This paper provides empirical evidence suggesting that fundamentals matter for stock price fluctuations once temporal instability underpinning stock price-relations is accounted for. Specifically, this study extends the out-of sample forecasting methodology of Meese and Rogoff (J Int Econ 14:3–24 (1983)) to the stock market after explicitly testing for parameter nonconstancy using recursive techniques. The predictive ability of a present value model based on Imperfect Knowledge Economics (IKE) is found to match that of the pure random walk benchmark at short forecasting horizons and to perform significantly better at medium to longer-run horizons based on conventional measures of predictability and direction of change statistics. In addition, the presence of a cointegrating relation is found only within regimes of statistical parameter constancy. Augmenting the MR methodology in a piecewise linear fashion yields empirical results in favor of a fundamentals-based account of stock price behavior overturning the recent results of Flood and Rose (2010).