Asymptotics of error probabilities of optimal tests

V. Kanišauskas ; K. Kanišauskienè

Teoriâ slučajnyh processov, Tome 25 (2020) no. 2, pp. 25-36.

Résumé

We consider first and second error probabilities of asymptotically optimal tests (Neyman-Pearson, minimax, Bayesian) when two simple hypotheses $H^{t}_{1}$ and $H^{t}_{2}$ parametrized by time $t \ge 0$ are tested under the observation $X^t$ of arbitrary nature. The paper provides details on the conditions of asymptotic decrease of probabilities of optimal criteria errors determined by $\alpha$ type Hellinger integral between measures $P^{t}_{1}$ and $P^{t}_{2}$, demonstrating that in the case of minimax and Bayesian criteria it is sufficient to investigate Hellinger integral, when $\alpha \in \left( 0, 1 \right)$, and in the case of Neyman-Pearson criterion it is observed only in the environment of point $\alpha=1$. Whereas Kullback-Leibler information distance is always larger than Chernoff distance; we discover that, in the case of Neyman-Pearson criterion, the probability of type II error decreases faster than in the cases of minimax or Bayesian criteria. This is proven by the examples of marked point processes of the i.i.d. case, non-homogeneous Poisson process and the geometric renewal process presented at the end of the paper.

Détail

Keywords: Hellinger integral, Neyman-Pearson test, minimax test, Bayesian test, Kullback-Leibler information, Chernoff information.

Bibliographie

[1] L. Birgé, “Vitesses maximales de décrvissance des erreurs et tests optimaux associés”, Z. Wahrsch. Verw. Geb., 55:2 (1981), 261–273

[2] N. N. Chencov, Statistical decision rules and optimal inference, Nauka, Moscow, 1972 (Russian)

[3] H. Chernoff, “A Measure of Asymptotic Efficiency for Tests of a Hypothesis Based on a Sum of Observations”, Ann. Math. Stat., 23 (1952), 493–507

[4] H. Chernoff, “Large sample theory$:$ Parametric case”, Ann. Math. Stat., 27:1 (1956), 1–22

[5] S. Ellis, Entropy, Large Deviations and Statistical Mechanics, Springer-Verlag, Berlin–Heidelberg–New York–Tokyo, 1985

[6] V. Kanišauskas, “An Asymptotic Formula of the Hellinger Transform for Multivariate Point Processes”, Lith. Math. J., 42:4 (2002), 322–327

[7] V. Kanišauskas, “Asymptotically minimax separation two simple hypotheses”, Lith. Math. J., 38:2 (1998), 169–184

[8] V. Kanišauskas, “Asymptotically minimax testing of simple hypotheses”, Lith. Math. J., 40:3 (2000), 313–320

[9] V. Kanišauskas, L. Dronova-Platbarzdė, “Asymptotical Separation of Two Simple Hypotesess Using the Most Powerful Criteria”, Journal of Young Scientists, 44:2 (2015), 105–109 (Lithuanian)

[10] V. Kanišauskas, K. Piaseckienė, “Prediction of the geometric renewal process”, Lith. J. of Stat., 56:1 (2017), 72–76

[11] F. Liese, I. Vajda, Convex Statistical Distances, Teubner publaddr Leipzig, 1987

[12] F. Liesem, I. Vajda, “f-divergences: sufficiency, deficiency and testing of hypotheses”, Advances in Inequalities from Probability Theory and Statistics, eds. Neil S. Barnett and Sever S. Dragomir, ed., Nova Publishers, Toronto, 2009, 131–173

[13] Yu. N. Lin'kov, Asymptotical Methods of Statistics for Stochastic Processes, Naukova Dumka, Kiev, 1993 (Russian)

[14] Yu. N. Lin'kov, L. A. Gabriel, “Large deviations in Bayes testing of the finite number of simple hypotheses”, Ukr. Math. J., 51:10 (1999), 1534–1542

[15] Yu. N. Lin'kov, L. A. Gabriel, “Large deviation theorems in Bayes testing problems”, Vestnik Volgograd. Univ. Ser. 1. Mathematics. Physics, 3 (1998), 62–68 (Russian)

[16] Yu. N. Lin'kov, “Large deviations in testing problems of counting processes”, Ukr. Math. J., 45:11 (1993), 1703–1712 (Russian)

[17] Yu. N. Lin'kov, “Large deviation theorems for extended random variables and some applications”, J. Math. Sci., 93:4 (1999), 563–573

[18] Yu. N. Lin'kov, “Large deviation theorems for extended random variables in the hypotheses testing problems”, Theory Stoch. Process., 5:21 (1999), 137–151

[19] Yu. N. Lin'kov, “Large deviation theorems in the hypotheses testing problems”, Exploring Stochastic Laws$:$ Festschrift in Honour of the 70th Birthday of Academician V. S. Korolyuk, eds. A. V. Skorokhod and Yu. V. Borovskikh, ed., VSP, Utrecht, 1995, 263–273

[20] Yu. N. Lin'kov, Lectures in mathematical statistics. 2, Istoki, Donetsk, 2001 (Russian)

[21] Yu. N. Lin'kov, “Limit theorems for the likelihood ratio in the hypotheses testing problems”, Random Oper. Stoch. Equ., 3:1 (1995), 23–40

[22] Yu. N. Lin'kov, “Limit theorems for the local density of measures in the hypotheses testing problems of counting processes”, Probability Theory and Mathematical Statistics, eds. B. Grigelionis and etc., ed., TEV, Vilnius, 1994, 497–515

[23] Yu. N. Lin'kov, L. A. Lunyova, “Large deviation theorems in the testing problems for Bernoulli sequences”, Random Oper. Stoch. Equ., 3 (1995), 231–243

[24] Yu. N. Lin'kov, M. I. Medvedeva, “Large deviation theorems for the logarithm of the likelihood ratio”, Teor. Veroyatn. Mat. Statist., 53 (1995), 87–96 (Ukrainian)

[25] Yu. N. Lin'kov, M. I. Medvedeva, “Large deviation theorems in testing problems for two simple hypotheses”, Ukr. Math. J., 47:2 (1995), 227–235 (Russian)

[26] Yu. N. Lin'kov, M. I. Medvedeva, “Large deviation theorems in the testing problems for the normal autoregressive processes”, Theory of Random Processes, 1:17 (1995), 71–81 (Russian)

[27] K. Pikturnaitė,, Asymptotic Separability of Probability Measures of Renewal Process, Master Thesis, Šiauliai, 2011 (Lithuanian)

[28] K. Pikturnaitė, V. Kanišauskas, “Asymptotic Separability of Probability Measures of Renewal Process”, Journal of Young Scientists, 2:31 (2011), 164–168 (Lithuanian)

[29] N. Shimkin, “Extremal Large Deviations in Controlled i.i.d. Processes with Applications to Hypothesis Testing”, Adv. Appl. Probab., 25:4 (1993), 875–894

[30] W.Stummer, I. Vajda, “On Bregman Distances and Divergences of Probability Measures”, IEEE Trans. Inform. Theory, 58:3 (2012), 1277–1288

[31] I. Vajda, “Distances and discrimination rates for stochastic processes”, Stochastic Process. Appl., 35 (1990), 47–57

[32] I. Vajda, “Generalization of discrimination-rate theorems of Chernoff and Stein”, Kybernetika, 26:4 (1990), 273–288