If one were using the rational substitute for forecasting (anticipatio) the best characterization, mathematically pseaking, would by to use the philosphy of history known as "a symmetric philosophy of history". Google: Danto, Columbia University, NYC The axis of symmetry is now give or take a few milliseconds. You first responder actually has it backwards with respect to a point in time if he even had a focus on time at all.
My point was that forecasting can be done either in linear fashion (regression, for example), or non-linear fashion (genetic algorthims, neural nets). The question did not address the specifics of what exactly "short-term" means. Are we talking point to point? Is it ticks (no time element), minutes, hours, days? I claim no expertise, but the research I've been made aware of seems to point to non-linear methods have better success forecasting price movement in the financial markets. I have yet to be convinced of that in a meaningful fashion, by the way, but neither am I convinced that linear methods do particularly well either.