We propose a general framework for asymptotic tests on single and multiple parameters using subseries statistics and a p-value aggregation method that is robust to any arbitrary form of dependence. The testing procedure have the advantage to guarantee good size and power close to any available unbiased asymptotic and bootstrap test. The procedure is: robust to diverging parameter space, computational cheap, does not require the choice of kernel unlike classical robust inference procedure does not require any knowledge of the variance of the parameters nor the model they arise from. It returns a consistent estimator of the variance along with the p-values of the individuals and joint tests. We also provide a procedure for FWER and FDR control. Applications include robust performance testing, test of homogeneity of risk measure, peer-performance evaluation, bucketing procedure. Simulation results show that the procedure performs well for both low and high frequency (stationary) data even under asynchronicity. The results suggest that the procedure can be used as an alternative to bootstrap based model confidence set procedures.