Order Determination of Large Dimensional Dynamic Factor Model

Abstract: Consider the following dynamic factor model:

m a t h b f R_{t} = s u m_{i = 0}^{q} m a t h b f {L a m b d a}_{i} m a t h b f f_{t - i} + m a t h b f e_{t}, t = 1, ..., T

, where

m a t h b f {L a m b d a}_{i}

is an

n i m e s k

loading matrix of full rank,

m a t h b f f_{t}

are i.i.d.

k i m e s 1

-factors, and

m a t h b f e_{t}

are independent

n i m e s 1

white noises. Now, assuming that

n / T o c > 0

, we want to estimate the orders

k

and

q

respectively. Define a random matrix

mathbf{Phi}_n( au)=frac{1}{2T}sum_{j=1}^T (mathbf{R}_j mathbf{R}_{j+ au}^* + mathbf{R}_{j+ au} mathbf{R}_j^*),

where

a u g e 0

is an integer. When there are no factors, the matrix

P h i_{n} (a u)

reduces to

mathbf{M}_n( au) = frac{1}{2T} sum_{j=1}^T (mathbf{e}_j mathbf{e}_{j+ au}^* + mathbf{e}_{j+ au} mathbf{e}_j^*).

When

a u = 0

,

m a t h b f M_{n} (a u)

reduces to the usual sample covariance matrix whose ESD tends to the well known MP law and

m a t h b f {P h i}_{n} (0)

reduces to the standard spike model. Hence the number

k (q + 1)

can be estimated by the number of spiked eigenvalues of

m a t h b f {P h i}_{n} (0)

. To obtain separate estimates of

k

and

q

, we have employed the spectral analysis of

m a t h b f M_{n} (a u)

and established the spiked model analysis for

m a t h b f {P h i}_{n} (a u)

.

This page was built for publication: Order Determination of Large Dimensional Dynamic Factor Model