A Study On Futures Price Prediction System With Genetic/Simulated Annealing Algorithm And Cubic Exponential Smoothing

The futures price can reflect variety of information related to the spot market, it's helpful for us to reduce the investment risk of futures market in maximum, thereby increasing the income, if we could predict the futures price in future accurately. The futures price prediction methods we used in everyday life are generally basic scale analysis method and technical analysis method, with the development of technology, artificial intelligent technology is gradually used to predict the futures price. This paper presents an adaptive cubic exponential smoothing prediction method based on the genetic algorithm and the simulated annealing algorithm. The cubic exponential smoothing method is a good time series prediction method, if can predict the future value only giving the last period predictive value and the actual value. It needn't much historical data, saving much data processing time and reducing the amount of data storage, so it's better for predicting the futures price. In the article a model was built using the cubic exponeontial smoothing method to predict the futures price. It's an effective method to predict the time series. However, its prediction accuracy has a very important relationship with the selection of smoothing initial value and smoothing parameter, in general case, the selection of smoothing parameter depends on experience, so prediction result may not accurate. The cubic exponential smoothing model belongs to one of the trend predictions, which is based on the historical data and has a certain usage limitation, so it needs to combine with other methods to achieve better prediction results. Therefore, we will introduce the genetic algorithm in the artificial intelligence algorithm to help select the smoothing parameter, meanwhile taking into account the limitation of genetic algorithm, and then the simulated annealing algorithm is added. Genetic algorithm is a good optimal search algorithm, in which population search and exchange information among individuals in the populations does not depend on gradient information and the problem itself, and it has better robustness, adaptability, parallelism and global searching ability. However, the local search ability of genetic algorithm is poor, which causes that a simple genetic algorithm wastes much time. The search efficiency is low in the late stage of evolution and it's easy to produce premature convergence. The simulated annealing algorithm has the "climbing" ability to get rid of getting into the local optimal, can find the global optimal value of objective function from the sense of probability with random searching skill. Therefore, the method combining genetic algorithm and simulated annealing algorithm automatically searches for the optimal smoothing parameter value, which does not depends on subjective judgment, so that the predictive value is much closer to actual value.