Evaluation of on-line trading systems: Markov-switching vs time-varying parameter models

Price forecasting for spot instances in Cloud computing

“…MRS-AR is a generalized version of TAR in which the regime switching is much more flexible [20]. In MRS-AR, a Markov stochastic process is used to model switching of regimes where regimes are considered as states of the Markov stochastic process.…”

“…Using all historical data to build a uniform model is usually not helpful to predict local trends [20]. Therefore, only data of a recent time window is and Section 7 concludes this research, pointing out future research directions.…”

“…Yi et al [10,11] generated the probability density function (PDF) of a failure (out-of-bid events) within a time interval given a starting price and a bid according to historical prices directly. Based on the PDF of a failure, the practical task to model many different time series with changing regimes such as exchange rates [21,40] and financial time series [20]. In MRS-AR, for each prediction step, the absolute probability distribution over regimes can be used to forecast regimes which is generated from the last conditional probability distribution over regimes multiplying transition probabilities among regimes [21].…”