This study develops a unified empirical framework to identify, characterize, and validate speculative (exuberant) episodes in the Romanian equity market, contributing to the growing literature on bubble dynamics in emerging markets. It aims to provide a comprehensive assessment of whether speculative behavior is persistent or episodic and to evaluate its implications for financial stability.
Using ten years of daily data for the six major Bucharest Stock Exchange indices, the study combines recursive right-tailed ADF tests, the nonlinear log-periodic power law singularity (LPPLS) model and a Markov-switching autoregressive specification. This multi-model approach enables the detection, timing and validation of exuberant episodes from complementary econometric perspectives.
The results reveal recurrent but non-persistent bubbles, with three major waves (2017–2018, 2019–2020 and 2023–2024) associated with global liquidity conditions, pandemic-related uncertainty and post-crisis recovery. Broad and total-return indices exhibit stronger and more persistent signals, while sectoral indices show shorter episodes. Speculative phases are episodic, synchronized in 2023–2024 and rapidly self-correcting, indicating systemic but contained dynamics without prolonged instability.
First, we depart from prior research by redirecting analytical attention from individual firms to the broader market, providing a genuinely aggregate perspective on speculative dynamics. Second, we offer the first systematic examination of bubble behavior across all major Romanian equity indices, allowing for a macro-level assessment of the structure, timing and persistence of exuberant episodes. Third, our contribution is reinforced by an exclusive and rigorous focus on bubble identification and verification, employing a comprehensive set of econometric procedures designed to enhance methodological reliability and cross-study comparability. Finally, our manuscript contributes to the literature by extending the scope of comparative bubble diagnostics beyond the standard pairing of an ADF-type measure and the LPPLS framework.
