A1500
Title: Computational methods and algorithms based on characteristic functions and their implementation in CharFunTool
Authors: Viktor Witkovsky - Slovak Academy of Sciences (Slovakia) [presenting]
Abstract: Characteristic-function-based (CF) methods provide a powerful analytical framework for statistical inference, yet their numerical implementation is often affected by oscillatory integrals, truncation, and loss of precision. This work presents stabilized computational techniques implemented in the open MATLAB framework CharFunTool, improving robustness and accuracy of CF evaluation and inversion. The approach combines Fourier inversion with exponential damping and smooth windowing, double-exponential (DE) quadrature for rapidly convergent integration over infinite domains, and adaptive quadrature guided by local oscillation analysis. Regularized and weighted estimators of empirical CFs further suppress high-frequency noise. The results demonstrate that numerically stabilized CF-based algorithms provide a coherent and robust framework for inference under possible instability and analytical intractability.
