Purpose>The authors examine the stability of dividend payout and the consistency in its predictability using sample of firms from 18 different countries amid their prevailing heterogeneous formal institutions (such as the legal system, corporate governance), the distinct state of economic development (developing vs developed) and changing times (during the crisis vs the noncrisis periods).Design/methodology/approach>The authors use tobit regression models with distinct specifications for the authors’ investigations. The authors alternately analyze the study’s results using Fama–Macbeth (FM) (1973) and generalized least square (GLS) regressions.Findings>The authors show a sharply declining stability in dividend payout with time using DeAngelo and Roll’s (2015) framework. In terms of predictive consistency, the authors report that only a few idiosyncratic factors predict dividends consistently, and these results hold qualitatively true across the robustness analysis. The firm's liquidity appears to be the most consistent predictor of dividends payout, whereas firm's size being on the other extreme. The results signify that the idiosyncratic factors that matter for firm's dividend policy are not country specific. Instead, it reveals commonality of predictors grounded on characteristics of countries such as legal environment, investor's protection, economic state (ES) and economic cycle.Originality/value>The authors contribute to the dividends literature by providing the evidence of dividend instability through time and disapproving the stylized fact of sticky dividends. Besides, the authors provide international evidence of inconsistent predictability of dividends.