Erica · PM Research MemoAll memos

PM RESEARCH MEMO

Talk: Why Every Trader on Earth is Watching the 10-Year Treasury Now | WDWL
Speakers: Josh Brown (host, Ritholtz Wealth / The Compound); Nick Colas, co-founder, DataTrek Research (guest). Jessica Rabe (DataTrek co-founder) out sick — mentioned, not on mic.
Date published: Monday, 24 August 2026, 5:00 PM ET
Duration: 42:48
Source URL: https://www.youtube.com/watch?v=WaKDHRCC3NE
Memo date: Monday, 24 August 2026
Source type: YouTube English auto-generated captions (ASR), flattened continuous prose. No per-cue VTT in the working dump (timedtext JSON3 was retrieved as prose; local curl of the same URL returned an empty 200; WEB player was login-walled). Timestamps in this memo are estimated from YouTube chapter markers plus narrative position / linear interpolation over 42:48. Treat as plus or minus a couple of minutes, wider inside the long 2:59-26:54 rates block.
ASR quality note: Auto-captions, not a human transcript. Names, tickers, and numbers are often garbled. Common corrections applied here where context is strong (Colas, Rabe, DataTrek, iShares, CAGR, AI-related, neutral rate, laddered, RIAs, 10-year / "tens," Volcker, Anthropic, Claude, Baidu, Alibaba / BABA, Tencent, MSCI ACWI, T. Rowe Price, roll risk, principal, equal-weight). Remaining uncertainties flagged in-line (notably "four 75 basis point rate hikes"). Sponsor read (FM Investments / SGVA ultrashort Treasury ETF; ASR "Troll Price" = T. Rowe Price) is not investment content. This memo contains no trade recommendations. Colas's "keep duration short / under five years" is his client advice as Source, not ours.

How to read this document: Restatements of the conversation are Source. Interpretive links and underwriting judgments are Inference. Any fact not spoken in the transcript is External check needed. Speaker-stated numbers are used as spoken and attributed; they are not independently verified. Brown vs Colas attributed throughout.

YouTube chapters (used as timestamp anchors, not as spoken content): 0:00 Ad Read · 1:30 Intro · 2:59 Long Term Treasury Yields Just Keep Moving Higher · 26:54 We Publish an S&P 500 Fair Value Grid Every 2-3 Weeks · 34:48 Grab Bag of Other Recent DataTrek Client Notes/Discussions · 42:48 end.


EXECUTIVE SUMMARY


SOURCE-ACCURATE SUMMARY

Chronological, faithful to what was said. Short quotes only where they carry the claim (max 20 words). All times estimated (plus or minus a couple of minutes). Brown = Josh Brown; Colas = Nick Colas.


SYSTEMS MAP / VALUE CHAIN ANALYSIS

Mark inference vs source throughout.

Key players (Source, with roles as they assigned them)

Node Role in their map Source locus (est.)
30-year UST / TLT (iShares) Exhibit of real-rate breakout; 2010s ~+8% CAGR vs 2020s -4.4%; failed "risk-off" hedge ~04:30-09:00
Inflation expectations (30y) Non-moving piece: ~1.5-2.5% for 15-16 years ~04:30-07:00
Real yields The moving piece: ~2-2.5% to pushing 3%; why nominal 5.2-5.3% ~04:30-07:00
10-year UST Equity "jitter" trigger at 5%; print 4.70-4.72 "today"; 2023 twitch at 5 even when orderly ~19:30-23:00
Fed / QE unwind / balance sheet First driver of high reals; "no bond buying," stable BS ~09:00-12:30
Neutral rate / US real economy Second driver: no recession through 2022 hikes, 2025 trade shock, oil 22 and 25 ~09:00-12:30, ~26:00
Federal deficit / Treasury credit Third driver; Brown: YTD 2026 already > all of 2025 ~09:00-14:30
AI IG issuers (Alphabet/Google, Amazon, Meta; Oracle as exception) Fourth driver: substitute for long UST inside RIA/PM books; high-grade except Oracle (rating "almost... junk," worst mega-cap tape) ~14:30-19:30, ~23:00-26:00
Hyperscaler cash flows (ads, high-end electronics) Still cyclical; must eventually service AI debt; private credit / PE / guarantees as new 2022-absent plumbing (Brown) ~23:00-26:54
S&P 500 / FactSet grid Rich vs 14-18x; tape = earnings revisions +15/+13, PE flat-to-down; 20x is the "payoff" line ~26:54-31:00
Oil / diesel / US-Iran Multiple-cap: 22x path; diesel 100+; Colas's base demoted ~6 weeks ago ~31:00-34:00
New Fed chair Co-named (with oil shock) as why multiples have not expanded ~31:00-34:00
AI labs / Anthropic / Gemini / coding vs non-coding 24x path requires customer-cited ROI; 2026 demand = coding; spend motive = AGI ~32:30-37:30
Chinese OSS / Baidu / BABA / Tencent / DeepSeek Fast/good but not the AGI endgame; constrained profit if they "win" ~36:30-42:30
Equal-weight vs cap-weight S&P; MSCI ACWI 14% vs ~38-40% tech; public markets as VC science project ~39:30-41:00
USD as reserve currency Innovation + tax-base expansion, not just army/navy; lose AI race then deficit capacity erodes ~17:30, ~41:00-42:30

Flow (Source, sequenced as they told it): Inflation expectations stay glued in a 1.5-2.5% band so the entire 30-year backup is a real-rate event. Reals are high because QE's artificial depression is over, the economy has not recessed through multiple shocks (so r-star is higher), deficits are large and growing (Brown: YTD already worse than 2025), and AI IG paper is a substitute good for long Treasuries in risk-averse PM/RIA books that just lived through TLT -4.4% CAGR. That substitution is rational at the credit level (Google as likely to repay as the sovereign) even if the corporate yield is higher for lack of Marines. Equities do not care until tens approach 5% (4.70-4.72 now), and they care more than in 2022 because the same mega-caps are now issuers with roll risk and AI reinvestment risk (Oracle as the live credit-tape exhibit). S&P YTD is an earnings-revision market at a ~20x multiple with no expansion cushion; upside fattens only if Iran/oil and/or AI-ROI land. Parallel implicit put: no recession for five years, which is why correlations and VIX are crushed and why hyperscalers will lever for AGI. That put is also why losing the US-China AGI race is, in Colas's telling, a reserve-currency problem.

Bottlenecks, leverage points, pricing power (Source then Inference)

  1. Bottleneck that is binding on the long end (real-rate / duration demand). Source: inflation expectations "dead flat"; reals "pushing up on 3%"; four demand/supply reasons, of which AI IG substitution is the "new topic." Inference: the marginal buyer of long UST is being bid away by IG paper that clears a "risk-averse, get-my-principal-back" screen. Pricing power on long yields sits with (a) whether r-star stays high and (b) whether the AI issuance calendar keeps offering a substitute. Short-end / under-5y duration is Colas's stated place to hide until growth breaks — report as his, not ours.

  2. Bottleneck that is not binding (inflation expectations, on his chart). Source: 15-16 years, 1.5-2.5%. Inference: an inflation-scare narrative for this backup is a misread on his evidence. If breakevens do break out, his decomposition is the first thing to re-underwrite.

  3. Bottleneck at 5.00% tens (equity multiple / growth). Source: 2020s jitter level; 4.70-4.72 today; 2023 twitch at 5 even orderly; 20x S&P discomfort; "maybe five, maybe six" for the real-economy break (autos, housing, spend, layoffs). Inference: about 30bp of 10-year backup is the distance from "orderly TV comfort" to his empirical trigger — a thin buffer. Velocity is a modifier, not a substitute for the level.

  4. Bottleneck inside mega-cap finance (issuance + roll + AI ROI). Source: buybacks to secondaries + bonds; Oracle nearly-junk / worst mega-cap; AI must "fund those debt issues." Brown: private credit, PE, illiquidity, hyperscaler guarantees. Inference: the 2022 "fortress net-cash quality" factor is no longer a free hedge against a rates shock. Pricing power has migrated from equity-holder residual (buybacks) to bondholder claim + capex.

  5. Bottleneck on the multiple (oil/Iran + Fed chair + unproven non-coding AI). Source: multiples stuck because of oil shock + new Fed chair; 22x needs cheaper oil; 24x needs AI value in tech earnings and customer citations; coding is not the economy. Inference: the S&P is priced as if G stays mid-teens; the option on 22-24x is a macro (Iran) plus a micro (non-coding ROI) dual key — either can stay lost.

  6. Hidden bottleneck (the five-year no-recession assumption). Source: low 10-plus-year sector correlations, low VIX, levered hyperscaler AGI spend, "believe that 100%" to be super-long; 38-40% of ACWI in the science project. Inference: this is the system's single point of failure. A garden-variety recession would re-correlate, lift vol, hit cyclical ad/device cash flows that service AI debt, and remove Colas's reason for staying short duration (he wants long duration when the economy weakens — a hedge that would then compete with forced de-risking).

Upstream / downstream implications (Inference, mapped onto Source)

Where constraints create investment hypotheses (not recommendations)

H1 (Source-led): If 30y inflation expectations stay in the 1.5-2.5% band, further long-end backups are real-rate / r-star / supply events. Research breakevens vs real yields as the first split of any "bond tantrum" headline. Do not treat CPI-scare language as confirmed by this source.

H2 (Source-led): If AI IG issuance stays at Brown's about $1.75T / +20-30% pace, long UST have a substitute-good problem inside non-dedicated accounts. Diligence the share of RIA/core-plus that can legally/behaviorally swap UST for GOOGL/AMZN/META paper.

H3 (Source-led): 10-year 5.00% is a harder equity event than 2022 if mega-cap balance sheets now carry issuance and roll risk. Oracle is the named exhibit; do not assume "quality growth" automatically dampens a 5% print.

H4 (Source-led): S&P path is an estimate path at a stuck ~20x. Track revision breadth (tech + energy) and strip SpaceX markups. Iran/oil is the multiple-expansion key Colas already downgraded; AI-ROI in customer earnings is the 24x key he still treats as a dream.

H5 (Source-led): Cap-weight vs equal-weight is a science-project concentration choice (38-40% vs 14% tech), not a style-box hobby. DataTrek's long-standing equal-weight skepticism is explicitly softening.

H6 (Inference): Colas's "keep duration under 5y until the economy weakens" and "be super long only if you 100% believe no recession for 5 years" are the same coin. A growth scare would be the signal to extend duration in his framework and the signal that the equity/AGI complex's implicit put is failing — i.e., the hedge and the risk asset can be forced at the same time. Research that collision; do not treat his duration advice as a free lunch.


SECOND AND THIRD-ORDER EFFECTS

Each chain: [Primary observation] then [Second-order] then [Third-order] then [Investment relevance]. Links tagged.

1) Real-rate breakout destroys the 2010s hedge (allocators, 60/40, TLT-like sleeves)

2) AI IG paper crowds the long end and rewires mega-cap factor exposure (credit PMs, equity PMs, Treasury)

3) 5% tens vs R-G: multiple compression unless growth outruns the discount rate (S&P, duration, GARP)

4) Five-year no-recession put underwrites low vol, low correlation, and levered AGI (vol, sector specialists, hyperscalers)

5) AGI race as reserve-currency / deficit-capacity insurance (USD, UST demand, US vs China AI names)


SCENARIO FRAMEWORK

They assigned no numeric probabilities. Colas did rank three upside S&P paths and explicitly parked downside as "well understood." Numeric weights below are analyst inference for research planning only, not theirs. Timeline anchors that are theirs: 10y 4.70-4.72 today, trigger 5%; 30y 5.2-5.3%; next-12-month S&P window "through '27"; Iran/oil demotion "about six weeks ago"; Anthropic IPO "this year"; "no recession for the next 5 years"; duration short until the economy weakens.

Bull — "Oil/Iran and/or AI-ROI unlock 22-24x; tens stay under 5; G outruns R"

Analyst inference probability: roughly 25%. Maps to their "better" (+13-28% at 22x) and "dream" (+12-39% at 24x). Colas already abandoned the 13-28 band as base six weeks ago; the dream "need[s] all three things." Brown's Anthropic/Gemini/customer-citation checklist is a high bar.

Base — "Stuck-20x earnings grind; reals stay high; duration stays a step stool"

Analyst inference probability: roughly 45%. Closest to their stated most-likely: S&P +6 to 16% over 12 months, multiples ~20, estimates still rising; Colas "we're in the six to 16 band"; summer rates orderly but not reversing; "keep duration pretty short, like under five years"; no signs of economic weakening; AI demand still coding-led; Iran unresolved.

Bear — "5% (or 6%) tens, estimate-cuts into a rich grid, and/or the five-year put fails"

Analyst inference probability: roughly 30%. They did not spec a downside S&P number ("thousand different ways"). This is Inference built from their own invalidation flags: 5% jitter, 20x discomfort, "not a lot of room for error if earnings miss," Oracle credit, cyclical hyperscaler cash flows, "maybe five, maybe six" for the real economy, "no natural cap" on yields, and the 100% no-recession belief required to stay super-long.

Invalidation hierarchy to put on a PM dashboard (mixed Source/Inference):

  1. 10-year through 5.00% — Source, hard (2020s jitter rule; 2023 precedent).
  2. Inflation expectations breaking the 1.5-2.5% 2010-present box — Inference as a falsifier of "it's reals."
  3. FactSet this-year / next-year revisions rolling over into a 20x tape — Source ("not a lot of room for error if earnings miss").
  4. Economy weakening — Source as the duration-extension trigger and (Inference) as the five-year-put fail.
  5. Oracle-to-complex credit contagion so IG is no longer the "safe substitute" — Source exhibit, Inference on spread.
  6. Non-coding AI ROI still absent by the Anthropic-IPO window — Source as why the 24x path is a dream; Inference as a clock on the science-project multiple.

COMPANY/ASSET WATCHLIST

No buy/sell. Hypotheses and watch items only. Metrics that were not in the talk are labeled External check needed. Speakers' portfolio-construction comments (duration under 5y, step stool) are their advice as Source, not recommendations of this memo.

UST 10-year

UST 30-year / long-duration beta (TLT as exhibit)

Short-end / ultrashort / under-5y duration (category)

Mega-cap AI issuers — Alphabet / Google, Amazon, Meta (IG); Oracle as the stressed exception

S&P 500 (fair-value grid)

Oil / diesel / US-Iran (macro overlay, not a commodity recommendation)

Anthropic (private; IPO "this year") and the customer-ROI complex

Equal-weight S&P vs cap-weight / MSCI ACWI tech share

China AI names — Baidu, Alibaba / BABA, Tencent, DeepSeek (color only)

Other names / items that appeared as color, not a thesis

iShares TLT (exhibit, above); FactSet (data vendor); SpaceX (Q2 markup at AMZN/GOOGL); Ben Carlson (1990s 5% average — colleague of Brown); Paul Volcker / 17% tens (history); new Fed chair (unnamed; identity External check needed); OpenAI, Microsoft, AWS (customer/provider list on the 24x path); ChatGPT / Claude / Gemini (intelligence/ROI); Moore's law (benchmark); VIX (low, with correlations); healthcare and financials (examples of what you can pick when correlations are low); Marines / Coast Guard / Navy (sovereign-vs-Google rhetoric); FM / SGVA / T. Rowe Price (sponsor). None of these is a watchlist primary except as already broken out.


DILIGENCE QUESTIONS & RESEARCH AGENDA

P1 — Must-do before any allocation-research decision

  1. Rates tape vs the prints they spoke (P1). Rebuild, as of recording/publish date (Mon 24 Aug 2026, with "today" 10y 4.70-4.72 and 30y 5.2-5.3%): constant-maturity 10y and 30y, 30y TIPS/breakeven/real split, whether 30y yield is in fact at 15- or 20-year highs, TLT total-return CAGR 2010-2019 vs 2020-YTD 2026 vs last six years (they used both "this decade so far" and "last 6 years"; Brown 7.8% vs Colas "almost 8%"). Data: Treasury H.15, Bloomberg, iShares TLT factsheet. Expert: rates strategist to replicate the exact inflation-expectation series on his chart (5y5y? 30y breakeven? — not specified).

  2. The 5.00% 10-year jitter rule (P1). Map 2020s S&P drawdowns vs 10y prints: did equities "jitter" at 5% this decade, and was 2023 an orderly 5% twitch as he said? Distance from 4.70-4.72 is ~28-30bp — option-implied probability of a 5% print over 1m/3m. Expert: equity-macro desk, not a duration PM only.

  3. Issuance substitution (P1). Verify Brown's $1.75T corporate issuance YTD and "20 or 30% ahead" of 2025. Split IG vs HY, mega-cap AI (GOOGL, AMZN, META, ORCL, others) vs the rest, coupons vs matched UST, who bought (RIAs, foreign official, pensions, dedicated credit). Without this, "every dollar not in Treasuries" is rhetoric. Expert: DCM / IG strategist; SIFMA / Bloomberg issuance.

  4. S&P grid identity (P1). Reproduce +12.1% YTD as of the Friday before 24 Aug 2026; FactSet NTM and this-year/next-year revision +15 / +13 (and 2027 ~13%); forward PE vs 14-26 grid; confirm "little PE contraction" and the SpaceX markup in AMZN/GOOGL Q2. If revisions are not +15/+13, the "100% earnings" punchline fails. Expert: house accounting for NTM EPS; AMZN/GOOGL 10-Q footnotes on SpaceX marks.

P2 — Needed to underwrite the changed mega-cap package and the r-star claim

  1. Oracle vs the IG complex (P2). Is Oracle in fact the worst-performing mega-cap, and is the rating "almost... junk"? Peer net-debt / EBITDA / interest coverage vs 2022 for META, AMZN, GOOGL, MSFT. Secondary-equity vs bond mix 2025-26. Brown's private-credit / PE / hyperscaler-guarantee claim needs a map (SPVs, leases, residual-value guarantees) — none quantified on mic. Expert: IG analyst + rating-agency primary.

  2. Deficit and r-star evidence (P2). Brown: YTD 2026 federal deficit already larger than all of 2025. CBO/Treasury Daily Treasury Statement. Colas's shock list: 2022 hike size (ASR "four 75 basis point" — reconcile to the 2022 path), 2025 trade-policy shock, oil 22 and 25 — did activity/unemployment actually "not notice"? Neutral-rate estimates (Laubach-Williams, NY Fed, FOMC longer-run) vs his qualitative "must be higher." Expert: fiscal analyst + Fed watcher.

  3. Iran / oil / diesel / Fed chair (P2). Spot vs his 55-60 start, 60-65 "reasonable," diesel 100+ to 70-80. Conflict timeline vs "six weeks ago" base-case demotion. Who is the new Fed chair? Unnamed on mic. Expert: energy geopolitics; Fed personnel.

  4. AGI-vs-OSS scoreboard (P2). artificialanalysis.ai (or equivalent) intelligence indices: ChatGPT +1300% since Nov 2022, Claude +76% last year, vs 41% Moore. Is that the site's own "intelligence" index, and does it mean what he thinks? Chinese OSS benchmark gaps vs US frontier. Anthropic IPO calendar. Expert: AI research engineer not on a lab IR call.

P3 — Process, identity, and long-horizon claims

  1. Correlation / VIX / ACWI 38% (P3 but high intellectual value). Replicate sector and pair-wise stock correlations at "10 plus year lows," VIX "very low," MSCI ACWI tech+big-tech 38% (he rounds to 40), equal-weight S&P 14% tech. If the 38% is a stretched definition of "tech and big tech," the science-project concentration is overstated. Expert: index specialist.

  2. ASR / identity cleanup (P3). Confirm DataTrek URL (datatrekresearch.com), 1,500+ briefing distribution, Jessica Rabe illness (color only), Ben Carlson 1990s-5% chart, "four 75 basis point" vs 475bp, "negative 4%" vs "-4.4%" TLT (he used both), 16+4 top global companies list. Keep unverified strings out of LP-facing lists.

  3. Downside they refused to spec (P3). They said downside is "well understood" and did not put numbers on it. Build an internal bear grid at 14/16/18x on cut estimates so we are not hostage to their upside-only slide. This is our work, labeled Inference, not theirs.


RISK ANALYSIS

Thesis risks (the framework is wrong)

Timing risks

Execution risks

External / policy / data risks

What would change the research view (actionable)

  1. 10-year sustained through 5.00% — treat his equity-jitter rule as live; re-open 2022-vs-now mega-cap credit transmission (Oracle first).
  2. 30y inflation expectations leaving ~1.5-2.5% — stop calling this a real-rate-only event.
  3. FactSet this-year/next-year revisions rolling over — the 6-16% base (which is all earnings) is invalid; grid says there is little PE cushion.
  4. Economic data weakening — in his framework, that is the duration-extension signal; in ours, it is also the five-year-put test. Research both, do not automatically buy long bonds because he would.
  5. AI IG issuance pace breaking down or Oracle-like stress spreading — substitution channel flips from "crowds UST" to "credit event."
  6. Iran resolution + oil/diesel into his 60-65 / 70-80 bands — 22x / 13-28% path comes back on the board (he already showed the mapping).
  7. Customer-cited AI earnings + Anthropic IPO with real revenue — 24x path graduates from dream; until then it is a checklist, not a base.
  8. Sector correlations and VIX lifting without a 5% tens print — the no-recession assumption is dying on its own; cap-weight science-project risk rises even if duration is quiet.

APPENDIX: SOURCE DISCIPLINE LOG


Prepared 24 August 2026 for internal PM-research use (AP). Educational summary of a public interview; not investment advice. Estimated timestamps only. ASR caveats apply. Single primary source: WaKDHRCC3NE transcript.

Desk copy · not a trade recommendation · Josh Brown + Nick Colas (DataTrek) · Compound · 24 Aug 2026