Bounds for tail probabilities of martingales using skewness and kurtosis (Q946798)

Let \(M_n= X_1+\dots+X_n\) be a sum of \(n\) independent random variables such that \(0 \leq X_k \leq 1, E(X_k)=0\) and \(E(X_k^2)= \sigma_k^2, 1 \leq k \leq n.\) \textit{W. Hoeffding} [J. Am. Stat. Assoc. 58, 13--30 (1963; Zbl 0127.10602)] obtained an upper bound on the tail probability \(P(M_n \geq nt), 0<t<1\) based on the variances \(\sigma_k^2, 1 \leq k \leq n.\) The authors obtain bounds for the tail probability using the skewness and kurtosis of the random variable \(X_k\) for \( 1 \leq k \leq n.\) The results continue to hold when \(X_k, 1 \leq k \leq n\) are martingale differences.