Real exchange rate dynamics and external balances: Econometric and artificial neural network analyses
本文结合向量误差修正模型和人工神经网络，研究实际汇率与外部平衡的动态关系，发现美国净外国资产驱动实际汇率，且神经网络模型在预测上优于传统方法。
We examine the dynamic relationship between the real exchange rate (RER) and external balances using both econometric and artificial neural network (NN) approaches. Our framework synthesizes a vector error correction model (VECM) and an NN model derived from it. Unlike traditional models such as vector autoregression (VAR), our models systematically incorporate economic relationships pertinent to the subject matter. These includes (i) the stock-flow relationship between the stock of net foreign assets (NFA) and current account flows, and (ii) the long-run, cointegrating relationship between the RER and NFA. After validating the framework, we apply the VECM and NN model to U.S. data, considering the country’s substantial net foreign debt and the associated dollar adjustments. Our VECM results differ markedly from previous VAR studies, revealing previously unidentified dynamics and indicating that the NFA position drives the RER. Our NN model, which appropriately handles both stationary and nonstationary data, outperforms alternative models in forecasting RER movements. Both modeling approaches highlight the importance of properly incorporating stock I (1) variables through cointegration and error correction.