Abstract: Electricity theft could cause escalation of power loss of associated feeder. As a consequence, there is a long-term dynamic interaction between the electricity usage of an anomaly user and the power loss of associated feeder. Vector auto regression (VAR) based approach was proposed to detect anomaly user connecting to feeder with high loss ratio. Firstly, the long-term equilibrium relationship between power loss of feeder and each user's electricity usage was analyzed with edge-limit co-consolidation test. Secondly, the VAR model of feeder loss and the electricity usage of users were constructed. The dynamic mechanism of feeder loss and user's usage was analyzed with the pulse response function. Thirdly, the contribution of the user's electricity usage on the feeder loss was analyzed with variance decomposition. Consequently, the user with significant impact on the loss of feeder could be identified as the anomaly one. The validity and accuracy of the approach proposed in the paper were verified with the test study and on-site inspection based on real world metering data of feeder with high loss ratio.