The estimator is built from ln(High/Low) squared, scaled by 1/(4 ln 2), then annualised. Because the daily range carries more information about how far the price wandered than the close alone, you get a similar-quality estimate from far fewer bars.
Its blind spot is the overnight gap. A stock that closes at 100 and opens at 90 after earnings, then trades in a tight 90 to 91 range, will show almost no Parkinson volatility for that day despite a 10% move having happened. For gap-prone instruments this understates risk badly.
It also assumes continuous observation. Real highs and lows are sampled from discrete trades, so the observed range is slightly narrower than the true one, biasing the estimate low, more so in thin markets.
Related: garman-klass-volatility, yang-zhang-volatility, close-to-close-volatility, realised-volatility