数据透明承诺Data transparency
这一节解释了模型的每一步——以及它做不到的事。
This section documents every step of the model — and what it cannot do.
v1.0.2 估值算法**真实使用**了 DEALS_DB 中 — 笔跨境授权交易(覆盖 2024-11 至 2026-02 公开披露数据,已剔除未披露金额条目)。其中约 50 笔为人工精选高质量条目(含完整临床阶段、备注),其余条目从公开数据批量摄入——**临床阶段标记为 `unknown`,药物类型 (modality) 通过药名规则推断**。可比交易回归与 rNPV 参数模型按 50/50 权重融合输出最终估值;当匹配样本 <3 时退化为纯 rNPV。提交数据 →
The v1.0.2 valuation algorithm **actually uses** the — cross-border licensing deals in DEALS_DB (publicly disclosed 2024-11 → 2026-02, with non-disclosed economics filtered out). About 50 entries are hand-curated with full clinical-stage detail and editorial notes; the remainder are bulk-ingested from public deal data — **clinical stage is recorded as `unknown` and modality is inferred from drug-name patterns**. Comp-regression and rNPV are blended 50/50; when fewer than 3 similar comps exist, the model falls back to pure rNPV. Submit data →
1 · 创新药 rNPV 模型(参数路径)1 · New drug rNPV (parametric path)
rNPV_licensor = PoS × Σ_t [Peak × Ramp(t) × ModPremium × CompMult × RoyaltyRate]
÷ (1 + WACC)^t
rNPV_asset = PoS × Σ_t [Peak × Ramp(t) × ModPremium × CompMult × NetMargin]
÷ (1 + WACC)^t
PoS · 阶段累计上市概率PoS · cumulative probability to approval
preclin 5% · Ph1 10% · Ph2 17% · Ph3 52% · NDA 88% · approved 100%
默认 WACC 11% · NetMargin 30% · 上市后 10 年专利期Default WACC 11% · NetMargin 30% · 10y post-approval patent runway
Ramp 曲线Ramp curve: 10/30/60/85/100/100/95/80/55/25 (% of peak by year)
ModPremium: sm 1.00 · mab 1.05 · bsab 1.25 · adc 1.40 · cart 1.50 · gt 1.55 · pep 1.10 · olig 1.20 · rc 1.45
CompMult: FIC 1.25 · me-too 1.00 · me-third 0.75
2 · 可比交易回归(数据路径)2 · Comparable-deal regression (data path)
// 在 DEALS_DB 中对每笔有披露金额的交易计算相似度权重// For each deal with disclosed economics, compute similarity weight
weight(d) = 0.05 (baseline)
+ 0.50 if d.indication == p.indication else 0.25 (adjacent) | 0.10 (same TA)
+ 0.30 if d.modality == p.modality else 0.10 × ratio(ModPremium)
+ 0.20 if d.stage == p.stage else 0.10 (±1) | 0.04 (±2)
// 仅保留 weight > 0.20 的样本; 若 N < 3, 跳过本路径// Keep samples with weight > 0.20; abort if N < 3
// 加权重采样后取分位数// Resample by weight, then take quantiles
total_implied = median(weighted_samples.total) × stage_correction × mod_correction × terr_factor
upfront_implied= median(weighted_samples.upfront) × (same correction)
P10 / P90 = 20%/80% quantiles of resampled distribution
stage_correction = StageMult(p.stage) / weighted_avg(StageMult(d.stage))
mod_correction = ModPremium(p.mod) / weighted_avg(ModPremium(d.mod))
terr_factor = 1.0 (global) / 0.85 (3+ regions) / 0.70 (2 regions) / 0.55 (single)
StageMult: preclin 0.35 · Ph1 0.7 · Ph2 1.0 · Ph3 1.6 · NDA 2.4 · approved 3.0
3 · 融合(blended) — 把两条路径合起来3 · Blended output — combining both paths
total_blended = 0.5 × rNPV_total + 0.5 × total_implied // N ≥ 3 时// when N ≥ 3
upfront_blended = 0.5 × rNPV_upfront + 0.5 × upfront_implied
P10 / P90 = union(rNPV_sensitivity, comp_quantile) // 取并集,更保守// take union, conservative
退化情况: 当 DEALS_DB 中匹配样本 < 3 时, 输出 = 100% rNPV(并在结果区显示退化提示)Fallback: when matched samples in DEALS_DB < 3, output = 100% rNPV (a fallback notice is shown in the result panel)
4 · 类似药利润分成模型4 · Biosimilar profit-share model
NPV_licensor_share = PoS × Σ_t [OrigPeak × (1 - Discount) × ShareRamp(t) × GM × ProfitShare]
÷ (1 + BS_WACC)^t
默认值:Defaults: share 35-45% · discount vs originator 25-35% · GM 55% · BS_WACC 9%
类似药 PoS:Biosim PoS: preclin 15% · Ph1 25% · Ph2 45% · Ph3 70% · NDA 85% · approved 100%
ShareRamp: 5/20/40/55/65/70 (% market share vs originator by year)
v1.0 类似药估值仅采用参数路径(不进行 comp-blended), 因为 DEALS_DB 中类似药交易样本极少。Biosim valuation uses the parametric path only (no comp-blending) in v1.0 — DEALS_DB carries too few biosim transactions to support the regression.
5 · 交易结构拆分5 · Deal structure decomposition
Upfront : Milestone : Royalty ≈ 8% : 40% : 52% (新药全球均值)(global new drug average)
按阶段首付占比 (参数模型):Upfront % by stage (parametric): preclin 4% · Ph1 7% · Ph2 11% · Ph3 18% · NDA 28% · approved 42%
royalty rate: Ph1 8-11% · Ph2 10-13% · Ph3 12-16% · Mkt 14-20%
P50 由参数+回归融合给出, P10/P90 由模型敏感性 ∪ 可比分位数 给出。P50 from blended path; P10/P90 = union of model sensitivity and comp quantiles.
6 · 买方匹配评分 (满分 100)6 · Buyer matching score (out of 100)
Score = 40·IndicationFit + 25·ModalityFit + 20·StageFit
+ 10·FinancialCapacity + 5·RecentActivity
- IndicationFit: 命中 1.0 / 邻接 0.6 / 同治疗领域 0.3 / 否则 0
- ModalityFit: 平台成熟 1.0 / 有管线 0.7 / 开放态度 0.4 / 否则 0.1
- StageFit: 偏好窗口命中 1.0 / 相邻 0.6
- FinancialCapacity: log(cash+revenue) 归一化
- RecentActivity: 近12月BD次数归一化- IndicationFit: direct hit 1.0 / adjacent 0.6 / same TA 0.3 / else 0
- ModalityFit: core platform 1.0 / active pipeline 0.7 / open stance 0.4 / else 0.1
- StageFit: preferred stage 1.0 / adjacent 0.6
- FinancialCapacity: log(cash+revenue) normalized
- RecentActivity: BD count last 12M, normalized
数据来源 & 已知局限Data sources & known limitations
数据来源Data sources
- PoS 默认值: BIO/QLS/Informa《Clinical Development Success Rates 2011-2020》行业基准。PoS defaults: BIO/QLS/Informa Clinical Development Success Rates 2011-2020 industry benchmark.
- 可比交易数据库 (DEALS_DB): 公司新闻稿、SEC 8-K/20-F 文件、BioPharma Dive / FiercePharma / Endpoints News 报道,人工录入,聚焦中国-海外跨境授权交易。当前 — 笔。Comp database (DEALS_DB): hand-curated from company press releases, SEC 8-K/20-F filings, BioPharma Dive / FiercePharma / Endpoints News reporting; focused on China–overseas cross-border licensing. Currently — deals.
- 买方财务: FY25 年报(MNC)或最近季报年化(中等市值);现金 = 现金+短期投资。Buyer financials: FY25 annual reports (MNCs) or latest quarterly annualized (mid-caps); cash = cash + short-term investments.
已知局限Known limitations
- 样本量小: 仅 — 笔可比交易,统计置信度有限,P10/P90 区间会偏宽。Small sample: only — deals — statistical confidence is limited; P10/P90 bands run wide.
- 选择偏差: 数据库偏向中国 license-out 交易,纯海外 in-licensing 覆盖较少。Selection bias: the database over-indexes China license-outs vs. pure overseas in-licensing.
- 未披露金额未纳入: 公开仅披露"undisclosed terms"的交易未参与回归。Undisclosed deals excluded: transactions with "undisclosed terms" are not included in regression.
- biobucks 膨胀: "总额"通常含发生概率较低的远期里程碑(销售里程碑常被高估)。Biobucks inflation: reported "total deal value" often inflates distant sales milestones with low realization probability.
- 类似药未做回归: 类似药交易在 DEALS_DB 中样本极少,v1.0 仅用参数路径。No biosim regression: too few biosim deals in DEALS_DB; v1.0 uses parametric only.
- 买方近期活跃度更新: 2025-Q4 起的并购/授权动作需要社区帮助持续刷新。Buyer recency: deal activity since 2025-Q4 needs ongoing community refresh.
本工具仅用于 BD 评估参考,不构成投资建议。所有源代码、参数默认值、公式实现均在 GitHub index.html 公开可见;欢迎挑战、修改、改进。
For BD evaluation reference only; not investment advice. All source code, parameter defaults and formula implementations are visible in index.html on GitHub. Challenge it, fork it, improve it.