This paper compared the total factor productivity (TFP) estimation methods
of growth accounting (GA) and the stochastic frontier model (SFM). In
specific, we were interested in comparing the TFP distribution with the
variance and skewness of these two methods. We used the Mining and
Manufacturing Survey from the Statistics Korea, and compared the
distributions of the TFP estimates obtained by GA and SFM. We found that
the two estimation methods resulted in significantly different findings. First,
GA produced larger variations in TFP than SFM. GA estimates TFP based on
a residual of a production function while SFM attempts to decompose an error
term into TFP and random noise. This decomposition in SFM may result in
smaller variations in the TFP estimates. Second, the two estimation methods
also identified different findings regarding the relationship between market
competition and TFP dispersion. According to GA, there is no clear
relationship between market competition and TFP dispersion. Alternatively,
SFM identified a negative relationship between the TFP dispersion and market
competition. This empirical finding of SFM is consistent with previous
empirical studies that suggested competitive environment forces less productive
firms to exit the market, resulting in narrower TFP distribution.