The Application Of NAR Neural Network And EMD Algorithm In Quantitative Trading

Since the birth of stock and futures market,people have been using all kinds of data models,machine learning and data mining methods to predict the future trend of stock prices,so as to get profits.The neural network learning algorithms are widely used,this is due to the highly self-learning neural network's stability and nonlinear fitting ability,compared to the mathematical model of statistics and econometrics,neural network is used to predict financial time series has more advantages.In the global stock market,the electronic trading platform plays a more and more important role,and the scale of quantitative transaction is increasing.Many investment institutions in western countries have received extensive attention in the use of quantitative transactions,showing the characteristics of procedural transactions.Take the T+1 Chinese stock market trading system,so this paper selects the futures market in the stock index futures as the research object,using neural network technology,EMD(Empirical Mode Decomposition)empirical mode decomposition quantitative modeling,proposed EMD-NAR quantitative timing strategy,and conducts the thorough research,and compared to measurement and trading system of double moving average trading system DualThrust,commonly used quantitative trading system,through the actual test found that EMD-NAR quantitative timing strategy in each index were significantly better than the double moving average trading system,DualThrust trading system,is a quantitative trading system is very suitable for short-term trading.In this paper,we break through the traditional inertia thinking of neural network for price prediction,and turn to look for the realization method of quantized trading strategy by using neural network.First,the EMD algorithm is used to decompose the original time series,and the continuous and derivable low frequency trend terms are combined from the decomposed intrinsic mode functions.The neural network is used to predict the low-frequency trend terms,which achieves excellent fitting results.Then,the high-frequency noise item is combined from the intrinsic mode function decomposed by the EMD algorithm,and its characteristic is analyzed,and its mean reversion characteristic is obtained,and it is further applied to the timing transaction.Finally,the combination of EMD algorithm and NAR neural network is designed to quantify the EMD-NAR timing strategy,the low-frequency trend was predicted using NAR neural network,the high frequency noise using the mean regression principle analysis,comprehensive factor of the two party issued trading signals quantitatived optional high and stable return of the winning strategy.