In this paper I estimate the nancial network among 8 types of traders across 5 dierent capital markets, which are stock, stock derivative, bond, bond derivative, and foreign exchange derivative market. The causal relationship between two traders are identified with nonlinear granger causality and each trader's connectedness measure is obtained by the network structure. In order to overcome the limit of VAR which most of previous literature assumes and re ect real trading decision making procedures, expectation forecasting of traders' net trading volume on next day is included in analyses. Expectation forecasting
values are predicted with LSTM (Long Short Term Memory) which is one of the most popularly used machine learning method.

In addition, the systemic risk spillover channels are investigated using network measures. I model 3-phased systemic risk spillover channel which is the link of the volatility of nancial indexes, traders network measure and traders daily net trading volumes. I nd that given the shock of nancial indexes, 3 traders among 40 traders become central and another 3
traders rapidly lose their in uence regardless of the sort of nancial indexes. Secondly, when the shock is given in those traders network measures, the traders in dierent markets are shown to have more sensitive responses. Finally with the shock on the sensitive traders daily net trading volumes, strong auto-correlation between impulse and response as well as the
phenomenon which traders from dierent markets respond actively, are also found. These are the evidences of systemic risk spillover channels through traders. 

This research can contribute to the previous research in that the nancial networks which reflect real trading environments are estimated, and that machine learning methods are applied to network estimation. Furthermore, the results of systemic risk spillover channels can be helpful to policy makers including nancial regulators and practitioners.