Inference for the mode of a log-concave density (Q2328065)

The authors propose a test for the location of the mode. For the model of a general log-concave density, they use the likelihood-ratio test of the unconstrained maximum-likelihood estimator and the estimator constrained to the hypothesis on the mode. Under the assumption that the second derivative of the logarithm of the density exists at the mode, they show that the asymptotic distribution of the likelihood-ratio is universal and thus universal quantiles can be used for this test statistic. The test is also consistent under the alternative. These results are based on the ocal limit theorem for the density estimator considered as a process in a neighborhood around the mode. In a simulation study it turns out that the distribution of the test statistic is not close to a chi-squared distribution, and empirical quantiles are given for the test statistic. Finally, the daily log returns of the S\&P 500 are discussed as a data example.