Tax-aware multi-account asset location

Assets are held across taxable, tax-deferred, and tax-exempt accounts with different account-level tax treatment, subject to household-level allocation constraints on stocks, bonds, and cash. The decision variable is placement: where each asset class is held.

It is a coupled allocation problem: moving bonds into one account forces an offsetting move elsewhere to satisfy the household constraint. And tax drag is path-dependent, driven by realized returns, rebalancing, and withdrawal timing, so placements cannot be ranked by a single-period calculation.

A stylized, deterministic allocation subproblem. Let $x_{a,k} \ge 0$ be the dollars of asset class $a$ held in account $k$:

$\mathcal{A}$

Asset classes: stocks, bonds, cash

$\mathcal{K}$

Household accounts, each with a tax treatment

$B_k$

Balance of account $k$

$\pi_a$

Household target weight of asset class $a$

$\tau_{a,k}$

Expected tax drag per dollar of asset $a$ in account $k$

The first constraint fills each account; the second holds the household allocation, and it is the one that couples the accounts. The problem is linear in this form and remains convex under standard extensions such as penalized rather than hard allocation targets.

1. Inputs. Account balances and tax treatment, household target allocation, capital-market assumptions.
2. Optimization. Solve (1)–(2) for the asset placement.
3. Stochastic evaluation. Simulate the optimized placement and the baseline placement policy over 30-year Monte Carlo paths.
4. Comparison. Annualized tax leakage, optimized placement versus baseline placement policy.

• Formulated and implemented in Python; solved with Gurobi and CVXOPT.
• Stochastic simulation for the 30-year evaluation.
• Architected and deployed at Fidelity while leading a four-person quant team.

The 12 bps/yr is the model result: annualized tax leakage relative to the baseline placement policy, from stochastic evaluation over a 30-year Monte Carlo horizon. The ≈3.5% is not a simulation output; it is the compounding implication of a persistent 12 bps annual reduction at a 4% annual return:

Because asset location changes where assets are held rather than what is held, the improvement comes without changing the household's allocation or market exposure.

• Capital-market assumptions. Results are conditional on the assumed return and risk inputs.
• Tax rules. Tax treatment is assumed to persist as modeled over the full horizon.
• Contributions and withdrawals. Cash-flow timing is an assumption, not an observation.
• Rebalancing. Results depend on the assumed rebalancing policy.
• Account mix. The benefit scales with how differently the household's accounts are taxed; a household with one account type has no location decision.
• Projection uncertainty. A 30-year projection is a distribution, not a forecast.