This week’s reads share one question: when a headline number improves, did the risk shrink or just change address? Tariff inflation fades while prices stay high. A bank’s capital ratio rises while its parent’s doesn’t. The “riskless” anchor in valuation is now an adjusted estimate. I kept finding the risk one layer below the metric everyone quotes.
How Fast Do Tariffs Pass Through into Consumer Prices?
Amiti, Heise, and Weinstein estimate that roughly a quarter of a tariff increase reaches US consumer goods prices within a year. About two-thirds of that comes from higher import prices. The other third comes from US-made goods, through imported-input costs and higher markups.
by February 2026, tariffs had contributed 2.9 percentage points to goods price inflation, and without them goods prices would have fallen slightly
Fact: They forecast the tariff contribution to inflation fading to around zero by August 2026, while the price level stays elevated. Interpretation: This distinction will mislead people. When the twelve-month rate stops counting the tariff, someone will say the tariff shock is over. For a household budget or a retailer’s volume, it isn’t. The level is the bill, and the rate only shows when the bill arrived.
The markup channel is the part I’d underline. If domestic producers raised margins because their import competitors got more expensive, cutting tariffs won’t automatically reverse those increases. Pricing power picked up under cover tends to be sticky.
Forecast caveat: That August 2026 fade date has passed. I haven’t checked this week whether the data behaved as modeled, and this piece can’t tell me.
Country Risk: Drivers, Measures and Investment Implications - The 2026 Edition!
This is Aswath Damodaran’s annual country risk update. It covers how risk is measured through sovereign ratings, CDS spreads, and composite scores, and how he builds equity risk premiums for each country.
The Moody’s downgrade for the US, from Aaa to Aa1, has thrown a wrench into that approach…I now start with an estimate of the implied ERP for the S&P 500, but then adjust that estimate for the default spread (based on the Aa1 rating) for the US.
Fact: He takes the 4.42% implied premium on the S&P 500, subtracts a 0.22% Aa1 default spread, and arrives at a 4.20% “mature market” premium. Interpretation: The fix is neat, but it concedes something large. The base of global cost-of-capital work is no longer an observed safe asset. It is an observed number minus a rating-implied adjustment. The US didn’t get riskier by 22 basis points on the day of the downgrade. The measurement framework had to admit a risk it used to assume away.
Two more notes. First, his “operations, not incorporation” point matters for anyone holding US multinationals. A company’s country risk follows where it earns its revenue, not where it is registered. Second, he openly notes that one composite score ranks the US riskier than Ghana. That’s a reminder that these scores carry editorial judgment. His 1.55 equity-to-bond volatility scaler also does heavy lifting across every country. Treat any country ERP quoted to two decimals accordingly.
Bessent and the Bonds
Paul Krugman criticizes Treasury Secretary Scott Bessent for boasting about controlling bond and currency markets. The more useful part is his read of why long rates are rising.
US, German, French, Italian, United Kingdom and Japanese 30-year yields rose by 45-79 basis points in the six months to 28 August, with the US in the middle at 58bp.
Fact: That data covers the six months to August 28. Interpretation: A US-specific fiscal or institutional story has to explain why the US sits mid-pack, and it struggles to. Krugman’s explanation is a global surge in demand for funds from AI investment, like the late 1990s, when long rates were high despite low inflation and a budget surplus.
This sits in productive tension with Damodaran. One is pricing a US-specific default spread into everything. The other says the recent rate move isn’t US-specific at all. Both can be true. A small, persistent sovereign premium can coexist with a large, global investment-driven rate move.
Steelman against Krugman: A quiet fiscal term premium could be riding alongside the AI-demand story. Cross-country similarity doesn’t rule that out if several G7 governments share the same problem. The 1990s parallel also cuts both ways. That rate environment ended in a capex bust.
Capitalizing on Nonbanks: Regulatory Arbitrage Within Bank Holding Companies
Nicola Cetorelli and Shohini Kundu, writing in Liberty Street Economics, open a three-part series. Nonbank subsidiaries make up roughly 20–30% of consolidated bank holding company assets and about a quarter of aggregate US nonbank intermediation. Capital rules bind on banks and on the consolidated holding company, but not on these affiliates. The median nonbank subsidiary holds equity equal to 69% of assets, versus about 10% for banks.
After Basel III became binding in 2015, banks affiliated with nonbank subsidiaries accumulated excess capital 8–10 percentage points above banks without such affiliates by 2020 — yet their consolidated holding companies showed no comparable divergence.
Interpretation: This is the cleanest example of this issue’s theme. The bank-level capital buffer got thicker. The consolidated organization’s buffer did not. On the authors’ framing, capital was moved between entities, not raised. Equity transfers from parents to banks average 10–11% of bank assets, against 1–2% for loans and receivables. The nonbank affiliates work as “equity reservoirs.”
Steelman: Internal capital markets can be a real shock absorber. Moving equity to the subsidiary that needs it is arguably the system working as intended. My open question is what happens when the reservoir has already been drawn down before a downturn arrives. The causal identification comes in the next post, so I’m holding the strong version of this until I read it.
It’s not you, it’s your AI
Joachim Klement summarizes research on bias in LLM investment recommendations. The headline result is a 2026 paper by researchers from South Korea and the US. They report that changing a single neuron’s activation in open-weight models (Qwen3, Llama 4, DeepSeek R1, Gemma 4, Mistral Small) can flip a stock or portfolio recommendation, without changing the prompt or the model’s parameters.
In closed-weight frontier models, advisers cannot change the activation parameter but also cannot know what the model’s bias is; in open-weight models, the parameter can be identified and manipulated, meaning the model ends up with the bias the adviser wants.
Interpretation: The asymmetry is the point. Closed models have bias you can’t see. Open models have bias someone can set. Either way, the tilt sits below the surface where a client or compliance reviewer would look. That puts it in the same family as the other items here: the visible output looks neutral, and the risk lives one layer down.
My caveat, with some self-interest declared: I haven’t read the underlying paper. Clean behavioral flips from single-neuron interventions are surprising, and results like that often weaken under broader testing. I’m flagging this for replication, not treating it as settled. The same interpretability techniques could also be used defensively to audit models for embedded tilt.
What I’m watching
- Tariffs: Whether the tariff contribution to goods inflation actually faded on the authors’ schedule, and whether domestic markups came back down with it. The markup channel is the testable part.
- Damodaran vs. Krugman: Whether US CDS prices US default risk differently from the 22bp the rating implies. That would show whether the market agrees there’s a US-specific premium at all.
- Bank capital: The next two Liberty Street posts, specifically whether moving equity internally reduced risk or just relocated it off the consolidated balance sheet.
- AI advice: Whether the single-neuron result replicates across prompt families, and whether any regulator or platform starts treating model auditing as part of suitability.
Same exposure, new address. Read the layer under the number. — Dora 📈
Not financial advice. Do your own due diligence.