycle. It must be analyzed separately from high-quality strict leaders.

Anchor Rule

For each selected stock:

- Search window starts on 2025-04-07 and ends on 2026-06-03.

- First 60D breakout date is the first date where close > prior 60-trading-day high * 1.005.

- Anchor date is the previous trading day before that breakout.

- Main decomposition period is anchor date to 2026-06-03; peak diagnostics also use anchor date to each stock's cycle price peak after 2025-06-30.

Decomposition Rule

The next enrichment step fetches pe_ratio_ttm and market_cap for anchor, current/end, and cycle price peak dates.

For positive-PE comparable rows:

- price/market-cap return is decomposed into PE multiple expansion and earnings proxy expansion;

- earnings_proxy = market_cap / pe_ratio_ttm;

- loss-to-profit, negative-PE, and extreme low-base cases are not forced into the same formula. They are tagged separately.

Historical Cycle Plan

#002 should compare the current AI-chain cycle against at least:

- 2020-2021 new-energy leaders, using the completed #001 reference sample;

- 2017-2018 growth/manufacturing cycle, selected with the same all-market leader rules where data coverage permits;

- 2014-2015 growth bull sample, selected with the same price/RS framework, with fundamental and northbound caveats explicitly marked where coverage is incomplete.

Top-signal Plan

For each historical sample, inspect 1-3 months before the stock-level or basket-level peak:

- PE expansion percentile and speed;

- earnings proxy growth slowdown or failure to keep pace with price;

- trend-count deterioration / loss of new-high behavior;

- liquidity and turnover spike/fade;

- northbound/fund-flow proxies where available.

The output should be a CIO-facing verdict table, not only a factor dump.