On Asymptotic Log-Optimal Buy-and-Hold Strategy

TL;DR

This paper demonstrates that in markets with a dominant asset, a buy-and-hold strategy on that asset is asymptotically optimal for logarithmic wealth growth, extending to model-agnostic scenarios and discussing high-frequency rebalancing conjectures.

Contribution

It proves the asymptotic log-optimality of buy-and-hold strategies in markets with a dominant asset, even without probabilistic models, and explores high-frequency rebalancing conjectures.

Findings

Buy-and-hold on a dominant asset is asymptotically log-optimal.

The result extends to markets without probabilistic return models.

Support for the high-frequency maximality conjecture is provided.

Abstract

In this paper, we consider a frequency-based portfolio optimization problem with $m \geq 2$ assets when the expected logarithmic growth (ELG) rate of wealth is used as the performance metric. With the aid of the notion called dominant asset, it is known that the optimal ELG level is achieved by investing all available funds on that asset. However, such an "all-in" strategy is arguably too risky to implement in practice. Motivated by this issue, we study the case where the portfolio weights are chosen in a rather ad-hoc manner and a buy-and-hold strategy is subsequently used. Then we show that, if the underlying portfolio contains a dominant asset, buy and hold on that specific asset is asymptotically log-optimal with a sublinear rate of convergence. This result also extends to the scenario where a trader either does not have a probabilistic model for the returns or does not trust a model obtained from historical data. To be more specific, we show that if a market contains a dominant asset, buy and hold a market portfolio involving nonzero weights for each asset is asymptotically log-optimal. Additionally, this paper also includes a conjecture regarding the property called high-frequency maximality. That is, in the absence of transaction costs, high-frequency rebalancing is unbeatable in the ELG sense. Support for the conjecture, involving a lemma for a weak version of the conjecture, is provided. This conjecture, if true, enables us to improve the log-optimality result obtained previously. Finally, a result that indicates a way regarding an issue about when should one to rebalance their portfolio if needed, is also provided. Examples, some involving simulations with historical data, are also provided along the way to illustrate the~theory.