The Comparative Study About Inter-commodity Arbitrage Strategy Based On BP Neural Network And NAR Dynamic Neural Network

With the continuous development of the national economy,the demand for steel is increasing day by day.At the same time,iron ore as the raw material for smelting steel has increased.The main source of China's iron ore imports from abroad,meanwhile,China is the world's largest consumer and importer of iron ore.Therefore,the fluctuations of iron ore price has a great influence on the development of China's national economy.However,China has been passive recipients of the international iron ore prices,which is not benificial to the development of the domestic industry.In terms of the smelting process,steel is the the smelting products of coke and iron ore.In terms of industrial chain,Steel futures as the upper reaches of the coke futures and iron ore futures products,which can be set to protect the transaction and resist market volatility risk.However,the existing research on the inter-commodity arbitrage of rebar industry is mainly focused on the future of coke and rebar futures.Therefore,this paper about the inter-commodity arbitrage of rebar industry is a useful supplement to the relevant theory,which refers to the future arbitrage of three futures rather than the traditional future arbitrage of two futures.This paper summarizes and clarifies the existing domestic and international research achievements of futures arbitrage.What's more,based on the view of the arbitrage mode and arbitrage strategies,this paper sums up the deficiency of the existing theoretical achievements and determines research direction.First of all,it introduces the basic concepts of future arbitrage theory and neural network model in detail,and describes the structure and operation principle of BP neural network model and NAR dynamic neural network model.Secondly,by describing the fluctuations of coke,iron ore and rebar of futures price,we use of Johnsan cointegration test method to verify the existence of the long-term relationship among them.Then,it introduces the setting of future position,transaction cost,margin,trading rules and the classification of sample.Finally,two different models of arbitrage strategies are compared by setting different opening and closing positions.The empirical results show that the BP neural network model arbitrage strategy and the NAR dynamic neural network model arbitrage strategy all have positive returns,but the difference of them is mainly reflected in the following aspects.First,for the fitting effect of the sample data,NAR neural network dynamic model fitting effect isbetter than that of BP neural network model in either sample data or sample data.Secondly,in terms of the accuracy of arbitrage,NAR dynamic neural network model arbitrage strategy has high precision.Thirdly,in terms of the data,the NAR dynamic neural network model is the least affected by the data,with the increasing of the sample data,the forecast effect is better.Finally,for terms of the total yield,whether it is in the sample data or out of the sample data,NAR dynamic neural network model arbitrage strategy is better than the BP neural network model arbitrage strategy.The innovation of this paper mainly focused on the following aspects.First,the existing futures research related to the object is mainly for agricultural futures,crude oil futures and derivatives futures,less for other futures studies.In addition,the Dalian Mercantile Exchange in 2013 launched the iron ore futures,so for the study of iron ore futures arbitrage less.Therefore,the research of this paper is a beneficial supplement to the arbitrage of iron ore futures.Secondly,the domestic futures arbitrage research strategy can be divided into two categories,the mean return arbitrage strategy and neural network model arbitrage strategy.The former focuses on the long-term equilibrium relationship between the futures contracts,which focuses on the short-term fluctuations of the futures contracts.In this paper,the above two kinds of arbitrage strategies are combined to study the long-term equilibrium relationship between the futures contracts and the short-term fluctuations of the futures contracts.Third,Domestic scholars few use neural network model to study the outcome of the futures arbitrage strategy.But the neural network arbitrage model of the existing research is relatively single,mainly based on the BP model.This paper will use the BP neural network model and the NAR dynamic neural network model to carry on the comparative study of the cross futures arbitrage strategies of the coke,iron ore and rebar.