Does Portfolio Resampling Really Improve Out-of-Sample Performance? Evidence from the Brazilian and Us Markets

Markowitz optimization, a key component of modern portfolio theory, is prone to estimation errors in mean vector and covariance matrix, leading to unrealistic portfolio weights, limited diversification, and weak out-of-sample performance. In response, a heuristic portfolio resampling approach proposed by Michaud (1998) offers potential for enhanced diversification and out-of-sample performance. This approach has drawn mixed feedback from scholars and practitioners, and its economic benefits remain unclear. To contribute to this discourse, we evaluate the performance of Michaud's resampling approach using recent data from Brazilian and US stock markets. Furthermore, we introduce a novel resampling scheme called factor-based portfolio resampling, which leverages the factor structure of stock returns. Our findings indicate that portfolio resampling, including factor-based resampling, can serve as a practical means to enhance diversification, lower transaction costs, and improve out-of-sample performance in both Brazilian and US market contexts