Beyond Forecasting: Recasting Volatility Control as a Routing Problem

Traditional volatility control often relies on a fixed estimator and a single, pre-specified control rule to manage portfolio exposure. This approach assumes that a single mapping from risk estimates to positions is sufficient across all market regimes. The authors question this assumption, arguing that volatility control is a decision-making problem where the optimal response to risk depends on the current market state. They propose VolRouter, a framework that treats volatility control as a routing problem, selecting the best estimator-controller pair from a library based on current market conditions.

VolRouter introduces a modular architecture that separates volatility control into three distinct stages: state inference, switch review, and pair selection. The system summarizes market data into a control-relevant state profile, which then informs the Router's decision to either maintain the current policy or switch to a more appropriate one. The Router can be implemented using various methods, including rule-based logic, machine learning, or even Large Language Models (LLMs). Crucially, the system evaluates candidate policies relative to one another and incorporates persistence constraints to prevent unnecessary, high-turnover switching.

Empirical testing across S&P 500, Multi-Asset, Bitcoin, and USDT datasets demonstrates that VolRouter consistently achieves higher Sharpe ratios compared to traditional fixed-rule approaches. For instance, in the S&P 500 setting, VolRouter improved the Sharpe ratio from 0.952 to 1.222 while simultaneously reducing maximum drawdown and daily CVaR. The results indicate that these performance gains stem from better risk allocation rather than simply increasing raw returns. Ablation studies confirm that the system's success relies on the ability to perform relative policy evaluation and maintain selective, persistent switching, rather than merely having a larger library of policies to choose from.