SI412: Portable Alpha: Asking the Questions That Matter ft. Harry Moore

Core argument

Harry Moore's central point: portfolios that appear diversified — equities, bonds, some private allocation — frequently are not diversified in the specific scenario investors are trying to insure against. If the AI capex trade cracks and the S&P falls 30-40% rather than 5-10%, Moore argues equities, private marks (which "slow your marks" but eventually take hits), and even bonds/gold (already correlated with equities earlier this year) offer little protection. Trend following, in his view, is structurally different because it holds no persistent long-term view — it can flip short equities, short bonds, or short the AI theme itself (semis, tech hardware) if the price trend reverses, offering protection specifically in that left tail rather than merely low average correlation.

The mechanism

Moore frames the AI trade as genuinely two-sided: comparisons to railroads and dot-com overbuild are a "gripping narrative," but he also cites real productivity gains inside Man Group's own quant research from AI tools, calling the uncertainty real rather than resolved. The immediate evidence he points to: a hedge fund run by Leopold Aschenbrenner reportedly grew to roughly $45bn in assets with limited track record and high leverage, then lost about two-thirds of that in a matter of weeks as the AI trade reversed in July — a book that looked market-neutral on paper but was effectively concentrated on one side of the AI theme. Separately, the yen saw what Moore and host Niels Kaastrup-Larsen estimated as a roughly 10-standard-deviation move after BOJ/Fed intervention near July month-end, illustrating how policy action, not just price momentum, can whipsat trend books.

On performance, Moore noted July's dispersion came from a stall in the AI/semis trade plus a rally in energy; managers long equities generally did well, those long semis/short software or short energy generally suffered. His own trend barometer read 27 (weak) into early August. Speed mattered: a standard 20/60/130/260-day lookback comparison showed 20- and 60-day models down year-to-date, 130-day slightly positive, 260-day more strongly positive — slower models benefited from staying persistently long stocks, long energy, and short bonds through the year's chop. Market universe also drove dispersion: alternative energy markets (Nordic and Italian power) stayed more range-bound than crude and heating oil, this year favoring traditional over alternative energy exposure; fixed income was harder the further from core government bonds (e.g., Polish rate swaps versus 10-year Treasuries); and in equities, cross-sectional constructions such as quality, value, and momentum factors, and semis-versus-software splits, drove a separate source of dispersion from simple long-index trend models.

Monthly figures cited: BTOP50 -0.67% in August (as of Tuesday), +7.17% YTD; SG CTA Index -0.88%, +7.15% YTD; SG Trend Index -0.94%, +6.89% YTD; SG Short-Term Traders Index +0.14%, +3.32% YTD. Traditional markets: MSCI World +2.77% in August, +12.45% YTD; the ex-US/Canada variant +1.83%, +11.8% YTD; S&P U.S. Aggregate Bond Index +0.64%, +0.31% YTD; S&P 500 Total Return +3.12% in August, +12.83% YTD.

What has to be true

Moore's featured research (a Man Group paper co-written with Jonathan Smith and Chris Pye, using SG Trend as the illustrative alpha) sets conditions for portable alpha structures to actually deliver on the diversification thesis rather than repeat 2008-style failures. Because trend needs only roughly $20-25 of margin per $100 of notional exposure, most of the capital is freed to fund an alpha layer atop a fully-held equity position — but that freed cash still needs a buffer. The paper's finding: roughly 40% total cash support for the beta leg (10% margin, 30% unencumbered) reduces the probability of a forced de-gearing event to about once every 100-200 years, based on Monte Carlo simulation of tail outcomes beyond the single historical path. In actual live/historical data over 26-27 years, a 20% buffer would have been fully drawn down exactly three times — the dot-com bust, the GFC, and COVID — and being forced to sell equities near a 2009 low would have missed roughly 29-30% of subsequent upside. Cutting the buffer to 20% "massively increases the chance" of being knocked out, per Moore; some investors with callable capital or credit lines may accept that risk, but most institutional allocators he deals with do not want ad hoc margin calls.

On rebalancing, the paper found frequency (monthly, quarterly, tolerance-based) makes little difference over the long run, but it mattered acutely in specific crises: in 2008, every rebalancing rule except a pure buy-and-hold underperformed, because rebalancing meant selling a rising trend allocation to buy more into a falling equity market. Moore's practical compromise is a tolerance-based rule (e.g., rebalance only if the mix drifts more than roughly 10 percentage points) rather than fixed-calendar rebalancing. On implementation, futures versus swaps showed no structural performance edge once implicit (futures) and explicit (swap) financing costs are accounted for — a no-arbitrage result — leaving the choice to operational preference and counterparty-risk tolerance.

On the leverage objection — that stacking alpha atop 100% beta simply adds risk — Moore's answer is that leverage is a portfolio-construction tool, not inherently dangerous: in a standard 50/50 equity/bond portfolio, roughly 85% of risk still comes from equities, so genuine risk balance requires leveraging the lower-return bond leg. For the alpha layer specifically, trend qualifies because its correlation to equities turns negative in the worst 5% of equity-return periods (not just on average), and because trend's volatility-scaling mechanically reduces position size and margin need as markets fall — the opposite of liquidity-provision strategies (short vol, merger arbitrage, treasury basis trades) that look attractive in normal times but can be hit hardest in the same liquidity crisis a tail hedge is meant to survive.

A secondary segment covered an academic paper by Nicholas Paulson and Vadim Sokolov linking the 1980s Turtle trading rules to later decision-theory literature: the Turtles' range-based volatility estimator maps to the Parkinson estimator; their unit-based position sizing is an early form of volatility-based risk parity; their stop-loss discipline (2N, roughly 2% of equity per position) sits deliberately to the conservative side of the Kelly criterion (citing Ed Thorp); correlation was managed via position caps rather than covariance matrices; and their drawdown rule (cut size 20% after a 10% loss) anticipates the Grossman-Zhou optimal drawdown-control literature of the 1990s.