In this work, we consider the outer Stefan problem for the short-time prediction of the spread of a volatile asset traded in a financial market. The stochastic equation for the evolution of the density of sell and buy orders is the Heat Equation with a non-smooth noise in the sense of Walsh, posed in a moving boundary domain with velocity given by the Stefan condition. This condition determines the dynamics of the spread, and the solid phase $[s^-(t),s^+(t)]$ defines the bid-ask spread area wherein the transactions vanish. We introduce a reflection measure and prove existence and uniqueness of maximal solutions up to stopping times in which the spread $s^+(t)-s^-(t)$ stays a.s. non-negative and bounded. For this, we use a Picard approximation scheme and some of the estimates of \cite{BH} for the Green's function and the associated to the reflection measure obstacle problem. Analogous results are obtained for the equation without reflection corresponding to a signed density. Additionally, we apply some formal asymptotics when the noise depends only on time to derive that the spread is given by the integral of the solution of a linear diffusion stochastic equation.