表題番号：2021C-749
日付：2022/05/12

研究課題A smoothed empirical beta copula

研究者所属（当時） | 資格 | 氏名 | |
---|---|---|---|

（代表者） | データ科学センター | 准教授 | 竇 暁玲 |

- 研究成果概要
- In modeling multivariate distribution with the empirical beta copula, we find that it is difficult to do the computation when sample size is larger than 500, because the empirical beta distribution is constructed by Gamma functions. We try to use Stirling's approximation formula for the Gamma functions and translate them back by logarithm functions. This makes the calculation possible, however it still takes time when the sample size N is large. To solve this problem, we introduce a smaller integer K << N to separate the data set into K times K grids. We then calculate the means of data in each grid. The number of the means, say M, is smaller than or equal to K^2. Instead of the original N data, we use the M<<N means as the new data in the multivariate distribution modeling with the empirical beta copula. This method can dramatically reduce computation time and make the empirical beta copula applicable for large sample size. This method is presented at The Japanese Joint Statistical Meeting 2021 (September 2021) and Waseda International Symposium Topological Data Science, Causality, Analysis of Variance & Time Series 2022 (March 2022).