Institutional investors commonly separate strategic asset allocation (SAA), which establishes a long-run policy or reference portfolio, from tactical asset allocation (TAA), which takes shorter-horizon active positions around that benchmark. We represent this familiar institutional architecture through a canonical two-step mean–variance formulation. The separation is useful for governance, delegation, risk budgeting, and implementation, but repeated sequential optimization need not be dynamically optimal when positions persist, signals evolve, and changing exposures is costly.

We formulate SAA–TAA as a benchmark-relative dynamic optimization problem with predictable returns and trading costs. The optimal policy trades partially toward a forward-looking aim portfolio and explicitly accounts for the continuation consequences of today’s trades. We derive exact policy-evaluation results that quantify the recursive policy-value loss from repeatedly following the conventional two-step SAA–TAA policy, whose tactical step applies the one-period TAA rule, relative to the dynamically optimal policy. We also derive an exact signal-persistence threshold separating attenuation from amplification of tactical views relative to conventional TAA and characterize the effects of future uncertainty and trading costs. Finally, we construct a simple, implementable family of modified two-step SAA–TAA rules and choose the rule that minimizes exact recursive policy-value loss within that family. The results show how an institution can preserve the familiar strategic–tactical architecture while recovering part of the value of dynamic optimization.