1: \begin{abstract}

2: An observed $K$-dimensional series $\left\{ y_{n}\right\} _{n=1}^{N}$

3: is expressed in terms of a lower $p$-dimensional latent series called

4: factors $f_{n}$ and random noise $\varepsilon_{n}$. The equation,

5: $y_{n}=Qf_{n}+\varepsilon_{n}$ is taken to relate the factors with

6: the observation. The goal is to determine the dimension of the factors,

7: $p$, the factor loading matrix, $Q$, and the factors $f_{n}$. Here,

8: it is assumed that the noise co-variance is positive definite and

9: allowed to be correlated with the factors. An augmented matrix, 

10: \[

11: \tilde{M}\triangleq\left[\begin{array}{cccc}

12: \tilde{\Sigma}_{yy}(1) & \tilde{\Sigma}_{yy}(2) & \ldots & \tilde{\Sigma}_{yy}(m)\end{array}\right]

13: \]

14: is formed using the observed sample autocovariances $\tilde{\Sigma}_{yy}(l)=\frac{1}{N-l}\sum_{n=1}^{N-l}\left(y_{n+l}-\bar{y}\right)\left(y_{n}-\bar{y}\right)^{\top}$,

15: $\bar{y}=\frac{1}{N}\sum_{n=1}^{N}y_{n}$. Estimating $p$ is equated

16: to determining the numerical rank of $\tilde{M}$. Using Rank Revealing

17: QR (RRQR) decomposition, a model order detection scheme is proposed

18: for determining the numerical rank and for estimating the loading

19: matrix $Q$. The rate of convergence of the estimates, as $K$ and

20: $N$ tends to infinity, is derived and compared with that of the existing

21: Eigen Value Decomposition based approach. Two applications of this

22: algorithm, i) The problem of extracting signals from their noisy mixtures

23: and ii) modelling of the S\&P index are presented.

24: \end{abstract}

25: