Is it true to say: $$a = o\Big(\mathrm{Var}(X)\Big) \implies a = o\Big(\sqrt{\operatorname{Var}(X)}\Big)$$ AS I think by the definition of $o$ notation we can have: \begin{align} a = o\Big(\mathrm{Var}(X)\Big) &\implies \frac{a}{\mathrm{Var}(X)} \rightarrow 0 \quad\text{and}\quad \operatorname{Var}(X)>\sqrt{\operatorname{Var}(X)}&\\ &\implies \frac{a}{\sqrt{\operatorname{Var}(X)}} \rightarrow0 &\\ &\implies a = o\Big(\sqrt{\operatorname{Var}(X)}\Big)& \end{align} Am I thinking right? any why?


I am not sure this exactly corresponds to your question, but here is a result that may be useful in that context

If $E(z_n^2)=O(a_n^2)$, $$z_n=O_p(a_n).$$

Proof: By assumption there is a $M_1$ and $n_\epsilon$, so that $$E(z_n^2)<M_1^2a_n^2$$ for $n\geqslant n_\epsilon$. From the Tschebychev inequality it follows that, for arbitrary $M_2>0$, $$ \Pr(|z_n|\geqslant M_2a_n)\leqslant\frac{E(z_n^2)}{M_2^2a_n^2} $$ Now choose, for some $\epsilon>0$, $M_2\geqslant M_1\epsilon^{-1/2}$, so that \begin{eqnarray*} \Pr(|z_n|/a_n\geqslant M_2)&<&\frac{M_1^2a_n^2}{M_2^2a_n^2}\\ &\leqslant&\frac{M_1^2}{M_1^2\epsilon^{-1}}\\ &=&\epsilon \end{eqnarray*}


If we have $$E(z_n-E(z_n))^2=O(a_n^2)$$ and $$E(z_n)=O(a_n),$$ then $$E(z_n^2)=E(z_n-E(z_n))^2+E(z_n)^2$$ implies that $$z_n=O_p(a_n).$$ The rate of the sequence of r.v.s then corresponds to the rate of the standard deviation.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.