This paper first reduces the problem of detecting structural breaks in a random walk to that of finding the best subset of explanatory variables in a regression model and then tailors various subset selection criteria to this specific problem. Of particular interest are those new criteria, which...