Ben F. MaierLittle side project now up on the arXiv. If you have data with a fat tail, you might want to fit a power-law tail without disregarding the shape of the "core", i.e. the support below a critical threshold. I derived maximum-likelihood estimators (or estimation methods) for a bunch of piece-wise Pareto distributions and wrote a Python library to perform the fits.
paper: https://arxiv.org/abs/2309.09589
github: https://github.com/benmaier/fincoretails
pip install fincoretails
This will remain a preprint.
Maximum-likelihood fits of piece-wise Pareto distributions with finite and non-zero core
We discuss multiple classes of piece-wise Pareto-like power law probability density functions $p(x)$ with two regimes, a non-pathological core with non-zero, finite values for support $0\leq x\leq x_{\mathrm{min}}$ and a power-law tail with exponent $-α$ for $x>x_{\mathrm{min}}$. The cores take the respective shapes (i) $p(x)\propto (x/x_{\mathrm{min}})^β$, (ii) $p(x)\propto\exp(-β[x/x_{\mathrm{min}}-1])$, and (iii) $p(x)\propto [2-(x/x_{\mathrm{min}})^β]$, including the special case $β=0$ leading to core $p(x)=\mathrm{const}$. We derive explicit maximum-likelihood estimators and/or efficient numerical methods to find the best-fit parameter values for empirical data. Solutions for the special cases $α=β$ are presented, as well. The results are made available as a Python package.