In this study, we analyze the topology of the global currency market using the Granger causality network and attempt to predict its links by utilizing the real effective exchange rate of 61 countries. In this context, we suggest two new link prediction methods using the eta squared as a weight of link. For the network analysis, we focus on the changes in cross-sectional topology and time-varying properties of the causality network during the sub-prime mortgage crisis, the European debt crisis, and the Chinese stock market turbulence. For the link prediction, we evaluate the prediction performance of the proposed method and those of other benchmarks. Based on the results, we observe significant increments in out-degrees and in-degrees of the originating continents of the global financial crisis. Also, we confirm the best prediction accuracy of the weighted causality method based on the statistical significance of higher area under curve in every aspect.