Résumé : Sizing frequency control ancillary service (FCAS) requirements is crucial for the reliable operation of power systems amid a continuous influx of variable renewable energy (VRE) generation. Reserve sizing is especially pertinent for the Philippine grids due to an expected transition to new FCAS classifications established by its Grid Code. In lieu of the existing deterministic formulation, this work proposes a dynamic approach for sizing secondary and tertiary reserves that accounts for the stochasticity and variability of load demand and VRE. We propose a method where historical power imbalances were calculated and clustered according to the time and day of week they occurred. The conditional probabilities of forecast and noise errors were characterized using kernel density estimation. Recursive convolution was performed to obtain the total reserve requirement probability distribution. The method was tested on Visayas grid's historical system operation data and used target reliability levels on the error distributions to size upward and downward reserve needs. Finally, the methodology was extended to demonstrate through a numerical experiment that sizing FCAS at temporal resolutions higher than one-hour, e.g., five-minute, provides the benefit of shrinking the required capacities by as much as 86.2\% compared to current deterministic FCAS sizing.