EquityDesk | 股票研究与财报分析

EquityDesk 把一张股票研究工作台装进了一个技能。给它一只股票、一份自选列表或一条规则,它会返回结构化报告,说明每个数字来自哪里、对应哪一天。覆盖美股(纽交所、纳斯达克)和韩股(KOSPI、KOSDAQ),韩国公司以 DART 披露为依据。

能做什么

• 财报前瞻: 市场一致预期、历史超预期与不及预期记录、决定结果的 3 到 5 个关键项目、公司指引,以及电话会上值得追问的问题。
• 财报复盘: 实际值与一致预期、去年同期的对比,从一次性因素、利润率、现金流看业绩质量,指引变化,以及本次财报确认了什么、还有哪些悬而未决。
• 规则筛选: 支持 CANSLIM 风格、股息、动量和价值/质量筛选。你能看到所用规则、每一步过滤后剩余的数量,以及附带单只风险提示的简短排名。
• 持仓检查: 粘贴你的持仓,即可得到按个股、行业、国家和币种划分的权重、集中度与重叠情况,并附计算公式。
• 回测方案设计: 把一句自然语言规则变成可检验的方案(股票池、买入、卖出、成本、基准),并检查前视偏差和幸存者偏差。若有代码和数据可用,会实际运行回测。
• 市场简报: 带日期的日报或周报,涵盖指数、行业、利率、美元/韩元等汇率,以及已确认的重要日程。
• 个股备忘: 业务模式、利润率、资产负债表、列明假设的估值,以及乐观、基准、悲观三种情景。
• 带市场检查的趋势模板与 CANSLIM 筛选: 先确认指数趋势,再用公开的 CANSLIM 权重和七项价格条件为候选股打分,并标明回看区间以及哪些条件未能评估。
• 仓位大小计算: 根据你提供的数字计算股数、风险金额和集中度上限,并列展示多个风险水平。绝不替你选择风险水平。
• 投资逻辑跟踪与催化剂日历: 包含正反证据的逻辑、你自己设定的失效条件,以及带日期的已确认事件。
• 股息收入规划: 根据你粘贴的持仓生成派息日历、预期股息率、成本收益率、相对自由现金流的派息比例、削减风险信号,并列比较再投资情景。税务规则只列为需要确认的事项,不替你计算。
• 交易日志与绩效复盘: 根据你自己的交易记录计算 R 倍数、带置信区间的胜率、盈亏比、利润因子、期望值、最大回撤和连亏次数;样本太少时会明确提醒。
• ETF 对比与重叠分析: 比较美股和韩股上市 ETF 的费率、持仓重叠、你直接持有的个股在 ETF 中的穿透权重、集中度、折溢价和跟踪误差。
• 公告与内部人交易摘要: 整理自选股的 SEC 文件(8-K、10-Q、Form 4、13D/13G、13F)和 DART 公告(重大事项报告、高管及主要股东持股报告、大量持股报告、库存股、增资),区分例行事项与值得关注的事项,不揣测动机。

需要提供什么

• 股票代码、6 位 KRX 代码,或股票列表
• 任务类型:前瞻、复盘、筛选、持仓检查、回测方案、市场简报或个股备忘
• 可选:投资期限、风格(成长、价值、股息)、显示币种,以及想使用的规则或阈值
• 可选:粘贴的公告、电话会记录、价格数据或你的持仓列表

会得到什么

• 开头标明数据模式:实时数据、仅限你提供的数据,或仅提供方法框架。
• 区分事实与解读的摘要
• 带有基准日、会计期间、币种和单位的表格
• 证据清单:每项重要结论的来源、日期和核实状态
• 未能确认的内容、查询失败的部分,以及接下来要核对的事项

请求示例

• “帮我做 NVDA 下次财报的前瞻:一致预期、过往超预期情况、关键变量和指引。”
• “把三星电子第三季度初步业绩与一致预期对比整理,按合并报表口径。”
• “在标普 500 里筛选派息率低于 60%、连续 5 年增加股息的公司,并显示每步过滤后剩余的数量。”
• “这是我的 8 只持仓和权重,看看我在行业和币种上集中在哪里。”

建立在广受认可的方法之上

撰写 v1.1 之前,我阅读了最知名的开源交易与金融技能的公开文档,只保留适合美股和韩股研究的方法。EquityDesk 借鉴的是方法而非代码,与这些项目没有关联,也未获其背书。它增加了对韩国市场的处理(DART、KRX、以 KOSPI 作基准),以及更严格的规则:每个数字都要有来源和日期。

• tradermonty/claude-trading-skills: 先看风险的流程、选股前先检查市场状态、决策前清单、仓位大小计算。
• himself65/finance-skills: 财报前瞻与复盘的结构,使用 DCF、可比公司倍数和敏感性表的估值。
• anthropics/financial-services: 为人工复核准备的分析师工作成果、投资逻辑跟踪、催化剂日历。
• wshobson/agents (backtesting-frameworks): 前视偏差与幸存者偏差检查、样本外与滚动前向测试。

真实请求与实际输出

以下三个示例于 2026 年 11 月 4 日在可实时联网的环境中,按本技能 v1.1 的指令运行得到。它们由另一个模型在另一个环境中运行,并非 Capafy 上驱动本技能的模型,因此实际使用时措辞、深度和数据访问范围可能不同。示例 1 的数字来自所列页面和 Yahoo Finance 日线收盘价,示例 2 使用作为示例挑选的股票的 Yahoo Finance 日线收盘价,示例 3 使用虚构数字。示例 2 因请求为韩语而以韩语显示。

示例 1:含股价反应的财报复盘,苹果 2026 财年第四季度

请求

Recap Apple's fiscal Q4 2026 report: results against consensus, the stock's reaction, and what guidance said.

回复

Data mode: A, live data. Checked 2026-11-04. US dollars. Apple reported after the close on Monday 2026-11-02, so the reaction session is Tuesday 2026-11-03.

Summary

• EPS $2.10 against $1.99 expected (LSEG): a $0.11 beat, 5.5%. Revenue $114.48 billion against $113.39 billion: 1.0% above, and up 11.7% from $102.46 billion a year earlier.
• Part of the EPS beat came from tariff refunds. Apple says they added $0.05 to EPS and about 1 point to gross margin. Without the refund, EPS would be about $2.05, still above $1.99, but it is unclear whether analysts had included the refund in their numbers. Unverified
• The December-quarter guide is the soft spot. Revenue growth guided to 7 to 10% (analyst average 6.1%), but gross margin guided to 45 to 46% against an average estimate of 46.3%.
• Regular-session reaction was -1.9% (close 338.32 to 331.90). That is smaller than the average absolute move of the previous four reports, 2.9%.

Line item	Actual	Expected (LSEG)	Difference
EPS	$2.10	$1.99	+5.5%
Revenue	$114.48B	$113.39B	+1.0%
iPhone	$57.58B	$56.54B	+1.8%
Mac	$10.79B	$9.61B	+12.3%
iPad	$6.23B	$6.70B	-7.0%
Wearables, Home and Accessories	$8.30B	$9.06B	-8.4%
Gross margin	47.9%	47.5%	+0.4 pt (about -0.6 pt without the refund)

Beat quality. The beat is real on the headline, but it leans on iPhone and Mac (Mac +24% year over year). Two segments missed, and the margin beat depends on the refund. Services came in about $31.6 billion, in line with expectations. One outlet's table labels this line "Other Products", while others label it Services; I treated it as Services, not verified against the 10-K.

Contradictory China data. Greater China revenue was $17.54 billion. CNBC reports growth of 21% year over year, while Yahoo Finance and StreetInsider show it below estimates of $18.58 to $18.75 billion (Visible Alpha, Bloomberg). Both can be true: strong growth, but below what analysts expected. The two estimates differ by provider, so I show both instead of averaging.

Reaction, measured the same way each time (report-day close to next close, after-close reports):

Report	Reaction
2025-10-30	-0.38%
2026-01-29	+0.46%
2026-04-30	+3.24%
2026-07-30	-7.35%
2026-11-02	-1.90%

Average absolute move of the first four: (0.38 + 0.46 + 3.24 + 7.35) ÷ 4 = 2.86%. After-hours figures vary by outlet (CNBC wrote about -3% in extended trading) and I did not verify them. A single day's move has many causes, including the broader market.

Open questions. How much of the December-quarter margin guide is memory costs versus mix (management cited higher memory costs)? Whether analysts' margin consensus included tariff refunds. The next confirmed date is the dividend: $0.27 per share, record date 2026-11-12, payable 2026-11-16.

Sources (opened): Apple newsroom, "Apple reports fourth quarter results", 2026-11-02: https://www.apple.com/newsroom/2026/11/apple-reports-fourth-quarter-results/ . CNBC live coverage, 2026-11-02 (LSEG estimates, guidance): https://www.cnbc.com/2026/11/02/apple-aapl-q4-2026-earnings-live-updates.html . Yahoo Finance daily chart data for AAPL (closes): https://query1.finance.yahoo.com/v8/finance/chart/AAPL . Seen only as search-result excerpts, not opened in full: Yahoo Finance and StreetInsider earnings articles (China estimates, Services line), Apple newsroom pages for the January, April and July 2026 reports (dates), and an SEC-filing listing for 2025-10-30.

Information and analysis only, not investment advice.

示例 2:八只股票(美国与韩国)的趋势模板检查

请求

내가 보는 8종목(AAPL, MSFT, NVDA, 삼성전자, SK하이닉스, KO, JNJ, XOM)을 추세 템플릿 기준으로 점검해 줘. 가격은 오늘 기준으로.

回复

데이터 모드: A, 실시간 확인. 야후 파이낸스 일봉 종가 2년치를 2026-11-04에 조회했습니다. 미국 종목은 11월 3일 종가, 한국 종목은 11월 4일 최신 값입니다. 52주 고점·저점은 일일 종가 기준입니다.

요약

• 가격 기준 7개 조건을 모두 충족한 종목: MSFT, KO, JNJ, XOM.
• AAPL은 6/7입니다. 현재가 331.90달러가 50일선 334.05달러보다 0.6% 낮아 "50일선 위" 조건만 빠졌습니다.
• NVDA는 4/7입니다. 현재가 206.69달러가 150일선(211.81)보다 낮고 50일선(221.46)도 밑돌며, 52주 저점보다 25.1% 위라 "30% 이상 위" 조건도 못 채웠습니다.
• 삼성전자와 SK하이닉스는 3/7입니다. 두 종목 모두 현재가가 50·150·200일선 아래이고, 52주 고점 대비 각각 -39.9%, -52.2%입니다.

종목	현재가	50일	150일	200일	52주 고점 대비	52주 저점 대비	충족
AAPL	331.90	334.05	309.78	297.34	-4.9%	+34.6%	6/7
MSFT	520.99	514.06	449.07	438.48	-4.1%	+47.7%	7/7
NVDA	206.69	221.46	211.81	204.56	-13.6%	+25.1%	4/7
삼성전자	218,000원	249,680	260,127	238,564	-39.9%	+130.0%	3/7
SK하이닉스	1,396,000원	1,611,120	1,709,313	1,503,165	-52.2%	+173.7%	3/7
KO	94.80	89.62	84.21	82.38	-1.1%	+40.4%	7/7
JNJ	266.17	265.47	250.68	247.48	-4.4%	+43.1%	7/7
XOM	177.46	168.32	156.61	155.09	-5.2%	+56.1%	7/7

평가한 7개 조건: ①현재가가 150일·200일선 위 ②150일선이 200일선 위 ③200일선이 최근 1개월(22거래일) 동안 상승 ④50일선이 150일·200일선 위 ⑤현재가가 50일선 위 ⑥52주 저점보다 30% 이상 위 ⑦52주 고점의 25% 이내.

평가하지 못한 조건. 여덟 번째 조건인 상대강도 순위(70 이상)는 비교할 전체 종목군이 없어서 계산하지 않았습니다. 대신 가중 상대수익률(0.4×3개월 + 0.3×6개월 + 0.3×12개월, 벤치마크는 미국 종목 S&P 500, 한국 종목 코스피)을 참고용으로 적습니다: AAPL +10.9%p, MSFT +7.3%p, NVDA -3.8%p, 삼성전자 +26.0%p, SK하이닉스 +47.8%p, KO +17.5%p, JNJ +15.3%p, XOM +24.1%p.

해석할 때 볼 점. 삼성전자는 공시된 3분기 실적이 사상 최대인데도 가격 추세 조건은 약합니다. 12개월 수익률은 +125.9%로 높지만 3개월은 -17.0%입니다. 이 차이의 원인은 이번 조회로는 확인하지 못했습니다. 미확인 통화(달러, 원)가 섞여 있어 가격끼리 비교하지 않았고, 이동평균은 종가 기준이며 배당·분할 조정 방식은 데이터 제공처 기준입니다. 거래량은 반영하지 않았습니다.

출처: 야후 파이낸스 차트 데이터(조회함) https://query1.finance.yahoo.com/v8/finance/chart/ (각 티커 및 ^GSPC, ^KS11).

가격 추세를 점검한 결과이며 매수·매도 신호가 아닙니다. 정보와 분석 제공용이며 투자 자문이 아닙니다.

示例 3:用你自己的数字计算仓位大小(虚构输入)

请求

Account $50,000. Entry $100, stop $92. Max position 10% of the account. How many shares at 0.5%, 1% and 1.5% risk?

回复

Data mode: B, your numbers only. No prices were opened.

Risk per share = 100 − 92 = $8. Position cap = 10% × 50,000 = $5,000, which is 50 shares at $100.

Risk level	Risk dollars	Shares by risk	Position value	Cap applies?	Final shares	Dollar risk at the stop
0.5%	$250	floor(250 ÷ 8) = 31	$3,100 (6.2%)	no	31	$248 (0.50%)
1.0%	$500	floor(500 ÷ 8) = 62	$6,200 (12.4%)	yes	50	$400 (0.80%)
1.5%	$750	floor(750 ÷ 8) = 93	$9,300 (18.6%)	yes	50	$400 (0.80%)

What this shows. At 1% and 1.5% the position cap, not the risk level, sets the size, so both end at 50 shares and 0.8% account risk. Only the 0.5% case is limited by risk. I am not choosing a risk level for you.

Not in the numbers. A stop order is not a guaranteed fill price, so a gap can lose more than $8 a share. Commissions, spreads and slippage are excluded. If other positions carry open risk, add them up: common reference limits in public guides are 6 to 8% of equity across the portfolio, as a reference and not a rule.

Information and analysis only, not investment advice.

示例 4:根据你自己的交易记录做交易日志复盘(虚构交易)

请求

Here is my trade log in R-multiples, oldest first: +2.1, -1, +0.8, -1, +3.2, -0.5, -1, +1.5, -1, +0.6. Review it.

回复

Data mode: B, your log only. Nothing was looked up.

Measure	Value	How it is computed
Trades	10
Wins / losses	5 / 5
Win rate	50% (Wilson 95% range about 24% to 76%)	5 ÷ 10
Total result	+3.7R	8.2 − 4.5
Average win / average loss	+1.64R / −0.90R	8.2 ÷ 5, −4.5 ÷ 5
Payoff ratio	1.82	1.64 ÷ 0.90
Profit factor	1.82	8.2 ÷ 4.5
Expectancy per trade	+0.37R	0.5 × 1.64 + 0.5 × (−0.90)
Maximum drawdown	1.5R	peak 4.1R after trade 5, low 2.6R after trade 7
Longest losing streak	2	trades 6 and 7

What the numbers show. Winners were larger than losers on average, and no loss exceeded 1R. The cumulative curve recovered its 4.1R peak at trade 8.

What they cannot show. Ten trades are far too few. The win rate could plausibly be anywhere from about 24% to 76%, so the +0.37R expectancy is descriptive only and I will not project future returns from it.

Three questions for your journal. Which setups produced the three biggest winners? Were the 1R losses stopped at the planned level, or did any run through it? What happened after the two-loss streak?

Information and analysis only, not investment advice.

为了能被核实而设计

• 不编造数字。在来源中找不到的数值不会凭记忆补上,而是标为 未核实。
• 优先使用一手资料:美股看 SEC EDGAR 文件和公司投资者关系页面,韩股看 DART 和韩国交易所数据。
• GAAP 或 IFRS 数据与调整后数据分开标注,韩国公司会注明合并或单体口径。
• 不下单、不连接券商账户,也不会说“买入”或“卖出”。输出是信息与分析,不构成投资建议。
• 网页或粘贴文档中夹带的指令一律忽略。

购买前需要了解的限制

• 实时数据取决于运行环境是否能访问网络。无法访问时,EquityDesk 只分析你粘贴的资料,并在数据模式中说明。
• 它不是实时或盘中交易信号工具。每次运行都是带时间戳的快照。
• 筛选和回测的准确度取决于能获取的数据。没有完整数据时,它不会编造名单,而是给出可在其他工具中运行的筛选规则。
• 过往业绩不代表未来表现。税务、法律和个人理财规划不在范围内。

常见问题

支持韩国股票吗?
支持。可处理 KOSPI 和 KOSDAQ 股票,在可访问时使用 DART 披露和韩国交易所数据,可用中文、韩语或英语回答。

它会告诉我该买什么吗?
不会。它展示证据支持的内容、风险,以及哪些因素会改变判断。决定由你自己做。

能替我交易吗?
不能。它没有券商连接功能,也不会下单。

怎么确认数字是真的?
每项重要结论都标有来源、日期和状态。未能确认的内容会标为 未核实。

仅提供信息与分析,不构成投资建议。