A new 45-industry, 23-year, dataset for Japan is used to investigate the duration of industry leadership. A new scaling relationship linking a firm´s current market share with the standard deviation of market share changes is reported. This relationship discriminates in a powerful way between rival candidate theoretical models of market share dynamics. It also makes possible a useful simplification in testing a benchmark model of a Markovian kind. Relative to that model, it is found that at least some industries display a `Chandlerian´ bias toward longer durations of leadership than would be present in the benchmark model.