Genghis,
velocity is not an observable in a GPS (neither are coordinates, for that matter, but let's not get picky), it is simply derived as the difference between two positions divided by time. From there, and having a precision figure for your time and position "measurements" (again, let's consider the position as a direct measurement) it's trivial to assert the precision of a speed readout by a simple error propagation computation.
Let A1 be your GPS position at time T1
Let A2 be your GPS position at time T2
Your velocity is V = (A2-A1) / (T2-T1)
Now:
Let EP1 be a measure of the uncertainty of A1
Let EP2 be a measure of the uncertainty of A2
Let ET1 be a measure of the uncertainty of T1
Let ET2 be a measure of the uncertainty of T2
An uncertainty value "EV" for V can be calculated as
EP = SQRT(EP1^2 + EP2^2)
ET = SQRT(ET1^2 + ET2^2)
EV = SQRT((EP/(A2-A1))^2 + (ET/(T2-T1))^2)
Notes:
* EP[12] is a suitable measure of positional uncertainty, such as SEP (Spherical Error Probable) or CEP (Circular Error Probable)
* ET[12] can probably be considered negligible, in which case the formula above is reduced to EV = SQRT(EP1^2 + EP2^2)/(A2-A1). Note that this is a relative error. Use SQRT(EP1^2 + EP2^2) for absolute error (assuming ET[12] ~= 0)
* I've done the error propagation above from memory. Needless to say, check the formulas! (is "formulae" too pedantic, btw?)
Hope this helped and got nothing too wrong