Model Predictive Control for Budget Allocation Under Non-Stationary Returns
A recent study on arXiv explores the allocation of finite-horizon budgets as a closed-loop economic control challenge, contrasting receding-horizon Model Predictive Control (MPC) with reactive budgeting strategies. Budgets are distributed periodically amid execution noise and operational limitations, and return efficiency may change over time. The researchers utilize a controlled simulation framework inspired by digital marketing to assess reactive pacing against MPC in increasingly non-stationary environments. Findings indicate that non-stationarity alone does not warrant predictive control; when return dynamics are either stationary or shift through unpredictable stochastic drift, MPC does not provide significant advantages over reactive methods. Conversely, when return efficiency shows a predictable pattern over the planning horizon, MPC effectively leverages intertemporal trade-offs, outperforming reactive budgeting.
Key facts
- Study evaluates MPC vs reactive budgeting for budget allocation under non-stationary returns.
- Budget allocation is framed as a closed-loop economic control problem.
- Simulation framework is motivated by digital marketing.
- Non-stationarity alone does not justify predictive control.
- MPC offers no systematic advantage when returns are stationary or drift unpredictably.
- MPC outperforms reactive budgeting when return efficiency has predictable structure.
- Predictable structure is captured through an underlying model.
- MPC exploits intertemporal trade-offs.
Entities
Institutions
- arXiv