Footnote:
Nach Informationen von SSRN wurde die ursprüngliche Fassung des Dokuments July 31, 2017 erstellt
Description:
An increasing number of firms buy commodities at spot markets characterized by price volatility. Due to different market regimes (e.g., bull and bear), spot price dynamics are non-stationary and only partially observable; neither the underlying stochastic price process, nor its parameters are known with certainty. To capture uncertainty in both price and price model, we exploit recent spot price observations to dynamically update (learning) probabilistic price regime information in the context of inventory control under stochastic demand and purchase price. By means of Bayesian dynamic programming, we prove that, if prices evolve according to doubly embedded stochastic processes described by hidden Markov regime switching (MRS) models, price(s)- and regime-belief-dependent (prior respectively posterior) base-stock policies, rather than price-dependent policies, are optimal. We distinguish between independent and Markovian price processes and demonstrate the difference concerning optimal base-stock functions and monotonicity properties. We numerically establish that ignoring regime shifts leads to suboptimal inventory decisions and we quantify the operational value of spot price models. We find that Bayesian regime belief updates (learning) can yield significant cost savings that are particularly high when demand volatility and inventory holding cost are low and regime persistence is high. However, in empirical environments, stochastic price models induce misleading speculation resulting in misspeculative inventory. Based on real spot market data, we show that price forecast accuracy and operational performance are not perfectly correlated and that it can be advantageous for inventory managers to ignore sophisticated price forecasting and instead follow a naïve strategy