Every random variable satisfies a certain nontrivial integrability condition (Q2507703)

The note draws attention on suitable properties of random variables satisfying a certain nontrivial integrability condition. Let \(F\) be the set of nonnegative and nondecreasing functions defined on \([0, \infty)\). Let \(G\) be the subset of \(F\) consisting of those functions \(g \geq 1\) such that \(g(x) \to \infty\) as \(x\to\infty\), and \(g(x) / x\) is nonincreasing on \((0, \infty)\). The main result of this article is stated by the following: Theorem 1. Let \(X\) be an arbitrary real-valued random variable and \(f \in F\). Assume that \(0 < Ef(|X|) <\infty\). Then, there exists \(g \in G\) such that \(Ef(|X|) \leq Ef(|X|)g(|X|) < \infty\). Refinements of this general property provide effective stronger results in connection with Mar\-kov's inequality and the central limit theorem. For instance, the general form of Markov's inequality provides the upper bound: \[ P(|X| \geq r) \leq Ef(|X|) / f(r),\quad r > 0, \tag{1} \] whenever \(f \in F\) satisfies \(Ef(|X|) < \infty\) and \(f(r) > 0, r > 0\). Theorem 1 above entails the following more general result: for \(f \in F\) satisfying the classical Markov's inequality, there is a further function \(g \in G\) with \(Ef(|X|)g(|X|) < \infty\) which provides the estimate: \[ P(|X| \geq r) \leq Ef(|X|)g(|X|) / f(r)g(r),\quad r> 0. \tag{2} \] Since \(g(r) \to \infty\) as \(r \to \infty\), the rate of convergence in the generalized Markov inequality (2) is strictly better than that in (1).