Abstract

Parameter uncertainty and time-varying dependencies among assets pose challenges to portfolio optimization in financial markets, particularly in emerging markets. This study evaluates an adaptive framework for portfolio optimization on the Tehran Stock Exchange by integrating the Black–Litterman model, a dynamic vine copula, and the Proximal Policy Optimization (PPO) algorithm. Within this framework, Black–Litterman is used to estimate expected returns, the vine copula is employed to model time-varying dependencies among assets, and PPO is used to learn a sequential portfolio-weight allocation policy. The dataset comprises 180 monthly observations for 112 firms listed on the Tehran Stock Exchange over the period 1390–1404. Following the initial estimation window, the performance of seven strategies was compared using 156 out-of-sample monthly observations. In addition, a paired block bootstrap with a six-month block length was employed to assess the uncertainty surrounding differences in Sharpe ratios, while an ablation analysis was conducted to examine the contributions of the framework’s components. The results show that DCBL_PPO achieved a Sharpe ratio of 0.743, a Sortino ratio of 1.205, and a compound annual growth rate of 15.16%, demonstrating more favorable point-estimate performance than the competing strategies. The strategy also exhibited a more favorable maximum drawdown. The ablation analysis further indicated that the performance improvement was observed primarily in the final configuration following the introduction of PPO. The empirical novelty of this study lies in the integrated evaluation of these three mechanisms within an adaptive framework, the comparison of its out-of-sample performance with alternative strategies, and the examination of result sensitivity to component composition and statistical uncertainty. Overall, the findings support the potential of this framework to improve portfolio performance in the market under investigation.