Fault prediction is a critical part in Prognostics and Health Management and can provide sufficient information for fault prevention and troubleshooting. When employing characteristic parameter monitoring method to implement radar fault prediction, the predictability measurement of the observation sequence should be solved first to ensure that the conclusion is valid and credible. Firstly, the composition of the radar performance monitoring sensor network, the characteristics of the performance parameter monitoring sequence and the applicable ARMA modeling prediction algorithm were introduced. Then a method was proposed to determine the predicting time advance applicable for the parameter observation sequences based on phase space reconstruction theory and chaos analysis. The method constructs the optimal embedding space and restores the dynamic characteristics of performance degradation of radar, by which the maximum predictable number of steps based on the largest Lyapunov exponent can be calculated to determine whether the performance observation parameters can be predicted. Furthermore, scientific basis of determining a valid time advance can be provided. Finally, the experimental analysis of radar signal-to-noise ratio parameter observation sequences was carried out, indicating that the proposed method is effective for radar equipment.