Data Prediction Research Based On Time Series

Time Series is very important in everyone's life. For example, the sales amount of the shop for several years and the mount of each year's polar bear and so on. If the mount varies with the time, we can call it a time series. Because time series has widely used in many aspect, it will benefit us a lot in our daily life if we figure out the deep patterns of time series.According to the stability of time series, time series is divided into stable time series and unstable time series. The stable time series is a fundamental time series. Based on autocorrelation and partial autocorrelation function of time series, we are able to choose a model that fits the time series best between autoregressive model(AR), moving average model(MA) and autoregressive moving average model(ARMA). The modeling procedure includes model identification, model order selection, model parameter estimation and model verification. After modeling the time series, we can use the model to predict future time series values.The pig is a very important domestic animal in China, and it's also a main source of the farmers'income. The pig price fluctuation is of great influence in farmer's life. If we can predict the pig price tendency, we can help farmers to earn more money. There are some papers discussing the method of using time series to predict the pig price of certain area, but the price doesn't represent the average level of the whole country. And also some papers don't analyze the prediction results, so we can't compare the real value with predicted value. In order to fix the problem, we use the average pig price time series to predict the price. We also divide the time series into two parts, one for error analysis and another for price tendency prediction. We use the data from 1995 to 2008 to build the model, after time series stabilizing, we get a ARMA(14,17) model. We use the model to predict the price from January to June 2009, and the result shows that the highest relative error is 7.3 percent; the lowest is 1 percent and the average relative error is 4.9 percent. So we can use the model in our real life to predict the pig price. In order to predict the price tendency, we use the data from January 1995 to August 2009 to build the model. After time series stabilizing, we get a ARMA(8,15) model. We use the ARMA model to predict the price tendency from January to June 2009, and the result shows that the price will stay still for 4 months and rise after that.Corn is the main food of pig in China, and it's also a very important farm produce. The corn price fluctuation can also affect the farmers'earnings. If we can predict the corn price, it will be very helpful in farmers'life. We use the corn price time series to predict the future price. We use the data from January 1995 to December 2008 to build the corn price time series model. We use ARMA(16,15) to predict the value from January 2009 to August 2009. The experiment shows that relative errors of the first five months are below 3 percent. At the same time, We use all the corn price data from January 1995 to August 2009 to build the ARMA(13,14) model for price tendency prediction of the next six month, and the results show that the corn price will rise slowly for four months and after that the price will decrease.Everything is affected by other things, so is the price time series. Univariate time series can only describe one variable at a time, but in the real world things always interact with others. So the univariate time series is not good enough to describe the price alone. Multivariate time series are able to describe the time series which has multi-variables more precisely, because multivariate time series can show how the variable interact with other variables.Because there is no paper using multivariate time series method to predict the average pig price, we propose a method to predict the pig price with multivariate time series. The multivariate time series are composed of two univariate time series, and one is pig price time series, the other is corn price time series. We use the data from January 1995 to December 2008 to build the multivariate time series model. Because the multivariate time series are unstable, we stabilize them and build ARMAV(15,17,15) model to predict the pig price. The results show that the average relative error of the first five month of 2009 is 4.6 percent.We compare the univariate time series prediction results with multivariate time series prediction results, and we can conclude that the average relative error of the multivariate time series is smaller than that of univariate time series. Because multivariate time series is composed of the pig price time series and the corn price time series, it contains more information than univariate time series, so the prediction result of multivariate time series is also better than univariate time series.As we know corn is the main food of the pig, but whether corn price time series is the Granger cause of the pig price time series has not been tested. We use Granger causality test method to verify the fact that the corn price time series is the Granger cause of the pig price time series when the max lag is seventeen. We can use the corn price time series to help predict the pig price time series.