According to Coleman's following paper (link is attached), the eigenvector corresponding to negative eigenvalue is the tangent vector to the contour at z=1, the question is how to make sure this eigenvector actually correspond to negative eigenvalue and what's the reason behind choosing this particular function. ( The argument in the paper is on page 4, first paragraph just below eq.2.21).
http://journals.aps.org/prd/abstract/10.1103/PhysRevD.16.1762