This paper focuses on the empirical analysis of performances between Bayesian estimation technique and certainty-equivalence-tangency technique varying sample size. In limited sample sizes (less than 60 when using monthly data, less than 160 when using weekly data), Bayesian estimation technique shows better performance measured by Sharpe ratio, though in extremely limited sample sizes, certainty-equivalence-tangency technique shows a little bit better performance when weekly data used. This is because estimation errors overweigh the performance differences.
This paper also examines the trade-off between sample size and population shift, During tested time span, 10 years ('91.9-'01.8), population shifts begin to occur in 6 months and show reverting tendency every 30 months. Consequently, using 6 month-sample period shows the highest performance.
In order to take full advantage of Bayesian estimation technique, we need to shift Bayesian estimation technique dominant sample periods to optimal sample period derived from trade-off between sample size and population shift by choosing optimal sample size per sample period.
