Hierarchical risk parity (HRP) methods have been recently proposed to solve some of the limitations of Markowitz’s classical mean-variance portfolio (MVP) selection model. Most of the evidence on the comparative performance is based on synthetic data and needs an assessment of the length of the past data required to construct the model. In this paper, we focus on real daily market data (S&P 500, Euro Stoxx 50) and synthetic data to provide additional empirical evidence on the relative performance of HRP models and three benchmarks such as the mean-variance model, an inverse variance approach, and an equally weighted portfolio. Using an out-of-sample performance comparison, we pay special attention to the estimation window used to construct the models. Our results show that the best out-of-sample Sharpe ratio is achieved when the estimation window includes five years of daily data. In addition, we find that HRP outperforms two usual benchmarks, such as the inverse variance approach and an equally weighted portfolio. However, the MVP is the best option for the Euro Stoxx 50 data regarding the Sharpe ratio and the best choice regarding usual risk measures.