Neglect of trends has unrecognized implications for interpretation of autoregressions. In nonparametric analysis of instructive samples of commodity price series, correspondences relating current price to the following price are linear, and returns appear independent of current price. After detrending current price, these relationships appear highly nonlinear, suggestive of the implications of a model of storage arbitrage. In such a model, any trend is not revealed in expected returns on positive stocks but in expected jumps from boom prices. We implement a new approach to consistent estimation of nonlinear empirical models with a trend in price that might not be zero.
分享
引用
