High-Dimensional Tail Index Regression
Organizations: Department of Economics, Vanderbilt University, VU Station B #351819, 2301 Vanderbilt Place, Nashville, TN 37235-1819 · Department of Economics, University of Washington, Box 353330, Savery 305, University of Washington, Seattle, WA 98195-3330 · Department of Economics, Lehigh University, Bethlehem, PA 18015
Abstract
Motivated by the empirical observation of power-law distributions in the credits (e.g., ``likes'') of viral posts on social media, we introduce a high-dimensional tail index regression model and propose methods for estimation and inference of its parameters. First, we propose a regularized estimator, establish its consistency, and derive its convergence rate. Second, we construct a debiased version of the regularized estimator to facilitate inference and prove its asymptotic normality. Monte Carlo simulation studies corroborate our theoretical findings. We apply these methods (1) to the text analysis of viral posts on X (formerly Twitter) and (2) to the analysis of abnormally high health expenditures.
Figures & tables
| Bias | SD | RMSE | 95% | ||||||
|---|---|---|---|---|---|---|---|---|---|
| 500 | 250 | Sparse | Gaussian | Pareto | 0.023 | 0.545 | 0.545 | 0.918 | |
| 500 | 250 | Exponential | Gaussian | Pareto | 0.012 | 0.494 | 0.494 | 0.936 | |
| 500 | 500 | Sparse | Gaussian | Pareto | 0.022 | 0.542 | 0.542 | 0.919 | |
| 500 | 500 | Exponential | Gaussian | Pareto | 0.007 | 0.487 | 0.487 | 0.942 | |
| 500 | 1000 | Sparse | Gaussian | Pareto | 0.033 | 0.520 | 0.521 | 0.927 | |
| 500 | 1000 | Exponential | Gaussian | Pareto | 0.000 | 0.483 | 0.483 | 0.938 |
| Word | Count | Tweets | Word | Count | Tweets | Word | Count | Tweets | |||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | lgbt | 23734 | 8133 | 11 | #lgbt | 4989 | 1280 | 21 | it’s | 3194 | 1842 | ||
| 2 | and | 18706 | 12357 | 12 | this | 4956 | 3532 | 22 | lgbt+ | 3165 | 578 | ||
| 3 | i | 10162 | 1559 | 13 | community | 4189 | 3640 | 23 | their | 3127 | 2641 | ||
| 4 | that | 9022 | 6931 | 14 | have | 4186 | 3605 | 24 | my | 2747 | 2003 | ||
| 5 | you | 8965 | 5374 | 15 | just | 3563 | 2900 | 25 | don’t | 2690 | 2195 | ||
| 6 | people | 7117 | 5495 | 16 | or | 3555 | 2903 | 26 | what | 2662 | 1840 |
| Word | SE | 95% CI | t | Word | SE | 95% CI | t | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | lgbt | -0.13 | 0.06 | [-0.26 | 0.00] | -1.99 | 16 | or | -0.05 | 0.10 | [-0.25 | 0.15] | -0.48 | |
| 2 | and | -0.01 | 0.06 | [-0.13 | 0.10] | -0.25 | 17 | so | 0.16 | 0.10 | [-0.05 | 0.36] | 1.49 | |
| 3 | i | 0.09 | 0.13 | [-0.17 | 0.36] | 0.70 | 18 | if | -0.08 | 0.12 | [-0.32 | 0.16] | -0.67 | |
| 4 | that | 0.04 | 0.07 | [-0.10 | 0.17] | 0.53 | 19 | all | 0.26 | 0.09 | [ 0.07 | 0.44] | 2.76 | |
| 5 | you | -0.01 | 0.08 | [-0.17 | 0.15] | -0.11 | 20 | who | -0.03 | 0.09 | [-0.21 | 0.16] | -0.28 | |
| 6 | people | 0.06 | 0.07 | [-0.09 | 0.20] | 0.75 | 21 | it’s | -0.11 | 0.14 | [-0.39 | 0.16] | -0.82 | |
| Condition | SE | 95% CI | t | ||
|---|---|---|---|---|---|
| Physical Limitations | -0.070 | 0.074 | [-0.214 | 0.075] | -0.941 |
| Work Limitations | -0.117 | 0.074 | [-0.262 | 0.028] | -1.587 |
| Difficulty Seeing | -0.225 | 0.114 | [-0.449 | 0.000] | -1.963 |
| Difficulty Hearing | -0.119 | 0.136 | [-0.386 | 0.148] | -0.873 |
| Difficulty Walking | -0.078 | 0.082 | [-0.240 | 0.083] | -0.950 |
| Cognitive Difficulties | -0.068 | 0.082 | [-0.229 | 0.093] | -0.832 |
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
| Bias | SD | RMSE | 95% | ||||||
|---|---|---|---|---|---|---|---|---|---|
| 500 | 250 | Sparse | Gaussian | Pareto | -0.102 | 0.592 | 0.601 | 0.944 | |
| 500 | 250 | Exponential | Gaussian | Pareto | -0.030 | 0.595 | 0.596 | 0.947 | |
| 500 | 500 | Sparse | Gaussian | Pareto | -0.103 | 0.592 | 0.601 | 0.942 | |
| 500 | 500 | Exponential | Gaussian | Pareto | -0.038 | 0.593 | 0.594 | 0.943 | |
| 500 | 1000 | Sparse | Gaussian | Pareto | -0.098 | 0.590 | 0.598 | 0.943 | |
| 500 | 1000 | Exponential | Gaussian | Pareto | -0.029 | 0.594 | 0.595 | 0.946 |
| Algorithm Stochastic optimization based estimation and confidence interval for HDTIR |
|---|
| Inputs: RADAR tuning specified below, confidence level , and deterministic fixed threshold . Nodewise penalties , where , is sufficiently large, and do not depend on or . |
| Initialize RADAR as specified below and initialize the retained data matrices and as empty. |
| for to do |
| Observe . If , discard this observation and continue. |
| Otherwise, append to and to . |
| Run one iteration of RADAR using the stochastic gradient at its current iterate , |
Heavy-Tailed Distributions
Heavy-tailed distributions assign unusually high probability to extreme values, shaping reliable modeling, inference, and decisions.
See all 60 papers