Bayesian nonparametric methods rely on Markov Chain Monte Carlo (MCMC) techniques, which assume a normal prior to derive posterior distribution under conditional probability. This assumption, however, often misrepresents actual prior beliefs because they are rarely normal and subject to adaptive decision-making. This paper proposes a novel Bayesian model for daily stock prices that integrates a Dirichlet Process Mixture (DPM) model into MCMC framework. The DPM component identifies the optimal number of clusters within the data and reveals latent sub-distributions while the MCMC framework provides robust estimation of transitional state probabilities in the posterior. Our results demonstrate that prior distributions are dynamically updated during estimation process, and we quantify the model’s sensitivity to evolving market conditions. Forecasting accuracy is evaluated and benchmarked against competitive models. Our integrated MCMC–DPM approach achieves substantial improvements in predictive accuracy, with RMSE reductions of 37.9% and MAPE reductions of 36.4% compared to naive benchmarks for the S&P index, and even larger gains of 44.3% and 52.2% respectively for more volatile stock prices. These results demonstrate that our proposed framework offers an enhanced estimation strategy capable of modeling stock prices characterized by complex, multi-modal distributions while provides more reliable forecasts for risk management and regulatory decision-making.