Built around a complete, working valuation model of Alibaba Group (BABA · 9988 HK). Every chart you see is real data from the workbook. Every concept is unpacked from the ground up — the way you wish someone had explained it the first time. Toggle the language anytime. 围绕阿里巴巴集团 (BABA · 9988 HK) 一份完整可运行的估值模型构建。您看到的每张图表都来自该模型的真实数据。每个概念都从零开始拆解 — 就像您当初希望别人为您解释的那样。随时切换语言。
Strip away the suits, the acronyms, and the Bloomberg terminals, and an investment bank is a remarkably old idea: a place that helps people who have money meet people who need money — and gets paid for the introduction. Everything else is detail. 脱去西装、缩写和彭博终端的外衣，投资银行其实是个非常古老的想法：一个帮助有钱的人遇见需要钱的人的地方 — 并因这次"介绍"而获得报酬。其余的，都是细节。
Imagine you start a tea-house in Hangzhou. Business is good. You want to open ten more. The problem: each new shop costs ¥1 million in equipment, deposits, and signs. You have ¥2 million. You need ¥8 million more. Who would lend it to you — or buy a slice of your future profits in exchange for the cash today?
想象您在杭州开了一家茶馆。生意红火。您想再开十家。问题来了：每家新店需要 100 万元的设备、押金和招牌。您只有 200 万元，还差 800 万元。谁会借给您这笔钱 — 或者愿意拿现在的现金，换取您未来利润的一部分？
In a village, you might ask your uncle. In a city of ten million, you cannot. So there has to be some institution that knows hundreds of "uncles" — savers, pensioners, mutual funds, sovereign wealth funds, insurance companies — and knows how to package your tea-shop story into a form those uncles can evaluate, price, and buy. That institution is an investment bank, and the document it produces is the prospectus, the pitch book, the research report. It is, in essence, a translator. 在村里，您也许可以问您舅舅。但在一个千万人口的城市里，这不现实。所以必须有这样一个机构，它认识成百上千个"舅舅" — 储户、退休人员、共同基金、主权财富基金、保险公司 — 并知道如何把您的茶馆故事包装成"舅舅们"能评估、能定价、能购买的形式。这个机构就是投资银行，而它产出的文件，就是招股说明书、推介书、研究报告。本质上，投资银行就是一个翻译者。
Modern investment banks are sprawling, but almost everything they do falls into four buckets. Understanding these four buckets is the price of admission for reading any financial news story. 现代投资银行规模庞大，但几乎所有业务都可以归入四类。理解这四类，是读懂任何财经新闻的入场券。
Helping companies and governments raise money. When Alibaba IPO'd on the NYSE in 2014, a consortium of investment banks priced the shares, marketed them to institutions, and underwrote — that is, guaranteed — that the deal would clear at the announced price. They earned a fee proportional to the size of the offering. 帮助公司和政府筹集资金。2014 年阿里巴巴在纽约证交所上市时，一群投资银行为股票定价、向机构推销，并承销 — 即保证 — 交易能在公布价格上完成。他们按发行规模收取费用。
Advising companies that want to buy other companies, or sell themselves. The bank builds the valuation, structures the deal, runs the auction, negotiates with the other side, and arranges the financing. The fee, in big deals, can be tens of millions of dollars. 为想收购其他公司或出售自己的公司提供建议。投行负责构建估值、设计交易结构、组织拍卖、与对方谈判，并安排融资。大型交易中，顾问费可能高达数千万美元。
Making markets in stocks, bonds, currencies, and derivatives. Buy-side clients (hedge funds, pension funds) tell the bank what they want to trade; the bank prices it, executes it, and takes a thin margin (the "bid-ask spread"). This is volume business — millions of trades per day at razor-thin margins. 在股票、债券、外汇、衍生品市场上做市。买方客户 (对冲基金、养老金) 告诉投行想交易什么；投行报价、执行、赚取一个薄薄的差价 ("买卖差")。这是规模生意 — 每天数百万笔交易，利润率极低。
Producing written analysis of companies, industries, currencies, and economies that helps the bank's clients decide what to buy and sell. This is the function our website unpacks. The output looks like a magazine article with charts; the work behind it is months of modeling, management meetings, and channel checks. 撰写关于公司、行业、汇率、经济的分析，帮助客户决定买什么、卖什么。这正是本网站要拆解的职能。成品看起来像一篇带图表的杂志文章；背后却是数月的建模、与管理层会面以及渠道调研。
The terms "sell-side" and "buy-side" trip up almost every beginner. They become obvious once you remember the basic geometry: the sell-side sells a service (advice, execution, research, underwriting) to the buy-side, which buys that service to make money on actual investments. "卖方"和"买方"几乎让所有新手头疼。但只要记住基本的几何关系，就一目了然：卖方出售一项服务 (建议、执行、研究、承销) 给买方，买方购买这项服务，并用它在真实投资中赚钱。 An equity research analyst at an investment bank is on the sell-side: she does not personally invest in Alibaba; she publishes reports that hedge funds, mutual funds, sovereign wealth funds, and family offices use to decide whether they will invest. Her career is built on the perceived usefulness of her work to the buy-side. If her recommendation is right and she said so first, she becomes valuable to the buy-side and to her firm; if she is consistently wrong, she becomes invisible. 投资银行的股权研究分析师属于卖方：她本人并不投资阿里巴巴；她发布的报告供对冲基金、共同基金、主权财富基金和家族办公室使用，这些机构据此决定它们是否投资。她的职业生涯建立在自己的工作对买方的有用性之上。如果她的建议正确且抢先说出来，她对买方、对她的雇主都很有价值；如果她持续错误，她就会消失在视野中。
Sell-side firms (investment banks, broker-dealers) produce services — research, trade execution, underwriting, M&A advice — and sell them to the buy-side. They are paid in fees and commissions. Buy-side firms (mutual funds, hedge funds, pension funds, insurance companies, sovereign wealth funds) invest other people's money in stocks, bonds, and other assets, and earn a management fee and (often) a share of profits. The two sides talk all day, every day; the sell-side is the buy-side's eyes and ears on the world. 卖方机构 (投资银行、券商) 生产服务 — 研究、交易执行、承销、并购建议 — 并出售给买方。它们以费用和佣金获得报酬。买方机构 (共同基金、对冲基金、养老金、保险公司、主权财富基金) 用别人的钱投资于股票、债券和其他资产，并赚取管理费和 (通常的) 利润分成。两方每天每时都在对话；卖方是买方在世界上的眼睛和耳朵。
The romantic image is a person staring at six monitors and shouting "buy!" The reality is more like a magazine writer with a calculator. An analyst typically covers between ten and twenty companies in a single industry — say, China internet, or US software, or European luxury — and is expected to know each of those companies the way a sports beat reporter knows a team: the lineup, the injuries, the locker-room politics, the schedule. She publishes: 浪漫的形象是：一个人盯着六块显示器大喊"买入！"现实更像是一个手持计算器的杂志撰稿人。分析师通常覆盖某一行业 10-20 家公司 — 比如中国互联网、美国软件、欧洲奢侈品 — 并被要求像体育记者了解一支球队那样了解每家公司：阵容、伤病、更衣室政治、赛程。她要发布：
Every piece of paper ends with three numbers: a rating (Overweight, Neutral, Underweight — sometimes called Buy/Hold/Sell), a price target (where she thinks the stock should trade in twelve months), and the implied return from today's price to that target. Those three numbers are the analyst's product. Everything else is the show that makes you believe them. 每份报告都以三个数字结尾：评级 (增持、中性、减持 — 有时叫"买入/持有/卖出")、目标价 (她认为股票 12 个月后应该交易在哪个价位) 和隐含回报率 (从今天的价格到那个目标价的回报)。这三个数字就是分析师的产品。其余一切，都是让您相信它们的"演出"。
A skeptic might ask: in an efficient market, isn't the current stock price the best estimate of value? Why does anyone publish another opinion? The honest answer is that markets are mostly efficient, but the words "mostly" and "in equilibrium" do a lot of work. Markets are efficient because thousands of analysts and investors are constantly hunting for mispricings; the moment they all stop hunting, prices stop tracking reality. Research is the unglamorous work that keeps prices honest. It is also, for the analyst, a competitive sport: the goal is not to be right about everything, but to be right about the few things before everyone else is. 怀疑论者可能会问：在有效市场里，当前股价不就是最好的价值估计吗？为什么还要别人发表观点？老实的回答是：市场大部分时候有效，但"大部分时候"和"均衡状态下"这几个字承载了很多含义。市场之所以有效，是因为成千上万的分析师和投资者不断地寻找定价偏差；他们一停止寻找，价格就开始脱离现实。研究就是让价格保持诚实的那项不光鲜的工作。对分析师而言，这也是一项竞技：目标不是事事正确，而是在其他人之前对少数几件事正确。
There is also a structural reason: most large institutional investors are required, by their mandates and by their regulators, to document their investment process. Buying a stock because "I felt like it" gets you fired. Buying it because "five sell-side analysts in our coverage universe have published Overweight ratings citing margin expansion and capital return; our own model triangulates within 10% of theirs; we are within risk limits" is defensible. Sell-side research is, in part, an audit trail. 还有一个结构性原因：大多数大型机构投资者，受其授权与监管要求，必须记录其投资过程。仅凭"感觉"就买入一只股票会让人丢掉饭碗。但如果说"我们覆盖范围内有五位卖方分析师发布了'增持'评级，理由是利润率扩张和资本回报；我们自己的模型与他们的结果差异在 10% 以内；我们在风险限额内"，这就站得住脚。卖方研究在某种程度上就是一条审计轨迹。
Everything that follows is structured as the journey a piece of research takes. We start with the three financial statements (the raw material), move to ratio and trend analysis (cleaning the raw material), build a forecast (extrapolating the cleaned material forward), and arrive at three independent valuation methods (turning the forecast into a price). At each step, we use Alibaba — a real, complicated, fascinating company — as the worked example. By the end, you will be able to read a sell-side research report not as a magic spell, but as a perfectly logical document, where every chart, table, and number has a defined home and a specific job. 接下来的所有内容，都按照一份研究报告的诞生路径组织。我们从三大财务报表 (原材料) 开始，进入比率和趋势分析 (清洗原材料)，构建预测 (把清洗好的原材料外推)，最后用三种独立的估值方法 (把预测变成价格)。每一步，我们都以阿里巴巴 — 一家真实、复杂、迷人的公司 — 作为实例。读完之后，您将能像阅读一份完全合乎逻辑的文档那样阅读卖方研究报告，而不是像看魔法咒语 — 其中每张图、每张表、每个数字，都有明确的归属和具体的工作。
If you flip through fifty sell-side reports, you discover they all have the same skeleton. Knowing the skeleton makes every individual report less mysterious — and lets you skip straight to the part you actually need. 如果您翻阅 50 份卖方报告，会发现它们都有相同的骨架。理解这副骨架，能让每份报告都不再神秘 — 您也可以直接跳到自己真正需要的那部分。
Almost every initiation report follows the same arc, refined over decades. Once you see it once, you see it everywhere: 几乎每一份首次覆盖报告都遵循相同的脉络，这套结构经过数十年打磨。一旦看过一次，您就能在所有地方认出来：
Notice that everything except the cover is, in a sense, a justification of the cover. The cover sets a claim ("this stock will go from $118 to $226 in twelve months, and here's the rating that reflects that view"); the rest of the report is the evidence. Reading a research report is, therefore, very much like reading a court filing or a scientific paper: assertion first, then evidence in support. 注意，除了封面之外，其他章节本质上都是对封面的论证。封面陈述了一个主张 ("这只股票将在 12 个月内从 118 美元涨到 226 美元，对应的评级如下")；其余部分是证据。因此，读一份研究报告很像读一份法庭诉状或科学论文：先有主张，然后有支持证据。
A finished report is the tip of an iceberg. Beneath the surface sit four artifacts an analyst maintains continuously, often updated every week: 成品报告只是冰山一角。水下是分析师持续维护、常常每周更新的四件"产物"：
The model is the analyst's microscope; the thesis is her hypothesis; the calendar is her clock; the network is her instrument for collecting evidence that is not yet in the public domain. Anyone can read 10-Ks. The analyst's edge is in marrying public filings with primary research — supplier conversations, web-traffic data, freight bookings, app downloads, channel checks at a Tesla showroom in Shanghai — to anticipate what the next 10-K will say before it is filed. 模型是分析师的显微镜；论点是她的假设；日历是她的时钟；人脉网是她收集尚未进入公开领域的证据的工具。每个人都能读 10-K 年报。分析师的优势在于将公开文件与一手研究 — 与供应商对话、网络流量数据、运费订单、App 下载量、对上海某特斯拉展厅的渠道调研 — 结合起来，在下一份 10-K 提交前预判它会怎么说。
A common misconception is that "price target $226" means "this stock will trade at $226." It does not. A price target is the analyst's central estimate of where the stock should trade in twelve months if her model assumptions prove correct. Around that point estimate is a distribution — the bull case ($283), the base case ($226), the bear case ($170) in our Alibaba example. The probability that the actual outcome lies within 10% of the point estimate is, empirically, only about 40%. Why publish a single number then? Because consumers of research demand one. A range is harder to act on than a number. The bank therefore picks the modal scenario and prints it on the cover; the conscientious analyst makes the whole distribution available in the back. 一个常见的误解是："目标价 226 美元" 意味着 "这只股票将在 226 美元交易"。它不是。目标价是分析师对股票 12 个月后应该交易在哪里的中心估计 — 如果她的模型假设正确的话。在这个点估计之外，还有一个分布 — 在我们的阿里巴巴例子里，牛市情景 283 美元，基准 226 美元，熊市 170 美元。实证上，实际结果落在中心估计 10% 范围内的概率，大约只有 40%。那为什么还要发布一个单一数字？因为研究消费者要求这样。一个区间比一个数字更难"采取行动"。所以银行选择最可能的情景，把它印在封面；负责的分析师，把整个分布留在后面。
When a buy-side investor reads a sell-side report, she is rarely looking for the price target itself. She is looking for: (a) the variant view — what does this analyst see that the market does not? (b) the data — is there a chart that I can lift into my own deck? (c) the channel-check color — what did supplier X tell the analyst that I didn't know? The price target is the marketing wrapper; the real product is the original thought inside. 当一位买方投资者阅读卖方报告时，她很少在意目标价本身。她在找的是：(a) 不同观点 — 这位分析师看到了市场没看到的什么？(b) 数据 — 有没有可以"搬"到我自己 PPT 里的图表？(c) 渠道调研的"色彩" — 供应商 X 对分析师说了什么是我不知道的？目标价是营销外壳；真正的产品是里面的原创思考。
Every public company in the world is summarized, every quarter, in three numerical tables: the income statement, the balance sheet, and the cash flow statement. Master these, and you can read any business in any country. The trick is to see them not as three independent objects, but as three views of the same thing — the way an architect's plan, elevation, and section drawings describe the same building. 全世界每一家上市公司，每个季度，都被三张数字表格所概括：利润表、资产负债表、现金流量表。掌握它们，您就能读懂任何国家、任何行业的公司。诀窍在于：不要把它们看成三个独立的东西，而是同一事物的三个视角 — 就像建筑师的平面图、立面图、剖面图，描述的是同一栋建筑。
Before we open Alibaba's books, let us build the intuition with a simpler example: a one-shop tea-house in Hangzhou over a single year. Every dynamic that appears in BABA's 941-billion-yuan income statement is present, in miniature, in this little shop. The math is identical; only the zeros change. 在打开阿里巴巴的账本之前，先用一个更简单的例子建立直觉：杭州一家茶馆在一年内的经营。BABA 那张 9410 亿元的利润表里出现的每一种动态，都以微缩形式存在于这家小茶馆中。数学是相同的；只是零的个数不一样。
Imagine you open Hangzhou Cha-Lou on 1 January. You put in ¥500,000 of your own savings as equity, and you borrow ¥300,000 from a bank as debt. You buy ¥400,000 of equipment and refit the storefront (this is capital expenditure — capex — and the equipment will last roughly five years, so its cost is spread across those years as depreciation). You stock ¥100,000 of tea leaves (inventory). On 31 December you check the till and find the year went like this: 想象您在 1 月 1 日开了一家"杭州茶楼"。您拿出 50 万元个人积蓄作为股本，并向银行借了 30 万元作为债务。您花 40 万元购买设备并翻新店面 (这是资本开支 — capex — 设备能用 5 年，所以成本分摊到这些年作为折旧)。您囤了 10 万元的茶叶 (库存)。12 月 31 日，您打开账册，发现这一年是这样的：
12,000 customers × ¥100 average ticket. This is the top of your income statement. 12,000 位顾客 × 100 元客单价。这是您利润表的最上一行。
The leaves you actually brewed, the cups, the gas. This is "cost of revenue" or "cost of goods sold." 您实际泡掉的茶叶、用掉的杯子、烧掉的煤气。这就是"营业成本"或"主营业务成本"。
Two part-time servers, the rent on the storefront, a few WeChat ads. These are "operating expenses" — overhead that doesn't scale one-for-one with cups sold. 两名兼职服务员、店铺租金、几条微信广告。这些是"营业费用" — 不会随杯数 1:1 增长的间接开支。
The ¥400k of equipment, spread across five years, is ¥80k per year of "wear and tear" expense — a non-cash charge. 40 万元设备摊到 5 年里，每年就是 8 万元的"损耗"费用 — 一种非现金支出。
Subtracting all the costs gives you operating profit of about ¥320k. Pay the bank ¥18k of interest. Pay the tax authority 25% of what's left (¥75k). The ¥227k that lands in your pocket is your net income. That sentence — revenue minus costs minus interest minus tax equals net income — is the income statement, top to bottom, in any language and any industry. Now the other two statements: 把所有成本减掉，您得到大约 32 万元的经营利润。付给银行 1.8 万元利息。再交给税务局剩下部分的 25% (7.5 万元)。落到您口袋里的 22.7 万元，就是您的净利润。这一句话 — 收入减去成本减去利息减去税收等于净利润 — 在任何语言、任何行业里，都是利润表的从头到尾。再看其他两张报表：
A tea-shop owner who keeps these three sets of books at the back of his till is, technically, doing financial reporting at the same level of abstraction as Apple's CFO. The only difference is scale — and a thousand pages of accounting rules to cover special cases. The rest of this chapter zooms into each of the three statements, then shows how they snap together. Once you can see the snapping, you have crossed the most important rubicon in financial analysis. 在收银台后面记着这三本账的茶楼老板，从技术上讲，做的是和苹果公司 CFO 同样抽象级别的财务报告。唯一的区别是规模 — 以及上千页用于覆盖特殊情况的会计准则。本章的剩余部分会放大每张报表，再展示它们如何"扣"在一起。一旦您能看到"扣合"的那一刻，您就跨过了财务分析里最重要的"卢比孔河"。
The income statement (also called the P&L, for profit and loss) tells you what a company earned and what it spent over a period — typically a quarter or a year. It is a story of inflows and outflows, designed to end at a single line: net income. 利润表 (也叫 P&L) 告诉您一家公司在一段时间内赚了多少、花了多少 — 通常是一个季度或一个财年。它讲述了流入和流出的故事，最终汇集到一个单独的数字：净利润。
Picture the income statement as a funnel. At the top, you pour in revenue — the gross value of everything the company sold. As you move down the funnel, you subtract costs in groups: first the costs that scale directly with selling more (cost of revenue), then the costs that support the business at a higher level (operating expenses), then interest paid to lenders, then tax. What drips out the bottom is net income — the amount that, in theory, belongs to the shareholders. 把利润表想象成一个漏斗。最上面，您倒入收入 — 公司卖出去的所有东西的总价值。沿着漏斗往下，您按组扣除成本：先是与销售直接挂钩的成本 (营业成本)，再是支撑业务的更高层级成本 (营业费用)，再是支付给债权人的利息，再是税。从漏斗底部滴出来的，就是净利润 — 理论上属于股东的金额。
Let's open Alibaba's actual P&L. The company has a March fiscal year-end (a hangover from its Asian listing schedule), so "FY25A" means the year ended 31 March 2025. The five years FY26E through FY30E are our forecast — labeled "E" for estimate. 让我们打开阿里巴巴真正的利润表。该公司财年截至 3 月底 (这是其亚洲上市时间表留下的"习惯")，所以 "FY25A" 指的是截至 2025 年 3 月 31 日的财年。FY26E 到 FY30E 是我们的预测年 — 用 "E" 表示估算。
CNY mn (bars, left axis) · margin % (lines, right axis) 百万人民币 (柱状，左轴) · 利润率 % (折线，右轴)
Below is a stripped-down income statement, presented in the layout you will see in any sell-side model. We walk through each row immediately after. 下面是一份精简版利润表，展示了任何卖方模型里都会用到的布局。表格之后，我们逐行讲解。
| CNY mn百万人民币 | FY22A | FY23A | FY24A | FY25A | FY26E | FY27E |
|---|---|---|---|---|---|---|
| Revenue收入 | 853,062 | 868,687 | 941,168 | 996,347 | 1,071,073 | 1,172,825 |
| YoY %同比% | +18.9% | +1.8% | +8.3% | +5.9% | +7.5% | +9.5% |
| Cost of revenue营业成本 | (539,450) | (549,695) | (586,323) | (598,285) | (637,288) | (686,103) |
| Gross profit毛利润 | 313,612 | 318,992 | 354,845 | 398,062 | 433,785 | 486,722 |
| Gross margin毛利率 | 36.8% | 36.7% | 37.7% | 40.0% | 40.5% | 41.5% |
| Sales & marketing销售与营销 | (132,225) | (121,616) | (127,058) | (129,525) | (139,239) | (146,603) |
| Product development产品开发 | (55,449) | (52,121) | (51,764) | (54,799) | (58,909) | (64,505) |
| General & administrative一般及行政 | (21,327) | (21,717) | (23,529) | (24,909) | (26,777) | (29,321) |
| Amortization & one-offs摊销与一次性 | (55,449) | (21,717) | (9,412) | (9,963) | (10,711) | (5,864) |
| Operating income (EBIT)营业利润 (EBIT) | 49,163 | 101,820 | 143,082 | 178,866 | 198,149 | 240,429 |
| EBIT marginEBIT 利润率 | 5.8% | 11.7% | 15.2% | 18.0% | 18.5% | 20.5% |
| Interest & other income利息及其他收入 | 5,000 | 22,000 | 12,000 | 38,000 | 18,000 | 14,000 |
| Tax所得税 | (14,353) | (28,479) | (33,343) | (46,626) | (46,472) | (54,702) |
| Net income to parent归母净利润 | 62,249 | 72,783 | 80,009 | 130,109 | 166,283 | 195,732 |
| Net margin净利率 | 7.3% | 8.4% | 8.5% | 13.1% | 15.5% | 16.7% |
| Diluted EPS (US$/ADR)稀释 EPS (美元/股 ADR) | 3.42 | 3.92 | 4.55 | 7.56 | 10.04 | 12.36 |
Walking this top-to-bottom, you should be able to tell a story in plain words: Alibaba's revenue grew from ¥853bn to ¥996bn between FY22 and FY25 (about 5% per year — modest, given pandemic and regulatory disruption). Cost of revenue grew slightly faster, so gross margin actually expanded from 36.8% to 40.0% as the mix moved toward higher-margin businesses (cloud, advertising). Operating expenses stayed roughly flat in absolute terms — sales and marketing actually came down — which means the savings from cost discipline flowed straight through to operating income, which more than tripled from ¥49bn to ¥179bn. Then non-operating income (interest on the ¥400bn cash pile, gains and losses on equity stakes in Ant Group and others) added another lift. After tax, net income for the parent shareholders nearly tripled. 从上到下走一遍，您应该能用大白话讲个故事：阿里巴巴收入从 FY22 的 8530 亿元增长到 FY25 的 9960 亿元 (年均约 5% — 鉴于疫情和监管扰动，已属不易)。营业成本增长稍快，所以毛利率反而扩张了，从 36.8% 升到 40.0%，因为产品组合向更高毛利的业务 (云、广告) 倾斜。营业费用绝对金额大致持平 — 销售与营销甚至下降 — 这意味着成本纪律省下的钱直接流到了营业利润，从 490 亿元增长到 1790 亿元，三倍有余。然后非经营收入 (4000 亿元现金堆产生的利息、对蚂蚁集团及其他股权投资的盈亏) 再添一笔。扣完税，归母净利润几乎翻了三倍。
Notice that we did not need to know what Alibaba is in order to read these numbers. Anyone with the four ratios (gross margin, EBIT margin, net margin, EPS) and one growth number (revenue YoY) can hold a passable conversation about the company's near-term financial health. The income statement is the great democratizer of financial analysis: it gives the reader a quick fluency in the operating story. 注意，要读懂这些数字，我们并不需要知道阿里巴巴是什么。掌握四个比率 (毛利率、EBIT 利润率、净利率、EPS) 和一个增长数据 (收入同比)，任何人都能在公司近期财务健康话题上谈得过去。利润表是财务分析的"民主化大师"：它让读者快速掌握公司经营故事的对话能力。
One: revenue is not cash. Revenue is "earned" when the company has delivered the product or service, regardless of whether the customer has paid. If you ship ¥1m of tea to a customer who promises to pay in 90 days, you book ¥1m of revenue today, but no cash. The unpaid bill sits on the balance sheet as accounts receivable. That is why a fast-growing company can look profitable on the income statement and yet desperately need cash. The cash flow statement (next subsection) exists precisely to disentangle this. 第一：收入不等于现金。当公司交付了商品或服务，收入就"实现"了，无论客户是否付款。如果您给一位 90 天后付款的客户运了 100 万元的茶叶，今天就要记 100 万元的收入，但没有现金。这张未付账单作为应收账款留在资产负债表上。这就是为什么高速增长的公司利润表上可能看着很赚钱，但现金周转捉襟见肘。现金流量表 (下一节) 的存在就是为了把这一切分清楚。
Two: depreciation and amortization are non-cash. When the tea-house owner buys a ¥400k espresso machine and writes off ¥80k of depreciation each year for five years, no cash actually leaves the till in years 2 through 5. The cash all went out the door in year 1. Depreciation is the accountant's way of matching the cost of the machine with the years in which the machine produces revenue. The cash flow statement adds depreciation back precisely because it was never a cash expense in the first place. 第二：折旧和摊销是非现金。当茶楼老板花 40 万元买了一台咖啡机，每年提 8 万元折旧、分摊 5 年，第 2-5 年实际并没有现金从收银台流出。所有现金都在第 1 年就花出去了。折旧是会计师把机器的成本匹配到它产生收入的年份的方式。现金流量表把折旧加回，正是因为它从一开始就不是真正的现金支出。
Three: stock-based compensation (SBC) is a real cost, even though it doesn't use cash. When Alibaba pays its engineers in stock, accountants record the fair value of those grants as an operating expense. No cash leaves the company. But shareholders are diluted: the same profit is now spread across more shares, so each existing share owns a smaller piece. SBC has averaged 4–8% of revenue at Alibaba in recent years — economically real, even if it never moved a dollar of cash. This is why sophisticated investors strip SBC out of "EBITDA" measures with care. 第三：股权激励 (SBC) 是真实成本，即便它不耗用现金。当阿里巴巴用股票支付工程师工资时，会计将这些授予的公允价值计入营业费用。公司没有现金流出。但股东被稀释了：同样的利润如今分布在更多的股份上，所以每一股老股的"份额"变小。近几年阿里巴巴的 SBC 平均占收入的 4-8% — 经济上是真实的，哪怕一分钱现金都没动。这就是为什么资深投资者在使用 "EBITDA" 类指标时，会小心地把 SBC 剔除。
"EBITDA" stands for Earnings Before Interest, Taxes, Depreciation, and Amortization. It is operating income with depreciation and amortization added back. Why care? Because it tries to approximate the cash a business is throwing off from operations, before financing decisions (interest) and tax structures get in the way. EBITDA is useful for comparing businesses with very different capital intensities — a software company has tiny D&A; a cloud-infrastructure company has enormous D&A; their operating cash characteristics are easier to compare on an EBITDA basis. It is also abused: EBITDA flatters companies whose depreciation expense is real economic obsolescence (e.g., AWS servers genuinely die in five years). Treat it as a sketch, not a fingerprint. "EBITDA"代表 Earnings Before Interest、Taxes、Depreciation 和 Amortization，即"扣除利息、税收、折旧、摊销前的盈利"。它是营业利润加回折旧与摊销。为什么有用？因为它试图近似一家企业从经营中产生的现金，把融资决策 (利息) 和税收结构暂时放在一边。EBITDA 对比较资本密度差异巨大的企业很有用 — 软件公司的折旧摊销很小；云基础设施公司的折旧摊销巨大；用 EBITDA 比较他们的经营现金特征更容易。它也常被滥用：EBITDA 会美化那些折旧实际上代表经济老化的公司 (例如 AWS 服务器真的会在 5 年内报废)。把它当作素描，而不是指纹。
A company that grows revenue 15% per year sounds wonderful — until you ask whether each new dollar of revenue is profitable. Some businesses can buy revenue: PDD subsidizes prices, Lazada subsidizes delivery, Uber subsidized rides. The income statement reveals whether the unit economics are healthy by separating revenue (top line) from the various margins below it. Three margins matter the most: 一家公司年收入增长 15% 听起来很美 — 但您要追问：每一元新收入是否赚钱？有些生意可以"买"收入：拼多多补贴价格、Lazada 补贴配送、Uber 曾补贴打车。利润表通过把收入 (顶行) 与下面各种利润率分开，揭示了单位经济性是否健康。最重要的三个利润率是：
A typical first question from an experienced investor is: "are margins expanding or contracting?" Expanding margins on flat revenue is almost always a positive story (cost discipline, mix shift, pricing power). Expanding margins on declining revenue is suspicious (cost cuts that may sacrifice the future). Contracting margins on flat revenue is a problem (competition, regulation, input cost pressure). Contracting margins on rapidly growing revenue can be fine, if the company is investing in the next growth wave. The chart above lets you toggle between the three margins and visually inspect Alibaba's history: notice the sharp 2022 trough (regulatory crackdown, mainland consumer weakness, restructuring), then the methodical recovery — a margin story rather than a revenue story, which is exactly what the bull thesis hinges on. 资深投资者的第一个问题往往是："利润率在扩张还是收缩？"收入持平时利润率扩张，几乎总是积极信号 (成本纪律、组合转移、定价能力)。收入下降时利润率扩张则可疑 (砍成本可能牺牲未来)。收入持平时利润率收缩是个问题 (竞争、监管、投入成本压力)。收入快速增长时利润率收缩可以接受，如果公司正在投资下一波增长。上面的图表允许您在三种利润率之间切换并目视检查阿里巴巴的历史：注意 2022 年的急剧低谷 (监管收紧、大陆消费疲软、内部重组)，然后是有条不紊的复苏 — 这是利润率故事而非收入故事，而这恰恰是看多论点所依赖的。
From CNY 996bn of revenue down to CNY 130bn of net income — every step shown. 从 9960 亿元收入降到 1301 亿元净利润 — 每一步都呈现。
Below the gross profit line sit the "operating expenses" — the overhead that supports the business but doesn't scale 1:1 with each unit sold. For a digital business like Alibaba, these line items reveal almost everything about strategy and discipline. 毛利润下方坐着"营业费用" — 支撑业务但不随每单位销售 1:1 扩张的间接开支。对像阿里巴巴这样的数字业务，这些行项目几乎揭示了战略与纪律的一切。
Sales & marketing. Promotional spend, advertising, sales-force compensation, user-acquisition subsidies. At Alibaba, S&M peaked at 15.5% of revenue in FY22 (a period of intense competition with PDD and Douyin) and has been disciplined down to 13% in FY25 and 12.5% in FY27E. A 1% reduction here equals about ¥10bn of incremental operating income — meaningful. The right question to ask: is the company under-investing in customer acquisition (storing up future weakness) or rationally accepting slower top-line growth to lift margins? 销售与营销 (S&M)。推广支出、广告、销售人员薪酬、用户获取补贴。在阿里巴巴，S&M 在 FY22 达到 15.5% 的收入占比峰值 (与拼多多、抖音激烈竞争时期)，被克制压到 FY25 的 13% 和 FY27E 的 12.5%。这里下调 1% 等于约 100 亿元增量营业利润 — 很显著。正确的问题：公司是在客户获取上投入不足 (为未来疲软埋种)，还是理性接受较慢的收入增长以提升利润率？
Product development (R&D). Engineers, product managers, designers, AI researchers. Alibaba spends about 5.5% of revenue on product development — a structurally important line because half of it goes to AI infrastructure and the Tongyi Qianwen foundation model effort. A company that cuts R&D to flatter near-term margins almost always pays a price 3-5 years out. We hold R&D / revenue flat in the forecast precisely because we want the bull case to be funded by genuine product progress, not by sacrificed innovation. 产品开发 (R&D)。工程师、产品经理、设计师、AI 研究员。阿里巴巴在产品开发上花费约 5.5% 的收入 — 结构上重要的一行，因为其中一半流向 AI 基础设施和通义千问基础模型工作。砍 R&D 以美化短期利润率的公司，几乎总在 3-5 年后付出代价。我们在预测中保持 R&D/收入持平，正是因为我们希望看多情景由真正的产品进步资助，而非牺牲创新。
General & administrative (G&A). Finance, legal, HR, executive compensation, audit fees. Around 2.5% of revenue and remarkably stable — a sign that the back-office is running efficiently. A G&A line that creeps up over time is often a symptom of unchecked acquisition activity (each acquired entity brings its own G&A overhead) or weakening management discipline. 一般及行政 (G&A)。财务、法律、人力、高管薪酬、审计费。约 2.5% 的收入且非常稳定 — 后台运行高效的标志。G&A 占比随时间蠕变上升，往往是失控并购活动 (每收购一个实体就带来自己的 G&A 间接成本) 或管理纪律弱化的症状。
Amortization of intangibles and impairments. When you acquire a company, you record the brands, customer relationships, and patents on the balance sheet as "intangible assets" and amortize them over their estimated useful life. The amortization expense reduces reported earnings but does not reduce cash. Impairments are sharper one-off charges when an acquired asset is judged to be permanently worth less. Alibaba took two enormous impairments in FY22 (-¥51bn) and FY23 (-¥17bn) related to the Sun Art and DingTalk businesses — the accounting clean-up of an over-aggressive acquisition spree. These are the items you should mentally "back out" when assessing recurring profitability. 无形资产摊销与减值。当您收购一家公司时，您把品牌、客户关系、专利在资产负债表上记为"无形资产"，并按估计使用年限摊销。摊销费用减少报告盈利但不减少现金。减值是更尖锐的一次性费用，当被收购资产被判定永久价值降低时计提。阿里巴巴在 FY22 (-510 亿元) 和 FY23 (-170 亿元) 计提了两笔巨额减值，与大润发和钉钉业务相关 — 一次过度激进收购的会计清理。这些是您在评估经常性盈利能力时应该"剔除"的项目。
Beneath gross profit, four operating metrics travel together: EBIT, EBITA, EBITDA, and adjusted EBITDA. Each is built by adding more items back to EBIT. The order: EBIT → EBITA (+ amortization of acquired intangibles) → EBITDA (+ depreciation as well) → adjusted EBITDA (+ stock-based compensation, restructuring, FX, sometimes legal fees). Each layer makes the company look more profitable. The honest reader looks at all of them and asks: which one best represents the cash a normal operating year throws off? 毛利润下方，四个经营指标一同出现：EBIT、EBITA、EBITDA、调整后 EBITDA。每个都通过往 EBIT 上加更多项目构建。顺序：EBIT → EBITA (+ 收购无形资产摊销) → EBITDA (+ 折旧也) → 调整后 EBITDA (+ 股权激励、重组、汇兑、有时还包括法律费用)。每一层都让公司看起来更赚钱。诚实的读者会同时看所有这些并问：哪一个最能代表一个正常经营年产生的现金？
Alibaba's preferred metric is adjusted EBITA by segment — EBIT plus amortization of acquired intangibles. This is reasonable for a company that has made many acquisitions where the GAAP amortization expense (¥5bn / year on a US$140bn business) is genuinely non-cash and not predictive of future operating performance. The trick is to keep stock-based compensation inside EBITA — which Alibaba does. That's important: SBC is a real cost (dilution of existing holders), and adjusted EBITDA frameworks that strip it out can systematically overstate the cash-generative power of equity-heavy companies. 阿里巴巴偏好的指标是按分部的调整后 EBITA — EBIT 加回收购无形资产摊销。这对一家做过很多收购、其 GAAP 摊销费用 (1400 亿美元业务上 50 亿元/年) 真正非现金且对未来经营表现无预测性的公司是合理的。诀窍是把股权激励保留在 EBITA 内 — 阿里巴巴这样做了。这很重要：SBC 是真实成本 (现有股东被稀释)，把它剔除的调整后 EBITDA 框架可能系统性高估权益重型公司的现金创造能力。
The Chinese statutory corporate tax rate is 25%. For high-tech companies designated as "advanced technology enterprises" (the bracket Alibaba's core technology subsidiaries qualify for), the rate is 15%. Add a 5-percentage-point Hong Kong withholding for offshore dividend repatriation, and you arrive at an effective tax rate of roughly 21–22% — which is what Alibaba has been reporting. We forecast 21.5% for the next five years. A change in tax policy (the abolition of the high-tech bracket, or a US-style global minimum tax on Chinese subsidiaries) could shift effective tax by 200-300bp, which would shave 3-4% off our forward EPS. Tax assumptions are easy to skip and dangerous to ignore. 中国法定企业所得税率为 25%。对被认定为"高新技术企业"的公司 (阿里巴巴的核心科技子公司所属的范畴)，税率为 15%。加上离岸股息汇回的 5 个百分点的香港预提税，您得到约 21-22% 的实际税率 — 这就是阿里巴巴一直报告的水平。我们预测未来五年为 21.5%。税收政策变化 (取消高新技术税率，或对中国子公司征收美式全球最低税) 可能让实际税率变动 200-300 个基点，会让我们前瞻 EPS 缩水 3-4%。税收假设容易跳过却危险不容忽视。
When a company owns 60% of a subsidiary, it consolidates 100% of the subsidiary's revenues and expenses into its own income statement (since it controls the subsidiary), and then shows a single deduction at the bottom called "net income attributable to non-controlling interest" — the 40% that legally belongs to the other shareholders. The real number the public-market shareholder cares about is the next line down: "net income attributable to parent shareholders." Always check both — a company with lots of partially-owned subsidiaries can show flattering consolidated net income while the actual share-holder-attributable income is much lower. 当一家公司持有子公司 60% 时，它把子公司 100% 的收入和费用合并到自己的利润表 (因为它控制子公司)，然后在底部显示一项扣除，叫"归属于少数股东的净利润" — 法律上属于其他股东的那 40%。公开市场股东真正关心的数字是下一行："归属于母公司股东的净利润"。永远同时检查两个 — 一家有许多部分持股子公司的公司可能显示令人满意的合并净利润，而实际归属股东的收入低得多。
Earnings per share is net income attributable to parent shareholders, divided by the diluted weighted-average share count. Two subtleties matter: (1) diluted means the count includes shares that would be issued if all outstanding options, restricted stock units, and convertible bonds were converted today — a higher count than the basic share count. Diluted EPS is therefore lower than basic EPS, and it is the metric the buy-side uses. (2) Weighted-average means the share count is computed over the reporting period — if a company bought back 5% of its float midway through the year, the average count for the year is somewhere between the starting and ending count. This is why current-period EPS doesn't fully reflect very recent buyback activity — the full benefit shows up in the next period. 每股盈利就是归母净利润除以稀释后加权平均股本数。两个微妙之处重要：(1) 稀释意味着股本数包括如果所有未行权期权、限制性股票单位、可转换债券今天全部转换会发行的股票 — 比基本股本数高。所以稀释 EPS 低于基本 EPS，是买方使用的指标。(2) 加权平均意味着股本数是按报告期计算 — 如果公司年中回购了 5% 的流通股，年内平均股本数介于期初和期末之间。这就是为什么当期 EPS 不完全反映非常近期的回购活动 — 完整效益会在下期显现。
Alibaba's diluted ADR share count has fallen from 2,700 mn in FY21 to ~2,390 mn in FY25 — a 12% reduction over four years. Mathematically, that 12% has lifted EPS by about 14% compounded — value created with cash that would otherwise have sat on the balance sheet at 5% nominal interest. This is what disciplined capital return looks like. 阿里巴巴的稀释 ADR 股本数已从 FY21 的 2700 百万降到 FY25 的约 2390 百万 — 四年降低 12%。从数学上讲，这 12% 复合提升 EPS 约 14% — 价值由本来会在资产负债表上以 5% 名义利率躺着的现金创造。这就是有纪律的资本回报的样子。
The income statement covers a period; the balance sheet captures a moment. It answers the question: as of 31 March 2025, what does this company own, what does it owe, and what is left over for the owners? 利润表覆盖一段时间；资产负债表则定格在某一时刻。它回答的问题是：截至 2025 年 3 月 31 日，这家公司拥有什么、欠什么、剩下多少属于股东？
There is a famous equation at the center of accounting, taught on the first day of every introductory course, and almost forgotten by the second week: 会计学的中心有一个著名的等式，每一门入门课的第一天都会教，又几乎在第二周就被遗忘：
Assets = Liabilities + Equity
资产 = 负债 + 权益 资产 = 负债 + 所有者权益
It is the single most important equation in finance, and once you internalize it, balance sheets stop looking like alphabet soup. Read aloud, it says: "everything the company has, was either lent to it or invested in it." That is it. There is no third source of value. Either someone gave you the money expecting it back (debt, accounts payable, deferred revenue — all liabilities) or someone gave it to you and accepted ownership in exchange (equity). The assets are what you did with the money. Every transaction a company conducts, no matter how exotic, ends up reshuffling these three buckets in a way that keeps the equation balanced — which is, of course, why it is called the balance sheet. 这是金融学最重要的一个等式。一旦您内化了它，资产负债表就不再像字母汤。读出来，它说："公司拥有的一切，要么是借来的，要么是投入的。"就是这样。没有第三个价值来源。要么有人把钱给您，期待将来收回 (债务、应付账款、递延收入 — 都是负债)，要么有人把钱给您，作为交换接受了所有权 (权益)。资产就是您用这些钱做了什么。任何交易，无论多复杂，最终都在这三个桶之间洗牌，并保持等式平衡 — 这就是为什么它叫"资产负债表"。
Almost every balance sheet has the same hierarchy, ordered from "easiest to turn into cash" to "hardest to turn into cash" on the asset side, and from "must be paid soonest" to "must be paid latest" on the liability side. We call these current (within one year) and non-current (more than one year). 几乎所有资产负债表都按同样的层级排列：资产侧按"最容易变现"到"最难变现"，负债侧按"最快要还"到"最晚要还"。我们称之为流动 (一年内) 与非流动 (一年以上)。
What the company owns or controls. 公司拥有或控制的东西。
Current (≤1 year):流动 (≤1 年)：
Non-current (>1 year):非流动 (>1 年)：
Who has a claim on the assets. 谁对资产拥有索取权。
Current liabilities (≤1 year):流动负债 (≤1 年)：
Non-current liabilities (>1 year):非流动负债 (>1 年)：
Equity:权益：
If the income statement tells you the year's story, the balance sheet tells you whether the company can survive the next storm. Alibaba's balance sheet, as of FY25A, is one of the strongest in the world. Of the ¥1.8 trillion in total assets, the company carries roughly ¥430bn in cash and short-term investments. Against that pile, it owes only about ¥230bn in debt. The difference — about ¥200bn, or US$28bn in cash — is the "net cash" figure that every analyst writes on the cover of every report. (The ¥86bn US$ figure quoted in the cover reflects long-term equity investments and other liquid items as well.) For comparison, the GDP of New Zealand is about US$245bn. Alibaba could roughly buy New Zealand for cash. This kind of balance-sheet position is what gives Alibaba its optionality: the company can spend on AI, return to shareholders, weather a recession, or fund acquisitions — without ever asking permission from a bond market. 如果说利润表讲述一年的故事，资产负债表则告诉您公司能否挺过下一场风暴。截至 FY25A，阿里巴巴的资产负债表是全球最强健之一。1.8 万亿元总资产中，公司持有约 4300 亿元的现金和短期投资。相对这堆现金，它只欠约 2300 亿元的债务。差额 — 约 2000 亿元，即 280 亿美元的现金 — 就是每位分析师写在每份报告封面上的"净现金"数字。(封面上引用的 860 亿美元数字还反映了长期股权投资和其他流动性项目。) 作为对照，新西兰的 GDP 约为 2450 亿美元。阿里巴巴大致可以现金买下新西兰。这种资产负债表状况赋予了阿里巴巴可选性：公司可以投资 AI、回馈股东、抵御衰退、为收购融资 — 全程无需向债券市场请示。
Stacked: assets vs. liabilities + equity (they must balance) 堆叠：资产 vs. 负债+权益 (必须平衡)
Divide current assets by current liabilities. If the answer is above 1.5, the company can almost certainly pay its short-term bills if something goes wrong. Below 1.0, it can't, and it needs to either raise cash or restructure. Alibaba's current ratio sits at 1.55 — exactly in the comfortable zone. Walmart sits around 0.9 (it pays suppliers slowly, collects from customers instantly, so it doesn't need a high ratio). Context matters; the heuristic is a first sniff. 用流动资产除以流动负债。如果商在 1.5 以上，公司几乎肯定能偿付短期账单 — 即便出了点问题。低于 1.0，则不能，公司要么筹资要么重组。阿里巴巴的流动比率为 1.55 — 正好在舒适区。沃尔玛大约 0.9 (它慢慢付供应商、立刻收顾客款，所以不需要高比率)。情境很重要；经验法则只是第一闻。
A common beginner's question is: "the balance sheet says Alibaba's equity is worth ¥1.09 trillion. The stock market says it's worth about US$280bn, or roughly ¥2 trillion. Which is right?" Both, in their own way. Book value reflects what the accountants recorded when each asset was acquired, adjusted for depreciation; it is largely a historical artifact. Market value reflects what investors believe the future cash-generation capacity of the assembled assets is worth today. For most companies, the two diverge widely. Coca-Cola's brand is worth tens of billions in market value but appears on the balance sheet at almost nothing — because the brand was built internally, not bought, and accounting rules forbid recognizing internally generated brands. Alibaba's Ant Group stake sits on the balance sheet at a written-down number; its market value, if Ant ever IPOs again, could be very different. This gap — between book and market — is what valuation chapters of this site exist to explain. 初学者常问："资产负债表说阿里巴巴的权益值 1.09 万亿元。股市说它值约 2800 亿美元，约 2 万亿元。哪个对？"以各自的方式，都对。账面价值反映的是会计师在每项资产被取得时记录的金额，再扣除折旧；它在很大程度上是一个历史产物。市场价值反映的是投资者认为这些资产组合在未来产生现金的能力今天值多少。对大多数公司，这两个数差异很大。可口可乐的品牌市场价值数百亿，但在资产负债表上几乎为零 — 因为品牌是内部建立的，不是买来的，会计准则禁止确认内部自创品牌。阿里巴巴持有的蚂蚁集团股权，在资产负债表上以减记的数字记录；如果蚂蚁哪天再次 IPO，其市值可能大不相同。账面与市场之间的差距 — 正是本网站估值章节存在的意义。
A profitable company can go bankrupt. A loss-making company can survive for decades. The reason is a difference between profit (an accounting concept) and cash (a thing you can spend). The cash flow statement is the bridge between the two. 一家盈利的公司可以破产。一家亏损的公司可以存活数十年。原因在于利润 (一个会计概念) 与现金 (一个您能花的东西) 之间的差异。现金流量表就是两者之间的桥梁。
Consider an obvious example. A construction company wins a ¥100 million contract to build an office tower. The customer will pay 10% on signing, 30% after the foundation is finished, 30% after the roof goes on, and 30% on handover, two years from now. Under accrual accounting (the system every public company uses), the construction company recognizes revenue as the building progresses — not as the cash arrives. So in the first year it might book ¥40m of revenue and ¥30m of cost, generating ¥10m of paper profit. Meanwhile, the cash it actually received was ¥10m on signing plus some of the milestone payments — and it spent ¥35m on labor and materials. On the income statement: profitable. In the bank: bleeding. Without the cash flow statement, you would never see the bleed. 考虑一个明显的例子。一家建筑公司拿到了 1 亿元合同，建一栋写字楼。客户在签约时付 10%、地基完成后付 30%、封顶后付 30%、两年后交付时付 30%。在权责发生制会计下 (所有上市公司都用这套体系)，建筑公司按工程进度确认收入 — 而不是按现金到账时间。所以第一年，它可能确认 4000 万元收入、3000 万元成本，产生 1000 万元的"账面利润"。同时，它实际收到的现金是签约时的 1000 万元加上部分里程碑付款 — 而它在人工和材料上花了 3500 万元。利润表上：盈利。账户里：失血。没有现金流量表，您永远看不到这个"失血"。
Every cash flow statement is split into three parts: 每张现金流量表都分成三部分：
Starts from net income, adds back non-cash items (depreciation, SBC, impairment), and adjusts for working-capital changes. The answer: how much cash did running the business actually generate? 从净利润起步，加回非现金项目 (折旧、股权激励、减值)，并调整营运资本变动。答案：经营业务实际产生了多少现金？
Capex (almost always negative), acquisitions (negative), divestitures (positive), and changes in long-term investments. This is the "what are we building for the future" line. 资本开支 (几乎总为负)、收购 (负)、出售 (正)、长期投资变动。这是"我们在为未来建设什么"那一段。
New borrowings (positive), debt repayments (negative), dividends paid (negative), share buybacks (negative), equity raises (positive). This shows how the company is moving cash to and from its capital providers. 新借款 (正)、还款 (负)、付股息 (负)、回购 (负)、增发 (正)。这表明公司在资本提供方之间如何调动现金。
Add the three together, adjust for FX, and you reconcile to the change in cash on the balance sheet. The cash flow statement is, in this sense, the proof that the balance sheet's cash line is correct. 把三段加起来，调整汇率，您就对上了资产负债表上现金余额的变动。从这个意义上说，现金流量表是资产负债表上现金科目的"证明"。
Operating, investing, financing flows (CNY bn) · free cash flow line 经营、投资、融资现金流 (十亿人民币) · 自由现金流折线
Almost everything we do in valuation is, at heart, a forecast of free cash flow (FCF). The reason is straightforward: a share of stock is a claim on future cash flows. Profit measures may be distorted by accounting choices; cash is cash. Most practitioners use a simple definition: 我们在估值中所做的几乎一切，本质上都是对自由现金流 (FCF) 的预测。原因很直接：一股股票是对未来现金流的索取权。利润度量可能因会计选择而扭曲；现金就是现金。大多数从业者使用一个简单定义：
FCF = Cash from operations − Capex
FCF = 经营现金流 − 资本开支 FCF = 经营现金流 − 资本开支
A more rigorous "unlevered FCF" — the version we'll use in the DCF chapter — also strips out the tax shield from interest, so as not to double-count financing decisions inside the operating estimate. But the intuition is the same: how much cash does this business throw off, after paying to maintain and expand the production base, that is genuinely available to give back to capital providers (debt and equity, combined)? 一个更严格的"无杠杆 FCF" — 我们将在 DCF 章节中使用 — 还会剔除利息税盾，以免在经营估算里重复计算融资决策。但直觉是一样的：这家公司在花钱维护并扩大生产基础之后，能产生多少真正可以返还给资本提供方 (债务+权益合并) 的现金？
A favorite trick of "growth at all costs" companies is to push more product into the channel at the end of each quarter, recognizing revenue but generating receivables instead of cash. Working capital (inventory + receivables − payables) balloons. Reported profit looks great; cash gets worse. The cash flow statement's "change in working capital" line is the place this shows up. If you see a company reporting strong earnings but a big negative working-capital line every period, you are looking at one of two things: (a) genuine growth pulling forward, or (b) channel stuffing. Earlier warnings of accounting fraud — Enron, WorldCom, Luckin Coffee — almost always show up first in cash flow long before they appear in headline earnings. "不惜代价增长"的公司最爱玩的把戏，是在每个季度末把更多产品压给渠道，确认收入却产生应收款而非现金。营运资本 (库存+应收-应付) 膨胀。报告利润看着很美；现金状况恶化。现金流量表上"营运资本变动"那一行，就是它现形的地方。如果您看到某家公司报告强劲盈利却伴随每期巨额负向营运资本变动，您看到的要么是 (a) 真实增长拉动，要么是 (b) 渠道压货。会计欺诈的早期信号 — 安然、世通、瑞幸咖啡 — 几乎总是先从现金流上出现，远早于头条盈利。
The income statement, the balance sheet, and the cash flow statement are not three separate documents. They are three views of one underlying reality, and any change to one of them forces specific, mechanical changes to the other two. Seeing those links is the difference between reading financial statements and understanding them. 利润表、资产负债表、现金流量表不是三份独立文件。它们是同一底层现实的三个视角，对其中一个的任何改动都会强制引发另两个的特定、机械的变动。看到这些联系，就是"读"财务报表和"懂"财务报表的区别。
There are five major links between the three statements. Memorize them, and you will be able to build any three-statement model from scratch: 三大报表之间有五大主要联系。记住它们，您就能从零搭建任何"三表模型"：
Net income from the P&L gets added to retained earnings on the balance sheet (after subtracting any dividend paid). This is why retained earnings grow year after year for profitable companies. 利润表上的净利润被加到资产负债表上的留存收益里 (扣除已付股息后)。这就是为什么盈利公司的留存收益会逐年增长。
Net income is the first line of the indirect-method cash flow statement, before non-cash adjustments. So every change to net income ripples directly into CFO. 净利润是间接法现金流量表的第一行，未做非现金调整前。所以每一次净利润变动都会直接传导到经营现金流。
Depreciation expense on the P&L reduces net PP&E on the balance sheet, but is added back on the cash flow statement (it was never cash). One expense, three statements. 利润表上的折旧费用减少资产负债表上的 PP&E 净额，但在现金流量表上被加回 (它从来不是现金)。一项费用，三张表都有体现。
Capital expenditure shows up as a cash outflow in CFI and a corresponding increase in PP&E on the balance sheet. Then it gets depreciated over its useful life, completing the loop. 资本开支在投资现金流中体现为现金流出，同时在资产负债表上增加 PP&E。然后按使用年限折旧，闭合循环。
The total of CFO + CFI + CFF, adjusted for FX, equals the change in the cash line on the balance sheet between two periods. This is the reconciliation that proves the model is internally consistent. CFO + CFI + CFF 的合计 (调整汇率) 等于资产负债表上现金科目在两期之间的变动。这就是证明模型内部一致的对账。
A share buyback is a cash outflow in CFF, an increase in treasury stock (a negative equity line) on the balance sheet, and a reduction in diluted shares outstanding that boosts EPS on the P&L. Three places, one transaction. 股票回购在融资现金流上是现金流出，在资产负债表上增加库存股 (一项负向权益)，同时减少稀释后流通股数从而推升利润表上的 EPS。三处发生，同一笔交易。
Click any arrow for a quick reminder of how the link works 点击任意箭头，快速回顾该联系如何运作
In FY24, Alibaba paid its first-ever annual dividend of US$1.00 per ADR. With ~2.4 billion ADRs outstanding, that's about US$2.4bn, or roughly ¥18bn at the prevailing exchange rate. Trace it through the statements: FY24，阿里巴巴第一次派发年度股息，每股 ADR 1.00 美元。当时约 24 亿股 ADR 流通，合计约 24 亿美元，按当时汇率约 180 亿元。让我们追踪它如何在报表中流动：
One transaction, three statements affected, every entry self-consistent. This kind of tracing — done thousands of times — is what an analyst means when she says "I built the model." A three-statement model is, mechanically, a giant set of these linked entries; when you change a single revenue cell, the model recomputes net income, cash flow, retained earnings, and ending cash in milliseconds, keeping every connection live. 一笔交易，影响三张表，每条分录自洽。这种追踪 — 重复数千次 — 就是分析师说"我搭好了模型"时的意思。一个三表模型在机械层面，就是这些联系分录的庞大集合；当您改动一个收入单元格，模型在几毫秒内重算净利润、现金流、留存收益和期末现金，保持每条联系实时生效。
A balance sheet of large numbers tells you almost nothing on its own. ¥1.8 trillion in assets sounds enormous, but is it? It depends on what those assets earn, what they are funded by, how quickly they turn over, and how they compare to last year and to peers. Ratios are the language we use to ask those questions, and they cluster into four families. 一张写着大数字的资产负债表，单独看几乎什么都说明不了。1.8 万亿元资产听起来很巨大，但真的巨大吗？这取决于这些资产能赚多少、由什么资金支持、周转多快、与去年和同行相比如何。比率就是我们用来提出这些问题的语言，可以分成四大家族。
Every ratio in finance is a sentence with two nouns and a slash, and every ratio is asking a question. The four families: 金融学里的每一个比率，都是"两个名词加一个斜杠"组成的一句话，每一个比率都在问一个问题。这四大家族是：
Is the business making money? Margins (gross, EBIT, net) tell you per ¥1 of revenue. Returns on capital (ROE, ROIC, ROA) tell you per ¥1 of money invested. 业务赚钱吗？利润率 (毛利、EBIT、净利) 告诉您每 1 元收入能赚多少。资本回报率 (ROE、ROIC、ROA) 告诉您每 1 元投入能赚多少。
Can the business pay its short-term bills? Current ratio, quick ratio, cash ratio, and CFO / current liabilities all answer this in different gradations of conservatism. 业务能付短期账单吗？流动比率、速动比率、现金比率、经营现金流/流动负债 — 用不同程度的保守性回答同一个问题。
How aggressively is the business financed by borrowed money? Debt-to-equity, debt-to-EBITDA, interest coverage. Higher leverage = higher expected returns and higher fragility. 业务使用借入资金有多激进？债务/权益、债务/EBITDA、利息保障倍数。杠杆越高 = 期望回报越高，脆弱性也越高。
How hard does each ¥1 of assets work? Asset turnover (revenue / assets), inventory days, receivable days, payable days — all measure operational tempo. 每 1 元资产工作有多努力？资产周转率 (收入/资产)、库存天数、应收天数、应付天数 — 都在衡量运营节奏。
If you ask any portfolio manager which ratios they actually look at first, two answers dominate: ROIC (return on invested capital) and the EBITDA margin trend. ROIC tells you whether the business creates value above its cost of capital — the technical definition of "is this a good business" — and the EBITDA margin trend tells you whether the operating leverage of the business is working for or against management. Both are easier to compute than you would think: 如果您问任何一位投资组合经理，他们最先看哪些比率，两个答案占据主导：ROIC (投入资本回报率) 与 EBITDA 利润率趋势。ROIC 告诉您业务是否能创造高于其资本成本的价值 — 这就是"是不是好生意"的技术定义 — EBITDA 利润率趋势告诉您业务的经营杠杆是为管理层服务还是反过来。两者都比您想象的更容易计算：
ROIC = NOPAT / (Debt + Equity − Cash)
where NOPAT = EBIT × (1 − tax rate)
其中 NOPAT = EBIT × (1 − 税率)
EBITDA margin trend = ΔEBITDA% / ΔRevenue%
positive = operating leverage; negative = cost creep
正 = 经营杠杆；负 = 成本蠕变
For Alibaba, FY25A ROIC was 9.5% (NOPAT of ¥140bn divided by invested capital of ¥1.5tn after deducting the cash pile). The company's WACC — the weighted average cost of capital, which we will derive in the DCF chapter — is 11.1%. ROIC below WACC means the company is, technically, destroying value at the margin: every new ¥1 invested earns less than the cost of the ¥1. This is the structural challenge Alibaba is wrestling with: the consolidated business has a return profile pulled down by AIDC and Local Services losses, even as the core TT and Cloud businesses earn very attractive returns individually. SOTP (Chapter 6) is the right tool for this. 对阿里巴巴而言，FY25A 的 ROIC 是 9.5% (NOPAT 1400 亿元除以扣除现金堆后的投入资本 1.5 万亿元)。公司的 WACC — 加权平均资本成本，我们将在 DCF 章节推导 — 是 11.1%。ROIC 低于 WACC 意味着技术上公司在边际上摧毁价值：每新增 1 元投入赚到的少于这 1 元的成本。这正是阿里巴巴面临的结构性挑战：合并业务的回报特征被 AIDC 和本地生活的亏损拉低，尽管核心的 TT 和云业务单独看回报非常有吸引力。SOTP (第六章) 才是处理这种情况的正确工具。
All ratios from the live Alibaba model · FY13A–FY30E 来自阿里巴巴动态模型的所有比率 · FY13A–FY30E
In the 1920s, DuPont's treasurer F. Donaldson Brown noticed something simple but profound: return on equity is mathematically the product of three smaller ratios — and that those three smaller ratios each tell you something different about the business. A century later, the framework still anchors how every analyst diagnoses what is changing inside a company. 1920 年代，杜邦公司财务主管 F. Donaldson Brown 发现了一个朴素却深刻的事实：股本回报率 (ROE) 在数学上等于三个更小比率的乘积 — 而这三个更小的比率，各自告诉您关于业务的不同信息。一个世纪后，这个框架仍然是每位分析师诊断公司内部变化的基石。
Start with the equation: 从等式开始：
ROE = (Net Income / Revenue) × (Revenue / Assets) × (Assets / Equity)
Net Margin × Asset Turnover × Equity Multiplier 净利率 × 资产周转率 × 权益乘数
Mathematically, the revenues cancel in the first two terms, and the assets cancel in the second and third. What you are left with is exactly net income / equity — ROE. So why bother decomposing? 数学上，前两项中收入相消，后两项中资产相消。最终剩下的，正是净利润/权益 — 也就是 ROE。那为什么还要拆解？
Because each of the three terms answers a different strategic question. A company can lift its ROE in three very different ways: 因为这三项分别回答不同的战略问题。一家公司可以用三种截然不同的方式提升 ROE：
Identifying which lever is moving is the analyst's job. A retailer (Walmart, JD.com) lives on the second lever: razor-thin margins (2–3%), high turnover (2–3×), modest leverage. A luxury brand (Hermès, Kweichow Moutai) lives on the first: enormous margins (30–50%), low turnover (0.6–1×), minimal leverage. A bank lives on the third: tiny margins on each transaction, huge leverage (often 10× or more equity multiplier — depositors are creditors). All three can have similar ROEs of 15–20%; the routes are entirely different. A change in ROE that you cannot trace to one of the three levers is a change you do not understand. 识别哪一个杠杆在动，是分析师的工作。零售商 (沃尔玛、京东) 靠第二个杠杆活：薄如刀片的利润率 (2-3%)、高周转 (2-3 倍)、温和杠杆。奢侈品品牌 (爱马仕、贵州茅台) 靠第一个杠杆活：巨大利润率 (30-50%)、低周转 (0.6-1 倍)、极少杠杆。银行靠第三个杠杆活：每笔交易微小利润、巨大杠杆 (权益乘数往往 10 倍以上 — 储户就是债权人)。三者都可以有相似的 15-20% ROE；路径完全不同。如果一项 ROE 变化您无法归因到这三个杠杆之一，那就是您没搞懂的变化。
Net margin × asset turnover × equity multiplier = ROE 净利率 × 资产周转率 × 权益乘数 = ROE
The three-step is a starting point. The five-step further decomposes net margin into "tax burden × interest burden × EBIT margin." Why? Because a company that improves ROE because tax rates went down is different from a company that improves ROE because it expanded operating margin. The former is policy luck; the latter is execution. In Alibaba's case, EBIT margin (the genuine operational lever) is the one we are betting on. Tax burden and interest burden have been roughly stable. 三步是起点。五步进一步把净利率拆解为"税负 × 利息负担 × EBIT 利润率"。为什么？因为一家公司因为税率下调而 ROE 上升，和因为经营利润率扩张而 ROE 上升，是两回事。前者是政策运气；后者是执行力。阿里巴巴的情况里，EBIT 利润率 (真正的经营杠杆) 才是我们押注的对象。税负与利息负担基本稳定。
ROE = Tax Burden × Interest Burden × EBIT Margin × Asset Turnover × Equity Multiplier
= (NI/Pretax) × (Pretax/EBIT) × (EBIT/Rev) × (Rev/Assets) × (Assets/Eq)
= (NI/税前) × (税前/EBIT) × (EBIT/收入) × (收入/资产) × (资产/权益)
All valuation is a forecast of the future. The forecast must be specific enough to be calculable and humble enough to be honest about what we don't know. We split the company into pieces small enough to model, attach a growth or margin assumption to each piece, and let the math reassemble the consolidated picture. 所有估值都是对未来的预测。预测必须足够具体到可以计算，又必须足够谦逊以诚实面对我们不知道的事。我们把公司拆成可建模的小块，给每块附上增长率或利润率假设，让数学把合并图景重新组装起来。
There are two philosophies for building a forecast. Top-down starts from the market: China e-commerce GMV will grow at x%, Alibaba's TT market share is y%, therefore TT revenue is GMV × share × take-rate. Bottom-up starts from the operating units: TT had ¥446bn of revenue last year, we think advertising can grow 8% (driven by the software-service-fee rollout) and customer management revenue 4% (modest growth in active sellers), so the segment grows 6% blended. Most professional models do both — bottom-up first, then cross-check against a top-down envelope. If your bottom-up forecast implies the company will gain 5 percentage points of market share in a saturated market, the top-down test catches the unrealistic ambition. 建立预测有两种哲学。自上而下从市场出发：中国电商 GMV 将增长 x%，阿里巴巴 TT 的市场份额是 y%，所以 TT 收入 = GMV × 份额 × 货币化率。自下而上从经营单元出发：TT 去年收入 4460 亿元，我们认为广告可以增长 8% (软件服务费推动)、客户管理收入增长 4% (活跃卖家温和增长)，所以分部综合增长 6%。大多数专业模型两种都用 — 自下而上先做，然后用自上而下的"信封"做交叉验证。如果您的自下而上预测意味着公司将在饱和市场获取 5 个百分点的市场份额，自上而下检验就能抓出不现实的雄心。
Every three-statement forecast rests on four sets of assumptions. Get these four right and the model takes care of itself: 每一份三表预测都建立在四组假设之上。把这四组做对，模型就会自洽：
By segment, geography, or product, depending on disclosure. Alibaba: six segments × five years = thirty growth assumptions. We start each one with the historical trend and adjust for known catalysts (TT take-rate change, Cloud AI demand) or risks (consumer weakness, regulatory drag). 按分部、地区或产品，取决于披露程度。阿里巴巴：六个分部 × 五年 = 30 个增长假设。我们以历史趋势作为起点，针对已知催化剂 (TT 货币化率变化、云 AI 需求) 或风险 (消费疲软、监管拖累) 进行调整。
Gross margin (mix shift, input costs), operating expenses (sales & marketing as % of revenue, R&D as % of revenue, G&A as % of revenue). The model lets each of these "step" toward a long-run target. 毛利率 (组合转移、投入成本)、营业费用 (销售营销占收入%、研发占收入%、管理占收入%)。模型允许每项"逐步"接近长期目标。
Capex / revenue (especially important for asset-heavy businesses), depreciation / revenue, working-capital intensity. Alibaba's capex / revenue is jumping from 4% to 12% as it builds AI infrastructure. 资本开支/收入 (对重资产业务尤其重要)、折旧/收入、营运资本密度。阿里巴巴的资本开支/收入正从 4% 跃升至 12%，因为它在建设 AI 基础设施。
How much will the company buy back / pay out / pay down? This is critical for EPS, share count, and net cash trajectory. Alibaba: ~US$13bn of buybacks per year + ~US$3bn of dividends. 公司将回购/分红/还债多少？这对 EPS、股本数、净现金轨迹至关重要。阿里巴巴：每年回购约 130 亿美元 + 派息约 30 亿美元。
Beginners often anchor too tightly on the recent historical trend ("revenue grew 8% last year, so let's assume 8% next year"). Experienced analysts know this fails at inflection points — both up (companies coming out of a cycle) and down (mature companies running out of TAM). The Alibaba model assumes the consolidated growth rate accelerates from FY25 (5.9%) into FY27 (9.5%) because the segment-level build (cloud reacceleration + AIDC growth + TT normalization) implies that, even though the headline trend would suggest deceleration. Building from the bottom prevents this trap. 初学者常常过度锚定最近的历史趋势 ("去年收入增长 8%，所以下一年也假设 8%")。经验丰富的分析师知道这在拐点 — 无论是向上 (公司走出周期) 还是向下 (成熟公司用尽 TAM) — 都会失效。阿里巴巴的模型假设合并增长率加速，从 FY25 的 5.9% 到 FY27 的 9.5%，因为分部级别的搭建 (云加速 + AIDC 增长 + TT 正常化) 意味着如此，尽管头条趋势可能暗示减速。从底层构建可避免这个陷阱。
The chart below is the segment build that drives the entire revenue line. The default values match our base-case model. Push any slider and watch consolidated FY27E revenue update in real time. This is essentially what an analyst does for hours every day: hypothesis, slider, observe, iterate. 下面的图表就是驱动整条收入线的分部搭建。默认值匹配我们的基准模型。拖动任意滑块，实时观察 FY27E 合并收入更新。这就是分析师每天数小时所做的：假设、滑块、观察、迭代。
Move the sliders. The chart and total below recompute live. 拖动滑块。图表与底部合计实时重算。
A valuation method converts a forecast of cash, profits, or growth into a single number that represents "what one share is worth today." There are three families of method, each with strengths and blind spots. Doing all three and triangulating is how professional analysts arrive at a price target they can defend. 一种估值方法把对现金、利润或增长的预测转化为一个单一数字，代表"今天一股值多少"。方法分为三大家族，各有优势与盲区。三种都做并三角验证，就是专业分析师得出可辩护目标价的方式。
The "purist" method. Forecast all the future cash flows of the company, discount them back to today at the cost of capital, sum them up, and that's the enterprise value. Add cash, subtract debt, divide by shares — that's the per-share value. The strength: it explicitly captures the time value of money and growth. The weakness: it is exquisitely sensitive to two long-tail assumptions — the discount rate and the terminal growth rate. "纯粹派"方法。预测公司未来所有现金流，按资本成本折现回今天，加总即为企业价值。加现金、减债务、除以股本 — 就是每股价值。优势：明确捕捉了货币时间价值与增长。劣势：对两个长尾假设 — 贴现率与终值增长率 — 极度敏感。
For multi-business companies (Alibaba, Amazon, Berkshire). Value each segment separately at the multiple appropriate for its business model (Cloud at a software multiple, Logistics at a logistics multiple, etc.), sum the pieces, add cash, subtract debt. Reveals "hidden" value masked by consolidated metrics. Vulnerability: cherry-picking comparables and conglomerate discount. 用于多业务公司 (阿里巴巴、亚马逊、伯克希尔)。按每个分部商业模式合适的倍数分别估值 (云用软件倍数、物流用物流倍数等等)，加总各块，加现金、减债务。揭示被合并指标掩盖的"隐藏"价值。脆弱性：可比公司挑选偏差与"集团折扣"。
The "market-anchored" method. Find peer companies, look at the multiples the market is paying for them (P/E, EV/EBITDA, EV/Revenue), and apply those multiples to the target. Strength: anchored to what the market is actually paying right now. Weakness: if the entire peer set is mispriced, comps just confirm the mispricing. "市场锚定"方法。找出同行公司，看市场为它们支付的倍数 (PE、EV/EBITDA、EV/收入)，把这些倍数应用到目标公司。优势：锚定于市场当前实际支付的价格。劣势：如果整个同行组都被错误定价，可比公司只能确认这一错误定价。
No single method is "right." The Nobel-prize winners of finance (Modigliani, Miller, Markowitz) would tell you DCF is the only logically defensible framework — but Modigliani also wrote about how impossible it is to estimate the perpetual growth rate. Practitioners settle for a discipline: do at least two methods, ideally three, and let the average inform the price target. When DCF and comps agree, you have a high-confidence view. When they disagree, you have a research question — and a research question is more valuable to a buy-side reader than a false-precision answer. 没有一种方法是"对的"。金融学的诺贝尔奖得主们 (莫迪利亚尼、米勒、马科维茨) 会告诉您 DCF 是唯一在逻辑上可辩护的框架 — 但莫迪利亚尼也写过永续增长率有多难估计。从业者退而求其次：至少做两种方法，理想是三种，让平均值告知目标价。当 DCF 和可比公司一致时，您拥有高置信度的观点。当它们分歧时，您有一个研究问题 — 而研究问题对买方读者比"虚假精确"的答案更有价值。
The discounted cash flow model is the closest finance has to a first-principles tool. Strip away the layers, and it is just: forecast the cash, weight it by time and risk, and add it up. Building one is a rite of passage for every analyst — and the model below is fully interactive. 现金流贴现模型是金融学最接近"第一性原理"的工具。剥开层层外壳，它就是：预测现金、按时间和风险加权、加起来。搭建一份 DCF 是每位分析师的"成人礼" — 下面的模型完全可交互。
The core insight of DCF is the time value of money. Would you rather have ¥100 today or ¥100 in five years? Today, obviously — because you could invest the ¥100, earn a return, and have more than ¥100 in five years. The mirror image: ¥100 in five years is worth less than ¥100 today. The amount less depends on the rate at which you could invest. That rate is your "discount rate." Mathematically: DCF 的核心洞见是货币的时间价值。您是想今天拿 100 元，还是五年后拿 100 元？显然今天 — 因为您可以投资这 100 元、赚取回报，五年后会有超过 100 元。镜像版：五年后的 100 元价值低于今天的 100 元。低多少，取决于您能用什么回报率投资。这就是您的"贴现率"。数学上：
PV = CF / (1 + r)t
CF = the cash flow, r = the discount rate per period, t = number of periods away CF = 现金流, r = 每期贴现率, t = 距今的期数
At a 10% discount rate, ¥100 five years from now is worth ¥100 / 1.105 ≈ ¥62 today. At a 5% discount rate, it would be worth ¥78. Notice how sensitive the answer is to the discount rate — and how the sensitivity grows with time. A cash flow 30 years away is barely worth anything at high discount rates; it becomes the dominant chunk of the valuation at low discount rates. This is why a 50-basis-point change in your discount rate can swing a DCF target by 20% or more. 在 10% 的贴现率下，五年后的 100 元今天值 100 / 1.105 ≈ 62 元。在 5% 贴现率下，则值 78 元。注意答案对贴现率有多敏感 — 这种敏感性随时间增长。30 年后的现金流在高贴现率下几乎不值钱；在低贴现率下却会成为估值的主导部分。这就是为什么贴现率变化 50 个基点就能让 DCF 目标价波动 20% 以上。
We start from operating income (EBIT), apply the tax rate to get NOPAT (net operating profit after tax — what the company would earn if it had no debt). Add back depreciation (non-cash), subtract capex (real cash going out), subtract the change in working capital (cash absorbed by inventory and receivables, less the float from payables and deferred revenue). The result is "unlevered free cash flow" — the cash available to all capital providers (debt and equity, combined) before any financing decisions. 从经营利润 (EBIT) 起步，应用税率得到 NOPAT (税后净营业利润 — 如果公司没有债务，它能赚的)。加回折旧 (非现金)、减去资本开支 (实际现金流出)、减去营运资本变动 (库存和应收款占用的现金，减去应付款与递延收入的"浮存")。结果就是"无杠杆自由现金流" — 在任何融资决策之前，对所有资本提供方 (债务+权益合并) 可用的现金。
| CNY mn百万人民币 | FY26E | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| EBIT | 198,149 | 240,429 | 278,634 | 317,837 | 358,520 |
| Tax on EBIT (21.5%)EBIT 税收 (21.5%) | (42,602) | (51,692) | (59,906) | (68,335) | (77,082) |
| NOPAT | 155,547 | 188,737 | 218,728 | 249,502 | 281,438 |
| + D&A+ 折旧摊销 | 69,620 | 76,234 | 84,238 | 91,820 | 99,165 |
| − Capex− 资本开支 | (96,397) | (99,690) | (97,198) | (91,820) | (91,537) |
| − Δ Working capital− 营运资本变动 | (3,736) | (5,088) | (6,157) | (5,832) | (5,650) |
| Unlevered FCF无杠杆 FCF | 125,033 | 160,193 | 199,611 | 243,670 | 283,416 |
The cash flows we just forecast are available to both debt and equity holders. So the discount rate has to be the blended cost of those two pools of capital — the weighted average cost of capital, or WACC. Two ingredients: 我们刚预测的现金流可供债务和权益持有人共同分享。所以贴现率必须是这两类资本的混合成本 — 加权平均资本成本，即 WACC。两个原料：
Weight each by the company's target capital structure (Alibaba is 92% equity / 8% debt — a very conservative balance sheet) and you get the WACC: 按公司的目标资本结构加权 (阿里巴巴 92% 权益/8% 债务 — 非常保守的资产负债表)，得到 WACC：
WACC = (E/V) × ke + (D/V) × kd × (1−t)
= 0.92 × 11.7% + 0.08 × 5.0% × (1−21.5%)
= 11.06%
Companies don't end in FY30. They continue. We deal with the infinity by assuming that, from year 6 onwards, the company grows at some modest perpetual rate g, and applying the Gordon growth formula to capitalize the cash flow forever: 公司不会在 FY30 终止。它们会继续。我们用以下方式处理"无限"：假设从第 6 年起，公司以某个温和的永续增长率 g 增长，并应用戈登增长公式将现金流永续资本化：
TV = CFt+1 / (WACC − g)
For Alibaba: 283,416 × (1.025) / (0.1106 − 0.025) ≈ ¥3,396 bn 阿里巴巴：283,416 × (1.025) / (0.1106 − 0.025) ≈ 3.396 万亿元
The terminal value, discounted back to today, often represents 70–80% of total enterprise value in a DCF. That's terrifying once you realize it. It also means the terminal growth rate (we used 2.5%) is the single most consequential assumption in the model — bigger than any year's revenue forecast. Convention: keep g well below long-run nominal GDP (3–4%). If your terminal growth is 5% and global GDP grows at 3%, you are implicitly saying your company will eventually own the world — usually wrong. 折回今天的终值，在 DCF 中往往占企业总价值的 70-80%。一旦意识到这点会让人不寒而栗。这也意味着终值增长率 (我们用了 2.5%) 是模型里最重要的一个假设 — 比任何一年的收入预测都大。惯例：把 g 保持在远低于长期名义 GDP (3-4%) 的水平。如果您用 5% 的终值增长率而全球 GDP 只有 3%，您隐含的意思就是您的公司最终会拥有整个世界 — 通常错的。
PV of FCFs + PV of terminal value = enterprise value (¥2,873bn for Alibaba). Then a "bridge" to equity value: FCF 现值 + 终值现值 = 企业价值 (阿里巴巴 28,730 亿元)。然后是到权益价值的"桥梁"：
| CNY mn百万人民币 | |
|---|---|
| PV of explicit-period FCF明确期 FCF 现值 | 754,733 |
| PV of terminal value终值现值 | 2,118,394 |
| Enterprise value企业价值 | 2,873,127 |
| + Cash & ST investments+ 现金及短期投资 | 428,093 |
| + Long-term equity investments (Ant et al.)+ 长期股权投资 (蚂蚁等) | 350,000 |
| − Total debt− 总债务 | (230,703) |
| − Non-controlling interest− 少数股东权益 | (68,535) |
| Equity value (CNY mn)权益价值 (百万人民币) | 3,351,982 |
| ÷ USD/CNY (7.20)÷ 美元/人民币 (7.20) | 465,553 |
| ÷ Diluted ADR shares (mn)÷ 稀释 ADR 股本 (百万) | 2,300 |
| Implied value per ADR (today)每股 ADR 隐含价值 (今天) | $202.41 |
| × 1-year forward roll (1+ke)× 1 年向前滚动 (1+ke) | 1.1168 |
| 12-month DCF price target12 个月 DCF 目标价 | $226.05 |
Done. The $226 price target you see on the cover page is the output of those four steps. Every chart that follows in our analyst's report is just an argument for some piece of this calculation — a defense of the WACC, a defense of the growth profile, a defense of the terminal value. Once you see the architecture, the chaos resolves into a building. 完成。封面上的 226 美元目标价就是这四步的产物。分析师报告里之后的每一张图，都是在为这道计算中的某一块辩护 — 为 WACC 辩护、为增长轨迹辩护、为终值辩护。一旦您看到这套架构，混乱就化为一座建筑。
All four levers from the live model — WACC, terminal growth, FCF growth, terminal multiple cross-check. 来自动态模型的四个杠杆 — WACC、终值增长、FCF 增长、终值倍数交叉验证。
A single DCF answer is fragile. Professional models almost always pair a "base case" output with a sensitivity table showing how the answer changes across a range of WACCs and terminal growth rates. Read the heat-map below diagonally: the upper-left cell is "low WACC + low g" and gives a fairly high PT; the bottom-right is "high WACC + high g" — also reasonable. The "wrong" cells are upper-right (low WACC + high g — implausible) and lower-left (high WACC + low g — defines the floor). Where in this map do you actually believe? That's the analyst's call. 单一的 DCF 答案是脆弱的。专业模型几乎总把"基准情景"输出与展示答案在一系列 WACC 与终值增长率下如何变化的敏感性表配对。下面的热力图按对角线读：左上格是"低 WACC + 低 g"，给出相当高的目标价；右下格是"高 WACC + 高 g" — 也合理。"错的"格在右上 (低 WACC + 高 g — 不太可信) 和左下 (高 WACC + 低 g — 定义了下限)。您实际相信这张图里的哪一格？这就是分析师的判断。
A consolidated income statement averages the genius of TT with the losses of AIDC and the capex of Cloud. The market then prices that average. SOTP undoes the averaging — values each business at the multiple it would command if it were standalone — and asks whether the sum is more than the consolidated whole. 合并利润表把 TT 的伟大与 AIDC 的亏损、云的资本开支平均起来。市场然后给这个平均值定价。SOTP 撤销了这种平均 — 用每个业务作为独立公司时应得的倍数来估值 — 然后问：分部之和是否大于合并整体？
SOTP is the right tool for "conglomerate" structures: holding companies, industrial groups, multi-platform internet companies. Anywhere the consolidated optic understates the value of the parts because their business models are too different to be priced on a single multiple. The classic example is Berkshire Hathaway — railroads + insurance + utilities + equity portfolio — but Alibaba is just as multi-headed: e-commerce + cloud + logistics + local services + media + investments. SOTP 是处理"集团"结构的正确工具：控股公司、工业集团、多平台互联网公司。任何场景下，合并视角因业务模式差异太大、无法用单一倍数定价而低估了各部分价值。经典例子是伯克希尔哈撒韦 — 铁路+保险+公用事业+股权组合 — 但阿里巴巴同样"多头"：电商+云+物流+本地生活+媒体+投资。
The art of SOTP is matching each segment to the right comparable set. A stable, cash-generating marketplace (TT) deserves an EV/EBITDA multiple in the 7–9× range (similar to Tencent's domestic ad/games business). A high-growth cloud business deserves an EV/Revenue multiple, since it isn't yet GAAP-profitable enough for an EBITDA multiple to be meaningful — 4–6× revenue, the kind of multiple Snowflake or AWS-pure-play businesses fetch. A barely-profitable cross-border e-commerce business deserves an EV/Revenue around 1×, akin to a logistics or distribution business. Pre-IPO logistics (Cainiao) might trade at 0.6–0.8× revenue, similar to ZTO or JD Logistics. SOTP 的艺术在于把每个分部匹配到正确的可比公司集合。稳定的、现金创造型的市场 (TT) 配得上 7-9 倍的 EV/EBITDA 倍数 (类似腾讯国内广告/游戏业务)。高增长云业务配得上 EV/收入倍数，因为它还未盈利到可以用 EBITDA 倍数 — 4-6 倍收入，类似 Snowflake 或 AWS 纯业务的倍数。一个勉强盈利的跨境电商业务配得上约 1 倍 EV/收入，类似物流或分销业务。Pre-IPO 物流 (菜鸟) 可能交易在 0.6-0.8 倍收入，类似中通或京东物流。
Click any wedge to see the underlying math 点击任意扇形，查看背后的计算
| Segment分部 | FY27E revFY27E 收入 | EBITAEBITA | Multiple倍数 | Type类型 | EV (US$bn)EV (十亿美元) |
|---|---|---|---|---|---|
| Taobao & Tmall淘宝天猫 | 531 | 231 | 7.0× | EV/EBITDA | 228 |
| AIDC | 229 | 9 | 1.0× | EV/Rev | 32 |
| Cloud Intelligence云智能 | 222 | 37 | 4.5× | EV/Rev | 140 |
| Cainiao菜鸟 | 151 | 8 | 0.8× | EV/Rev | 17 |
| Local Services本地生活 | 109 | 4 | 1.0× | EV/Rev | 15 |
| DME数字媒体 | 25 | 1 | 0.6× | EV/Rev | 2 |
| Others其他 | 121 | (5) | 0.8× | EV/Rev | 14 |
| Total operating EV经营 EV 合计 | 448 | ||||
| + Cash & ST inv (FY27E)+ 现金及短投 (FY27E) | +115 | ||||
| + Equity investments+ 股权投资 | +46 | ||||
| + Cainiao IPO carry+ 菜鸟 IPO 留存 | +7 | ||||
| − Debt & minorities− 债务及少数股东 | −39 | ||||
| Equity value (US$bn)权益价值 (十亿美元) | 577 | ||||
| ÷ 2,170 mn ADR · × roll = PT÷ 2,170 百万 ADR · × 滚动 = 目标价 | $297 | ||||
$297 from SOTP versus $226 from DCF. The DCF answers "if I owned the consolidated cash flows, what would I pay?" SOTP answers "if I could split the company into six and sell each separately, what would I get?" The $71 gap is the implicit "conglomerate discount" — the value the market may eventually unlock if Cainiao IPOs, if Cloud is spun out, if AIDC reaches profitability and the segments are explicitly recognized. The blended price target of $262 in our report is the average of the two, weighted by confidence in each approach. SOTP 的 297 美元 vs. DCF 的 226 美元。DCF 回答的是"如果我拥有合并现金流，我会付多少？"SOTP 回答的是"如果我能把公司拆成六块分别卖，我能拿到多少？"71 美元的差距，就是隐含的"集团折扣" — 如果菜鸟 IPO、云被分拆、AIDC 达到盈利、分部被明确认可，市场最终可能释放的价值。我们报告里 262 美元的混合目标价，就是两者按对每种方法的信心加权后的均值。
Comps are the most quoted, least understood form of valuation. They are quick — fifteen minutes to assemble — and intuitive: "JD trades at 8× FY26 P/E; Alibaba trades at 9.4×; pricing seems consistent." But trade comparables conceal a great deal of nuance, and a sloppy comp set will mislead you faster than any other method. 可比公司是被引用最多、被理解最少的估值形式。它们快 — 15 分钟可以搭好 — 而且直观："京东 FY26 P/E 8 倍；阿里巴巴 9.4 倍；定价看起来一致。" 但交易可比公司隐藏着大量微妙之处，一份草率的可比集合会比任何其他方法都更快地误导您。
Alibaba versus its peer set: 阿里巴巴对比其同行集：
All multiples FY26E · BABA highlighted in burnt orange 所有倍数 FY26E · BABA 高亮为橙色
Reading the chart: BABA trades at a clear discount to the global average on every multiple. 9.4× P/E versus the 16× peer median (a 41% discount). 4.5× EV/EBITDA versus 9.5× (a 53% discount). 1.4× EV/Revenue versus 2.0× (a 30% discount). The market is implicitly saying that Alibaba's growth is structurally lower than peers, its risk is structurally higher (China regulatory, US-listing, RMB-translation), or both. The bullish view is that this discount overshoots — that Alibaba's improved capital structure, capital return discipline, and re-acceleration in Cloud justify a multiple closer to peers. The bearish view is that the discount is fundamentally deserved and unlikely to close. Buy-side investors take a position on this very question; sell-side analysts present the data and recommend a side. 解读图表：BABA 在每个倍数上都明显低于全球均值。9.4× P/E vs. 同行中位数 16× (折价 41%)。4.5× EV/EBITDA vs. 9.5× (折价 53%)。1.4× EV/Revenue vs. 2.0× (折价 30%)。市场隐含的意思是阿里巴巴的增长结构性低于同行、风险结构性高于同行 (中国监管、美国上市、人民币换算)，或两者兼有。看多观点是这种折价"过头" — 阿里巴巴改善的资本结构、资本回报纪律、云业务加速，理应支撑接近同行的倍数。看空观点是该折价根本上应得且难以缩窄。买方投资者就这一问题站队；卖方分析师呈现数据并推荐立场。
A common beginner error: take the peer median P/E of 16×, multiply by Alibaba's FY26E EPS of $10.04, and conclude the price target should be $161. This is wrong because (a) Alibaba's geography, growth, and risk are not the peer median's, and (b) it ignores the cash on Alibaba's balance sheet (which represents ~$30 of the share price in pure liquidity, not profit). The correct move is either to pick a single best comp and adjust for differences (e.g., apply Tencent's multiple but discount for Alibaba's slower growth) or to do enterprise-value-based comps (EV/EBITDA) and then bridge through net cash to get to equity value per share. 初学者常见错误：取同行中位数 P/E 16×，乘以阿里巴巴 FY26E EPS 10.04 美元，得出目标价应为 161 美元。这是错的，因为 (a) 阿里巴巴的地区、增长、风险不是同行中位数；(b) 它忽视了阿里巴巴资产负债表上的现金 (代表了股价里约 30 美元的纯流动性，不是利润)。正确做法要么挑一个最佳可比公司并对差异进行调整 (例如应用腾讯的倍数但因阿里巴巴增长较慢而打折)，要么做基于企业价值的可比 (EV/EBITDA)，然后通过净现金桥接至每股权益价值。
A CEO does only two things, and one of them is capital allocation: deciding how every dollar of free cash flow is deployed. Re-invest in the existing business? Acquire a competitor? Pay a dividend? Buy back stock? Pay down debt? Sit on it? Each choice has compounding consequences over a decade. The Alibaba capital allocation story over the past three years is one of the most important transitions in modern Chinese corporate finance. CEO 只做两件事，其中一件就是资本配置：决定每一元自由现金流如何部署。再投资现有业务？收购竞争对手？派股息？回购股票？偿还债务？囤在手里？每个选择都有十年期复利后果。过去三年阿里巴巴的资本配置故事，是现代中国企业财务里最重要的转型之一。
Every dollar of free cash flow that lands in the corporate treasury has to be sent somewhere. The destinations cluster into five buckets, each with a different risk-return profile and different signaling value to the market. 每一元落入公司账户的自由现金流，都必须被送往某处。目的地分成五个桶，每个有不同的风险回报特征，并向市场传递不同的信号价值。
Buy more servers, build more warehouses, hire more engineers. The "default" choice for a growing company. ROI = incremental cash flows / incremental capex. If ROI > WACC, it creates value. 买更多服务器、建更多仓库、招更多工程师。成长公司的"默认"选择。ROI = 增量现金流 / 增量资本开支。若 ROI > WACC，创造价值。
Faster than building, more expensive than buybacks. Academic literature suggests roughly 60% of M&A destroys value for the acquirer. Successful M&A requires the buyer to overpay less than the cost savings or revenue synergies are worth. 比自建快，比回购贵。学术文献显示约 60% 的并购摧毁了收购方价值。成功的并购要求收购方的超付低于成本节约或收入协同的价值。
A fixed, recurring cash distribution to shareholders. Hard to cut without market punishment, so management treats the dividend as a long-term commitment. Alibaba initiated its first dividend in FY24 ($1.00/ADR) — a watershed signaling event. 向股东进行固定、经常性的现金分配。一旦削减就会被市场惩罚，所以管理层把分红视为长期承诺。阿里巴巴在 FY24 启动首次分红 ($1.00/ADR) — 一个分水岭式信号事件。
More flexible than dividends — turn on and off without long-term commitment. Mechanically: company repurchases shares, retires them or holds in treasury, reduces share count, mechanically raises EPS. Most effective when the stock is undervalued. Alibaba has retired ~600 million ADRs since FY21. 比分红更灵活 — 可以开关而不必长期承诺。机制：公司回购股票、注销或留作库存股、减少股本、机械性提升 EPS。在股价被低估时最有效。FY21 起阿里巴巴已注销约 6 亿股 ADR。
The "safety" choice. Reduces fragility but locks in the after-tax cost of debt as a return. For companies with already-low leverage like Alibaba, the marginal value of additional cash is low — which is why the company has been returning capital aggressively. "安全"选项。降低脆弱性但把税后债务成本锁定为回报。对像阿里巴巴这种已经低杠杆的公司，增加现金的边际价值很低 — 这就是公司一直在激进返还资本的原因。
The market reads capital allocation like tea leaves. Aggressive buybacks at depressed prices say "management thinks the stock is cheap." A first-ever dividend says "this is now a mature, predictable business." Heavy capex says "we see growth opportunities ahead." The signaling content is sometimes more valuable than the cash itself. 市场把资本配置当茶叶占卜。在低价时激进回购说"管理层认为股价便宜"。首次分红说"这现在是成熟、可预测的业务"。重资本开支说"我们看到前方的增长机会"。信号含义有时比现金本身更有价值。
Between FY21 and FY25, Alibaba shifted from a company that spent capital aggressively on acquisitions (Sun Art, Yintai, NetEase Music stakes, Cainiao build-out) to a company that returns capital aggressively to shareholders. The total spend on buybacks plus dividends has gone from US$3.7bn in FY21 to US$17.5bn in FY25 — roughly a 5× increase. Cumulatively over the four years, Alibaba has returned about US$60bn to shareholders. For context, the total market cap is around US$280bn — meaning the company has retired or paid out roughly one-fifth of its own market value in four years. This is a scale of capital return rarely seen in Chinese internet, and it is the single most important behavioral change underpinning the bull case. 从 FY21 到 FY25，阿里巴巴从一家在收购上激进花费资本 (大润发、银泰、网易云音乐入股、菜鸟搭建) 的公司，转变为一家激进向股东返还资本的公司。回购加分红的总支出从 FY21 的 37 亿美元增长到 FY25 的 175 亿美元 — 大约 5 倍。四年累计，阿里巴巴已向股东返还约 600 亿美元。作为对照，总市值约 2800 亿美元 — 意味着公司在四年内注销或支付了相当于自身市值约五分之一。这是中国互联网很少见的资本返还规模，也是支撑看多论点最重要的行为转变。
Imagine a company with ¥100bn of net income and 1 billion shares outstanding. EPS = ¥100. Now the company buys back 50 million shares (5% of float) at the market price. Net income is unchanged (no operations are affected). Share count is now 950 million. EPS = ¥105.26. The EPS just grew 5.26% without the business doing anything. If the multiple is constant, the stock price rises by 5.26% too. Over five years, retiring 4-5% of shares per year compounds to a 20-25% mechanical EPS lift — independent of any operating progress. 想象一家公司有 1000 亿元净利润和 10 亿股流通股。EPS = ¥100。现在公司按市场价回购 5000 万股 (流通量的 5%)。净利润不变 (经营不受影响)。股本数现在是 9.5 亿。EPS = ¥105.26。EPS 在业务什么都没做的情况下增长了 5.26%。如果倍数不变，股价也上涨 5.26%。五年内，每年注销 4-5% 股本，复利累积成 20-25% 的机械 EPS 提升 — 与任何经营进展无关。
There is, however, a deep subtlety: buybacks only create value if the stock is purchased below intrinsic value. Buying back stock at $300 when intrinsic value is $150 transfers wealth from the company (and remaining shareholders) to the sellers. The Warren Buffett rule on buybacks: only when the stock trades below conservative intrinsic value. For Alibaba at $118 with our DCF-derived intrinsic of $226, every dollar of buyback today creates approximately one dollar of value for remaining shareholders — exactly the kind of opportunity disciplined CFOs hunt for. 不过有一个深层微妙：回购只在股票低于内在价值时创造价值。在内在价值 150 美元时按 300 美元回购股票，将财富从公司 (和剩余股东) 转移给卖方。沃伦·巴菲特关于回购的规则：仅当股票交易低于保守内在价值时。对于 118 美元的阿里巴巴和我们 DCF 推导出的 226 美元内在价值，今天的每一美元回购为剩余股东创造约一美元的价值 — 这正是有纪律的 CFO 寻找的机会。
A buyback funded from operating cash flow is unambiguously good when the stock is cheap. A buyback funded by issuing new debt is more controversial: it raises financial risk in exchange for the EPS lift. A buyback that offsets stock-based compensation (i.e., the company is dilution-neutral, not actually retiring shares) is the lowest-quality form — it merely prevents per-share value destruction rather than creating per-share value. Read the cash flow statement carefully: net share count change tells you what's actually happening. 来自经营现金流的回购，在股价便宜时无疑是好事。来自新增债务的回购更具争议：以提高金融风险换取 EPS 提升。抵消股权激励的回购 (即公司只是稀释中性，并未真正注销股本) 是最低质量的形式 — 它只是防止每股价值毁损，而非创造每股价值。仔细阅读现金流量表：净股本数变化告诉您真正发生了什么。
Everything we've built so far has been a toolkit. Now we point the toolkit at one company in detail. Alibaba is the perfect case for a beginner: it has six segments at very different stages of maturity, a transformative recent management change, a fortress balance sheet, a beaten-down stock, and a real, fundamental debate about its future that is impossible to resolve without the techniques in this site. 至此我们搭建的都是工具箱。现在把工具箱对准一家公司展开。阿里巴巴是初学者完美的案例：它有六个处于截然不同成熟阶段的分部、近期一次变革性的管理层更替、一份堡垒般的资产负债表、一只被打压的股票，以及一场关于其未来的真实而根本的辩论 — 这场辩论没有本网站的工具是无法解决的。
Alibaba Group Holding Ltd. was founded in 1999 by Jack Ma and a small group of co-founders, in a small apartment in Hangzhou, China. Twenty-five years later, it is the consolidated parent of the largest e-commerce franchise in the world (by GMV), the third-largest public cloud computing infrastructure in the world (after AWS and Azure), a top-five international online marketplace, the largest food-delivery / local-services platform in China by some measures, China's largest pure-play smart logistics network, and a streaming media business with hundreds of millions of users. As of the most recent fiscal year (FY25A, ended March 2025), the company reported consolidated revenue of ¥996bn — roughly $140bn — and net income of ¥130bn (about $18bn). The market cap floats around US$280bn at the time of writing. 阿里巴巴集团控股有限公司由马云与一小群联合创始人于 1999 年在中国杭州的一间小公寓里创立。25 年后，它是世界最大电子商务特许经营 (按 GMV 计算) 的合并母公司、世界第三大公有云基础设施提供商 (仅次于 AWS 和 Azure)、全球前五大国际在线市场之一、按某些口径中国最大的外卖/本地生活平台、中国最大的纯智能物流网络，以及一项拥有数亿用户的流媒体业务。截至最近财年 (FY25A，2025 年 3 月结束)，公司报告合并收入 9960 亿元 — 约 1400 亿美元 — 净利润 1300 亿元 (约 180 亿美元)。撰写本文时市值约 2800 亿美元。
In 2023, after several years of regulatory pressure and a sweeping reorganization, Jack Ma's chosen successor Daniel Zhang stepped down. Joe Tsai (the long-serving co-founder) became chairman, and Eddie Wu (a low-profile founding engineer) became CEO. The new team's mantra: refocus on the consumer; refocus on the cloud; refocus on capital discipline. The market initially didn't believe it. The stock fell to multi-year lows. The reorganization split Alibaba into six "business groups," each with the structural option to raise its own capital or eventually go public. The six segments we will now walk through were the direct organizational consequence of that 2023 reset. 2023 年，经过数年监管压力与全面重组，马云钦定的继任者张勇辞任。蔡崇信 (长期任职的联合创始人) 接任董事长，吴泳铭 (低调的创始工程师) 接任 CEO。新团队的口号：重新聚焦消费者；重新聚焦云；重新聚焦资本纪律。市场起初并不相信。股价跌至数年低点。重组将阿里巴巴拆为六个"业务集团"，每个集团都拥有自筹资本或最终上市的结构性选项。我们接下来要走的六个分部，就是 2023 年那场重塑的直接组织后果。
Each segment has its own competitive position, its own growth trajectory, its own margin profile, and its own valuation multiple. Treating Alibaba as a single business is the single biggest source of analytical error in coverage of this name. Walking through the six is the antidote. 每个分部都有自己的竞争地位、增长轨迹、利润率特征和估值倍数。把阿里巴巴当成单一业务来看，是覆盖这只股票最大的分析错误来源。逐个走完六个分部，就是解药。
TT is the original Alibaba: a two-sided marketplace where hundreds of millions of Chinese consumers transact with millions of merchants. Taobao is the C2C / SME-merchant marketplace; Tmall is the more curated, brand-store marketplace. Combined, TT processes more than ¥8 trillion of gross merchandise value (GMV) annually — about 35% of all online retail in China — and converts that GMV to revenue at a "take-rate" of 4–5% (advertising + customer-management fees + a newly introduced 0.6% software-service fee on confirmed GMV, rolled out in 2024). TT 是原始的阿里巴巴：一个双边市场，数亿中国消费者与数百万商家在这里交易。淘宝是 C2C / 中小商家市场；天猫是更"精选"的品牌店市场。两者合并每年处理超过 8 万亿元的商品交易总额 (GMV) — 约占中国所有在线零售的 35% — 并以 4-5% 的"货币化率"将 GMV 转化为收入 (广告 + 客户管理费 + 2024 年新推出的对确认 GMV 收取 0.6% 软件服务费)。
In FY25A, TT generated ¥446bn of revenue and ¥202bn of EBITA — a 45% segment margin, capital-light and structurally enviable. The investment thesis hinges on this segment's "re-monetization" — the take-rate stabilizing after years of compression, and the software-service fee adding a couple of points of incremental margin over time. The bear case argues that PDD and Douyin have permanently changed the competitive landscape and TT take-rates will continue to compress. The chart below tells the story so far: TT revenue growth turned positive in late FY26 after a multi-year flatline, and the third quarter (Dec 2025) print of +10.1% YoY was the first clear signal of the inflection. 在 FY25A，TT 产生了 4460 亿元收入与 2020 亿元 EBITA — 45% 的分部利润率，轻资本且结构上令人羡慕。投资论点的基础就是这个分部的"重新货币化" — 货币化率在多年压缩后稳定，软件服务费随时间增加几个百分点的边际利润率。看空者主张拼多多与抖音永久改变了竞争格局，TT 货币化率会继续压缩。下面的图表讲述了到目前的故事：TT 收入增长在 FY26 末由多年的"持平"转正，第三季度 (2025 年 12 月) 同比 +10.1% 是拐点的首个明确信号。
Stacked area · the cake is growing, but the slice composition is changing 堆叠面积 · 蛋糕在变大，但切片构成在变化
Alibaba International Digital Commerce is the consolidated umbrella for everything Alibaba does outside greater China: Lazada (Southeast Asia), AliExpress (cross-border, especially Europe, MENA, LatAm), Trendyol (Turkey), Daraz (South Asia). In FY25A it was the fastest-growing segment (+28-31% YoY for most of FY25) and still loss-making — generating a ¥13bn EBITA loss. The bull case has two pillars: (a) Lazada management has guided to FY27E EBIT breakeven and the trajectory is on track; (b) AliExpress is benefiting from cross-border tailwinds (Chinese exporters selling direct to international consumers) that may sustain double-digit growth for years. The bear case: a softening global consumer (especially in MENA) and intense local competition (Shopee in SEA, Trendyol's local rivals in Turkey) could push profitability out beyond FY27E. 阿里国际数字商业 (AIDC) 是阿里巴巴大中华区之外所有业务的合并伞：Lazada (东南亚)、速卖通 (跨境，主要是欧洲、中东北非、拉美)、Trendyol (土耳其)、Daraz (南亚)。在 FY25A 它是增长最快的分部 (FY25 大部分时间同比 +28-31%)，仍然亏损 — 产生 130 亿元 EBITA 亏损。看多观点有两根支柱：(a) Lazada 管理层指引 FY27E EBIT 盈亏平衡，轨迹符合预期；(b) 速卖通受益于跨境顺风 (中国出口商直销给国际消费者)，可能维持多年双位数增长。看空观点：全球消费者疲软 (尤其中东北非) 与激烈的本地竞争 (东南亚的 Shopee、土耳其的 Trendyol 本地对手) 可能把盈利时间表推到 FY27E 之外。
CIG is the third-largest public cloud infrastructure provider globally, with a dominant position in mainland China (~37% of public-cloud spend by IDC's measurement). After a 2023 reorganization that emphasized profitability over growth, CIG's revenue growth slowed from 30%+ to 0-5% in late FY23. The AI workload boom of 2024-25 changed everything: Q3 FY26 (Dec 2025) revenue was +21% YoY, with AI-related revenue (~mid-teens % of CIG) growing triple-digits. The Tongyi Qianwen foundation model and the new AI-native services (model training, inference, vector databases) are now the structural growth driver. We forecast CIG accelerating to 25%+ growth in FY27E with positive operating leverage — the segment with the most asymmetric upside. CIG 是全球第三大公有云基础设施提供商，在中国大陆占据主导地位 (按 IDC 测算约 37% 的公有云支出)。2023 年的一次重组强调盈利优先于增长，CIG 的收入增长率从 30%+ 跌到 FY23 末的 0-5%。2024-25 年的 AI 工作负载繁荣改变了一切：FY26 Q3 (2025 年 12 月) 收入同比 +21%，AI 相关收入 (约占 CIG 的十几个百分点) 三位数增长。通义千问基础模型和新的 AI 原生服务 (模型训练、推理、向量数据库) 现在是结构性增长驱动。我们预测 CIG 在 FY27E 加速至 25%+ 增长且具有正向经营杠杆 — 这是上行空间最不对称的分部。
Cainiao is China's largest smart-logistics network — not a traditional courier (it doesn't own trucks like SF Express), but a platform connecting carriers, fulfillment centers, and merchants with real-time data and routing. In FY25A revenue was ¥106bn with a small EBITA profit. Cainiao has had a long-rumored IPO timeline, refiled at various times, and remains a "real option" in our SOTP: a successful IPO (likely 2026 or 2027) would crystallize a value the consolidated model doesn't fully capture. We assign a modest EV/Revenue of 0.8× pending the IPO event. 菜鸟是中国最大的智能物流网络 — 不是传统的快递商 (它不像顺丰那样拥有卡车)，而是一个用实时数据和路由连接承运商、履约中心、商家的平台。在 FY25A 收入 1060 亿元，EBITA 小幅盈利。菜鸟有一个被传闻已久的 IPO 时间表，多次重新申报，在我们的 SOTP 中仍是一个"实物期权"：成功上市 (可能在 2026 或 2027 年) 会实现合并模型未完全捕捉的价值。在 IPO 事件兑现前，我们给予一个温和的 0.8 倍 EV/Revenue 估值。
LSG houses Ele.me (food delivery) and Amap (mapping). It is fundamentally a competition story versus Meituan, the dominant local-services player in China. After years of losses, the FY25 print showed a meaningfully narrowing loss, and we forecast LSG to turn EBITA-positive by FY27E. The market values Meituan at roughly 1.5× revenue; we use 1.0× for LSG given its smaller scale and weaker network density. Even at 1.0×, the segment is worth approximately US$15bn — modest, but real, and growing. LSG 包含饿了么 (外卖) 与高德 (地图)。它本质上是与美团 — 中国本地生活的主导玩家 — 竞争的故事。多年亏损后，FY25 数据显示亏损实质性收窄，我们预测 LSG 在 FY27E 实现 EBITA 转正。市场给美团约 1.5 倍收入估值；考虑到 LSG 规模较小、网络密度较弱，我们给 1.0 倍。即使按 1.0 倍，该分部价值约 150 亿美元 — 温和但真实，且在增长。
DME is the smallest segment — primarily Youku (streaming video) plus an entertainment business. It runs roughly breakeven and contributes only 2% of revenue. The market essentially discounts the segment entirely, awarding a 0.6× EV/Revenue multiple. We don't disagree; the bull case here is mostly about reducing the drag on consolidated margins as management trims non-core operations. DME 是最小的分部 — 主要是优酷 (流媒体视频) 加一项娱乐业务。它大致盈亏平衡，仅贡献 2% 的收入。市场基本完全打折该分部，给予 0.6 倍 EV/Revenue 倍数。我们不反对；这里的看多观点主要是随着管理层削减非核心业务，减少对合并利润率的拖累。
Looking at the segment-level EBITA waterfall (below) makes the story unmistakable: TT generates ~¥200bn of EBITA on its own; everything else is, on a consolidated basis, a small positive or a small negative. The implication is that, today, Alibaba is essentially Taobao & Tmall plus a portfolio of options. The bull case is that the options pay off (Cloud, AIDC, Cainiao). The bear case is that the options' losses persist, and TT is increasingly stranded as a single great business inside a complicated holding structure. Either way, looking only at the consolidated number obscures what is really happening. 看分部级的 EBITA 瀑布图 (下方) 让故事变得清晰：TT 自己产生约 2000 亿元 EBITA；其他一切在合并基础上是小幅正或小幅负。含义是：今天，阿里巴巴本质上等于 TT 加一组期权。看多观点是这些期权会兑现 (云、AIDC、菜鸟)。看空观点是这些期权的亏损持续，TT 越来越孤立成一个复杂控股结构内的"单一伟大业务"。无论哪种，只看合并数字都会掩盖真实情况。
Annual numbers are the year's average. Quarterly numbers are the heartbeat. Reading eight quarters in sequence is how an analyst spots inflections — the moment when something genuinely changes, before the annual revision catches up. 年度数字是一年的平均。季度数字是心跳。按顺序读 8 个季度，是分析师发现拐点的方法 — 在年度修订赶上之前，某件事真正发生改变的那一刻。
The most important read from the trailing 8 quarters is the segment growth divergence. TT growth was negative or near zero through most of FY25, then suddenly accelerated to +10.1% in Q3 FY26 (Dec 2025). Cloud growth, meanwhile, accelerated from +3% to +21% over the same period. AIDC growth decelerated from +29% to +12% as base effects compressed. Three different stories, hidden inside a single consolidated +5.4% line. The chart below lets you flip through each segment to see the cadence. 从过去 8 季度最重要的读取是分部增长分化。TT 增长在 FY25 大部分时间为负或接近零，然后突然在 FY26 Q3 (2025 年 12 月) 加速到 +10.1%。云增长则在同一时期从 +3% 加速到 +21%。AIDC 增长因基数效应压缩从 +29% 减速到 +12%。三个不同的故事，藏在合并 +5.4% 的单一数字里。下面的图表允许您切换每个分部以观察节奏。
Alibaba's stock price doesn't only respond to the company — it responds to China. Two years of weak consumer demand, deflationary pressure, and a depreciating yuan have shaped the multiple investors are willing to pay. Understanding the macro overlay is part of fundamental research, not separate from it. 阿里巴巴的股价不仅响应公司本身 — 它响应中国。两年的疲弱消费需求、通缩压力、贬值的人民币塑造了投资者愿意支付的估值倍数。理解宏观叠加是基本面研究的一部分，而非独立的部分。
Five macro indicators matter for any China-domestic consumer name: (1) real GDP growth — the headline; (2) CPI inflation — whether nominal pricing power exists; (3) M2 money supply growth — liquidity tail-wind or head-wind; (4) PMI — leading indicator of cyclical demand; (5) USD/CNY — translation effect for ADR investors. 对任何中国境内消费类股票，五个宏观指标重要：(1) 实际 GDP 增长 — 头条；(2) CPI 通胀 — 是否存在名义定价能力；(3) M2 货币供应增长 — 流动性顺风或逆风；(4) PMI — 周期性需求的领先指标；(5) 美元/人民币 — ADR 投资者的换算效应。
There are three transmission channels. First, demand: a weak consumer means slower TT GMV growth and slower take-rate normalization, both pillars of the bull case. Second, the multiple: a slowing China is correlated with foreign investor de-risking from China internet — this is what drove BABA from 30× P/E in 2020 to below 10× today. Third, the discount rate: China country risk premium (the 1.5% adder we use in the WACC) reflects exactly this — it rises when geopolitical or domestic risks heighten, mechanically lowering DCF-derived value. A 50bp CRP increase shaves ~10% off our DCF price target. 有三条传导通道。第一，需求：疲弱的消费者意味着 TT GMV 增长放缓、货币化率正常化放缓，二者都是看多论点的支柱。第二，估值倍数：放缓的中国与外资从中国互联网"去风险化"相关 — 这就是 BABA 从 2020 年 30× P/E 跌到今天 10× 以下的原因。第三，贴现率：中国国家风险溢价 (我们在 WACC 里用 1.5% 加项反映了这一点) — 在地缘政治或国内风险升高时上升，机械性压低 DCF 推导价值。CRP 上升 50 个基点，会让我们的 DCF 目标价缩水约 10%。
A picture of the share price tells you what investors believed at each moment. Alibaba's chart is one of the most dramatic in modern Asian capital markets — a peak in October 2020 followed by a 75% drawdown, then a partial recovery in early 2025. Knowing the catalysts that drove each leg is part of understanding the stock. 一张股价图告诉您每个时刻投资者相信什么。阿里巴巴的图表是现代亚洲资本市场最具戏剧性的之一 — 2020 年 10 月顶峰，随后 75% 回撤，2025 年初部分复苏。理解每一段是由什么催化剂驱动，是理解这只股票的一部分。
Every investment thesis is a probabilistic statement, and the bull case is only as good as the analyst's honesty about the bear case. The risk section in a sell-side report is not boilerplate — it is the analyst's stress test of her own conviction. 每个投资论点都是概率陈述，看多观点的好坏，仅取决于分析师对看空观点的诚实度。卖方报告里的风险章节不是样板文 — 它是分析师对自身信念的压力测试。
A standard risk matrix scores each risk on two axes: likelihood (how probable is the bad outcome over the forecast horizon?) and impact (how badly would it hurt the price target?). The product is a "tier" — high, medium, low. The discipline of being explicit about both axes forces the analyst to think clearly. Below is the matrix for Alibaba. 标准风险矩阵在两个轴上为每项风险评分：可能性 (预测期内不良结果发生的概率多大？) 与影响 (它对目标价的伤害有多大？)。乘积即为"等级" — 高、中、低。在两个轴上都明确表态的纪律，迫使分析师清晰思考。下面是阿里巴巴的矩阵。
We've walked through the toolkit, the case, the segments, the macro, the risks. Now we close the loop with the actual investment thesis — the five reasons we believe Alibaba is worth significantly more than where it trades today, and the catalyst calendar that should reveal whether we are right. 我们走过了工具箱、案例、分部、宏观、风险。现在用真正的投资论点闭环 — 我们相信阿里巴巴价值显著高于今天交易价的五个理由，以及应当揭示我们是否正确的催化剂日历。
Take-rate stabilization plus the 2024 software-service fee rollout creates a structural revenue lift, returning TT EBITA growth to high-single-digits and providing the cash anchor for the rest of the portfolio. 货币化率稳定加上 2024 年软件服务费推出，创造结构性收入提升，TT EBITA 增长回归高个位数，为投资组合其余部分提供现金锚。
AIDC reaches EBIT breakeven by FY27E as Lazada scales and AliExpress unit economics improve, removing a key drag on consolidated margins. AIDC 在 FY27E 实现 EBIT 盈亏平衡，因 Lazada 规模化和速卖通单位经济性改善，消除合并利润率上的关键拖累。
CIG growth re-accelerating to 25%+ on AI workloads. Tongyi Qianwen ecosystem and inference demand drive structural mix-shift supporting multiple expansion. CIG 增长因 AI 工作负载重新加速至 25%+。通义千问生态与推理需求驱动结构性产品组合转移，支持估值倍数扩张。
US$26bn residual buyback authorization plus a second consecutive annual dividend supports an 8-9% shareholder yield, materially supportive of EPS growth and the equity story. 剩余 260 亿美元回购授权加上连续第二年的年度股息，支持 8-9% 的股东收益率，显著支撑 EPS 增长和股权故事。
Net cash > US$80bn plus US$50-60bn of unrealized equity investments (Ant ~33% + others) anchor a constructive risk/reward and provide downside support. 净现金 > 800 亿美元 + 500-600 亿美元未实现股权投资 (蚂蚁 33% + 其他) 锚定建设性的风险/回报，并提供下行支撑。
Bull $215 (25%) × Base $172 (55%) × Bear $95 (20%) = $167 weighted PT, with DCF-derived $226 implying 91% upside on the central forecast. 牛市 $215 (25%) × 基准 $172 (55%) × 熊市 $95 (20%) = $167 加权目标价，DCF 推导的 $226 在中心预测下隐含 91% 上行空间。
A scenario matrix forces the analyst to attach probabilities to outcomes — the discipline that prevents thesis drift. Below is our probability-weighted table. 情景矩阵迫使分析师为结果附加概率 — 防止论点漂移的纪律。下面是我们的概率加权表。
| Scenario情景 | Probability概率 | FY27E Rev (CNY bn)FY27E 收入 (十亿) | EBIT %EBIT 率 | EPS (US$)EPS (美元) | PT (US$)目标价 (美元) | Upside上行 |
|---|---|---|---|---|---|---|
| Bull | 25% | 1,300 | 18.0% | $2.10 | $215 | +81% |
| Base | 55% | 1,180 | 16.5% | $1.80 | $172 | +45% |
| Bear | 20% | 1,050 | 14.0% | $1.40 | $95 | -20% |
| Probability-weighted PT概率加权目标价 | $167 | +41% | ||||
A thesis without catalysts is a hope. The catalyst calendar lays out the events over the next twelve to eighteen months that will, mathematically, either confirm or refute the thesis. Each is a moment where the market will reprice based on a specific data point. 没有催化剂的论点是希望。催化剂日历列出未来 12-18 个月内将在数学意义上确认或否定论点的事件。每一项都是市场基于特定数据点重新定价的时刻。
TT take-rate print, AIDC margin progression, and the first FY27 capex guidance — a quintessential triple-input event. TT 货币化率数据、AIDC 利润率进展，以及首次 FY27 资本开支指引 — 一个典型的三输入事件。
Annual product showcase for Cloud / Tongyi Qianwen. Look for major partnership announcements, model-release cadence, customer deployments — directly tied to the cloud re-rating thesis. 云/通义千问年度产品展示。关注重大合作伙伴公告、模型发布节奏、客户部署 — 直接关联云重估论点。
If Cainiao refiles for HK listing, SOTP unlocks ~US$5-8bn of trapped value visible on the consolidated balance sheet. 如果菜鸟重新申报香港上市，SOTP 在合并资产负债表上释放约 50-80 亿美元的被困价值。
The single biggest read on TT GMV trajectory. A "good" 11.11 (mid-to-high single-digit YoY) confirms the FY27 forecast; a weak print would invalidate the TT pillar. 对 TT GMV 轨迹的单一最大读取。"良好"的 11.11 (同比中到高个位数) 确认 FY27 预测；疲软的数据将否定 TT 支柱。
Strategy refresh under the Wu / Tsai leadership. Watch for changes in capital-return commitments, capex framework, segment-disclosure granularity. 吴/蔡领导下的战略更新。关注资本回报承诺、资本开支框架、分部披露粒度的变化。
The mother of all updates — a year of catalysts now scored. The price target gets re-tested in earnest. 所有更新之母 — 一年催化剂在此被打分。目标价此时被认真重新测试。
A company's long-run profitability is not determined primarily by management quality — it is determined by the structure of the industry it competes in. Michael Porter's Five Forces framework, published in 1979, remains the most useful one-page model for analyzing that structure. Applied to Alibaba's six segments, it shows why some look so promising and others look so hard. 一家公司的长期盈利能力主要不取决于管理层质量 — 而取决于其所在行业的结构。迈克尔·波特 1979 年发表的"五力模型"仍是分析该结构最有用的"一页"模型。应用到阿里巴巴的六个分部，它展示了为什么有些看起来那么有希望、有些那么艰难。
Chinese e-commerce is a strong example of a "good business in a tough neighborhood." Network effects (more shoppers attract more merchants, which attract more shoppers) create high barriers to new entry. Switching costs for merchants (storefront, reviews, fulfillment integrations) are real. And the suppliers (the millions of independent merchants) are individually weak. But: rivalry is intense. PDD has demonstrated that a low-price-led alternative can take meaningful share. Douyin's livestream commerce is a substitute for traditional search-led shopping. Customer bargaining power is high — Chinese consumers will jump platforms for a 10% discount. The net effect is that TT operates with very high margins (45% EBITA) and a very stable market position, but its growth rate has been pulled down to single digits by rivalry and substitutes. The Porter analysis lets you see, structurally, why this is so. 中国电商是"在艰难街区里的好生意"的强例。网络效应 (更多购物者吸引更多商家，再吸引更多购物者) 创造了高新进入壁垒。商家的转换成本 (店铺、评价、履约集成) 是真实的。供应商 (数百万独立商家) 单个看是弱的。但是：对抗激烈。拼多多证明了一个低价驱动的替代方案能拿走可观份额。抖音的直播电商是传统搜索式购物的替代品。客户议价能力高 — 中国消费者会为 10% 折扣换平台。净效应是 TT 以非常高的利润率 (45% EBITA) 和非常稳定的市场地位运营，但其增长率被对抗与替代品拉到个位数。波特分析让您结构性地看到为什么如此。
Public-cloud infrastructure is one of the most attractive industries to participate in: extremely high barriers (capex, talent, scale economics), low supplier power (chip vendors are concentrated but cloud players negotiate hard), low customer bargaining power for individual SMEs (large enterprise customers, by contrast, have real leverage), and few real substitutes (private-cloud / on-prem is harder, slower, and more expensive at scale). The bad news: rivalry is brutal among the giants (AWS, Azure, Google in the US; Alibaba, Tencent, Huawei in China), and that rivalry has historically suppressed margins. But the industry is consolidating around 3-4 winners in each region, and consolidation tends to relax rivalry. The Porter analysis suggests CIG margins should structurally improve over the next decade — exactly what our model assumes. 公有云基础设施是参与最有吸引力的行业之一：极高壁垒 (资本开支、人才、规模经济)、低供应商权力 (芯片供应商集中，但云玩家硬议价)、对个别中小企业的低客户议价能力 (反之，大型企业客户有真正杠杆)、少有真正替代品 (私有云/本地部署在规模上更难、更慢、更贵)。坏消息：巨头之间对抗残酷 (美国 AWS、Azure、谷歌；中国阿里巴巴、腾讯、华为)，对抗在历史上压制了利润率。但行业正在每个地区围绕 3-4 个赢家整合，整合往往松弛对抗。波特分析提示 CIG 利润率在未来十年结构性改善 — 正是我们模型假设的。
AIDC is structurally the hardest of Alibaba's six segments. Southeast Asia, MENA, LatAm, and Turkey each have aggressive local competitors (Shopee in SEA, Mercado Libre in LatAm, Trendyol's local rivals in Turkey) plus the threat of Amazon expansion. Network effects are partially localized — meaning Alibaba can't simply transplant its China dominance. Customer switching costs are low. Supplier power varies by market. Net: lower margins than TT for the foreseeable future, with high competitive intensity. The Porter analysis explains why AIDC trades at a 1× EV/Revenue multiple in our SOTP, vs. TT at 7× EV/EBITDA. The market is correct to assign a lower multiple structurally; the bull case has to be that scale and operating leverage eventually win, but it will take years. AIDC 在结构上是阿里巴巴六个分部中最难的。东南亚、中东北非、拉美、土耳其，各有激进的本地竞争者 (东南亚的 Shopee、拉美的 Mercado Libre、土耳其 Trendyol 的本地对手) 加上亚马逊扩张的威胁。网络效应部分本地化 — 意味着阿里巴巴不能简单移植其在中国的主导地位。客户转换成本低。供应商权力因市场而异。净效应：可预见将来的利润率低于 TT，竞争强度高。波特分析解释了为什么 AIDC 在我们的 SOTP 中以 1× EV/Revenue 倍数交易，而 TT 是 7× EV/EBITDA。市场结构性地给予较低倍数是正确的；看多论点必须是规模和经营杠杆最终胜出，但需要数年。
A first-rate analyst spends 20% of her time on the company and 80% on the industry. A first-rate management team in a structurally bad industry (think traditional newspapers) will generate weak long-run returns. A mediocre management team in a structurally great industry (think AWS in 2010) will generate spectacular returns regardless. Always ask the structural question before judging the manager. 一流的分析师 20% 的时间在公司上、80% 在行业上。在结构性差行业的一流管理团队 (想想传统报纸) 会产生疲弱的长期回报。在结构性好行业的平庸管理团队 (想想 2010 年的 AWS) 不论如何会产生壮观回报。在评判经理之前，永远先问结构问题。
ESG used to be a polite afterthought in equity research. It is now a primary input for many institutional investors — both because of investor mandates and because environmental, social, and governance risks have proven to be real economic risks. For Alibaba, the "G" is by far the most material; the "E" is improving; the "S" is complicated. ESG 过去在股权研究中是一个礼貌性的事后想法。它现在是许多机构投资者的主要输入 — 既因为投资授权，也因为环境、社会、治理风险已被证明是真实的经济风险。对阿里巴巴，"G" 是迄今最重要的；"E" 在改善；"S" 复杂。
Alibaba reports Scope 1 + 2 emissions of approximately 6.6 million tonnes of CO₂ equivalent (FY24), and has committed to carbon neutrality by 2030 across its own operations and to net-zero across the value chain by 2050. The bulk of emissions are concentrated in CIG data centers — the same data centers that are rapidly expanding for AI workloads. Reconciling rapid capex growth with the net-zero commitment requires aggressive renewables sourcing. The company has signed multiple long-term PPAs (power-purchase agreements) for solar and wind generation in China and is investing in liquid-cooled data center designs that improve power-usage-effectiveness (PUE) by 20-30%. This is a structurally good story for institutional investors with ESG mandates. 阿里巴巴报告范围 1+2 排放约 660 万吨二氧化碳当量 (FY24)，并承诺到 2030 年实现自身运营的碳中和、到 2050 年实现价值链的净零。排放的主体集中在 CIG 数据中心 — 也正是为 AI 工作负载快速扩张的同一批数据中心。将快速资本开支增长与净零承诺协调起来，需要激进的可再生能源采购。公司已签订多份长期 PPA (购电协议) 用于中国的太阳能和风能发电，并投资于把电源使用效率 (PUE) 改善 20-30% 的液冷数据中心设计。这对有 ESG 授权的机构投资者是结构性好故事。
Alibaba's workforce of ~205,000 has shrunk by roughly 7% over FY23-25 as the company restructured and offloaded non-core operations. The social dimension extends to the 10 million+ merchants who depend on Alibaba's platforms for their livelihood — a stakeholder group whose interests sometimes conflict with shareholder interests (e.g., higher take-rates lift Alibaba revenue but pressure merchant margins). The 2021 antitrust crackdown was, in part, a regulatory expression of merchant pushback against Alibaba's exclusivity policies. The company has since made structural concessions and is monitored closely by SAMR (the State Administration for Market Regulation). 阿里巴巴约 20.5 万员工，在 FY23-25 期间随着公司重组并剥离非核心业务，已收缩约 7%。社会维度延伸到 1000 多万依赖阿里巴巴平台为生的商家 — 一个其利益有时与股东利益冲突的利益相关方群体 (例如，更高的货币化率提升阿里巴巴收入但压低商家利润率)。2021 年反垄断收紧，部分是对阿里巴巴排他性政策的商家反弹的监管表达。公司自那以来做出了结构性让步，并受到国家市场监督管理总局 (SAMR) 密切监管。
Alibaba's corporate governance has three distinctive features that an investor must understand: 阿里巴巴的公司治理有三个鲜明特征，投资者必须理解：
Net assessment: Alibaba's governance is more complicated than a typical S&P 500 name, but not worse — it is differently structured. The market's persistent "China internet discount" partially reflects governance complexity; the question is whether the discount is too large or about right. Our view: about right, leaving the upside dependent on the operating thesis rather than on governance rerating. 净评估：阿里巴巴的治理比典型的标普 500 公司更复杂，但不更差 — 它结构不同。市场持续的"中国互联网折价"部分反映了治理复杂性；问题是折价过大还是大致合适。我们的观点：大致合适，让上行空间取决于经营论点而非治理重估。
You now have the entire grammar of equity research — statements, ratios, forecasts, three valuation methods, scenario, risk, and catalyst frameworks. To go from here to actually generating original calls, the work is hands-on: pick a name in your circle of competence, build a model, defend a price target. Repeat for ten companies. By the end, the reflexes are yours. 您现在掌握了股权研究的整套语法 — 报表、比率、预测、三种估值方法、情景、风险、催化剂框架。要从这里走到真正生成原创判断，工作是动手的：在您能力圈里选一只股票，搭模型，为目标价辩护。重复 10 家公司。到那时，反射就是您的了。
Equity research is a craft, not a science. Models give you precision; judgment gives you accuracy. The craft is in combining the two without confusing one for the other. 股权研究是手艺，不是科学。模型给您精度；判断给您准确性。手艺就在于把两者结合，而不要把其中一个当成另一个。
A naïve view of equity markets imagines a calm, rational price that perfectly reflects the present value of future cash flows. The reality is messier — and more interesting. Understanding the gap between "theoretical value" and "market price" is where most of the actual money in research is made. 一种天真的股票市场观，想象一个完美反映未来现金流现值的、平静理性的价格。现实更乱 — 也更有趣。理解"理论价值"与"市场价格"之间的差距，正是研究中实际赚钱的地方。
A stock price at any moment reflects three things in superposition: (1) fundamental cash-flow expectations (what we built the DCF around); (2) flows (passive index buying, retail momentum, hedge-fund crowding, foreign capital flight); and (3) narrative (a story that motivates the marginal buyer or seller). In efficient-market theory, only the first matters; in the real world, all three. The relative weight depends on the stock. 任何时刻的股票价格都反映三件事的叠加：(1) 基本面现金流预期 (我们围绕其构建 DCF)；(2) 资金流 (被动指数购买、散户动量、对冲基金扎堆、外资外流)；(3) 叙事 (激励边际买家或卖家的故事)。在有效市场理论中，只有第一项重要；在真实世界中，三者都重要。相对权重取决于股票。
For Alibaba, the cash-flow story (DCF, SOTP) supports a value materially above the spot price. So why doesn't the spot rise? Three reasons. First, narrative: "China is uninvestable" was the dominant narrative for foreign capital in 2022-24, and narratives are sticky. Second, flows: global passive funds reduced China weight over the same period; the steady selling pressure suppressed the price independent of fundamentals. Third, risk premium: investors demanded a higher return for holding the stock (the country risk premium), which mechanically lowers the price for the same cash flows. The bull thesis is implicitly a thesis on all three returning to normalcy: the narrative shifting, flows turning, and the risk premium falling. None of these are forecastable with precision — but recognizing that all three forces are at play makes the price gap less mysterious. 对阿里巴巴而言，现金流故事 (DCF、SOTP) 支撑了一个远高于现价的价值。那为什么现价不上涨？三个原因。第一，叙事："中国不可投资"在 2022-24 是外资的主导叙事，叙事是有粘性的。第二，资金流：全球被动基金在同一时期降低了中国权重；持续的卖压独立于基本面压制价格。第三，风险溢价：投资者要求更高回报来持有该股票 (国家风险溢价)，在相同现金流下机械性压低价格。看多论点隐含的是这三者回归常态：叙事转变、资金流逆转、风险溢价下降。这些都不能精确预测 — 但认识到三股力量都在起作用，使得价格差距不再神秘。
John Maynard Keynes is credited with this maxim, and it sits at the center of every value-investor's bookcase. It captures the asymmetry between "being right about value" and "being right about price." A stock you believe is worth $226 can stay at $90 for years if the narrative and flows refuse to cooperate. Hedge funds that levered long positions against this thesis blew up during the 2022 China rout — the value was there, but the holding period was longer than the borrowed capital allowed. 这句格言归于约翰·梅纳德·凯恩斯，坐落在每位价值投资者的书架中央。它捕捉了"对价值正确"与"对价格正确"之间的不对称。您相信值 226 美元的一只股票，如果叙事和资金流拒绝配合，可能多年停留在 90 美元。在 2022 年中国大跌中，对这一论点加杠杆做多的对冲基金爆仓 — 价值是真实的，但持有期长于借入资本允许的期限。
This is why the catalyst calendar (chapter 14) matters so much. The catalyst is the "trigger" that converts value into price. Without identifiable catalysts, a value-investing thesis is "buy and pray." With clear catalysts, the thesis becomes "buy, and within 12-18 months, the following events will force the market to revalue." 这就是为什么催化剂日历 (第十四章) 如此重要。催化剂是把价值转化为价格的"扳机"。没有可识别的催化剂，价值投资论点就是"买入并祈祷"。有清晰的催化剂，论点就变成"买入，并在 12-18 个月内，以下事件将迫使市场重估"。
A powerful diagnostic question: at the current price, what is the market implicitly assuming about future growth? You can back this out from the DCF formula. If the spot is $118.50 and net cash plus equity investments adds up to ~$30 per ADR, the market is paying $88.50 for the operating cash flows. At a 10× P/E, that implies $8.85 of "normalized" EPS — but the company will earn over $10 in FY26E and over $12 in FY27E. So the market is implicitly assuming either (a) earnings will decline from here, (b) the multiple should be lower than 10× (high required return), or (c) some combination. Either of these is hard to justify on the visible fundamentals. 一个强大的诊断问题：在当前价格下，市场隐含对未来增长的假设是什么？您可以从 DCF 公式反向推导。如果现价 118.50 美元，净现金加股权投资合计每股 ADR 约 30 美元，市场为经营现金流支付 88.50 美元。按 10× P/E，隐含 8.85 美元的"正常化"EPS — 但公司在 FY26E 将赚超 10 美元，FY27E 超 12 美元。所以市场隐含假设要么 (a) 盈利将从这里下滑，要么 (b) 倍数应低于 10× (高要求回报)，要么 (c) 两者组合。在可见基本面上，二者都难以证立。
When the implied scenario is harsh and the company doesn't deliver according to that harsh scenario, the gap closes. This is, mechanically, how undervalued stocks rerate: not via management doing something spectacular, but via management not screwing up while the market's pessimistic assumptions are revealed as wrong. 当隐含情景严苛而公司未按那个严苛情景兑现时，差距闭合。机械上，这就是被低估股票重估的方式：不是通过管理层做出某件惊艳之事，而是通过管理层不搞砸，同时市场的悲观假设被揭示为错误。
The phrase "variant view" is sell-side shorthand for: "here is something I believe that the rest of the market does not believe." A buy-side reader does not pay for the consensus; they pay for the differentiated insight. Three patterns produce variant view: "变体观点"是卖方的简称，意思是："这里是我相信而市场其他人不相信的某件事。"买方读者不为共识付费；他们为差异化洞察付费。三种模式产生变体观点：
An analyst with one strong variant view earns her seat for the year. The price target is a wrapper; the variant view is the product. 拥有一个强有力变体观点的分析师，整年都有座位。目标价是包装；变体观点是产品。
Every public company holds a one-hour conference call after each quarterly earnings release. The structure is identical across companies; the information density is enormous; and an experienced analyst can extract three or four model-moving inputs in a single hour. Knowing how to listen is half the skill. 每家上市公司在每个季度业绩发布后召开一场一小时的电话会议。各公司的结构相同；信息密度巨大；一位有经验的分析师能在一小时内提取三到四个能撬动模型的输入。知道如何"听"是技能的一半。
Almost every earnings call follows the same arc: 几乎每场电话会都遵循同样的脉络：
Executives rarely say "we're going to miss the quarter" or "we're worried about competition." They speak in a code that the practiced ear learns to translate. A few common decodes: 高管很少说"我们这季度会未达预期"或"我们担心竞争"。他们用一种密码说话，老练的耳朵能学会翻译。一些常见解码：
After every Alibaba quarterly call, an analyst should walk away with a half-dozen specific data points to feed into the model: 每场阿里巴巴季度电话会后，分析师应该带走六个左右的具体数据点，喂给模型：
Listen to the call once live, then read the transcript twice. The transcript reveals the precise wording the management used (often softened from the spoken version). Then compare your notes against the printed press release; gaps between what the press release emphasizes and what the call emphasized often reveal where the company would like the focus to be (versus where it actually was). The skill of reading a transcript is much like reading a court ruling: the words chosen, in the order chosen, are doing specific legal and signaling work. 实时听一遍电话会，然后读两遍文字记录。文字记录揭示管理层使用的精确措辞 (常常比口语版本柔和)。然后把您的笔记与新闻稿对比；新闻稿强调什么与电话会强调什么之间的差距，往往揭示公司希望焦点在哪里 (相对实际在哪里)。读文字记录的技能很像读法庭判决：所选的词、所选的顺序，都在做特定的法律与信号工作。
The introductory DCF chapter gave you the mechanics. Here we go deeper on the assumptions that actually drive the answer, the structural weaknesses of the model, and the techniques practitioners use to keep themselves honest. 入门 DCF 章节给了您机制。这里我们深入真正驱动答案的假设、模型的结构性弱点，以及从业者用来保持诚实的技术。
In our Alibaba DCF, the terminal value contributes ¥2,118bn out of ¥2,873bn enterprise value — 74%. This is not unusual; in most DCFs, the terminal value is the dominant chunk. Which means that the modest-sounding "g = 2.5%" assumption is doing the heaviest analytical lifting in the entire model. Two important sanity checks practitioners apply: 在我们的阿里巴巴 DCF 中，终值在 28730 亿元企业价值中贡献 21180 亿元 — 74%。这并不罕见；在多数 DCF 中，终值是主导部分。这意味着听起来温和的"g = 2.5%"假设，在整个模型中承担了最重的分析重量。从业者应用两个重要的健全性检查：
Sanity check 1: the implied exit multiple. If you take your terminal-year FCF, divide it by (WACC − g), you get the terminal value. Divide that by the terminal-year EBITDA, and you get the "implied exit multiple" — what the market is paying for one dollar of perpetual EBITDA in your final year. For Alibaba: terminal FCF $283bn × 1.025 / (0.1106 − 0.025) = $3,396bn TV / FY30E EBITDA of $458bn ≈ 7.4× EV/EBITDA. This is the multiple your model implicitly assumes the market will award Alibaba in year 5. Compare this to today's 4.5× and the peer median of 9.5×: our implied exit multiple is well within the range, suggesting the terminal value is not aggressive. If your implied exit multiple lands at 15× — well above peers — your model is implicitly assuming a multiple expansion that is not justified by the cash flows themselves. 健全性检查 1：隐含退出倍数。用您的终值年 FCF 除以 (WACC − g)，得到终值。除以终值年 EBITDA，得到"隐含退出倍数" — 市场为您最终年的一元永续 EBITDA 支付多少。阿里巴巴：终值 FCF $283bn × 1.025 / (0.1106 − 0.025) = $3,396bn 终值 / FY30E EBITDA $458bn ≈ 7.4× EV/EBITDA。这是您的模型隐含假设市场将在第 5 年给阿里巴巴的倍数。把这个对照今天的 4.5× 和同行中位数 9.5×：我们的隐含退出倍数落在区间内，提示终值不激进。如果您的隐含退出倍数落在 15× — 远高于同行 — 您的模型隐含假设了不被现金流本身证立的倍数扩张。
Sanity check 2: the GDP-share test. If your forecast says a company grows 12% per year forever, in 30 years its revenue will be larger than its country's GDP. Obviously impossible. The conservative practitioner ensures terminal growth is below long-run nominal GDP — typically 2-3% in mature economies, slightly higher in emerging markets. Our 2.5% for Alibaba is well below long-run China nominal GDP (~5%), leaving the company room to keep growing but without claiming an impossible share of the economy. 健全性检查 2：GDP 份额测试。如果您的预测说一家公司永远以每年 12% 增长，30 年后它的收入将大于其国家的 GDP。显然不可能。保守的从业者确保终值增长低于长期名义 GDP — 通常在成熟经济体中 2-3%，在新兴市场略高。我们对阿里巴巴的 2.5% 远低于中国长期名义 GDP (~5%)，给公司留出继续增长的空间，但不主张占据经济中不可能的份额。
There are two conventions for assigning a discount factor to each year of cash flow. "End-year" assumes the entire year's cash arrives on December 31 — convenient and pessimistic. "Mid-year" assumes the cash arrives evenly over the year, which mathematically means using a discount period of (year − 0.5). Mid-year discounting raises the valuation by about 5%, because you're not waiting as long to receive the cash. Our model uses mid-year. Be aware that some shops use end-year; the choice will modestly affect the answer. 为每年现金流分配贴现因子有两种惯例。"年末"假设全年现金在 12 月 31 日到达 — 方便且悲观。"年中"假设现金在年内均匀到达，数学上意味着使用 (年份 − 0.5) 的贴现期。年中贴现将估值提高约 5%，因为您等待现金的时间没那么长。我们的模型用年中。要意识到有些机构用年末；选择会温和影响答案。
A single-stage DCF (5 years explicit + perpetual terminal) works for mature businesses. For a high-growth tech company, the "shape" of cash flows is wrong: the company will grow 30% for the first five years, 15% for the next five, then maybe 8% for another five, then settle into 3% terminal. Forcing this onto a single-stage model overstates the terminal value materially. The fix is a two-stage or three-stage DCF, with explicit transitions between growth phases. 单阶段 DCF (5 年明确+永续终值) 对成熟业务有效。对高增长科技公司，现金流的"形状"是错的：公司将在头 5 年增长 30%，下 5 年 15%，再之后 5 年也许 8%，然后落定为 3% 终值。把这个塞进单阶段模型会显著高估终值。修正是双阶段或三阶段 DCF，在增长阶段之间显式过渡。
For Alibaba, we use single-stage because the company is large and mature enough that a 5-10% growth path settles quickly into a terminal regime. For a Tongyi Qianwen-only carve-out — if that ever existed as a standalone company — you would want a three-stage model with hyper-growth for years 1-5, fade for years 6-12, and a modest terminal. 对阿里巴巴，我们用单阶段，因为公司足够大且成熟，5-10% 的增长路径会很快沉淀到终值制度。对一个仅通义千问的剥离 — 如果它曾以独立公司存在 — 您会想要三阶段模型：第 1-5 年超高增长、第 6-12 年衰减、温和终值。
CAPM ("kapem") computes the cost of equity as risk-free rate + beta × equity risk premium. It is the most-used formula in corporate finance and also the most-criticized. The criticisms are valid: beta is unstable, the equity risk premium is unobservable, and the formula assumes a perfect capital market that doesn't exist. Yet CAPM survives because the alternatives (Fama-French three-factor, arbitrage-pricing theory, build-up methods) are similarly arbitrary at the inputs. In practice, we treat CAPM as an organizing framework — a defensible way to write down a cost of equity number that we then triangulate with comparables and reality checks (e.g., what dividend yield + growth rate does the market embed in mature businesses with similar risk?). CAPM ("卡-pem") 把权益成本计算为：无风险利率 + 贝塔 × 股权风险溢价。它是企业财务中使用最多的公式，也是最被批评的。批评是合理的：贝塔不稳定、股权风险溢价不可观察、公式假设一个不存在的完美资本市场。但 CAPM 存活下来，因为替代方案 (Fama-French 三因子、套利定价理论、构建法) 在输入上同样武断。实践中，我们把 CAPM 视为一个组织框架 — 一种可辩护的写下权益成本数字的方式，然后我们用可比公司和现实检查 (例如，市场在类似风险的成熟业务中嵌入了什么股息率+增长率？) 来三角验证。
Two practical refinements: (1) use a regression-based beta over 5 years of weekly returns — not 1-year daily, which is too noisy for forecasting purposes; (2) for emerging-market companies, add a "country risk premium" — our 1.5% for China is conservative; some practitioners argue for 2-3% during periods of political stress. The CRP is the single most impactful WACC lever for Alibaba. 两个实用细化：(1) 用基于回归的 5 年周频回报贝塔 — 不是 1 年日频，对预测而言太嘈杂；(2) 对新兴市场公司，加上"国家风险溢价" — 我们对中国的 1.5% 是保守的；一些从业者主张在政治压力期 2-3%。CRP 是阿里巴巴 WACC 中最具影响的杠杆。
A vanilla DCF treats the future as a single forecast trajectory. In reality, management has options — to expand if things go well, to retreat if they don't, to pivot if a new opportunity emerges. These optionalities have economic value not captured by a single-path DCF. For Alibaba: (1) the option to IPO Cainiao, worth perhaps US$5-8bn in unlocked value; (2) the option to scale Tongyi Qianwen as a separate cloud-AI platform; (3) the option to deploy ¥400bn of cash into a transformative AI acquisition; (4) the option to retreat from AIDC if losses don't trend down, freeing up management bandwidth. None of these show up in the DCF. A sophisticated valuation adds an "option value" line — often 5-10% of the consolidated valuation — to reflect this. 普通 DCF 把未来视为单一预测轨迹。现实中，管理层拥有期权 — 如果事情顺利就扩张、不顺就撤退、新机会出现就转向。这些可选性具有单路径 DCF 未捕捉的经济价值。对阿里巴巴：(1) 菜鸟 IPO 的期权，也许价值 50-80 亿美元的解锁价值；(2) 把通义千问作为独立云-AI 平台规模化的期权；(3) 把 4000 亿元现金部署到变革性 AI 收购的期权；(4) 如果亏损不下降就从 AIDC 撤退的期权，释放管理层带宽。这些都不显示在 DCF 中。资深估值会加上"期权价值"行 — 常常占合并估值 5-10% — 以反映这些。
A DCF can produce a number like "$226.05." Resist the temptation to take the cents seriously. The model is precise but not accurate. The 95% confidence interval on the DCF, with reasonable parameter uncertainty, is something like $150 to $310. A price target of "$226" is shorthand for "we believe the central value lies in the upper half of that range." Treating the number as having three significant figures of precision is the most common analytical mistake. The two-way sensitivity table on our DCF page is the antidote: read the entire surface, not the single cell. DCF 可以产生像 "$226.05" 这样的数字。抵制把分认真对待的诱惑。模型精确但不准确。在合理参数不确定性下，DCF 的 95% 置信区间大约是 $150 到 $310。目标价 "$226" 是 "我们相信中心值在该范围上半部" 的简称。把数字视为具有三位有效精度，是最常见的分析错误。我们 DCF 页上的双向敏感性表是解药：读整张表面，而不是单个单元格。
Our DCF says Alibaba is worth $226. Our SOTP says $297. Trading comparables suggest $130-150 if you anchor to the peer median. Three answers, three frameworks. Which one wins, and how do you defend the choice? 我们的 DCF 说阿里巴巴值 $226。我们的 SOTP 说 $297。交易可比公司在锚定同行中位数时显示 $130-150。三个答案、三种框架。哪一个胜出，您如何为选择辩护？
Each method has well-known biases. The disciplined practitioner doesn't just average the three; she weights them according to what each method does well and poorly for the specific company. 每种方法都有众所周知的偏差。有纪律的从业者不只是把三者平均；她根据每种方法对特定公司做得好与做得差的地方进行加权。
The DCF and SOTP both produce values above the spot price, while the comps suggest the spot is roughly fair. The divergence is informative: (a) the DCF and SOTP are framework-driven — they say "if the future plays out as we forecast, the cash flows are worth more than the stock" — while (b) the comps are market-driven — they say "the market is currently pricing similar-looking businesses at multiples that imply $130-150 for Alibaba." Both can be true simultaneously: the cash flows are valuable, but the market is unwilling to pay for them today because of macro/narrative/flow forces. DCF 和 SOTP 都产生高于现价的价值，而可比公司提示现价大致合理。分歧具有信息：(a) DCF 和 SOTP 是框架驱动 — 它们说"如果未来按我们的预测展开，现金流值更多"— 而 (b) 可比公司是市场驱动 — 它们说"市场目前为看起来类似的业务支付的倍数，对阿里巴巴隐含 $130-150。" 两者可以同时成立：现金流有价值，但市场今天不愿为它们付钱，因为宏观/叙事/资金流。
A research analyst's job is to take a side. We weight DCF and SOTP at 70% combined (the "fundamentals win in the long run" view) and comps at 30% (anchored to current market reality). That mix produces our blended $262 price target. A different shop might use 50/50 and get a target closer to $200. A purely tactical trader, who only cares about the next 3 months, might use 90% comps and arrive at $130 — and would be neither wrong nor right; just operating on a different time horizon. The clarity of saying which weighting we used and why is what separates a defensible recommendation from a hunch. 研究分析师的工作是选边。我们把 DCF 和 SOTP 合计 70% 加权 ("基本面长期获胜" 观)，可比公司 30% (锚定当前市场现实)。这一组合产生我们 $262 的混合目标价。另一家机构可能用 50/50 并得到接近 $200 的目标价。一个纯战术交易者只关心未来 3 个月，可能用 90% 可比公司并到达 $130 — 既不对也不错；只是在不同时间视角操作。说清我们用了哪种加权及为什么，是可辩护建议与拍脑袋之间的区别。
An analyst who can't defend a price target in a 60-second elevator pitch hasn't done the work. Here is the test: 不能用 60 秒电梯演讲为目标价辩护的分析师，没有完成工作。这是测试：
"Our $226 price target on Alibaba is a 70/30 blend of DCF ($226) and SOTP ($297), de-weighted by 30% trading comparables ($150). Implied 12-month upside is 91% from the current $118.50 spot. The base case requires three things: TT take-rate normalization (sup\xC2\xA0ported by the late-FY26 quarter's +10% GMV print), Cloud reacceleration (Q3 FY26 confirmed at +21% YoY), and continued capital return (US$13bn buyback per year on the ¥400bn cash pile). The bear case at $95 assumes a softer consumer for two more years and a 100bp WACC widening on geopolitical risk. The risk/reward is asymmetric: 90% upside to base, ~20% downside to bear. We rate Overweight." "我们对阿里巴巴 $226 的目标价是 DCF ($226) 和 SOTP ($297) 的 70/30 混合，按 30% 交易可比公司 ($150) 加权调整。相对当前现价 $118.50 隐含 12 个月上行 91%。基准情景需要三件事：TT 货币化率正常化 (受 FY26 末季度 +10% GMV 数据支撑)、云重新加速 (FY26 Q3 同比 +21% 已确认)，以及持续的资本回报 (基于 4000 亿元现金堆每年 130 亿美元回购)。$95 的熊市情景假设消费再疲弱两年、地缘政治风险下 WACC 扩大 100 个基点。风险/回报不对称：基准上行 90%，熊市下行约 20%。我们评级'增持'。"
Notice the structure: methods, blend, key assumptions, asymmetry, rating. Every word does work. No filler. A buy-side PM, having heard this, can decide in another 60 seconds whether to engage further. That is the standard. 注意结构：方法、混合、关键假设、不对称、评级。每个词都在工作。没有填充。一位买方 PM 听完后，能在另一个 60 秒内决定是否进一步介入。这就是标准。
The most overlooked line in any financial model is "change in net working capital." It can swing free cash flow by tens of billions for a company the size of Alibaba — and yet a generation of analysts treats it as a residual plug. Done well, working-capital analysis is the difference between a model that survives the first quarter and one that doesn't. 任何财务模型中最容易被忽视的一行是"营运资本净额变动"。它能让一家阿里巴巴规模的公司的自由现金流摆动数百亿 — 然而一代分析师把它视为"残差"填塞。做得好的话，营运资本分析是模型能撑过第一个季度与撑不过之间的区别。
Working capital = current operating assets − current operating liabilities. In plain words: the cash tied up in inventory and unpaid customer bills (receivables), minus the cash you've borrowed from suppliers (payables) and from customers who pre-paid (deferred revenue). For a growing company, working capital usually grows — more sales means more inventory and receivables — and the cash is "absorbed" by the balance sheet rather than landing in the bank. The cash flow statement makes this visible as a negative working-capital change. 营运资本 = 流动经营资产 − 流动经营负债。大白话：被库存和未付客户账单 (应收) 占用的现金，减去您从供应商 (应付) 和预付款客户 (递延收入) 那里"借"来的现金。对成长公司而言，营运资本通常增长 — 更多销售意味着更多库存和应收 — 现金被资产负债表"吸收"而不是落到银行。现金流量表把这显示为负向的营运资本变动。
The cash conversion cycle (CCC) = DSO + DIO − DPO. For Alibaba, low DSO (~30 days) + low DIO (~40 days for the direct-purchase tail) − high DPO (~200 days) = a negative CCC of roughly −130 days. In plain words: Alibaba gets paid by customers, holds suppliers' cash for nearly 7 months, and only then pays the supplier. The float earns interest. For a company doing ¥1tn of business per year, even a 5% interest rate on 130 days of float is worth ¥18bn — a hidden, structural advantage. 现金转换周期 (CCC) = DSO + DIO − DPO。对阿里巴巴，低 DSO (~30 天) + 低 DIO (~40 天，直采尾部) − 高 DPO (~200 天) = 负 CCC 约 −130 天。大白话：阿里巴巴先从客户收款，把供应商的钱持有近 7 个月，然后才付供应商。浮存金赚利息。对一家每年做 1 万亿元业务的公司，即便对 130 天浮存金 5% 利息，价值 180 亿元 — 一个隐藏的、结构性的优势。
In our DCF, we model "change in NWC" as 5% of the change in revenue — a benign assumption suitable for a mature platform. If we had instead assumed 15% (closer to a typical goods-retail business), the cumulative working-capital absorption over five years would be three times larger, shaving roughly 8% off the DCF target. The assumption is easy to overlook because it's a single number; getting it wrong, in either direction, materially moves the answer. 在我们的 DCF 中，我们把"营运资本变动"建模为收入变动的 5% — 适合成熟平台的温和假设。如果我们改假 15% (更接近典型商品零售业务)，五年累计营运资本吸收会大三倍，把 DCF 目标价削掉约 8%。该假设容易被忽视，因为它是单个数字；任一方向算错，都会显著改变答案。
A common beginner's mistake is to value a company on its earnings alone — ignoring the cash on the balance sheet and the debt. For some companies, this barely matters; for others, like Alibaba, the balance-sheet items are 25% of the equity value. Doing valuation without the balance sheet is doing valuation with one eye closed. 初学者的常见错误是仅以盈利估值一家公司 — 忽视资产负债表上的现金和债务。对一些公司这无关紧要；对另一些，如阿里巴巴，资产负债表项目占权益价值的 25%。不带资产负债表做估值，等于闭着一只眼做估值。
Both DCF and SOTP produce an enterprise value — the value of the operating business, before financing decisions. To get to equity value (what shareholders own), you adjust: DCF 和 SOTP 都产生一个企业价值 — 经营业务在融资决策前的价值。要得到权益价值 (股东拥有的)，您调整：
For Alibaba, the cash and equity investments alone add about US$110 of value per ADR — meaning that even if the operating business were worth zero, the stock would be worth $110 from the balance sheet items. That balance-sheet floor is one of the reasons the bear case at $95 is a "stress" case rather than an "expected" case. 对阿里巴巴，仅现金和股权投资每股 ADR 增加约 110 美元的价值 — 意味着即使经营业务一文不值，股价也会因资产负债表项目值 110 美元。这个资产负债表底线是 $95 熊市情景属于"压力"情景而非"预期"情景的原因之一。
Alibaba owns approximately 33% of Ant Group, the parent of Alipay. Ant was scheduled to IPO in October 2020 at a valuation that would have made it the largest IPO in history — over US$300bn equity value. The IPO was suspended at the last moment by Chinese regulators, and Ant's value has been written down to roughly US$70-80bn on Alibaba's books today. The 33% stake is therefore worth about US$24-26bn on the carrying value, but the "true" market value is uncertain — somewhere between the written-down level and the original IPO price. Most sell-side models use a range of US$10-20bn for the Alibaba share. Our SOTP uses the conservative end. 阿里巴巴持有蚂蚁集团约 33%，蚂蚁是支付宝的母公司。蚂蚁原定 2020 年 10 月上市，估值会让它成为历史上最大的 IPO — 权益价值超 3000 亿美元。IPO 在最后时刻被中国监管机构暂停，蚂蚁的价值在阿里巴巴账上已减记至约 700-800 亿美元。33% 股权按账面价值约值 240-260 亿美元，但"真实"市场价值不确定 — 介于减记水平和原 IPO 价之间。多数卖方模型使用阿里巴巴份额 100-200 亿美元的区间。我们的 SOTP 使用保守端。
If Ant ever IPOs at $200bn (a plausible level — Tencent's WeChat Pay is comparable scale and Tencent trades at $620bn market cap), the Alibaba share would be worth US$66bn — adding roughly US$28 per ADR to our SOTP. Yet none of this is in the current price. This is one of the embedded "real options" that makes Alibaba's risk/reward asymmetric. 如果蚂蚁有朝一日以 2000 亿美元上市 (合理水平 — 腾讯的微信支付规模相当，腾讯市值 6200 亿美元)，阿里巴巴的份额值 660 亿美元 — 给我们的 SOTP 增加每股 ADR 约 28 美元。然而这些都不在当前价格里。这是让阿里巴巴风险/回报不对称的嵌入式"实物期权"之一。
The hands-on craft of building a model is impossible to learn by reading. But it is possible to provide the scaffolding and the discipline. Here is the playbook one of the world's top equity-research teams gives a fresh hire on day one. 搭模型的动手手艺无法仅靠阅读学会。但提供脚手架与纪律是可能的。这是世界顶级股权研究团队第一天给新员工的剧本。
Pull at least 5 years of historical financial statements from the company's annual reports (10-K for US filers, 20-F for foreign private issuers like Alibaba, annual report for HKEX issuers). Type them into a clean Excel sheet, one column per year. Do not paste-link — type them, line by line. The slowness is the point: typing forces you to read every number, and reading every number reveals everything from accounting policy changes to one-off charges to category reclassifications. By the time you finish a 5-year history, you know the company better than 95% of the people who claim to follow it. 从公司年报中拉取至少 5 年的历史财务报表 (美国申报人的 10-K、像阿里巴巴这样的外国私人发行人的 20-F、港交所发行人的年报)。把它们打到一份干净的 Excel 表里，每年一列。不要粘贴链接 — 一行一行地打。慢就是要点：打字迫使您读每一个数字，读每一个数字会揭示一切 — 从会计政策变化、到一次性费用、到类别重分类。当您完成 5 年历史时，您对这家公司的了解超过 95% 自称追踪它的人。
A separate tab — call it "Assumptions" — that holds every input you might want to flex. Forecast revenue growth, gross margin, opex as % of revenue, capex / revenue, working-capital intensity, tax rate, share count, FX. Color-code the inputs (we use blue for scenario-flexible inputs, black for in-sheet calculations, green for cross-sheet links). This convention has been standard on Wall Street for thirty years; you should adopt it. 一个单独的标签页 — 叫"假设" — 持有您可能想要调整的每一个输入。预测收入增长、毛利率、营业费用占收入%、资本开支/收入、营运资本密度、税率、股本数、汇率。给输入着色 (我们用蓝色表示情景可调输入、黑色表示表内计算、绿色表示跨表链接)。这一惯例在华尔街已是 30 年标准；您应该采用。
Build forward years from the assumptions. Revenue = prior year × (1 + growth rate). Gross profit = revenue × gross margin. Each opex line = revenue × the appropriate ratio. Sum them to EBIT. Add the small non-operating items. Subtract tax (revenue-based estimate). Subtract minorities. You are at net income. Compute EPS = net income / diluted share count. The forward P&L is now a function of your assumptions. 从假设构建预测年。收入 = 上年 × (1+增长率)。毛利润 = 收入 × 毛利率。每条营业费用 = 收入 × 适当比率。加总得 EBIT。加上小的非经营项目。减税 (基于收入的估计)。减少数股东。您到达净利润。计算 EPS = 净利润/稀释股本数。前瞻利润表现在是您假设的函数。
Each line of the balance sheet is forecast either as a % of revenue (most operating items), held flat (debt), or driven by the cash flow statement (cash, retained earnings, PP&E). The art is in keeping the assumptions consistent: if revenue grows 10%, receivables should grow roughly 10% (not 30%). Use turnover days as a sanity check. 资产负债表的每一行都被预测为：要么收入的%(大多数经营项目)、要么持平 (债务)、要么由现金流量表驱动 (现金、留存收益、PP&E)。艺术在于保持假设一致：如果收入增长 10%，应收账款应大致增长 10% (不是 30%)。用周转天数作为健全性检查。
Start from net income (link to the P&L). Add D&A (the same line that reduced PP&E on the BS). Add SBC. Subtract change in working capital (computed from the BS changes). That's CFO. Subtract capex (input from assumptions; flows to PP&E). Subtract M&A and other investing. That's CFI. Subtract dividends paid (input from assumptions; reduces retained earnings). Subtract buybacks (input from assumptions; increases treasury stock). That's CFF. Sum CFO + CFI + CFF = net change in cash. The ending cash on the cash flow statement must equal the cash line on the balance sheet. If it doesn't, your model has an error — find it. 从净利润起步 (链接到利润表)。加 D&A (同一行在资产负债表上减少 PP&E)。加 SBC。减营运资本变动 (从资产负债表变动计算)。这是 CFO。减资本开支 (来自假设的输入；流向 PP&E)。减并购和其他投资。这是 CFI。减已付股息 (来自假设的输入；减少留存收益)。减回购 (来自假设的输入；增加库存股)。这是 CFF。CFO + CFI + CFF = 现金净变动。现金流量表上的期末现金必须等于资产负债表上的现金行。如果不等，模型有错 — 找出来。
From your P&L, pull EBIT for each forecast year. Compute NOPAT. Pull D&A and capex from elsewhere. Compute working-capital change. That gives you unlevered FCF for each year. Discount each year by (1 + WACC)year. Compute terminal value via the Gordon formula. Discount terminal. Sum to enterprise value. Bridge to equity value via cash, debt, minority interest. Divide by shares to get per-share value. Done. 从您的利润表中拉取每个预测年的 EBIT。计算 NOPAT。从其他地方拉取 D&A 和资本开支。计算营运资本变动。这给您每一年的无杠杆 FCF。把每年按 (1+WACC)年 折现。通过戈登公式计算终值。折现终值。加总到企业价值。通过现金、债务、少数股东权益桥到权益价值。除以股本得到每股价值。完成。
Now break your model. What happens to the price target if revenue growth is 200bps lower than your base case for the entire forecast period? What if gross margin is 100bps lower? What if WACC is 50bps higher? Build a two-way data table that flexes the two assumptions that matter most. This is the discipline that separates a model from a fortune-telling exercise. 现在打破您的模型。如果收入增长在整个预测期低于基准 200 个基点，目标价会怎样？如果毛利率低 100 个基点呢？如果 WACC 高 50 个基点呢？构建一份双向数据表，调整最重要的两个假设。这种纪律把模型与算命练习区分开。
The best Excel models all share one feature: they use named ranges instead of cell references in formulas. Instead of =C9*Assumptions!$Q$13, write =Revenue_FY26 * GrossMargin_FY26. Your formulas become readable in plain English, and a year from now when you come back to the model, you will thank yourself. The same trick: never paste-link to external workbooks; copy the data in as values with a "Source" cell next to it. External links break and embarrass you in front of clients.
最好的 Excel 模型都共享一个特征：在公式中使用命名范围而非单元格引用。不要写 =C9*Assumptions!$Q$13，写 =Revenue_FY26 * GrossMargin_FY26。您的公式变得用大白话可读，一年后您回到模型时，您会感谢自己。同样的诀窍：永远不要粘贴链接到外部工作簿；把数据作为值复制进来，旁边放一个"来源"单元格。外部链接会断开并让您在客户面前出丑。
Every chart and number on this site is derived from a real, working Alibaba spreadsheet model — 23 sheets, 14 years of historical statements, 6 years of forecast, full DCF, SOTP, sensitivity, and the segment build. The English version of the workbook is in the project root; we also produced a fully translated Mandarin version. Open the model, flex an assumption, watch the entire chain of dependents update. The discipline of building one from scratch — using these as references — is the single best way to internalize the material in this site. 本网站上的每张图和每个数字都来自一份真实可运行的阿里巴巴电子表格模型 — 23 个标签页、14 年历史报表、6 年预测、完整 DCF、SOTP、敏感性、和分部搭建。英文版工作簿在项目根目录；我们还制作了完整翻译的中文版。打开模型，调整一个假设，看整条依赖链更新。从头搭一份 — 用这些做参考 — 是把本网站的材料内化的最佳方式。
Some mistakes are so common they form a folk taxonomy passed from senior analyst to junior. Knowing the list does not immunize you against the mistakes, but it shortens the time to recovery when (not if) you make them. 有些错误如此常见，已形成从资深分析师传给初级的民间分类法。知道清单并不会让您免疫，但在您 (不是"如果") 犯这些错误时，会缩短恢复时间。
If you avoid these ten, you will avoid 80% of the analytical mistakes the next ten cohorts of equity-research juniors will make. The other 20%? You will discover yourself, the hard way, and add to the list. 如果您避开这十个，您就避开了未来十届股权研究初级会犯的 80% 分析错误。剩下的 20%？您会以艰难的方式自己发现，并加到清单上。
A single company's history in the public markets can teach more about IPOs, secondary listings, conglomerate discounts, and regulatory risk than any textbook chapter. Alibaba's twelve-year arc from 2014 to 2026 has touched every major theme in modern equity markets — and it's not over. 一家公司在公开市场的历史能教会您比任何教科书章节更多的 IPO、二次上市、集团折价、监管风险知识。阿里巴巴从 2014 年到 2026 年的十二年弧线触及了现代股票市场的每一个主要主题 — 而且尚未结束。
Alibaba listed on the New York Stock Exchange on 19 September 2014. The deal priced at US$68 per ADR, raised US$25 billion (later upsized to US$25 billion after the green shoe), and gave the company a market cap of US$170 billion on day one — making it the largest IPO in history at that moment. The lead underwriters were Credit Suisse, Deutsche Bank, Goldman Sachs, Morgan Stanley, and a consortium of others. The deal was 100× oversubscribed and traded up 38% on the first day to close at US$93.89. For context: at that valuation, Alibaba was already worth more than Facebook (at IPO), Amazon (at IPO), and Google (at IPO) combined. The IPO was the moment Chinese internet officially arrived as a global asset class. 阿里巴巴 2014 年 9 月 19 日在纽约证交所上市。交易定价每股 ADR 68 美元，融资 250 亿美元 (绿鞋后扩大至 250 亿美元)，公司首日市值 1700 亿美元 — 当时史上最大 IPO。主承销商包括瑞信、德意志银行、高盛、摩根士丹利及其他财团。交易超额认购 100 倍，首日上涨 38%，收于 93.89 美元。作为对照：在那个估值下，阿里巴巴已经比 Facebook (IPO 时)、亚马逊 (IPO 时)、谷歌 (IPO 时) 合起来还值钱。该 IPO 是中国互联网作为全球资产类别正式登场的时刻。
In November 2019, Alibaba completed a secondary listing on the Hong Kong Stock Exchange under ticker 9988. The deal raised US$13bn, making it the largest HK listing in nine years. Strategic rationale: provide an Asian-time-zone listing accessible to mainland Chinese investors via the Stock Connect program, and create a "safe harbor" in case the US ADR listing came under regulatory pressure. The 1 ADR : 8 ordinary share fungibility was preserved — investors could (and can) move between the two listings via depositary conversion. This optionality became extremely valuable during the 2022 PCAOB tensions. 2019 年 11 月，阿里巴巴在港交所完成二次上市，代码 9988。交易融资 130 亿美元，是九年来香港最大上市。战略理由：提供一个亚洲时区上市、可通过沪深港通让大陆中国投资者接触，并在美国 ADR 上市遭遇监管压力时创造"安全港"。1 ADR : 8 普通股的可互换性被保留 — 投资者可以 (现在仍可) 通过托管转换在两个上市之间移动。这一可选性在 2022 年 PCAOB 紧张局势期间变得极为宝贵。
In late October 2020, Ant Group — Alibaba's 33%-owned affiliate that operates Alipay — was 48 hours from completing the largest IPO in history (US$34bn). On 2 November, Chinese regulators suspended the IPO, citing changes in the regulatory environment for financial technology. The Alibaba stock fell 8% on the news. Two weeks later, the State Administration for Market Regulation launched an antitrust investigation against Alibaba. In April 2021, Alibaba was fined US$2.8bn — at the time the largest antitrust fine in Chinese history. The Ant IPO has never been refiled. The episode marked the start of a multi-year regulatory recalibration that compressed China-internet valuations across the sector. 2020 年 10 月末，蚂蚁集团 — 阿里巴巴 33% 持股的关联公司，运营支付宝 — 距完成史上最大 IPO (340 亿美元) 仅有 48 小时。11 月 2 日，中国监管机构以金融科技监管环境变化为由暂停 IPO。阿里巴巴股票当日下跌 8%。两周后，国家市场监督管理总局对阿里巴巴启动反垄断调查。2021 年 4 月，阿里巴巴被罚款 28 亿美元 — 当时中国历史上最大的反垄断罚款。蚂蚁 IPO 从未重新申报。这一事件标志着一场多年监管再校准的开始，压缩了整个中国互联网板块的估值。
In March 2023, Alibaba announced its largest reorganization in history: splitting into six business groups (TT, Cloud, AIDC, Cainiao, LSG, DME), each with its own CEO, board, and the structural ability to raise external capital or eventually IPO. The market interpreted the move two ways: (a) a defensive response to regulatory pressure to "de-platformize" the conglomerate; and (b) an offensive move to unlock the SOTP discount by making each segment separately valuable. Within weeks, plans were announced to IPO Cainiao (later delayed) and to spin off CIG (later reversed in 2024 as the AI boom made management want to keep cloud inside). The reorganization is the structural backdrop against which every current Alibaba investment thesis is built. 2023 年 3 月，阿里巴巴宣布其历史上最大的重组：拆分为六个业务集团 (TT、云、AIDC、菜鸟、本地生活、数字媒体)，每个集团拥有自己的 CEO、董事会、以及对外融资或最终 IPO 的结构能力。市场以两种方式解读：(a) 对监管压力 "去平台化" 集团的防御性回应；和 (b) 通过使每个分部分别有价值来释放 SOTP 折价的进攻性举措。几周内宣布了菜鸟 IPO 计划 (后被推迟) 和分拆 CIG (后在 2024 年 AI 繁荣使管理层希望把云留在内部时撤销)。重组是当前每一项阿里巴巴投资论点构建的结构性背景。
In September 2023, Daniel Zhang — Jack Ma's chosen successor since 2019 — stepped down as CEO and chairman. Joe Tsai (co-founder, long-serving vice chair) became chairman; Eddie Wu (a low-profile founding engineer) became CEO. The new team's first major statement was a 180-degree pivot on capital allocation: from "growth at all costs" to "capital discipline + shareholder return." The first-ever annual dividend was announced in November 2023. The buyback program was supercharged. The market took 18 months to fully digest the change in posture; by early 2025, the stock had begun a meaningful rerating that continues today. 2023 年 9 月，张勇 — 自 2019 年起马云钦定的继任者 — 辞去 CEO 与董事长职务。蔡崇信 (联合创始人、长期副董事长) 接任董事长；吴泳铭 (低调的创始工程师) 接任 CEO。新团队的第一个重大声明是资本配置上的 180 度转向：从"不惜代价增长"到"资本纪律+股东回报"。2023 年 11 月宣布了史上首次年度股息。回购计划被提速。市场用了 18 个月才完全消化姿态的变化；到 2025 年初，股价开始了显著的重估，并延续至今。
Through 2024, the global AI investment narrative shifted from "GenAI applications are coming" to "GenAI infrastructure is being built right now." Alibaba's Tongyi Qianwen foundation model, combined with the CIG cloud-infrastructure footprint, suddenly became one of three credible China-domestic AI stacks (alongside Baidu and Tencent). Cloud revenue accelerated from low-single-digit to mid-twenties-percent growth. The market began awarding Alibaba a cloud multiple inside the SOTP. By late 2024, the stock had risen ~80% from its September 2024 trough — though it remained well below its 2020 peak. The story of the next 18 months is whether the cloud rerating broadens out to the full company or remains a partial repricing. 2024 年贯穿全年，全球 AI 投资叙事从"GenAI 应用即将到来"转向"GenAI 基础设施正在建设中"。阿里巴巴的通义千问基础模型，结合 CIG 云基础设施足迹，突然成为中国国内三个可信 AI 堆栈之一 (与百度、腾讯一道)。云收入从低个位数加速到 25% 左右增长。市场开始在 SOTP 中给阿里巴巴一个云倍数。到 2024 年末，股价从 2024 年 9 月低点上涨约 80% — 尽管仍远低于 2020 年顶峰。未来 18 个月的故事在于：云的重估是否扩散到整家公司，还是仍是局部重定价。
Twelve years after the IPO, the central questions for Alibaba are surprisingly familiar to those who follow Asian tech: (1) Can the platform business — TT — defend its take-rate against PDD and Douyin? (2) Can Cloud and AI sustain reaccelerated growth? (3) Will the international expansion (AIDC) become profitable, or remain a perpetual investment? (4) Will the geopolitical overhang ease, or worsen? The answers will determine whether the next twelve years look like a quiet compounder or another period of regulatory and competitive volatility. Equity research exists to make money by getting these questions less wrong than the consensus. IPO 12 年后，阿里巴巴的核心问题对跟踪亚洲科技的人来说出乎意料地熟悉：(1) 平台业务 — TT — 能否对抗拼多多与抖音守住其货币化率？(2) 云与 AI 能否维持重新加速的增长？(3) 国际扩张 (AIDC) 会变得盈利，还是继续无尽投资？(4) 地缘政治阴云会缓解还是恶化？答案将决定接下来十二年看起来是一个安静的复利者，还是另一段监管与竞争波动期。股权研究通过比共识更少出错地回答这些问题来赚钱。
Alibaba's competitive context is not just other e-commerce companies. It is a five-front war across e-commerce, cloud, local services, logistics, and international markets — with a different formidable opponent on each front. 阿里巴巴的竞争背景不只是其他电商公司。它是横跨电商、云、本地生活、物流、国际市场的五线作战 — 每条战线都有不同的强大对手。
Founded in 2015, PDD built its initial scale through a viral "group-buying" mechanism that mobilized lower-tier-city Chinese consumers. By 2024, PDD's domestic GMV had grown to a level that approaches half of TT's. More importantly, PDD's international business (Temu) has rapidly scaled in the US and Europe — putting pressure on both TT's domestic share and AliExpress's international growth. PDD's economics are extraordinary: net margin of 22.7% (vs. Alibaba consolidated 13%) and ROE of 27%. The market awards PDD a forward P/E of 8× — actually below Alibaba — because investors are nervous about the sustainability of Temu's subsidy-led growth. 2015 年成立的 PDD 通过病毒式"团购"机制建立初始规模，动员了中国下沉市场消费者。到 2024 年，PDD 国内 GMV 已增至接近 TT 一半的水平。更重要的是，PDD 的国际业务 (Temu) 在美国和欧洲快速规模化 — 同时对 TT 国内份额和速卖通国际增长施压。PDD 的经济性非凡：净利率 22.7% (相对阿里巴巴合并 13%)、ROE 27%。市场给 PDD 前瞻 PE 8 倍 — 实际上低于阿里巴巴 — 因为投资者对 Temu 补贴驱动增长的可持续性紧张。
Tencent and Alibaba are sometimes framed as opposites — social vs. commerce, games vs. trade — but in practice they are two faces of the same Chinese internet duopoly. Tencent's WeChat Pay competes with Alipay in payments; its Mini-Programs platform competes with Taobao for merchant attention; Tencent Cloud is the #2 Chinese cloud player after Alibaba Cloud. The relationship has historically been complementary at the top of the funnel (each company tolerates the other's dominance in its home turf), competitive at the application layer. Tencent trades at 18× P/E versus Alibaba's 9.4× — a multiple gap that partly reflects perceived growth quality and partly reflects social-media-platform premiums. 腾讯和阿里巴巴有时被描绘为对立 — 社交 vs. 商务、游戏 vs. 贸易 — 但实际上它们是同一个中国互联网双寡头的两面。腾讯的微信支付与支付宝在支付上竞争；其小程序平台与淘宝竞争商家注意力；腾讯云是中国第二大云玩家，仅次于阿里云。这种关系在历史上在漏斗顶端互补 (每家公司容忍对方在其主场的主导地位)，在应用层竞争。腾讯交易 18× PE 相对阿里巴巴 9.4× — 这一倍数差距部分反映感知增长质量、部分反映社交媒体平台溢价。
JD is the only large-scale Chinese e-commerce player that operates a direct-purchase, owned-inventory, owned-logistics model — closer in spirit to Amazon than to a marketplace. The model is capital-intensive and lower-margin (16% gross vs. Alibaba's 40%) but produces a different competitive moat: speed of delivery, control over product authenticity, and a captive merchant base that values JD's logistics. JD trades at the cheapest multiple of the China-internet group — about 8× P/E — partly because growth has been mid-teens and partly because the operating model is structurally lower-margin. 京东是唯一一家运营直采、自有库存、自有物流模式的大规模中国电商玩家 — 精神上比集市更接近亚马逊。该模式资本密集、利润率较低 (16% 毛利率 vs. 阿里巴巴 40%)，但产生不同的竞争护城河：配送速度、产品真伪控制、重视京东物流的"被锁"商家基础。京东交易于中国互联网组中最便宜的倍数 — 约 8× PE — 部分因为增长一直在中段、部分因为运营模式结构性较低利润率。
ByteDance is not yet public (so no stock price to reference), but its impact on the Chinese commerce landscape may be the most consequential of all. Douyin's livestream-commerce GMV reportedly grew from near-zero in 2020 to over RMB 2.5 trillion in 2024. The format — entertainers and KOLs (key opinion leaders) selling products live to followers — is structurally different from search-led commerce on Taobao. Where Taobao is "I know what I want; let me search," Douyin is "I'm watching; let me buy what's on screen." Different attention pattern, different conversion economics. Whether livestream represents 10% or 30% of total Chinese e-commerce in 2030 is one of the largest single uncertainties in our model. 字节跳动尚未上市 (所以没有股价可参考)，但它对中国商务格局的影响可能是所有人中最重大的。抖音的直播电商 GMV 报道从 2020 年接近零增长到 2024 年的 2.5 万亿元以上。该格式 — 表演者和 KOL (关键意见领袖) 向粉丝直播销售产品 — 在结构上不同于淘宝上的搜索式商务。淘宝是"我知道我想要什么；让我搜索"，抖音是"我在看；让我买屏幕上的东西"。不同的注意力模式、不同的转化经济学。直播在 2030 年代表中国电商总量 10% 还是 30%，是我们模型中最大的单一不确定性之一。
Amazon is not directly a domestic competitor to Alibaba in China (it exited the Chinese marketplace in 2019), but it is the dominant comparison for valuation purposes. Amazon's market cap of US$2.2tn is roughly 8× Alibaba's despite revenues only 5× larger and EBIT margins roughly equivalent. The multiple gap reflects the market's confidence in AWS (the cloud growth runway, the margin profile, the international expansion), Prime ecosystem stickiness, and the perceived governance and regulatory premium of a US-domiciled company. The implicit question for Alibaba investors: how much of the Amazon multiple gap is structural (and permanent) versus cyclical (and closable)? 亚马逊不直接是阿里巴巴在中国的国内竞争对手 (它于 2019 年退出中国市场)，但它是估值目的最主要的比较对象。亚马逊 2.2 万亿美元市值大约是阿里巴巴 8 倍，尽管收入只大 5 倍、EBIT 利润率大致相当。倍数差距反映市场对 AWS (云增长跑道、利润率特征、国际扩张) 的信心、Prime 生态的粘性、以及对美国注册公司的感知治理与监管溢价。对阿里巴巴投资者的隐含问题：亚马逊倍数差距有多少是结构性 (永久) 的，有多少是周期性 (可闭合) 的？
If you take only one thing from this site, take this checklist. It is the synthesis of everything we've covered, compressed to one page, applicable to any name in any industry. 如果您只从本站带走一件事，请带走这份清单。它是我们覆盖一切的综合，压缩到一页，适用于任何行业的任何股票。
Fifteen questions. Half a day of work for a name you've never looked at. A career's worth of practice for the names you cover. Equity research, at the end of the day, is the discipline of asking these questions before you put money to work — and the humility of revisiting them every quarter, in case any of the answers changed. 十五个问题。对您从未看过的股票，半天的工作。对您覆盖的股票，一职业生涯的练习。归根到底，股权研究是在投入资金之前问这些问题的纪律 — 以及每个季度重温它们的谦卑，以防任何答案改变了。
Financial models assume rational investors making consistent decisions based on probability-weighted expected values. Real investors include retirees facing forced selling at retirement, hedge-fund analysts whose careers depend on the next quarter, sovereign-wealth-fund committees voting under political pressure, and retail traders pursuing TikTok-driven momentum. The collective behavior of these heterogeneous actors is the market. Understanding it adds a layer of analytical sophistication that no model alone captures. 财务模型假设理性投资者根据概率加权的期望值做出一致决策。真实投资者包括退休时面临被迫卖出的退休者、职业取决于下一个季度的对冲基金分析师、在政治压力下投票的主权财富基金委员会、以及追逐 TikTok 驱动动量的散户。这些异质行动者的集体行为就是市场。理解它，会增加任何模型本身都无法捕捉的一层分析复杂度。
George Soros's most influential analytical concept is "reflexivity": the observation that, in some cases, the market price itself influences the fundamentals it is supposedly trying to value. A bank stock that falls sharply may trigger a depositor flight that creates an actual liquidity crisis — making the falling price self-fulfilling. A tech company whose stock collapses cannot use stock as currency for acquisitions or recruiting — making the collapse operationally damaging. A real-estate developer whose stock falls cannot refinance bonds at reasonable rates — making default more likely. The DCF imagines a world where price reflects value; reflexivity reminds us that price sometimes creates value (positive or negative). 索罗斯最有影响力的分析概念是"反身性"：观察到在某些情况下，市场价格本身影响它本应试图估值的基本面。一只急跌的银行股可能引发储户挤兑，造成实际流动性危机 — 使下跌价格自我实现。一家股价崩塌的科技公司无法用股票作为收购或招聘的货币 — 使崩塌对运营造成损害。一家股价下跌的房地产开发商无法以合理利率为债券再融资 — 使违约更可能。DCF 想象一个价格反映价值的世界；反身性提醒我们价格有时创造价值 (正或负)。
For Alibaba, reflexivity has been mostly benign — the company is far too well-capitalized to suffer operational damage from price weakness. But for smaller Chinese tech names, the 2022 sell-off triggered exactly the kind of feedback loop reflexivity predicts: lower stock prices → tighter access to capital → forced asset sales → lower future earnings → lower fundamental value. 对阿里巴巴而言，反身性大多是良性的 — 公司资本太充足，不会因价格疲软而遭受运营损害。但对较小的中国科技公司，2022 年的抛售触发了反身性所预测的反馈循环：更低的股价 → 资本接入收紧 → 被迫资产抛售 → 更低的未来盈利 → 更低的基本面价值。
Robert Shiller's research on "narrative economics" shows that stories — sometimes loose, viral, sticky stories — drive equity prices over horizons of years, not just hours. The "FAANG" narrative drove US tech multiples to 30-40× for half a decade. The "China-tech is uninvestable" narrative drove Chinese-internet multiples to single digits for three years. Neither narrative was perfectly true; both were strong enough to overwhelm fundamental analysis for extended periods. An analyst with a great DCF and no narrative awareness is bringing a calculator to a knife fight. 罗伯特·席勒关于"叙事经济学"的研究表明，故事 — 有时是松散的、病毒式的、有粘性的故事 — 在年而非小时的视角上驱动股票价格。"FAANG"叙事将美国科技倍数推到 30-40 倍长达半个十年。"中国科技不可投资"叙事将中国互联网倍数压到个位数长达三年。两个叙事都不完全真实；两者都足够强大，在长时间里压倒了基本面分析。一位拥有出色 DCF 但没有叙事意识的分析师，是带着计算器去刀战。
A growing share of global equity capital is allocated by passive index funds. When Alibaba enters or exits an MSCI index, when index weights are rebalanced quarterly, when a tracking ETF receives or returns capital, mechanical buying and selling flows through the market — independent of any fundamental opinion. Estimates suggest passive flows account for 40-50% of US equity volume. The implication: a fundamental analyst can be entirely correct about value and entirely wrong about price over a 6-12 month horizon, simply because the passive flow direction is opposite to the fundamental signal. 越来越大份额的全球股权资本由被动指数基金分配。当阿里巴巴进入或退出 MSCI 指数、当指数权重季度性再平衡、当追踪 ETF 接收或返还资本时，机械的买卖通过市场流动 — 独立于任何基本面观点。估计显示被动资金流占美国股票成交量的 40-50%。含义：一位基本面分析师在 6-12 个月视角上可能完全正确地判断价值却完全错误地判断价格，仅仅因为被动资金流方向与基本面信号相反。
For Alibaba specifically, the steady reduction in China weights inside global emerging-market indices over 2022-24 created persistent selling pressure that no fundamental improvement could fully offset. The 2025 stabilization (and partial reversal) in those weightings is one of the technical tailwinds that supports our bullish thesis — independent of the operating story. 具体到阿里巴巴，2022-24 期间全球新兴市场指数中持续降低中国权重，造成任何基本面改善都无法完全抵消的持续卖压。2025 年那些权重的稳定 (与部分逆转) 是支撑我们看多论点的技术顺风之一 — 独立于经营故事。
The most experienced research professionals are usually the most paranoid about their own biases. The discipline of writing things down (in handoff notes, in committee memos, in the model itself) creates an external check against the rationalizations that the mind constructs on the fly. 最有经验的研究专业人士通常对自己的偏差最为偏执。把事情写下来的纪律 (在交接笔记里、在委员会备忘录里、在模型本身里) 创造了对头脑临时构建的合理化的外部检查。
A stock that is a screaming buy on a 5-year view can be a poor short-term holding. A stock that's exhausted in the long-run might be a great 3-month trade. Time horizon is not an afterthought — it's the lens through which every other analytical input is filtered. Different time horizons require different analytical frameworks. 在 5 年视角上是尖叫买入的股票，可能是短期糟糕持有。在长期已耗尽的股票，可能是 3 个月伟大交易。时间视角不是事后想法 — 它是过滤所有其他分析输入的镜头。不同的时间视角需要不同的分析框架。
Position closed within a day to a week. Driven by news flow, technicals, momentum, options-related flows. Fundamental research is largely irrelevant at this horizon. Traders care about catalysts and crowd positioning. 仓位在一天到一周内关闭。由新闻流、技术、动量、期权相关资金流驱动。基本面研究在此视角下大多不相关。交易者关心催化剂和拥挤仓位。
Bought before an earnings print, sold after. Or held into a specific event (FDA approval, central bank decision). Fundamental research used to identify which catalysts are mispriced. 在业绩公布前买入，公布后卖出。或持有至特定事件 (FDA 批准、央行决定)。基本面研究用于识别哪些催化剂被错误定价。
The classical sell-side "price target" horizon. Built around earnings progression, multiple normalization, catalyst calendar. This is the home turf of sell-side research. 经典卖方"目标价"视角。围绕盈利进展、倍数正常化、催化剂日历搭建。这是卖方研究的主场。
Warren Buffett's home. Built around durable competitive advantages, sustained returns on capital, long-run secular trends. Quarterly fluctuations are noise. 沃伦·巴菲特的主场。围绕持久竞争优势、持续资本回报、长期长周期趋势搭建。季度波动是噪音。
The same stock — Alibaba at $118.50 — looks different at each horizon: 同一只股票 — 118.50 美元的阿里巴巴 — 在每个视角下看起来不同：
The most expensive mistakes in investing come from confusing horizons. A trader who reads a 5-year DCF and puts on a short-term position is asking for trouble: the fundamental signal won't manifest in days. A long-term investor who panics on a -10% week and sells is acting on a horizon mismatch: the weekly move is noise relative to the 5-year cash flow profile. Knowing which horizon you are operating on, and matching the analysis to that horizon, is half the discipline. 投资中最昂贵的错误来自混淆视角。一位交易者读 5 年 DCF 并下短期仓位是在自找麻烦：基本面信号不会在数天内显现。一位长期投资者在一周 -10% 上恐慌并卖出，是在视角错配上行动：周线波动相对于 5 年现金流特征是噪音。知道您在哪个视角上操作，并把分析与那个视角匹配，是纪律的一半。
At a 5-year horizon, the central question for Alibaba is: does China remain a valuable, accessible market for global capital? If yes, Alibaba is one of the cheapest ways to own a fortress balance sheet and a still-dominant set of platform franchises. If no — if geopolitical separation widens, if regulatory pressure resumes, if domestic competition continues to compress take-rates — then the operating cash flows we forecast will be real but valued at lower multiples. Our 12-month thesis is implicitly betting that the 5-year outlook is more constructive than the spot price implies. A buy-side investor with a 5-year mandate would weight that bet more heavily than a 12-month investor. 在 5 年视角上，阿里巴巴的核心问题是：中国是否仍是对全球资本有价值的、可接入的市场？如果是，阿里巴巴是拥有堡垒资产负债表和仍占主导的平台特许经营组合的最便宜方式之一。如果不是 — 如果地缘政治分离扩大、监管压力恢复、国内竞争继续压缩货币化率 — 那么我们预测的经营现金流将是真实的，但被以较低的倍数估值。我们 12 个月的论点隐含地押注 5 年前景比现价隐含的更建设性。具有 5 年授权的买方投资者会比 12 个月投资者更重地加权那个押注。
A first-time reader, faced with Alibaba's 350-page 20-F, will give up by page 50. A practiced reader extracts 80% of the value in 90 minutes by knowing exactly which sections to read and which to skim. Here's the playbook. 第一次阅读者，面对阿里巴巴 350 页的 20-F，会在第 50 页放弃。熟练读者通过准确知道读哪些章节、略哪些章节，在 90 分钟内提取 80% 的价值。这是剧本。
Foreign private issuers like Alibaba file Form 20-F annually with the SEC. The form has a prescribed structure (since 1999): Part I is descriptive (key information, business overview, risk factors), Part II is financial (statements, MD&A, controls), Part III contains exhibits. Most of the value, especially for a returning reader, is in three places. 像阿里巴巴这样的外国私人发行人每年向 SEC 提交 20-F 表格。该表格有规定结构 (自 1999 年起)：第一部分是描述性 (关键信息、业务概述、风险因素)，第二部分是财务 (报表、MD&A、控制)，第三部分包含附件。大部分价值，尤其对回访读者，在三个地方。
Identity of directors and senior management (Item 6). Useful only if there has been turnover. Major shareholders and related-party transactions (Item 7). Important to read once; check for new related-party transactions in subsequent years. Controls and procedures (Item 15). Mostly boilerplate; look for "material weaknesses" only — if there are any, the entire model is suspect. 董事及高级管理层身份 (项目 6)。仅在有人员变动时有用。主要股东与关联交易 (项目 7)。读一次很重要；在后续年份检查新的关联交易。控制与程序 (项目 15)。大多是样板文；只看"重大缺陷"— 如果有，整个模型可疑。
The most powerful way to read 20-Fs is in pairs: this year and last year, side by side. The diff tells you what changed. New risk factor language, modified accounting policies, expanded segment disclosures, additional named related parties — these signal evolving company priorities and risks. A senior analyst can spot the diff between two 20-Fs in 20 minutes and convert it into a single "what changed and why it matters" memo. That memo is the entire output of the analytical exercise. 阅读 20-F 最有力的方式是成对：今年与去年并排。差异告诉您什么变化了。新的风险因素措辞、修改的会计政策、扩展的分部披露、新增的关联方 — 这些信号公司不断演变的优先事项与风险。资深分析师能在 20 分钟内识别两份 20-F 之间的差异，并将其转化为单一的"什么变化以及为什么重要"备忘录。该备忘录是整个分析练习的全部输出。
The FY25 20-F (filed July 2025) showed three material diffs from the FY24 20-F that we believe were under-appreciated by the market: FY25 20-F (2025 年 7 月提交) 显示了三项相对 FY24 20-F 的实质性差异，我们认为被市场低估：
Each of these is a small signal individually. Together they tell a story: management is hardening its commitment to capital return, improving transparency, and conservatively positioning the Ant exposure. None of this shows up in headline EPS or revenue. All of it shows up in a careful reading of the 20-F. This is, mechanically, the kind of variant view that justifies an analyst's seat. 每一项单独看是一个小信号。合在一起讲述一个故事：管理层正在强化对资本回报的承诺、改进透明度、保守地定位蚂蚁敞口。没有一项显示在头条 EPS 或收入中。所有这些都显示在对 20-F 的仔细阅读中。机械上，这是证立一位分析师座位的那种变体观点。
For a US-listed Chinese ADR, geopolitics is not a separate consideration — it is woven through the discount rate, the multiple, the available investor base, and the executability of corporate strategy. A serious analyst tracks half a dozen specific bilateral and policy threads as carefully as she tracks revenue growth. 对一只在美国上市的中国 ADR，地缘政治不是独立考虑 — 它编织在贴现率、倍数、可触及的投资者基础、以及企业战略可执行性中。一位严肃的分析师跟踪六条特定的双边和政策线，就像她跟踪收入增长一样仔细。
An analyst can't forecast Treasury sanctions any more than she can forecast next week's weather, but she can incorporate the risk into the model. The standard mechanisms are: (1) widen the country risk premium in the WACC (we use 1.5%; some practitioners use 2.5%-3% in periods of stress); (2) discount the long-tail terminal value by an additional factor; (3) build "tail-risk" scenarios where a specific bad event reduces the valuation by a specific amount. Our bear-case scenario at $95 incorporates all three. 分析师无法预测财政部制裁，就像她无法预测下周的天气一样，但她可以把风险纳入模型。标准机制是：(1) 扩大 WACC 中的国家风险溢价 (我们用 1.5%；一些从业者在压力期用 2.5%-3%)；(2) 对长尾终值打折 — 增加一个折扣因子；(3) 构建"尾部风险"情景，特定坏事件让估值减少特定金额。我们 $95 的熊市情景包含三者。
In a worst-case PCAOB / delisting scenario, US ADR investors could find themselves holding paper that can no longer trade on the NYSE. The HK secondary listing is the structural safety valve: 1 ADR can be converted to 8 HK shares via the depositary, and the HK shares trade freely on HKEX. The depositary infrastructure has handled hundreds of millions of ADR-to-HK conversions; the operational risk of a forced conversion is real but bounded. The existence of this safety valve is one of the reasons the bear-case for Alibaba is not as severe as it would be without the dual listing — a structural detail that has billions of dollars of value embedded in it. 在最坏的 PCAOB/退市情景中，美国 ADR 投资者可能发现自己持有再也不能在纽交所交易的纸张。香港二次上市是结构性安全阀：1 ADR 可以通过托管转换为 8 股港股，港股在港交所自由交易。托管基础设施已处理过数亿股 ADR 到港股的转换；强制转换的操作风险真实但有界。这一安全阀的存在是阿里巴巴熊市情景没有像没有双重上市那样严重的原因之一 — 一个嵌入了数十亿美元价值的结构性细节。
Alibaba is a representative case of the platform economy that emerged between 2005 and 2020. Understanding where platforms came from — and where they go next — is essential to a long-term thesis on any of them. 阿里巴巴是 2005 年至 2020 年期间涌现的平台经济的代表案例。理解平台从何而来 — 以及它们接下来去向何处 — 对任何一只平台股的长期论点至关重要。
The platforms of 2005-2015 — Google, Facebook, Amazon, Alibaba, Tencent — were discovery platforms. They monetized the act of finding things: products, friends, information. The economic model was "two-sided network": more consumers attract more producers, who attract more consumers. Once a network achieved critical mass, the moat compounded itself. Margins were high; growth was high; reinvestment was modest. 2005-2015 的平台 — 谷歌、Facebook、亚马逊、阿里巴巴、腾讯 — 是发现平台。它们将寻找东西的行为货币化：产品、朋友、信息。经济模式是"双边网络"：更多消费者吸引更多生产者，再吸引更多消费者。一旦网络达到临界质量，护城河会自我复利。利润率高；增长高；再投资温和。
The platforms of 2015-2023 moved beyond pure matching to vertical integration. Amazon built logistics and devices. Alibaba built Cainiao logistics and Hema brick-and-mortar. Google launched hardware. The reasoning: control of the full stack defends against future competition and creates new monetization layers. Capex rose. Margins came under pressure. But the network effects of the original platform protected the cash flows long enough to fund the new investments. 2015-2023 的平台超越纯粹匹配走向垂直整合。亚马逊建物流和设备。阿里巴巴建菜鸟物流和盒马实体店。谷歌发布硬件。理由：控制完整堆栈防御未来竞争并创造新货币化层次。资本开支上升。利润率承压。但原始平台的网络效应足以保护现金流以资助新投资。
2023 onwards is the AI re-platforming. The original discovery platforms — including Alibaba — are racing to build AI infrastructure not because cloud-revenue growth is the prize (it is, but it's modest) but because the next decade's commerce, search, and creative workflows will run on top of an AI stack. Whoever owns the foundation model + the cloud infrastructure + the user / merchant base will own the next platform layer. This is why Alibaba's Tongyi Qianwen investment is strategically existential, not a side project. 2023 年起是 AI 重新平台化。原始发现平台 — 包括阿里巴巴 — 正在赛跑构建 AI 基础设施，不是因为云收入增长是奖品 (它是，但温和)，而是因为下一个十年的商务、搜索、创意工作流将运行在 AI 堆栈之上。谁拥有基础模型 + 云基础设施 + 用户/商家基础，谁就拥有下一个平台层。这就是为什么阿里巴巴的通义千问投资在战略上是关乎存亡的，不是一个副项目。
If Alibaba executes on the third wave, the company in 2035 looks materially different from 2025: TT continues to throw off ¥250-300bn of annual EBITA (a stable cash anchor), CIG becomes a US$50-80bn-revenue cloud-and-AI business with 20-25% EBIT margins, AIDC reaches profitable scale across 5-6 international corridors, and Cainiao trades as a public company at a US$40-60bn market cap. The consolidated entity is worth perhaps US$700-900bn — three times today's market cap. If Alibaba does not execute, the AI capex becomes stranded, the platform franchises slowly erode under Douyin/PDD pressure, and the company becomes a value trap held up by buybacks and dividend yield. Both futures are inside the bear/base/bull distribution; the next five years will decide which one we are heading toward. 如果阿里巴巴在第三波上执行，2035 年的公司看起来与 2025 年大不相同：TT 继续产生 2500-3000 亿元年度 EBITA (稳定现金锚)，CIG 成为 500-800 亿美元收入的云和 AI 业务，EBIT 利润率 20-25%，AIDC 在 5-6 个国际走廊达到盈利规模，菜鸟作为公开公司交易于 400-600 亿美元市值。合并实体可能值 7000-9000 亿美元 — 今天市值的三倍。如果阿里巴巴未执行，AI 资本开支变成搁浅资产，平台特许经营在抖音/拼多多压力下缓慢侵蚀，公司变成被回购和股息收益率支撑的价值陷阱。两种未来都在熊/基准/牛分布内；未来五年将决定我们走向哪一个。
This is why equity research, when done well, is not a forecasting exercise; it is a probabilistic exercise. We are not saying "Alibaba will be worth $226"; we are saying "the probability-weighted distribution of outcomes points to a central estimate of $226, with an asymmetric upside, supported by a balance sheet that limits downside." That is the disciplined statement an investor can act on, and it is the synthesis of every chapter on this site. 这就是为什么股权研究做得好时，不是预测练习；它是概率练习。我们不是在说"阿里巴巴会值 226 美元"；我们是在说"结果的概率加权分布指向 226 美元的中心估计，具有不对称的上行空间，由限制下行的资产负债表支撑。"那是投资者可以行动的有纪律陈述，也是本网站每个章节的综合。
In the tech industry, between 5% and 25% of total compensation is paid in stock — restricted stock units (RSUs), performance shares, options. The accounting is non-cash, but the economic effect on existing shareholders is unmistakably real: every share issued to an employee is a share that didn't go to existing owners. Modelling SBC carefully is one of the most common ways junior analysts get the long-run intrinsic value wrong. 在科技行业，总薪酬的 5% 到 25% 以股票支付 — 限制性股票单位 (RSU)、绩效股、期权。会计上是非现金，但对现有股东的经济影响明白无误地真实：每发给员工一股，就是没发给现有股东的一股。仔细建模 SBC 是初级分析师把长期内在价值搞错最常见的方式之一。
A typical engineer at Alibaba is offered, on hire, a 4-year RSU grant — say, $200,000 of stock vesting 25% per year. The accountants take the fair value of those shares at the grant date and expense it ratably over the vesting period: $50,000 per year for four years. No cash leaves the company. But four years later, the engineer owns shares that existing shareholders' shares didn't grow into. The dilution is real; the expense is real; the cash flow is unchanged. 阿里巴巴典型工程师在入职时被授予 4 年期 RSU — 比如 20 万美元股票，每年归属 25%。会计师将那些股份在授予日的公允价值费用化，按比例分摊到归属期：四年每年 5 万美元。没有现金离开公司。但四年后，工程师拥有现有股东股份未增长成的股票。稀释是真实的；费用是真实的；现金流不变。
Practitioners split into two camps on SBC treatment. Camp 1: SBC is an expense, full stop. Strip it out of "adjusted" metrics at your peril; "adjusted EBITDA ex-SBC" is a misleading number that flatters equity-heavy companies. Camp 2: SBC is a recruiting cost paid in equity rather than cash; the cash flow statement is correct to add it back; valuation should reflect the cash a business genuinely throws off. Both are defensible; the only indefensible position is to use one framework in one part of the analysis and the other in another part. Pick one and apply consistently. 从业者在 SBC 处理上分成两个阵营。阵营 1：SBC 是费用，句号。把它从"调整后"指标中剔除有风险；"调整后 EBITDA 剔除 SBC"是美化权益重型公司的误导性数字。阵营 2：SBC 是以权益而非现金支付的招聘成本；现金流量表把它加回是正确的；估值应反映企业真正产生的现金。两者都可辩护；唯一不可辩护的立场是在分析的一部分使用一个框架、在另一部分使用另一个。选一个并一致应用。
A company that issues SBC continually and buys back stock continually has a share count that depends on the netting of the two flows. For Alibaba, SBC issued in FY25 was approximately ¥20bn / $2.8bn — equivalent to about 26 million ADRs at the current price. Buybacks executed were $15bn — equivalent to about 127 million ADRs. Net retirement: roughly 100 million ADRs, or 4.2% of the float. The net share count change is the right metric for tracking real dilution; never look at SBC or buyback in isolation. 持续发行 SBC 且持续回购股票的公司，其股本数取决于两个流的净额。对阿里巴巴，FY25 发行的 SBC 约 200 亿元/28 亿美元 — 相当于当前价约 2600 万股 ADR。回购执行 150 亿美元 — 相当于约 1.27 亿股 ADR。净注销：约 1 亿股 ADR，即流通的 4.2%。净股本变化是追踪真实稀释的正确指标；永远不要孤立地看 SBC 或回购。
A company that pays 15% of revenue in SBC is implicitly betting that its stock price will hold up over the vesting period. If the stock falls 50%, the company has to either issue 2× as many shares to deliver the same dollar value to employees (massive dilution) or accept employee turnover (operational risk). This is why SBC-heavy companies amplify both the upside and downside of any thesis on the stock. Alibaba's SBC has fallen from a peak of 9% of revenue in FY15 to 2% in FY25 — a structural de-risking of the equity story that the market has under-appreciated. 支付 15% 收入作为 SBC 的公司，隐含地押注其股价在归属期内保持。如果股价跌 50%，公司必须要么发行 2 倍数量的股份来向员工交付相同的美元价值 (巨大稀释)，要么接受员工流失 (运营风险)。这就是为什么高 SBC 公司放大股票任何论点的上行与下行。阿里巴巴的 SBC 已从 FY15 的 9% 收入峰值降到 FY25 的 2% — 一种市场低估的股权故事结构性去风险化。
If you've made it this far through the site, you've earned a peek behind the curtain. What does an actual day at a sell-side equity research desk look like — the meetings, the inbox, the cold reads, the late-night model edits? 如果您已读到这里，您赢得了幕后一瞥。卖方股权研究台的真实一天看起来是什么样的 — 会议、邮箱、冷读、深夜模型编辑？
The analyst wakes up and reaches for the phone before her feet hit the floor. Overnight: Alibaba's Q3 print landed at 7 p.m. ET. Bloomberg consensus expectations have already been published. Buy-side clients in Asia have been emailing for the past five hours asking for a flash read. The first job is to type a 300-word "first look" note that goes to the trading desk and to the sales force by 6 a.m. The note has to be right enough to be useful but fast enough to be timely. 分析师醒来，脚还没落地就伸手拿手机。一夜之间：阿里巴巴 Q3 业绩美东时间晚 7 点公布。彭博一致预期已经发布。亚洲的买方客户过去 5 小时一直在邮件询问快速解读。第一项工作是打 300 字的"第一观察"笔记，要在早上 6 点前到交易台和销售人员手中。笔记必须足够正确以有用，足够快以及时。
All analysts gather (in person or by Zoom) for a 30-minute morning call. Each presents the most important overnight development on the names they cover. Order is by sector seniority and by news intensity. The analyst gets 2-3 minutes to summarize the Alibaba print and her revised thinking. Sales people listen and decide which clients to call first; trading desks listen and decide which names to expect flow in. 所有分析师聚集 (现场或 Zoom) 开 30 分钟早会。每人汇报覆盖股票最重要的隔夜进展。顺序按行业资历和新闻强度。分析师有 2-3 分钟总结阿里巴巴业绩和她修订的想法。销售人员听并决定先打哪些客户电话；交易台听并决定预期哪些股票有资金流。
The analyst opens the live Alibaba model. The print shows a 2% revenue beat (driven by Cloud), a 50 bps EBIT margin miss (driven by AIDC subsidy), and a meaningful capex guide for FY26 (¥125bn vs. consensus ¥110bn). Each input flows into the model: revenue updated, margin assumptions revised, capex line increased. The DCF target moves from $226 to $231 — small but directional. A revised "post-print" note goes out at 10 a.m. with the new model snapshot. 分析师打开动态的阿里巴巴模型。业绩显示收入超预期 2% (由云驱动)、EBIT 利润率不及预期 50 个基点 (由 AIDC 补贴驱动)、以及一项对 FY26 有意义的资本开支指引 (1250 亿元 vs. 一致预期 1100 亿元)。每个输入流入模型：收入更新、利润率假设修订、资本开支行增加。DCF 目标从 226 美元移到 231 美元 — 小但有方向性。修订的"业绩后"笔记 10 点带新模型快照发送。
Alibaba's CFO and head of IR host a 90-minute analyst call. The analyst is one of about 25 sell-side analysts on the call. She gets one question, asked in order of seniority. Her question is pre-planned: a specific clarification on the AIDC subsidy trajectory that, if answered, will resolve the largest single uncertainty in her model. Management's answer is partial; she takes notes, plans a follow-up at the after-call group meeting with IR scheduled for 11 a.m. local Beijing time. 阿里巴巴 CFO 和 IR 主管主持 90 分钟分析师电话会。分析师是约 25 位卖方分析师之一。她按资历顺序问一个问题。她的问题是预先计划的：关于 AIDC 补贴轨迹的具体澄清，如得到回答，会解决她模型中最大的单一不确定性。管理层的回答是部分的；她做笔记，计划在计划 11 点 (北京时间) IR 后续小组会议跟进。
The next 90 minutes are spent on the phone with five buy-side portfolio managers, in sequence. Each call is 15-20 minutes. The PMs ask sharper, more pointed questions than the company call — "If take-rate compresses 50 bps in FY27, where does your target go?" The analyst's job is to communicate her view clearly, defend the model, and listen to feedback. A few of the PMs will mention they're trimming or adding to the position; the analyst quietly notes this — buy-side positioning is a market signal. 接下来 90 分钟用电话与五位买方投资组合经理依次通话。每次通话 15-20 分钟。PM 们问比公司电话会更尖锐、更尖刻的问题 — "如果货币化率在 FY27 压缩 50 个基点，您的目标会去哪里？" 分析师的工作是清晰传达她的观点、为模型辩护、聆听反馈。一些 PM 会提到他们正在减仓或加仓；分析师默默记下 — 买方仓位是市场信号。
分析师转向更广泛的研究：阅读 PDD、京东、腾讯的隔夜报告，看是否有跨股票主题正在出现。她阅读 SAMR 关于电商反垄断的中文新闻稿，因为 Alibaba 是关键当事方。她为下周的客户演示拉取一些数据点 (PDD 的 Temu GMV、Tencent 的 WeChat Pay 量) — 一份将这五家股票放在一起的横截面观点。 The analyst turns to broader research: reading overnight notes on PDD, JD, Tencent, checking whether cross-stock themes are emerging. She reads a Chinese-language SAMR press release on e-commerce antitrust, because Alibaba is a key party. She pulls a few data points (PDD's Temu GMV, Tencent's WeChat Pay volume) for next week's client presentation — a cross-cutting view that puts the five stocks in a single frame.
The full post-print research note — 12 pages, with updated charts, updated forecast tables, updated price target, refined investment thesis, refined risks — has to clear by 5 p.m. for distribution. The analyst writes it; the editor cleans it; the compliance officer reviews the disclosures; the head of research signs off. Then it goes out to thousands of buy-side accounts simultaneously. 完整的业绩后研究笔记 — 12 页，带更新图表、更新预测表、更新目标价、精炼的投资论点、精炼的风险 — 必须在下午 5 点前清空以分发。分析师写它；编辑清理它；合规官员审查披露；研究主管签字。然后同时发送到数千个买方账户。
A 60-minute meeting with a small-account hedge fund that wants the analyst's detailed views on Cloud. Then 45 minutes with a sovereign wealth fund. Then 30 minutes with the firm's portfolio strategist, who wants the analyst's view on China weighting in the bank's recommended model portfolio. 与一家小账户对冲基金 60 分钟会议，他们想要分析师对云的详细看法。然后 45 分钟与一家主权财富基金。然后 30 分钟与公司的投资组合策略师，他想要分析师对银行推荐模型组合中国权重的看法。
Dinner at the desk. The analyst finishes integrating the day's new data points into the model and updates the scenario tables. Sets the agenda for tomorrow: a call with Alibaba's IR to follow up on the AIDC subsidy question, a call with a supply-chain source in Shanghai about Singles Day prep, a call with the sales force about which themes to push to clients tomorrow morning. 在桌前吃晚餐。分析师完成把当日新数据点整合到模型中并更新情景表。设定明天议程：与阿里巴巴 IR 通话跟进 AIDC 补贴问题、与上海一位供应链消息人士通话了解双 11 备货情况、与销售人员通话明早向客户推哪些主题。
The reader of a sell-side research note is consuming the compressed output of this kind of day. Behind every chart on this site, behind every paragraph of the Alibaba investment thesis, is hundreds of hours of model-building, conversation-having, news-reading, source-cultivating, and writing. The toolkit you have learned to read in this guide is also, in reverse, the toolkit you would need to produce. Read research with the awareness that the work is harder than it looks — and the discipline of doing your own work, even imperfectly, is more valuable than passively consuming what others have done. 读卖方研究笔记的读者，正在消费这种一天的压缩输出。本网站上每一张图后面、阿里巴巴投资论点的每一段后面，是数百小时的模型搭建、对话进行、新闻阅读、消息源培养、写作。您在本指南中学会阅读的工具箱，反过来也是您生产所需的工具箱。阅读研究时要意识到工作比看起来更难 — 而做您自己工作的纪律，即便不完美，比被动消费别人做过的更有价值。
Every domain hides behind its vocabulary. Memorize this list and you can sit in a sell-side morning meeting without nodding politely at words you don't understand. 每个领域都藏在它的词汇背后。记住这份清单，您就能坐在卖方早会里，不会对听不懂的词点头敷衍。