Short-term Stock Price Prediction Model Research Based On Echo State Network

As we know, stock plays an important role in our national economy. For our national economy, how to maintain market stability and avoid stock market disaster is a problem to consider. For investors, how to avoid the risk and maximizing investors’ profit is the problem that investors need to think about. Thus, it is very necessary to predict stock price in order to make the best decisions. Our stock market is a very complex nonlinear system. For such a complex system, the researchers used neural network to predict the stock price and have achieved good results. But there are some problems such that the training process is complex when using traditional neural network to predict stock price. In addition, the stock price is affected by many factors, that make stock price presents a different trend, as a result, a single model is hard to meet predict requirements. At the same time, most of stock price prediction model lacks of regional industry commonality, that is, every time when we predict a stock’s price we must build separate models for each stock and this model contains only the information of this stock and it does not include the information of these stocks in the same industry and in the same area. This makes the model can not cover all the factors that affect the stock price, leading to the prediction accuracy not high, the prediction process cumbersome and complex.Currently, most of the stock price prediction model lacks regional industry commonality and BP network training process is complex. Based on these problem, this paper use ESN to bulid short-term stock price prediction general model of Shanghai real estate industry. This simplify the training process, after training the model can predict any stocks in Shanghai real estate industry. And compared with the single models, the prediction accuracy of the regional industry general model improved significantly. Using a single nonlinear model for short-term stock price prediction methods to predict the effect is not ideal, based on general model, this paper presents short-term stock price prediction general model of Shanghai real estate industry model based on KMeans-ESN. By selecting different clustering indexes, this paper presents KMeans-ESN short-term stock price forecasting general model of Shanghai real estate industry based on data volatility clustering and KMeans-ESN short-term stock price forecasting general model of Shanghai real estate industry based on data volatility and trends clustering. These three models are compared. And we find these models’ suitable type of data.Meanwhile, the ESN requires a lot of initial parameters set in advance, and these parameters need to have some experienced researchers or by trial to determine. But the two ways have problems of require manual intervention, time-consuming and low efficiency. For the existing problems above, this paper presents GSA optimization model for ESN. In this paper, we use GSA to optimize parameters of ESN. We optimized from two angles:a single shrinkage factor and the spectral radius optimization, multi-value shrinkage factor and the spectral radius optimization. And we explore every angle selection of GSA fitness function. The results showed that the optimization results of using a single shrinkage factor and the spectral radius is better than the optimization results of using multi-value shrinkage factor and the spectral radius. Using a 10-fold cross validation error of the average value of the training and testing error of the average value as the GSA fitness function is the best choice.