主管:中国科学院
主办:中国优选法统筹法与经济数学研究会
   中国科学院科技战略咨询研究院

Chinese Journal of Management Science ›› 2020, Vol. 28 ›› Issue (5): 1-13.doi: 10.16381/j.cnki.issn1003-207x.2020.05.001

• Articles •     Next Articles

Testing the Nonlinear Cointegration Relation of Monetary Models of Exchange Rate Determination ——An Analysis Based on the Deep GRU Neural Network

LU Xiao-qin1,2, FENG Ling1, DING Jian-ping1,3   

  1. 1. School of Finance, Shanghai University of Finance and Economics, Shanghai 200433, China;
    2. Jiaxing University, Zhejiang 314001, China;
    3. Shanghai Institute of International Finance and Economics, Shanghai 200433, China
  • Received:2018-06-13 Revised:2018-10-17 Online:2020-05-30 Published:2020-05-30

Abstract: The monetary model of exchange rate has been the focus of academic that many scholars have conducted linear cointegration test on it, which the results are not satisfactory.In this paper, three versions of monetary models of exchange rate determinations (Flexible price, Forward-looking and Real Interest Differential Models) are tested for six selected countries with floating exchange rate regimes, by applying the nonlinear Johansen cointegration tests, facilitated by the Gated Recurrent Unit (GRU) neural network technique. The GRU technique has the advantages of intelligent memory, autonomous learning and strong approximation ability in deep learning. Based on country-by-country analysis, evidence of nonlinear cointegration between exchange rates and macroeconomic fundamentals is found. This suggests the validity of monetary models and the advantage of advanced deep learning tools in testing economic theory.
The concrete steps of the nonlinear cointegration test in this paper are: First, the long memory characteristics of the sequences are tested, because if there is a nonlinear cointegration relationship between the data sequences, it means that the sequence data must have the long memory characteristics. Secondly, the deep GRU neural network method is used to construct the nonlinear cointegration function, which has the transfer memory function that the ordinary neural network does not have, and has the advantage on the time series data mining. Finally, test whether the residuals of the constructed GRU model are short memory sequences (SMM). If the residuals are short memory sequences, the GRU can extract the non-linear characteristics between the sequences and prove the existence of nonlinear cointegration relationship between the sequences.
The results show that: (1) The original sequence after normalization has the feature of long memory, which is suitable for the discussion with the nonlinear cointegration theory. (2) The residual of the nonlinear cointegration model constructed by GRU neural network is short memory sequence, which found evidence of nonlinear cointegration between exchange rates and macroeconomic fundamentals.
In this paper, advanced deep GRU technology is introduced into nonlinear cointegration test, which can play a role in attracting valuable contributions to expand the tool box of cointegration empirical test.

Key words: exchange rate, monetary models, nonlinear cointegration, the deep GRU neural network

CLC Number: