Time series data often exhibit variation, clustering, and persistence in uncertain environments. Based on an uncertain measure, existing studies have primarily focused on conditional mean structures while ignoring variance and persistent dynamics. To address this gap, this paper develops an uncertain generalized autoregressive conditional heteroskedasticity (UGARCH) model that incorporates lagged conditional variance terms. We first prove sufficient conditions to ensure mean stationarity of the variance process. Then, we propose recursive constrained least squares to estimate the model parameters and use cross validation to determine the orders of the mean and variance equations. In addition, we also employ uncertain hypothesis tests to determine whether a time series exhibits a UGARCH effect and to assess the goodness of fit. Point and interval forecasts are subsequently constructed from the recursively obtained next-period conditional variance and the expected value of the mean equation. Numerical simulations demonstrate the parameter estimation performance of the proposed method and show that UGARCH achieves smaller conditional-variance fitting errors than UARCH under the considered settings. In the empirical analysis, we apply the proposed method to West Texas Intermediate crude oil futures closing prices, which exhibit time-varying volatility and support short-term point and interval forecasts. Keywords: Uncertain measure, Uncertain time-series analysis, UGARCH model, hypothesis test

翻译：暂无翻译