A $100MM discretionary global macro book run across five asset classes for five weeks. Top of all twelve groups, and concentrated enough that the process matters more than the number.
A $100MM notional book over five weeks, 27 April to 29 May 2026, with a requirement to trade all five core asset classes: equities, futures, rates and credit, FX and options.
The process was discretionary top-down for ideas and systematic for risk. Form the macro view judgmentally, express it through whichever instrument in the cross-asset stack is most efficient, then size it by volatility rather than notional. Conviction sets the direction, volatility sets the size.
Two reflexive narratives anchored nearly all positioning: the AI-infrastructure capex cycle running from hyperscaler spending through silicon, memory and storage, and the Strait of Hormuz oil shock, traded for its second-order effect of a supply-driven inflation impulse keeping rates higher for longer.
Cumulative profit of $15,596,764, or +15.60%, the highest of twelve groups and roughly 1.8x the second-placed book. Profitable in all five weeks, so the equity curve never had a peak-to-trough drawdown.
Week 2 was the strongest at +$7.85M on the semiconductor earnings catalyst, led by Micron at +$4.98M and AMD at +$2.63M. Week 3 was deliberately flat at +$0.02M: with a lead established, the three semiconductor single names were closed after their run and the capital rotated into a diversified macro book.
Mean weekly return was 3.12% against a 3.19% standard deviation, a weekly Sharpe of 0.98. Trade hit rate was 57%, twenty-one of thirty-seven.
Two trades, Dell at +$5.43M and Micron at +$4.98M, account for roughly two thirds of cumulative profit. That is the honest headline and it belongs above the rank, not beneath it.
Dell was the largest single position: its Infrastructure Solutions Group reported revenue up 181% year on year to $29.0bn, with AI-optimised server revenue up 757% to $16.1bn and a $9.7bn Pentagon contract behind it. The thesis was right and the sizing was aggressive, and both of those facts count.
The same week produced the worst trade in the book. A Salesforce straddle was meant to be direction-neutral, but the put leg collapsed for -$4.84M when the stock held into the print and implied volatility was crushed. It was oversized for a convex, time-decaying instrument. Only the Dell winner kept the week positive, which is not a risk framework, it is a coincidence.
With five weekly observations, the annualised Sharpe and the zero realised drawdown are properties of a short favourable sample. I would not present either as structural.
Sizing moved from equal notional in the early weeks to explicit volatility parity by Week 5: notional equals target volatility times AUM divided by the instrument volatility, with a 6% portfolio target and the sign taken from the discretionary view. High-volatility names are scaled down, low-volatility names scaled up, and each position contributes roughly equal risk.
Fixed income volatility came from modified duration times annualised yield volatility. The 30-year leg at 15.35 times 64bp gives about 9.8% price volatility. A single-trade leverage cap mattered: the 2s30s steepener long leg screened at a raw vol-adjusted size near $400MM and was capped at $100MM.
Four pillars held it together: volatility parity, a conviction premium allowing higher-conviction ideas a larger share after the vol adjustment, catalyst tagging so every trade was anchored to a dated event rather than a feeling, and correlation-weighted caps on overlapping themes.
The themes were nominally independent and were not. AI, oil, rates and events were all expressions of a single risk-on, oil-inflation regime. Directional co-movement between theme-level P&L ran 0.77 between oil and rates and 0.74 between AI and event-driven.
So the book behaved as one bet wearing five asset-class labels. Diversification delivered less protection than the asset-class count implied, and when something turned, losses would have spread rather than stayed isolated. In this sample nothing turned. That is luck, not design.
The fixes are specific: a standing tail hedge such as index puts rather than relying on directional books for protection, correlation-aware caps so clustered themes share one risk budget, and scenario testing (oil down 15%, an AI de-rating, yields plus or minus 50bp) run before the downside arrives rather than after.
The S&P 500 returned roughly +4.3% over the window, so +15.60% is about 3.7x equity beta, and the outperformance came from single-name selection rather than index exposure. Cash returned +0.43%.
One qualification belongs on the record: the S&P GSCI was up roughly 38.7% year to date to end-May, energy-led, so part of the energy P&L is commodity beta available to any long-commodity position. The equity and event-driven books were the genuine alpha.
The game assumed mid execution with no costs, so I tested that. A deliberately conservative flat 0.25% on roughly $1.1bn of traded notional costs about $2.75M and still leaves +12.85%. An asset-class-aware estimate, about 4bp blended, costs roughly $0.44M for +15.16%. Under either, the book stays ahead of the second-placed +$8.68M, so the result is not an artefact of frictionless execution.