The Statistical Problem

There is a deep and underappreciated statistical problem with detecting genuine alpha. Returns are noisy. Even a manager with true positive alpha of 2% per year will, in any given year, underperform or outperform by much more than 2% due to random variation. Separating signal from noise requires many years of data.
For a fund with annual return volatility of 15%, detecting a true alpha of 2% per year at conventional statistical significance (p < 0.05) requires roughly 55 years of data. Even a fund with a strong alpha of 4% per year would need around 14 years of data before that outperformance could be statistically distinguished from luck. Very few fund managers have track records long enough to make this determination.
This means that when you look at the distribution of fund returns at any point in time, you are largely looking at the distribution of luck, not skill. The few managers who appear at the top of performance tables may be genuinely skilled, or may simply have drawn lucky random outcomes.