为了更加准确地预测人民币汇率,提出一个双阀值LSTAR模型,并对该模型的非线性检验进行了研究.结果表明,不同于传统的单阀值STAR模型需要对转换函数进行三阶Taylor展开,双阀值LSTAR模型只需对转换函数进行一阶Taylor展开即可,这大大节省了自由度.随后的预测实验表明,双阀值LSTAR模型不仅比传统的单阀值LSTAR模型更契合无行动区间理论,而且有更强的预测能力.

The paper proposes an LSTAR model with two thresholds in order to forecast RMB exchange rate more accurately. The research on its nonlinearity testing shows that to test the nonlinearity   of this kind of LSTAR models, we need only the transition  function with its first-order Taylor approximation, which can help
   us save many degrees of freedom. Then the simulation result  shows that the LSTAR model with two thresholds is more reasonable  and has higher precision than the traditional STAR models with only one threshold.