Estimating the Pair of Factor Numbers via Eigenvalue Ratios or Rank Minimization.

Description

The function is to estimate the pair of factor numbers via eigenvalue-ratio corresponding to RMFA method or rank minimization and eigenvalue-ratio corresponding to Iterative Huber Regression (IHR).

Usage

KMHFA(X, W1 = NULL, W2 = NULL, kmax, method, max_iter = 100, c = 1e-04, ep = 1e-04)

Arguments

Details

If method="P", the number of factors k_1 and k_2 are estimated by

\hat{k}_1 = \arg \max_{j \leq k_{max}} \frac{\lambda _j (\bold{M}_c^w)}{\lambda _{j+1} (\bold{M}_c^w)}, \hat{k}_2 = \arg \max_{j \leq k_{max}} \frac{\lambda _j (\bold{M}_r^w)}{\lambda _{j+1} (\bold{M}_r^w)},

where k_{max} is a predetermined value larger than k_1 and k_2. \lambda _j(\cdot) is the j-th largest eigenvalue of a nonnegative definitive matrix. See the function MHFA for the definition of \bold{M}_c^w and \bold{M}_r^w. For details, see He et al. (2023).

Define D=\min({\sqrt{Tp_1}},\sqrt{Tp_2},\sqrt{p_1 p_2}),

\hat{\bold{\Sigma}}_1=\frac{1}{T}\sum_{t=1}^T\hat{\bold{F}}_t \hat{\bold{F}}_t^\top, \hat{\bold{\Sigma}}_2=\frac{1}{T}\sum_{t=1}^T\hat{\bold{F}}_t^\top \hat{\bold{F}}_t,

where \hat{\bold{F}}_t, t=1, \dots, T is estimated by IHR under the number of factor is k_{max}.

If method="E_RM", the number of factors k_1 and k_2 are estimated by

\hat{k}_1=\sum_{i=1}^{k_{max}}I\left(\mathrm{diag}(\hat{\bold{\Sigma}}_1)>P_1\right), \hat{k}_2=\sum_{j=1}^{k_{max}}I\left(\mathrm{diag}(\hat{\bold{\Sigma}}_2) > P_2\right),

where I is the indicator function. In practice, P_1 is set as \max \left(\mathrm{diag}(\hat{\bold{\Sigma}}_1)\right) \cdot D^{-2/3}, P_2 is set as \max \left(\mathrm{diag}(\hat{\bold{\Sigma}}_2)\right) \cdot D^{-2/3}.

If method="E_ER", the number of factors k_1 and k_2 are estimated by

\hat{k}_1 = \arg \max_{i \leq k_{max}} \frac{\lambda _i (\hat{\bold{\Sigma}}_1)}{\lambda _{i+1} (\hat{\bold{\Sigma}}_1)+cD^{-2}}, \hat{k}_2 = \arg \max_{j \leq k_{max}} \frac{\lambda _j (\hat{\bold{\Sigma}}_2)}{\lambda _{j+1} (\hat{\bold{\Sigma}}_2)+cD^{-2}}.

Value

Author(s)

Yong He, Changwei Zhao, Ran Zhao.

References

He, Y., Kong, X., Yu, L., Zhang, X., & Zhao, C. (2023). Matrix factor analysis: From least squares to iterative projection. Journal of Business & Economic Statistics, 1-26.

He, Y., Kong, X. B., Liu, D., & Zhao, R. (2023). Robust Statistical Inference for Large-dimensional Matrix-valued Time Series via Iterative Huber Regression. <arXiv:2306.03317>.

Examples

set.seed(11111)
   T=20;p1=20;p2=20;k1=3;k2=3
   R=matrix(runif(p1*k1,min=-1,max=1),p1,k1)
   C=matrix(runif(p2*k2,min=-1,max=1),p2,k2)
   X=array(0,c(T,p1,p2))
   Y=X;E=Y
   F=array(0,c(T,k1,k2))
   for(t in 1:T){
     F[t,,]=matrix(rnorm(k1*k2),k1,k2)
     E[t,,]=matrix(rnorm(p1*p2),p1,p2)
     Y[t,,]=R%*%F[t,,]%*%t(C)
   }
   X=Y+E
   
   KMHFA(X, kmax=6, method="P")
   
   KMHFA(X, W1 = NULL, W2 = NULL, 6, "E_RM")
   KMHFA(X, W1 = NULL, W2 = NULL, 6, "E_ER")
   

Estimating the Pair of Factor Numbers via Eigenvalue Ratios Corresponding to \alpha-PCA

Description

The function is to estimate the pair of factor numbers via eigenvalue ratios corresponding to \alpha-PCA.

Usage

KPCA(X, kmax, alpha = 0)

Arguments

Details

The function KPCA uses the eigenvalue-ratio idea to estimate the number of factors. In details, the number of factors k_1 is estimated by

\hat{k}_1 = \arg \max_{j \leq k_{max}} \frac{\lambda _j (\hat{\bold{M}}_R)}{\lambda _{j+1} (\hat{\bold{M}}_R)},

where k_{max} is a given upper bound. k_2 is defined similarly with respect to \hat{\bold{M}}_C. See the function alpha_PCA for the definition of \hat{\bold{M}}_R and \hat{\bold{M}}_C. For more details, see Chen & Fan (2021).

Value

Author(s)

Yong He, Changwei Zhao, Ran Zhao.

References

Chen, E. Y., & Fan, J. (2021). Statistical inference for high-dimensional matrix-variate factor models. Journal of the American Statistical Association, 1-18.

Examples

   set.seed(11111)
   T=20;p1=20;p2=20;k1=3;k2=3
   R=matrix(runif(p1*k1,min=-1,max=1),p1,k1)
   C=matrix(runif(p2*k2,min=-1,max=1),p2,k2)
   X=array(0,c(T,p1,p2))
   Y=X;E=Y
   F=array(0,c(T,k1,k2))
   for(t in 1:T){
     F[t,,]=matrix(rnorm(k1*k2),k1,k2)
     E[t,,]=matrix(rnorm(p1*p2),p1,p2)
     Y[t,,]=R%*%F[t,,]%*%t(C)
   }
   X=Y+E
   
   KPCA(X, 8, alpha = 0)

Estimating the Pair of Factor Numbers via Eigenvalue Ratios Corresponding to PE

Description

The function is to estimate the pair of factor numbers via eigenvalue ratios corresponding to PE method.

Usage

KPE(X, kmax, c = 0)

Arguments

Details

The function KPE uses the eigenvalue-ratio idea to estimate the number of factors. First, obtain the initial estimators \hat{\bold{R}} and \hat{\bold{C}}. Second, define

\hat{\bold{Y}}_t=\frac{1}{p_2}\bold{X}_t\hat{\bold{C}}, \hat{\bold{Z}}_t=\frac{1}{p_1}\bold{X}_t^\top\hat{\bold{R}},

and

\tilde{\bold{M}}_1=\frac{1}{Tp_1}\hat{\bold{Y}}_t\hat{\bold{Y}}_t^\top, \tilde{\bold{M}}_2=\frac{1}{Tp_2}\sum_{t=1}^T\hat{\bold{Z}}_t\hat{\bold{Z}}_t^\top,

the number of factors k_1 is estimated by

\hat{k}_1 = \arg \max_{j \leq k_{max}} \frac{\lambda_j (\tilde{\bold{M}}_1)}{\lambda _{j+1} (\tilde{\bold{M}}_1)},

where k_{max} is a predetermined upper bound for k_1. The estimation of k_2 is defined similarly with respect to \tilde{\bold{M}}_2. For details, see Yu et al. (2022).

Value

Author(s)

Yong He, Changwei Zhao, Ran Zhao.

References

Yu, L., He, Y., Kong, X., & Zhang, X. (2022). Projected estimation for large-dimensional matrix factor models. Journal of Econometrics, 229(1), 201-217.

Examples

   set.seed(11111)
   T=20;p1=20;p2=20;k1=3;k2=3
   R=matrix(runif(p1*k1,min=-1,max=1),p1,k1)
   C=matrix(runif(p2*k2,min=-1,max=1),p2,k2)
   X=array(0,c(T,p1,p2))
   Y=X;E=Y
   F=array(0,c(T,k1,k2))
   for(t in 1:T){
     F[t,,]=matrix(rnorm(k1*k2),k1,k2)
     E[t,,]=matrix(rnorm(p1*p2),p1,p2)
     Y[t,,]=R%*%F[t,,]%*%t(C)
   }
   X=Y+E
   
   KPE(X, 8, c = 0)

Matrix Huber Factor Analysis

Description

This function is to fit the matrix factor models via the Huber loss. We propose two algorithms to do robust factor analysis. One is based on minimizing the Huber loss of the idiosyncratic error's Frobenius norm, which leads to a weighted iterative projection approach to compute and learn the parameters and thereby named as Robust-Matrix-Factor-Analysis (RMFA). The other one is based on minimizing the element-wise Huber loss, which can be solved by an iterative Huber regression algorithm (IHR).

Usage

MHFA(X, W1=NULL, W2=NULL, m1, m2, method, max_iter = 100, ep = 1e-04)

Arguments

Details

For the matrix factor models, He et al. (2021) propose a weighted iterative projection approach to compute and learn the parameters by minimizing the Huber loss function of the idiosyncratic error's Frobenius norm. In details, for observations \bold{X}_t, t=1,2,\cdots,T, define

\bold{M}_c^w = \frac{1}{Tp_2} \sum_{t=1}^T w_t \bold{X}_t \bold{C} \bold{C}^\top \bold{X}_t^\top, \bold{M}_r^w = \frac{1}{Tp_1} \sum_{t=1}^T w_t \bold{X}_t^\top \bold{R} \bold{R}^\top \bold{X}_t.

The estimators of loading matrics \hat{\bold{R}} and \hat{\bold{C}} are calculated by \sqrt{p_1} times the leading k_1 eigenvectors of \bold{M}_c^w and \sqrt{p_2} times the leading k_2 eigenvectors of \bold{M}_r^w. And

\hat{\bold{F}}_t=\frac{1}{p_1 p_2}\hat{\bold{R}}^\top \bold{X}_t \hat{\bold{C}}.

For details, see He et al. (2023).

The other one is based on minimizing the element-wise Huber loss. Define

M_{i,Tp_2}(\bold{r}, \bold{F}_t, \bold{C})=\frac{1}{Tp_2} \sum_{t=1}^{T} \sum_{j=1}^{p_2} H_\tau \left(x_{t,ij}-\bold{r}_i^\top\bold{F}_t\bold{c}_j \right),

M_{i,Tp_1}(\bold{R}, \bold{F}_t, \bold{c})=\frac{1}{Tp_1}\sum_{t=1}^T\sum_{i=1}^{p_1} H_\tau \left(x_{t,ij}-\bold{r}_i^\top\bold{F}_t\bold{c}_j\right),

M_{t,p_1 p_2}(\bold{R}, \mathrm{vec}(\bold{F}), \bold{C})=\frac{1}{p_1 p_2} \sum_{i=1}^{p_1}\sum_{j=1}^{p_2} H_\tau \left(x_{t,ij}-(\bold{c}_j \otimes \bold{r}_i)^\top \mathrm{vec}(\bold{F})\right).

This can be seen as Huber regression as each time optimizing one argument while keeping the other two fixed.

Value

The return value is a list. In this list, it contains the following:

Author(s)

Yong He, Changwei Zhao, Ran Zhao.

References

He, Y., Kong, X., Yu, L., Zhang, X., & Zhao, C. (2023). Matrix factor analysis: From least squares to iterative projection. Journal of Business & Economic Statistics, 1-26.

He, Y., Kong, X. B., Liu, D., & Zhao, R. (2023). Robust Statistical Inference for Large-dimensional Matrix-valued Time Series via Iterative Huber Regression. <arXiv:2306.03317>.

Examples

set.seed(11111)
   T=20;p1=20;p2=20;k1=3;k2=3
   R=matrix(runif(p1*k1,min=-1,max=1),p1,k1)
   C=matrix(runif(p2*k2,min=-1,max=1),p2,k2)
   X=array(0,c(T,p1,p2))
   Y=X;E=Y
   F=array(0,c(T,k1,k2))
   for(t in 1:T){
     F[t,,]=matrix(rnorm(k1*k2),k1,k2)
     E[t,,]=matrix(rnorm(p1*p2),p1,p2)
     Y[t,,]=R%*%F[t,,]%*%t(C)
   }
   X=Y+E
   
   #Estimate the factor matrices and loadings by RMFA
   fit1=MHFA(X, m1=3, m2=3, method="P")
   Rhat1=fit1$R 
   Chat1=fit1$C
   Fhat1=fit1$F
   
   #Estimate the factor matrices and loadings by IHR
   fit2=MHFA(X, W1=NULL, W2=NULL, 3, 3, "E")
   Rhat2=fit2$R 
   Chat2=fit2$C
   Fhat2=fit2$F
   
   #Estimate the common component by RMFA
   CC1=array(0,c(T,p1,p2))
   for (t in 1:T){
   CC1[t,,]=Rhat1%*%Fhat1[t,,]%*%t(Chat1)
   }
   CC1
   
   #Estimate the common component by IHR
   CC2=array(0,c(T,p1,p2))
   for (t in 1:T){
   CC2[t,,]=Rhat2%*%Fhat2[t,,]%*%t(Chat2)
   }
   CC2

Projected Estimation for Large-Dimensional Matrix Factor Models

Description

This function is to fit the matrix factor model via the PE method by projecting the observation matrix onto the row or column factor space.

Usage

PE(X, m1, m2)

Arguments

Details

For the matrix factor models, Yu et al. (2022) propose a projection estimation method to estimate the model parameters. In details, for observations \bold{X}_t, t=1,2,\cdots,T, the data matrix is projected to a lower dimensional space by setting

\bold{Y}_t = \frac{1}{p_2} \bold{X}_t \bold{C}.

Given \bold{Y}_t, define

\bold{M}_1 = \frac{1}{Tp_1} \sum_{t=1}^T \bold{Y}_t \bold{Y}_t^\top,

and then the row factor loading matrix \bold{R} can be estimated by \sqrt{p_1} times the leading k_1 eigenvectors of \bold{M}_1. However, the projection matrix \bold{C} is unavailable in practice. A natural solution is to replace it with a consistent initial estimator. The column factor loading matrix \bold{C} can be similarly estimated by projecting \bold{X}_t onto the space of \bold{C} with \bold{R}. See Yu et al. (2022) for the detailed algorithm.

Value

The return value is a list. In this list, it contains the following:

Author(s)

Yong He, Changwei Zhao, Ran Zhao.

References

Yu, L., He, Y., Kong, X., & Zhang, X. (2022). Projected estimation for large-dimensional matrix factor models. Journal of Econometrics, 229(1), 201-217.

Examples

   set.seed(11111)
   T=20;p1=20;p2=20;k1=3;k2=3
   R=matrix(runif(p1*k1,min=-1,max=1),p1,k1)
   C=matrix(runif(p2*k2,min=-1,max=1),p2,k2)
   X=array(0,c(T,p1,p2))
   Y=X;E=Y
   F=array(0,c(T,k1,k2))
   for(t in 1:T){
     F[t,,]=matrix(rnorm(k1*k2),k1,k2)
     E[t,,]=matrix(rnorm(p1*p2),p1,p2)
     Y[t,,]=R%*%F[t,,]%*%t(C)
   }
   X=Y+E
   
   #Estimate the factor matrices and loadings
   fit=PE(X, k1, k2)
   Rhat=fit$R 
   Chat=fit$C
   Fhat=fit$F
   
   #Estimate the common component
   CC=array(0,c(T,p1,p2))
   for (t in 1:T){
   CC[t,,]=Rhat%*%Fhat[t,,]%*%t(Chat)
   }
   CC

Statistical Inference for High-Dimensional Matrix-Variate Factor Model

Description

This function is to fit the matrix factor model via the \alpha-PCA method by conducting eigen-analysis of a weighted average of the sample mean and the column (row) sample covariance matrix through a hyper-parameter \alpha.

Usage

alpha_PCA(X, m1, m2, alpha = 0)

Arguments

Details

For the matrix factor models, Chen & Fan (2021) propose an estimation procedure, i.e. \alpha-PCA. The method aggregates the information in both first and second moments and extract it via a spectral method. In detail, for observations \bold{X}_t, t=1,2,\cdots,T, define

\hat{\bold{M}}_R = \frac{1}{p_1 p_2} \left( (1+\alpha) \bar{\bold{X}} \bar{\bold{X}}^\top + \frac{1}{T} \sum_{t=1}^T (\bold{X}_t - \bar{\bold{X}}) (\bold{X}_t - \bar{\bold{X}})^\top \right),

\hat{\bold{M}}_C = \frac{1}{p_1 p_2} \left( (1+\alpha) \bar{\bold{X}}^\top \bar{\bold{X}} + \frac{1}{T} \sum_{t=1}^T (\bold{X}_t - \bar{\bold{X}})^\top (\bold{X}_t - \bar{\bold{X}}) \right),

where \alpha \in [-1,+\infty], \bar{\bold{X}} = \frac{1}{T} \sum_{t=1}^T \bold{X}_t, \frac{1}{T} \sum_{t=1}^T (\bold{X}_t - \bar{\bold{X}}) (\bold{X}_t - \bar{\bold{X}})^\top and \frac{1}{T} \sum_{t=1}^T (\bold{X}_t - \bar{\bold{X}})^\top (\bold{X}_t - \bar{\bold{X}}) are the sample row and column covariance matrix, respectively. The loading matrices \bold{R} and \bold{C} are estimated as \sqrt{p_1} times the top k_1 eigenvectors of \hat{\bold{M}}_R and \sqrt{p_2} times the top k_2 eigenvectors of \hat{\bold{M}}_C, respectively. For details, see Chen & Fan (2021).

Value

The return value is a list. In this list, it contains the following:

Author(s)

Yong He, Changwei Zhao, Ran Zhao.

References

Chen, E. Y., & Fan, J. (2021). Statistical inference for high-dimensional matrix-variate factor models. Journal of the American Statistical Association, 1-18.

Examples

   set.seed(11111)
   T=20;p1=20;p2=20;k1=3;k2=3
   R=matrix(runif(p1*k1,min=-1,max=1),p1,k1)
   C=matrix(runif(p2*k2,min=-1,max=1),p2,k2)
   X=array(0,c(T,p1,p2))
   Y=X;E=Y
   F=array(0,c(T,k1,k2))
   for(t in 1:T){
     F[t,,]=matrix(rnorm(k1*k2),k1,k2)
     E[t,,]=matrix(rnorm(p1*p2),p1,p2)
     Y[t,,]=R%*%F[t,,]%*%t(C)
   }
   X=Y+E
   
   #Estimate the factor matrices and loadings
   fit=alpha_PCA(X, k1, k2, alpha = 0)
   Rhat=fit$R 
   Chat=fit$C
   Fhat=fit$F
   
   #Estimate the common component
   CC=array(0,c(T,p1,p2))
   for (t in 1:T){
   CC[t,,]=Rhat%*%Fhat[t,,]%*%t(Chat)
   }
   CC