On the exponential inequalities for strictly stationary and negatively associated random variables (Q2272096)

Let \((X_i)\) be a sequence of strictly stationary and negatively associated random variables. The author proves an exponential inequality for \[ E\left(\left|\sum_{i=1}^nX_i-E(X_i)\right|>\varepsilon_n\right) \] with an appropriate sequence \(\varepsilon_n\to\infty\). For this purpose the random variables are appropriately decomposed into a bounded and an unbounded part and with a high degree of technicality corresponding exponential inequalities are established for these to achieve the main result. The inequalities improve corresponding ones given recently by [\textit{H. J. Nooghabi} and \textit{H. A. Azarnoosh}, Stat. Pap. 50, No. 2, 419--428 (2009; Zbl 1247.60039)]. As a consequence of the main result, in case of geometrically decreasing covariances \(|\text{Cov}(X_1,X_{n+1})|=C\cdot\rho^n\) for some \(C>0\) and \(0<\rho<1\) the author provides the rate of convergence \(n^{-1/2}(\log\log n)^{1/2}(\log n)^{3/2}\) in a corresponding strong law of large numbers, which is close to the optimal rate achieved for independent random variables by the law of the iterated logarithm.