The portfolio is constructed from explicit, scenario-conditional views of the world. Each asset's expected return is derived from a fair-value model anchored to macro inputs. The allocation that emerges is the one that minimises regret across scenarios, subject to factor and risk constraints.
A small set of forward macro states drives everything downstream. Bear, Base, and Bull are not equal-weighted: the central case carries the bulk of probability mass, with tails sized to reflect current risk distribution around the central case.
Each scenario fixes a coherent 12-month forward state for growth, inflation, policy, and risk appetite. The portfolio is then optimised for blended performance across all three.
Macro scenarios feed asset-level fair-value models. Equity returns are predicted by an ElasticNet regression on macro levels, year-on-year changes, earnings yield and realised volatility. Credit spreads come from a three-sub-model ensemble: a cycle/labour panel, a rates/curve panel, and a stress panel with data-driven hockey-stick triggers.
The output is a 12-month expected return for every position under every scenario. This is what the optimiser sees, rather than historical averages.
One hundred thousand candidate portfolios are simulated under group and factor constraints. Each is evaluated by its regret (the gap between its outcome and the best feasible outcome) under every scenario. The chosen allocation balances worst-case regret against probability-weighted regret.
The frontier traced below is the locus of regret-best portfolios across CVaR buckets. The selected portfolio is the point on the frontier whose factor exposures, group caps, and target return all bind cleanly.
Two assets that look different on a holdings sheet (a high-yield bond and a small-cap equity, say) often share the same underlying risk. The optimiser decomposes every position into its factor exposures and caps how much portfolio variance any single factor can contribute.
This is what stops a "diversified" book from being a leveraged equity bet wearing different labels.
The numbers shown here are illustrative.