估值模型 1c9067e
Commit：1c9067e3ae0663816503acaaae205910374a4fbb
LLM：openai/gpt-6.1-sol。
此版本統一收錄版本制度建立時正在採用的研究；歷史各次研究的原始 LLM 未逐筆追溯。
估值方法
以共同估值日的獲利與合理倍數建立 Bear／Base／Bull 情境，依經營結果的相對可能性加權；同版本、同目標日的公司分別排名。
加權目標價 = Σ（情境機率 × 情境目標價）；預期報酬 = 加權目標價 ÷ 現價 − 1。
估值指令
- Read the report chunks for demand, customer/project progress, capacity, product mix, margins and earnings revisions. Use sentiment as context for expectations alongside business evidence.
- Describe plausible Bear/Base/Bull business outcomes, choose earnings and valuation anchors supported by the reports, and explain the chosen targets briefly. A coherent scenario set from one report can be the starting point.
- Compare broker assumptions before choosing anchors; align earnings periods and target dates. Different broker targets alone do not define scenarios.
- Use the actual valuation method, such as EPS × P/E. When a report supplies scenario targets without a full earnings/multiple decomposition, cite the targets and their business assumptions; leave unavailable decomposition in model detail.
- Assign coarse probabilities from the research judgement and explain the relative weighting in one sentence. There is no fixed default weighting. Distinguish report forecasts from researcher assumptions in the model/source detail.
- Keep the workflow lightweight: include inputs that affect the valuation, with evidence gaps recorded once where relevant.