As a former component engineer and failure analyst for electronic circuits/assemblies, I’d like to comment on this topic. What’s being discussed here is known as voltage derating, and for aluminum electrolytic capacitors (the most common power supply bulk filtering capacitors) derating of more than 20% of maximum continuous voltage is rarely necessary unless the capacitor is likely to see frequent surges in excess of its surge rating. Even in high temperature applications it’s hardly ever necessary to derate more than 50%. So, for example, a capacitor on a 12V input (that might see occasional excursions to 15V) would require 15V/0.8 = 18.75V (a 20V part would be appropriate) for 20% derating, and at 50% derating a 35V rating is more than enough (15V/0.5 = 30V).
Selecting a part with significantly higher voltage is acceptable if larger size, higher ESR, and higher dissipation factor (with its accompanying higher leakage current) are also acceptable. The aluminum electrolytic capacitor’s dielectric is a very thin aluminum oxide layer that is maintained by the applied voltage, so after a short period of operation a 100V part operated at 20V will have essentially the same dielectric thickness (and strength) of a 25V part operating at the same 20V. There’s generally no need to select a part rated for more than twice the highest anticipated long-term DC voltage.
If it is necessary to guard against momentary high voltages spikes/surges, zener diodes and TVS (transient voltage suppressors) are much better choices.
(See this Application guide from Cornell Dubilier for more criteria for selecting aluminum electrolytic capacitors: http://www.cde.com/resources/catalogs/AEappGUIDE.pdf )
(For additiional capacitor selection guidance, see this