I disagree. The point of the article was the show the steps required to develop a statistical advantage in the market place. If you develop a robust model and are very diligent in how it executes and learns, you can be successful. However, what you say about market structure is true. It goes through periods of stability, followed by abrupt changes. Any model that a trader has developed has been developed on such a short time-scale of market activity, that it can turn out to be a bad sample size. Depends on the scale of time and trades.