美股跌到位了吗？· 7.20-7.24 本周回顾

📒 本周回顾：哪些对了，哪些错了

一句话：发过的判断，对了回来对，错了也回来认。

先说错的。

周三那封《财报开牌前，聪明钱在买翅膀》里，我们押了两条：谷歌财报之夜会比特斯拉动得大；两只都不会冲出市场赛前划好的波动带。

第一条，错。实际是特斯拉动得大。

第二条，财报当晚看起来对了：两只都乖乖待在带内。但第二天，特斯拉 −14.5%，谷歌 −7.1%，到周五收盘双双跌穿带子下沿（谷歌收 319.74，带下沿 329.17；特斯拉收 313.03，下沿 359.85）。也错。

两个错其实是同一个错：财报的账单，是第二天才付的。我们把当晚的带子，当成了整段行情的围栏。

再说对的。三条，全是同一类。

英伟达那笔约 3 亿美元的大单：我们周三当天判断是老仓搬家、不是新赌注（判断台账 07-22 当天立案，先写判断，后看开奖），并且把第二天该看到的两个数字提前写下来：旧仓应减约 8 万张，新仓应增约 8.8 万张。第二天开奖：−78471 和 +88009。两个数都对上了。

三巨头同一天各冒出 1.1 万张的五条腿：我们判断是同一家机构的程序单。第二天五条腿的持仓齐刷刷各增约 1.1 万张。

财报周被炒到天价的短期期权，开盘后必然大幅泄气：最狠的一只，一夜从 100 上下泄到 41。

本周回顾：猜价格往哪走，0 对 2；看懂这笔钱在干什么，3 对 3。我们不猜牌，我们看筹码。

🙋 问得最多的两个问题

一句话：我不预测价格，但钱的每个动作都记了账。

跌到位了吗？

不知道，也不猜。但这一周有三笔钱的动作，值得你知道。

第一笔：保险没涨价。 你的车要是出了事故，第二年续保，保费多半要涨：保险公司认定这条路以后还会出事，才敢多收钱。这一周市场天天挨打，短期的怕上蹿下跳，来回跑了 12.4%；可给未来一年买保护的价钱，全周只动了 1.9%。出了事故，保费却没涨。 卖保险的是全市场最会算账的一批人：他们没提价，意思是在他们的账本上，这一周是一场事故，不是路况变了。你纠结割肉还是加仓的那一刻，这个数不告诉你该怎么做；它只告诉你：天天跟恐惧做买卖的那批人，这周没改价。

第二笔：有人押会动，不押方向。 有人在 8 月底到期的合约上花了五六百万美元，同时买涨和买跌，两边距离现价对称。这笔单 99% 留仓过夜，是真仓；到期日正好罩住下周的美联储会议和整个财报季。

第三笔：极端的险有人上。 周一有人花一万二美元，买了 3000 张 SPY 跌 86% 才赔的保险，4 美分一张。第二天几乎全部留仓。

三笔连起来读：主流的钱没在慌，会动的钱在加价，极端的险一直有人买。 这不是预测，是记账。

存储还跌吗？

先把一个反直觉的事实放正：周五被血洗的存储，其实是本周最强的板块。 美光这一周 +8.5%（周五 −7%），闪迪 +6%（周五 −10.8%）。周五吐回去的，少于这一周涨出来的。

然后看钱怎么动的：

• 周三收盘前最后半小时，有人花约 1 亿美元买了美光的看涨，这笔单赚钱的门槛是月底站上 996。周五美光收 920.95，低于那道门槛 75 美元。下周五开奖，赢输我们都晒。
• 周五大跌当天收盘前，又有人花 960 万美元，买美光回到 1000 的票。跌下来的当天，就有钱进场赌反弹。
• 但想上车的先看票价：买它们期权的成本，闪迪排在过去一年的 96 分、美光 88 分（100 分就是这一年最贵那天）。便宜的是股票，不是上车的成本。
• 长期的票价没慌：闪迪一年多之后到期的期权价格，一周从 118 降到 114，还在轻微降温。

钱的表态翻译成一句：短线在对赌，长线没改主意，入场费仍在最贵档。

🗓 下周看点：钱已经表态的两场

一句话：我不猜结果，只报市场已经付钱买下的预期。

第一场：Meta 财报，周三盘后，同一晚撞上美联储决议。一晚两颗雷。

钱的表态：市场给它划的围栏 ±8.01%（558 到 655 美元），四家巨头里最宽，是苹果（±3.9%）的两倍。翻译：市场为 Meta 准备了全场最大的意外空间。微软同晚开牌，围栏 ±6.62%。

周四苹果、亚马逊接棒。亚马逊的账本歪着：押它涨的持仓是押跌的 3.4 倍。每张合约都有买方和卖方，这个倍数不代表谁会赢，只代表如果结果朝反方向走，被打脸的人多得多。

第二场：存储的两次终审。

周五 07-31：美光那张 1 亿美元大单开奖，996 见分晓。
08-05：闪迪财报。到时候它 96 分的票价，就要接受检验。

我们的挂账（美光 07-31、诺基亚 08-28）到期都会回来对，赢输都晒。

这封信不告诉你该怎么做。它只保证一件事：我们说过的每句话，都会回来对账。

📒 The scorecard: what we got right, what we got wrong

One line: every call we publish comes back for settlement — wins and losses alike.

The misses first.

On Wednesday we made two calls: Google would move more than Tesla on earnings night, and neither would break out of the range the market had priced in.

First call: wrong. Tesla moved more.

Second call looked right that night — both stayed inside their bands. Then came the next day: Tesla −14.5%, Google −7.1%, and by Friday's close both had fallen through the bottom of their bands (Google closed 319.74 vs. a floor of 329.17; Tesla 313.03 vs. 359.85). Wrong again.

Both misses are the same miss: earnings bills come due the next day. We mistook one night's range for the whole ride's fence.

And one piece of homework we failed: a call required checking open interest before Friday's open. We missed the window; the contracts expired and settled to zero. Permanently unverifiable. Logged as cannot-judge. No excuses.

Now the hits. Three, all the same kind.

That roughly $300M Nvidia print: we called it a position moving house, not a new bet — logged the same day, judgment first, verdict later — and wrote down the two numbers the next morning should show: old position down ~80k contracts, new one up ~88k. The verdict: −78,471 and +88,009. Both matched.

Five legs of ~11.2k contracts each, across three mega caps, in one day: we called it one institution's programmed trade. Next morning, all five legs' open interest rose in lockstep.

Earnings-week options priced at absurd levels must deflate at the open: the worst one went from around 100 to 41 overnight.

The scorecard in one line: guessing where prices go, 0 for 2. Reading what the money is doing, 3 for 3. We don't guess cards. We count chips.

One more confession: we also cited a wrong closing price this week (an after-hours print instead of the official close). The ledger now carries a dated correction; the verdict stands. We publish our bookkeeping errors too — this ledger is worth something precisely because it can't be quietly rewritten.

🙋 The two questions we get asked most

One line: we don't predict prices. We keep books on what the money does.

Done falling?

We don't know, and we won't guess. But three money moves this week are worth knowing.

First: insurance didn't get more expensive. Crash your car and your premium jumps at renewal — the insurer has decided your road is dangerous. The market took hits all week; short-term fear swung 12.4% top to bottom. Yet the price of one-year protection moved just 1.9%. An accident, and no premium hike. The people selling insurance are the best accountants in the market: no reprice means that in their books, this week was a crash, not a change in the road. The moment you're debating whether to cut or add, this number won't tell you what to do — it tells you the people who trade fear for a living didn't change their price this week.

Second: someone is betting on movement, not direction. Someone spent $5–6M on end-of-August contracts, buying calls and puts in equal size, symmetric around the current price. 99% held overnight — real positions. The expiry covers next week's Fed meeting and the rest of earnings season.

Third: tail risk still has buyers. On Monday someone paid $12,000 for 3,000 contracts of SPY protection that only pays if the market falls 86%. Four cents apiece. Nearly all of it held the next day.

Read together: mainstream money isn't panicking, movement money is paying up, and crash insurance keeps finding buyers. That's not a forecast. That's bookkeeping.

Is memory done falling?

First, a fact that cuts against Friday's tape: the stocks bleeding on Friday were the strongest group of the week. Micron finished the week +8.5% (Friday −7%); SanDisk +6% (Friday −10.8%). Friday gave back less than the week added.

Now the money:

• In the last half hour of Wednesday, someone spent about $100M on Micron calls that only pay if it stands above 996 by month-end. Friday's close: 920.95, 75 dollars below the line. Verdict next Friday; we post it win or lose.
• During Friday's selloff, before the close, another $9.6M bought tickets on Micron returning to 1,000. The day it fell, money showed up to bet on the bounce.
• But check the ticket price before boarding: options on SanDisk cost more than on 96% of days in the past year; Micron, 88%. The stock got cheaper. Boarding didn't.
• Long-dated prices aren't panicking: SanDisk's year-plus options drifted from 118 to 114 over the week. Cooling, slightly.

The money's statement in one line: the short term is a knife fight, the long term hasn't changed its mind, and the entry fee is still near the top of the range.

🗓 Next week: where the money has already voted

One line: we don't guess outcomes. We report the expectations the market has already paid for.

Act one: Meta reports Wednesday after the close — the same night as the Fed decision. Two grenades, one evening.

The market's fence for Meta: ±8.01% ($558–$655), the widest of the four mega caps reporting next week — twice Apple's (±3.9%). Translation: the market has budgeted Meta the most room for surprise. Microsoft reports the same night, fenced at ±6.62%.

Thursday, Apple and Amazon take the stage. Amazon's book is lopsided: bets on up outnumber bets on down 3.4 to 1. Every contract has a buyer and a seller — the ratio doesn't say who wins, only that if it goes the other way, far more faces get slapped.

Act two: two verdicts on memory.

Friday 07-31: that $100M Micron bet settles. 996 is the line.
08-05: SanDisk reports. Its 96th-percentile ticket price goes on trial.

Open cases (Micron 07-31, Nokia 08-28) all come back for settlement — posted win or lose.

This letter won't tell you what to do. It guarantees one thing: everything we say comes back for settlement.