Multi-Period Portfolio Optimization and Execution Costs

Direct answer: What is multi-period portfolio optimization?

Multi-period portfolio optimization plans changes across time instead of forcing the entire target in one trade list. It balances future targets, transaction costs, taxes, liquidity, drift, and execution risk.

How Celestice helps

Celestice can run candidate optimization methods with shared inputs, constraints, taxes, stress tests, and diagnostics. In practice, Celestice helps teams compare the trade-offs, choose deliberately, and preserve evidence before an optimized portfolio becomes an action.

One snapshot is not a strategy

Almost every method in this series optimizes a single rebalance: given today's state, find today's best portfolio. That's necessary but incomplete. Investing is not one decision — it's a sequence of them, each affecting the next: today's trade changes tomorrow's starting point, trading costs accumulate, and forecasts evolve. Trade aggressively toward the optimum today and you may incur costs that a patient, multi-step path would have avoided. The portfolio you should hold today depends on where you expect to go tomorrow.

The execution-aware family treats optimization as a trajectory over time rather than a snapshot. It is the most operationally sophisticated corner of the gallery, and as always, its output is reviewed research behind the approval gates.

Single-period optimization with transaction costs

The bridge from snapshot to trajectory is single-period optimization with transaction costs. It still optimizes one rebalance, but accounts honestly for the cost of getting there — spread, commission, and market impact priced into the objective so the optimizer only trades when the benefit exceeds the cost. This alone fixes the most common naive-optimizer mistake: churning the book chasing a marginally better allocation that the trading costs more than erase. When trading is costly, the optimizer rationally tolerates some drift from the ideal rather than chasing it.

The engine builds this on a dedicated execution-aware backend with a proper execution model — covariance and return forecasts, leverage and turnover limits, and a transaction-cost term — so the "optimal" trade is the one that's optimal after you pay to make it.

Multi-period optimization

Multi-period optimization is the full idea: plan a sequence of rebalances over a horizon, optimizing the whole trajectory at once and recognizing that the best move today depends on where you expect to be tomorrow. This matters whenever the path has structure — a known cash inflow to invest gradually, a position to unwind without moving the market, a glide path toward a target allocation, a position that needs trimming for tax reasons next quarter, or a forecast that decays over time.

The classic intuition: a single-period optimizer might make a large trade today and reverse part of it next month, paying costs twice; a multi-period optimizer sees the round trip coming and trades more patiently. It produces not a single target but a plan — trade this much now, more next period, converge to the target over the horizon. Over many rebalances, that foresight compounds into materially lower costs and steadier risk. It's the difference between "where should I be?" and "how should I get there?"

Execution-aware backtesting

A plan is only trustworthy if you can test it under realistic conditions, which means simulating it the way it would actually have traded. Execution-aware backtesting runs a strategy through a market simulator that applies real-world frictions — transaction costs, market impact, slippage, and the actual realized path of prices — rather than assuming frictionless fills at closing prices. This is the crucial discipline that separates a real backtest from a fantasy: the difference between "here is the return if trading were free and instantaneous" and "here is the return after the frictions you would really have paid." A strategy that looks brilliant frictionlessly and mediocre under execution-aware simulation is mediocre — the simulation is telling the truth.

The review discipline is crucial here: a backtest is evidence, not a promise. Check the cost and slippage assumptions, and never confuse a backtested path with a guaranteed future one.

Log-growth (Kelly) sizing

For long-horizon compounding, the gallery offers Kelly-criterion (log-growth) sizing, which sizes positions to maximize the expected logarithm of wealth — the growth-optimal bet size over many periods — rather than single-period mean-variance utility. It's powerful and theoretically elegant, but aggressive: full-Kelly portfolios endure stomach-churning drawdowns. In practice it's used fractionally (half-Kelly, quarter-Kelly) or with risk controls, as a long-horizon growth lens rather than a literal prescription. The review focus is making sure the sizing matches the investor's genuine tolerance for the wild ride full Kelly implies.

What to review

The execution-aware family rewards careful reading of its assumptions:

• Are the cost assumptions realistic? The whole value of these methods is honest cost modeling; optimistic spread or impact estimates make them worse than useless. Confirm the inputs.
• Does the horizon match the decision? Multi-period only helps when the path genuinely has structure; for a one-off rebalance, single-period-with-cost is the right, simpler tool.
• Is the backtest's realism credible? Slippage, impact, and the realized price path drive the result — review them before trusting the equity curve.

When the trajectory matters

Reach for execution-aware methods when how you trade materially affects the outcome: large positions that move the market, gradual deployment of new capital, unwinding without signaling, cost-sensitive high-turnover strategies, or any mandate where the trading path — not just the destination — is part of the job. For a small, liquid, low-turnover or buy-and-hold portfolio, single-period optimization captures most of the value; for everything else, the trajectory is the strategy.

The takeaway

Real investing is a sequence, not a snapshot. Single-period-with-cost optimization stops the churn, multi-period optimization plans the whole trajectory, execution-aware backtesting tests it against trading reality, and log-growth sizing addresses long-run compounding. Honest cost assumptions are everything. In the finale: how to choose among the dozens of methods in this series, compare them fairly, and govern the result — turning a gallery of tools into a disciplined decision.