中文摘要：

焦炭、铁矿石是冶炼钢铁的主要原料,中国钢铁企业生产所需要的铁矿石主要依赖于进口。国外铁矿石市场的价格制定规则很大程度上阻碍了钢铁企业的发展。选取焦炭、铁矿石、螺纹钢期货作为研究对象,运用单位根检验、协整等方法研究焦炭、铁矿石、螺纹钢三者价格之间的长期均衡关系。在此基础上,通过设置不同的开平仓阀值,运用BP神经网络模型和NAR动态神经网络模型在样本区间内进行套利策略对比研究。实证结果表明：NAR动态神经网络模型的预测能力更强,其套利策略在螺纹钢、铁矿石和焦炭三者间进行跨品种套利效果更好。

英文摘要：

Coke and iron ore are the main raw materials to produce steel, but China＇s steel enterprises largely depend on the import of iron ore to meet the need of production. Therefore, the development of the steel enterprises is greatly hindered by the price setting rules of foreign iron ore markets. The authors choose coke, iron ore and rebar futures as the objects of the study to study the long term equilibrium relationships among the three futures by using unit root test and cointegration methods. By setting different threshold values of opening and offsetting positions, a comparison study is made on the arbitrage strategies of BP neural network model and NAR dynamic neural network model. Empirical results show that NAR dynamic neural network model has better prediction ability, so its arbitrage strategies have better performance in the cross - commodity arbitrage of Coke, iron ore and rebar.