# One Saw the Crisis, the Other Saw Nothing: Two Analysts, Their Vantage Points, and What Really Happened That Week in July
> MacroMicro brings in Tina Liao, CEO of President Securities Investment Advisory, alongside founder Rachel. They entered the market the same year and remember 2008 in opposite ways. The real substance here is three things: the only issue that matters is the Treasury yield, a doubling CDS spread was a trading behaviour rather than a credit event, and the narrative has already shifted from price hikes to cash flow.
Published: 2026-08-09
Locale: en
Tags: macromicro, podcast-notes, macro, federal-reserve, semiconductors, capex, risk-management
TL;DR: Whether an event matters isn't about how dramatic it looks — it's about whether it moves the one variable the decision-maker is pinned to. In this episode that variable is the US Treasury yield. And the sell-off that week in July had almost nothing to do with fundamentals; what changed was leverage and forced selling.
> *From the side, a whole range; head-on, a single peak.*
> *Far, near, high, low — no two views alike.*
> *I cannot tell the true shape of Lu Shan,*
> *only because I am standing inside the mountain.*
> —— Su Shi, "Written on the Wall of West Forest Temple" (1084)
Su Shi wrote this quatrain on a temple wall, and it isn't really about the mountain. It's about position. Seen from the side, the same mountain is a long ridge; seen head-on, it's a single peak. You can't make out its true shape — not because you aren't looking hard enough, but because you are standing inside it.
The most interesting stretch of this episode has exactly that shape: two analysts who entered the market the same year remember the same financial crisis in almost opposite ways.
## What this episode is about
This is MacroMicro's "Finance Friends" special, hosted by Roger, with two guests: Rachel, MacroMicro's founder, and Tina Liao, CEO of President Securities Investment Advisory. They started their careers less than a year apart, both somewhere between 2007 and 2008 — and went in completely different directions. One began in macro and later founded a data platform; the other began in electronics and worked her way up to running an advisory firm.
The tone is set in the first three minutes. The host introduces "two senior analysts," and Tina immediately objects to the word "senior" — she'd prefer "research girl." Later, when Roger asks her which AI or semiconductor name to buy, her response is to freeze on the spot.
Their reasons for entering the industry are worth keeping too. Rachel studied economics through her master's and loved it because it contains mathematics, psychology and even physics all at once; what actually decided her career was an evening watching television with her uncle, when a strategist (she recalls Cheng Shu-fen) was breaking down the economy on a finance channel, and her uncle said the job seemed like a fit for her. Tina's version is more worldly and more honest: her mother was a broker, so she wanted into the securities industry — "I wanted to make a lot of money." The original goal was to become a fund manager; she ended up an analyst almost by accident.
**Original episode**: MacroMicro Podcast, "From the Financial Crisis to the AI Wave: Senior Analysts on Reading the Market | Finance Friends Special," published 9 August 2026 (recorded 6 August per the episode), with Rachel (founder, MacroMicro) and Tina Liao (CEO, President Securities Investment Advisory).
## The notes I took
**People haven't gone numb to bad news — they've learned to separate real issues from fake ones, and this episode names only one real issue.** Rachel's argument is that the market's fading reaction to tariffs, the Iran conflict and fiscal deficits isn't habituation; it's recognition that most of these are one-off events. At the peak of tariff panic last year, MacroMicro pointed out that the numbers didn't add up: a 50–60% rate made no sense, and roughly 10–15% would already cover the deficit created by the big tax-and-spending bill. The actual crux is the Treasury yield — and whether you believe the President cares about it. She offers a checkable record: each time the yield pushed above 4.5%, something softened. April brought a sixty-day pause on striking Iran; June brought a signed memorandum of understanding; recently, with the yield at 4.6–4.7%, another sixty-day framing appeared. The test isn't how dramatic an event is. It's whether it touches the variable the decision-maker is pinned to.
**Tina comes at it from intent and lands on the same two indicators.** Her reading: the purpose of tariffs is capacity relocation — so even though the market has stopped caring about the tariff headline, capacity relocation is still grinding forward. The purpose of the Iran conflict is control over energy, and energy matters because of the AI build-out. Both threads converge on two things to watch: oil and Treasuries. She adds a neat validation — with each successive round of similar headlines, the move in oil gets smaller. That decay is itself evidence that the market is learning to sort signal from noise.
**They entered the same year; one saw the crisis, the other saw nothing.** Rachel joined in 2007 covering macro, and watched mortgage delinquency rates, housing inventory and new home sales all start to move — she saw a full cycle turn, and that's ultimately why she founded MacroMicro. Tina's version: the 2008 crisis barely registered, because she was covering solar and LEDs, two sectors that dipped briefly and recovered fast, then got stronger. She names this herself — the industry analyst's blind spot. You only look at your own patch, so you don't notice the whole landscape shifting; she says she still sees it in the analysts she manages today. What actually shook her wasn't 2008 but 2011, the European debt crisis and the US downgrade; she was more senior by then, took the hit, changed firms because of it, and started building macro coverage from that point on. **The point here isn't "macro matters more." It's that your vantage point determines which kind of disaster you are capable of seeing.**
**Warsh's line — watch the ball, not the referee — is worth more than any rate forecast.** July's meeting held rates 9–3, which itself signals internal disagreement, but Rachel thinks the key moment was in the press conference: he said he was glad that, with less forward guidance, the market is learning to face the ball rather than the referee. The ball is the data and the events; the referee is the Fed. The implication is that the run-up in oil and yields has already done some of the tightening for him, so he doesn't need to move rates yet and can keep optionality while watching. On the data, the episode notes June's core inflation fell quickly once fighting paused — core CPI at -0.02% month-on-month and core PCE at 0.13% — with no sign of the sticky, second-wave inflation people feared, against the New York Fed's Williams having said that roughly 0.2% monthly core is a safe pace.
**Tina then maps this onto her own job, which is the most candid passage in the episode.** Warsh's position, she says, is much like an analyst's: say too much and you constrain yourself, and you and the market start influencing each other — as during the Ukraine war, when the Fed had already committed to holding rates for a period and then couldn't respond quickly. In her own life, the question she gets constantly is "Tina, I saw you on TV, what should I buy right now," and she finds it genuinely difficult, because the moment she answers she has anchored someone. And things change: she agrees the long-term AI story is intact, yet a one-to-two-month Rubin delay changes revenue and changes how the stock trades — "I can't report back to you in real time." So what she wants to transmit is the logic, the argument, and the observable data, and let people judge. She's honest about the cost: without forward guidance, the dispersion in judgment widens and the volatility of being wrong goes up. The upside is that it forces you to build your own view.
**The sell-off that week in July was almost entirely not about fundamentals; what changed was leverage and forced selling.** Tina's reconstruction is worth following in full. Normally in earnings season, the first company prints well and everything that prints well afterwards rallies. This time everything printed well and everything fell. The differentiating detail in Google's 17 July report was that free cash flow turned negative faster than expected. Tight cash flow means you may need to issue debt — and precisely then the war escalated, oil went up, and people started worrying about hikes. Credit default swap spreads widened; she notes some cloud service providers went from thirty-odd to sixty or seventy, a doubling. That produced a wave of "AI is a bubble, the money is about to dry up" commentary online. She reads it instead as trading behaviour, using Taiwan's convertible bond market as the analogy: a good company isn't going to fail, but when its convertible drops from a hundred-something to ninety-eight, the risk system starts flashing and tells you to cut. In a falling market there are no bids, so forced selling makes the drop worse. Layer on deleveraging and thin liquidity in small and mid caps that same week. Her conclusion: fundamentals did change, but only marginally — and if you insist on naming the flaws, they're the Rubin delay (visible in June, and by August the two months have passed) and cost pressure from the shortage components.
**The narrative has already moved from price hikes to cash flow — and part of the capex upgrade is price, not volume.** That's Tina's call, and it leads straight to her inference: through the second half and into next year, chasing the price-hike names should carry less weight. She notes some suppliers are already adjusting specifications in response to supply and demand — trimming memory capacity, for instance — but you cannot read that as demand weakening, because without the trim the supply wouldn't be there at all. Rachel completes the picture across three layers. Upstream remains genuinely tight, though inventories are rising — and rising inventory has two opposite readings, active building ahead of a strong season or passive accumulation because goods aren't moving. Her read of the current data favours the former. In the middle layer, all five major cloud providers raised capex and three of them now have negative free cash flow (Google for the first time); MacroMicro's two tests are AI monetisation coverage — using remaining performance obligations and backlog against capex, which was actually better in Q2 than Q1 — and rates, with AAA corporate yields around 5.2–5.3% looking reasonable given a 3.75% policy rate and a ten-year near 4.5%; what to watch is whether that AAA level starts climbing meaningfully. Downstream, non-AI consumer and automotive remain soft, but fiercer competition among AI applications is a good thing, since it pushes down cost per unit of compute and broadens adoption — the Jevons paradox. As for what to watch technically in the second half, Tina's answer is packaging: die sizes are growing, agentic AI needs larger context memory which pushes memory stacking architecture, and optical transmission later on would add an optical engine on top. The difficulty only goes one way.
**Rachel's two differences versus 2000 are both testable.** The first is valuation: she recalls Cisco trading near 200 times earnings back then, whereas the episode cites NVIDIA at roughly 30 times trailing and under 20 times forward, and TSMC at about 30–31 times trailing and around 22 times forward. The second is the direction of the industry narrative. In 2000, supply came first — build the infrastructure, then wait ten or twenty years for e-commerce, streaming, social and video demand to actually boom. Now demand comes first, to the point where supply can't be brought online, which is exactly why HBM and advanced packaging are constrained. Tina's bullishness runs on a different argument: compute supply is far behind demand, and some agentic AI deployments aren't stalled by unwillingness but by tokens. Her example is her own morning meeting, where the team was debating whether to bring an analyst-facing AI agent into the advisory business — that one deployment alone would burn an enormous number of tokens, and if every firm does it, the order of magnitude changes entirely.
**The part AI can't take over gets described very concretely.** Tina's example is mature-node capacity: TSMC's earnings call touched on it, that fed through to UMC, and then Vanguard came out with its own comments on mature-node demand. Feed all three sets of call notes to a model and it struggles to extract the core thread; an analyst can turn it into a coherent story — mature nodes really aren't that tight, but certain applications (she names power management ICs) are growing, so suppliers exposed to that pocket benefit. And the client question she relays is the sharpest line in the episode: "I asked AI these five questions and got five different answers — tell me which one is true." Rachel echoes it from the product side: with the AI service they launched in July, the hardest part wasn't connecting the database but arguing out how each type of question should be answered, prompt by prompt and skill by skill. What the user is buying, in other words, is those thousands of words of embedded judgment — and the fact that someone stands behind whether they're right.
## Extended thoughts
**1. "What's the real issue" is really asking a very operational question: whose pain point is pinned down.**
The test this episode offers isn't "which event is bigger." It's "does this event move the variable the decision-maker is stuck with." The Treasury yield qualifies because it binds at both ends: for policymakers it's a thermometer of funding cost and market confidence; for companies it decides whether the capex build can keep being debt-financed. Two roads to the same number — that's what earns it the lead role.
Why tariffs demoted to a fake issue is the more interesting inversion. It isn't that they don't matter; Tina says capacity relocation is still under way. It's that the ceiling has already been set by yields: go too high and you damage confidence and push yields up, which is the one place the decision-maker can't afford to go. **Once an issue is capped by another issue, it stops being the variable that sets the tone.**
What makes this framework usable is that it can be falsified. If yields climb meaningfully and the policy stance doesn't soften — priorities shift, or tolerance rises — then the "soft spot" assumption is void and the whole chain needs rewriting, not discounting. Applied to my own holdings, the question becomes: which variable is this story pinned to, and where is that variable heading?
**2. The CDS passage is a textbook case of one number with two readings — and the mistake I make most often.**
Thirty-odd to sixty or seventy, described as "doubling," sounds alarming. But sixty or seventy is still a very low absolute level. The layer Tina adds with the convertible bond analogy is the real mechanism: the company is fine, but once the price breaks the risk limit, forced stops trigger — and in a falling market there are no bids, so the selling compounds. The pressure is mechanical and has nothing to do with the company's health.
So the test isn't "therefore don't worry." It's three things read together: **absolute level rather than percentage change; whether a forced-selling mechanism sits at that price; and what liquidity conditions look like at the same moment.** If all three point to mechanism, it's most likely trading behaviour. If the absolute level has entered genuinely expensive territory, or if a debt issue actually fails, then it's a different story. Tina leaves that line explicitly: the day capex actually gets cut because of cash flow is a different story. That sentence is worth more than any price target, because it can be observed and either confirmed or refuted.
The same disease shows up twice in this episode; the second instance is inventory. Rising inventory can be active building into a strong season or passive accumulation because product isn't moving, and **on the financials the two look identical**. The number won't tell you which. You need the context — shipment schedules, how tight supply is, and whether the inventory converts to revenue over the next quarter or two. That's exactly where I cut corners reading financials: inventory up, mark it down; margin down, mark it down — skipping the step that costs effort.
**3. The part AI can't take over is precisely the part of my own process that's least systematised.**
Tina's mature-node example looks like information aggregation and isn't. The reason TSMC's, UMC's and Vanguard's comments on the same subject have to be read together is that **the three sit in different positions with different interests, so each statement warrants a different discount** — and that judgment has to come first for the aggregation to mean anything. Hand a model three sets of notes and it produces an average, and the right answer usually isn't the average. It's "this one company, for this specific reason, is more credible this time."
That client's line — five questions, five answers, tell me which is true — locates the value precisely. It isn't producing content. It's carrying the judgment. And the only way to prove you're carrying it is to leave a record someone can check afterwards.
Which connects to something I keep insisting on: any conclusion has to come with "under what circumstances would I admit I was wrong." This episode is generous on that front, leaving a whole row of things to mark against later — whether policy softens again the next time yields climb, whether upstream inventory converts to revenue in the second half, whether three negative free cash flows force a capex cut, whether AAA yields start climbing meaningfully. Compare that with "long-term AI remains intact," a sentence that can never be wrong — **and therefore carries no information**.
## Worth reading further
- Original episode: MacroMicro Podcast, "From the Financial Crisis to the AI Wave: Senior Analysts on Reading the Market | Finance Friends Special" (9 August 2026), with Rachel and Tina Liao
- The Fed publishes FOMC statements, the vote record and press conference transcripts — the 9–3 split and the forward-guidance language can be checked directly
- US CPI (Bureau of Labor Statistics) and PCE (Bureau of Economic Analysis) monthly figures are public; the June core readings cited here can be verified
- The US Treasury publishes the daily yield curve, and the St. Louis Fed's FRED database carries AAA corporate bond yield series for checking the 5.2–5.3% figure
- Cloud providers disclose capex, free cash flow and remaining performance obligations in their quarterly filings and decks — "three turned negative" and the coverage ratio can be recomputed independently
- CME's FedWatch tool publicly displays the probability distribution implied by rate futures
- Company multiples and the Philadelphia Semiconductor Index are checkable from public quote sources; the figures in the episode were spoken at a moment in time and move afterwards
- The Su Shi quatrain at the top is my own footnote while listening, not part of the episode
---
**Disclaimer**: This is a personal listening note and study record, published as educational content. It **does not constitute investment advice, an offer, or a solicitation**. Companies, institutions, policies, figures and prices mentioned come from the public episode and public sources, were spoken at the time of recording, and may since have changed. Nothing here recommends any specific security, provides price targets or entry and exit timing, or forecasts rates, oil, currencies or any commodity price. Investing carries risk; judge independently according to your own financial situation and risk tolerance, and consult a qualified professional where appropriate. Copyright in the quoted material belongs to the original programme — please listen to it and support its creators.