Portfolio experiment

Claude allocates, you execute. Started 2026-07-20 · day 60 · ends 2026-12-31
Value
63,842 kr
Since start
+1.05%
+2.71% excl. transition
ACWI
+1.85%
vs ACWI
-0.80pp
+1.16pp excl. transition
The headline is your money from 2026-07-20, so it carries the cost of switching portfolios. Excl. transition measures from 2026-07-21 13:43, when the portfolio was first fully held, and is therefore gross of every trading cost paid so far: there has been no turnover since. Returns do not change with account size; only the cost drag does, and `bin/scale.py` shows it barely moves because the 0.25% FX conversion is proportional.

Execution checklist

Target portfolio fully held. Nothing outstanding.

Thesis

Own the broadening, not the crowded leader.
The AI earnings are real, but positioning is the risk. A record 82% of managers were crowded long semis with cash at 3.6%, the lowest since February and under the 4% that triggers BofA's sell signal, and the trade broke within a week. Overweight instead what the data says is broadening, plus the two cheapest sectors, with energy doubling as a hedge against an unresolved Hormuz.

Positions

InstrumentQtyOpenLastCurP/L DKK%Value DKK
Amundi Prime All Country World Acc UCITS ETF358.0012.7512.89EUR396+1.2%34,421
iShares MSCI World Small Cap UCITS ETF140.009.018.96EUR-51-0.5%9,358
Equinor ASA26.00362.20419.00NOK1,019+15.7%7,518
Danske Bank A/S17.00362.90375.40DKK212+3.4%6,382
BNP Paribas SA8.00104.22100.74EUR-208-3.3%6,010
Total1,36963,689

What would prove this wrong

Underweight AI / semis
Semis stop selling off on good news. Concretely: SOX regains and holds its 17 July level of 11,674, and a hyperscaler capex raise or a memory earnings beat is met with a rising share price rather than a falling one.
Most likely way this portfolio underperforms. Stated up front. Rewritten 29.07.2026: the original watched the capex number, which by then had come in strong while semis fell anyway, so read literally it would have fired against a position that was working.
Energy 20%
Durable Hormuz settlement plus demand weakness sends oil into the $60s.
The hedge then costs money in a rising market.
Financials 20%
Inflation falls fast enough that the Fed and ECB cut instead of hike, compressing margins.
June CPI at 3.5%, down from 4.2%, is already partial evidence against this leg.
The broadening 15%
Breadth reverses and mega-cap concentration resumes, as it did through 2023-2025.
Equal-weight vs cap-weight is the tell.

Calendar

2026-07-29FOMC decision
First read on Warsh's guidance; rate itself expected unchanged.
2026-09-15FOMC + fresh dot plot
Most important policy event in the window.
2026-09-29100% US tariff on patented pharmaceuticals
Why Novo was sold. Watch European pharma.
2026-09-30US government funding deadline
Shutdown risk weeks before the election.
2026-11-03US midterms
Split government is historically the most equity-favourable outcome.
2026-11-09US affiliates-rule suspension and China Nexperia counter-restrictions both expire
Next genuine policy inflection. Zero Entity List actions all year, so the truce has held until now.
2026-12-31Experiment ends
Judged against ACWI, the old portfolio, and OMXC25, net of costs.

Decision log

WhenKindSummary
2026-09-10T14:55reviewWeeks 5-7 catch-up review (due 26.08, 02.09, 09.09; run 10.09): NO TRADE. No falsifier fired at any of the four dates, largest sleeve drift +1.77pp of a 5pp trigger.
2026-08-19T12:15reviewWeek 4 review: NO TRADE. No falsifier fired and every one moved AWAY. energy moved furthest: the US-Iran MoU EXPIRED 17.08 with no deal and Brent is $91.52, so the tail risk the sleeve hedges is live - and it was deliberately not added to, because rule 2 says confirming evidence is not a reason to trade. Largest sleeve drift +1.05pp of a 5pp trigger. First down week: -0.89% against ACWI -1.08% and the old portfolio -2.64%.
2026-08-12T15:33tradeai-underweight response EXECUTED: sold 5 Chevron at 196.69 USD (15:20:57Z), bought 68 WEBN at 13.008 EUR (15:22:09Z). Cost 62.79 DKK, 0.097% of the account, inside the 60-75 DKK pre-committed estimate. Energy now 10% (Equinor alone), global core 54.2%.
2026-08-12T09:19reviewWeek 3 review: ai-underweight FIRED - holds test passed 5 of 5 (SOX closes 05-11.08 all above 11,674, week low 11,707.78). Pre-committed response recommended to owner: sell 5 CVX in full (~6,354 DKK) into WEBN (~64 shares). No other falsifier fired; energy moved away (Brent +12.9% on the stalled deal); sleeve drift max 0.68pp.
2026-08-05T09:42reviewWeek 2 review: no trade. ai-underweight is ARMED - every clause of it carrying a number has fired, and it resolves 12.08.2026 on a 'holds' test defined before the US open.
2026-07-29T13:34reviewWeek 1 review: no trade. No falsifier fired, no sleeve drifted past 1.6pp of a 5pp trigger.
2026-07-21T13:43tradeTransition complete: all 6 target positions held, fully invested.
2026-07-21T13:36tradeTransition placed: sold all 8 original holdings, bought the 6-position target.

Machine

Disk free
101.5 GB
474.6 GB / 600.6 GB used (79%)
Database (bolig)
173.4 GB
RAM available
15.8 GB
14.8 GB / 30.6 GB used
Swap used
3.9 GB
of 32.0 GB
CPU now
2%
load 1.0 / 3.0 / 4.0 · 16 cores
Uptime
40d 8h
Load is the 1 / 5 / 15-minute average; on this box anything under the core count is comfortable. Measured 2026-09-18 19:15 UTC.
Plausible range to 31 December is roughly -15% to +20%, and where it lands inside that is mostly luck. The tilts might add a point or two of expected return. They cannot manufacture a high one. A single 5-month result is not evidence of skill in either direction.

Rendered 2026-09-18 19:15 UTC from 2602 snapshots. Prices are 15 minutes delayed. Not investment advice.