On probability and moment inequalities for supermartingales and martingales (Q5901239): Difference between revisions

The author obtains Burkholder-type inequalities for martingales which are products of independent random variables. More precisely, let \(V_1,\dots, V_n\) be independent random variables having mean zero and let \(S_j:= V_1(V_2+ 1)\cdots (V_j+ 1)\), \(j= 1,\dots, n\). Then \((S_j)\) is a martingale with respect to the filtration \(({\mathcal F}_j)\) \((j= 0,\dots, n)\) generated by \(V_1,\dots, V_n\). Put, for \(j= 1,\dots, n\), \(\sigma^2_j= E[X^2_j\mid{\mathcal F}_{j-1}]\), where \(X_j= S_j- S_{j-1}\), and let \(B^2_n= \sigma^2_1+\cdots+ \sigma^2_n\), \(A_t= E[|X_1|^t]+\cdots+ E[|X_n|^t]\). Theorem 1. Let \(E[V^4_j]<\infty\) and \(E[V^3_j]= 0\), \(j= 1,\dots, n\). Then \[ E[S^4_n]< 816A_4+ 893 \sum_{1\leq j< k\leq n} E[\sigma^2_j \sigma^2_k]< 816A_4+ 447E[B^4_n]. \] Now let \(S_j:= (V_1+ 1)\cdots(V_j+ 1)\) \((j= 1,\dots, n)\) and put \(\beta_t(n):= E[|X_n|^t]\). Theorem 2. For all \(t\geq 2\), \[ E[|S_n|^t]= (t(t- 1)/2)^{t/2} \Biggl(\sum^n_{j=1} \beta^{2/t}_t(j)\Biggr)^{t/2}. \]