The autoregressive process takes an important part in predicting problems leading to decision making. In practice, we use the least squares method to estimate the parameter θ̃ of the first-order autoregressive process taking values in a real separable Banach space B (ARB(1)), if it satisfies the following relation: $X{̃}_t=\theta ̃X{̃}_{t-1}+\epsilon {̃}_t$. In this paper we study the convergence in distribution of the linear operator $I(\theta {̃}_T,\theta ̃)=(\theta {̃}_T-\theta ̃)\theta {̃}^{T-2}$ for ||θ̃|| > 1 and so we construct inequalities of Bernstein type for this operator.

For almost all infinite binary sequences of Bernoulli trials $(p,q)$ the frequency of blocks of length $k\left(N\right)$ in the first $N$ terms tends asymptotically to the probability of the blocks, if $k\left(N\right)$ increases like ${\log }_{\frac{1}{p}}N-{\log }_{\frac{1}{p}}N-\psi \left(N\right)$ (for $p\le q$) where $\psi \left(N\right)$ tends to $+\infty$. This generalizes a result due to P. Flajolet, P. Kirschenhofer and R.F. Tichy concerning the case $p=q=\frac{1}{2}$.

Passing from Cesàro means to Borel-type methods of summability we prove some ergodic theorem for operators (acting in a Banach space) with spectrum contained in ℂ∖(1,∞).