The Mathematics of Total Portfolio Allocation: Payoff-Space Invariance, Shadow Prices, and Dynamic Capital Allocation

The total portfolio approach (TPA) asks an institutional investor to allocate capital, risk, liquidity, and organizational attention according to each opportunity's contribution to the total fund rather than to an asset-class sleeve. The practice is increasingly important, but its mathematical content is often described as a mindset rather than a theory. This paper supplies a formal foundation. The primitive object is an attainable surplus payoff in a function space; securities, factors, strategies, and asset classes are treated as alternative coordinate systems. This formulation yields a payoff-space invariance principle: economically equivalent presentations produce the same implemented payoff. A convex-analytic converse identifies payoff and cost equivalence from the values of all continuous linear allocation problems, and an exact descent criterion characterizes which polyhedral mandates are representation invariant. It also exposes why non-aggregated bucket and position constraints can make strategic asset allocation depend on arbitrary labels. Within a liability-aware quadratic model, the total-fund solution is obtained in closed form and interpreted as a metric projection. The resulting distance identity measures the certainty-equivalent cost of governance restrictions.A canonical lossless compression of TPA into strategic asset allocation is obtained when the sleeve-level objective equals the optimized within-sleeve value function; an approximation theorem bounds the loss from imperfect compression.Further results derive a shadow-adjusted acceptance rule for new opportunities, Euler allocations for total risk, a quantile rule for liquid reserves, an ellipsoidal robust counterpart, stability bounds under estimation error, and a nonseparable dual-price decentralization result for multi-team institutions. A dynamic formulation connects total portfolio decisions to Bellman, Hamilton--Jacobi--Bellman, and induced Fokker--Planck equations and yields an exact two-state linear--quadratic--Gaussian policy with a stochastic predictor.A reproducible hypothetical example shows how the framework separates the economic benefit of flexibility from turnover and constraint costs. TPA thereby becomes a testable mathematical program: optimize in payoff space, price every scarce resource, and treat governance as part of the feasible set rather than as an informal overlay.