Under a pre-specified number of groups and the number of common factors, this function implements clustering for N individuals in the panels. Each of individuals in the group are subject to the group-specific unobserved common factors.
PDMIFCLUST(X, Y, NGfactors, NLfactors, Maxit = 100, tol = 0.001)The (NT) times p design matrix, without an intercept where N=number of individuals, T=length of time series, p=number of explanatory variables.
The T times N panel of response where N=number of individuals, T=length of time series.
A pre-specified number of common factors across groups (see example).
A pre-specified number of factors in each groups (see example).
A maximum number of iterations in optimization. Default is 100.
Tolerance level of convergence. Default is 0.001.
A list with the following components:
Label: The estimated group membership for each of the individuals.
Coefficients: The estimated heterogeneous coefficients.
Lower05: Lower end (5%) of the 90% confidence interval of the regression coefficients.
Upper95: Upper end (95%) of the 90% confidence interval of the regression coefficients.
GlobalFactors: The estimated common factors across groups.
GlobalLoadings: The estimated factor loadings for the common factors.
GroupFactors: The estimated group-specific factors.
GroupLoadings: The estimated factor loadings for each group.
pval: p-value for testing hypothesis on heterogeneous coefficients.
Se: Standard error of the estimated regression coefficients.
Ando, T. and Bai, J. (2016) Panel data models with grouped factor structure under unknown group membership Journal of Applied Econometrics, 31, 163-191.
Ando, T. and Bai, J. (2017) Clustering huge number of financial time series: A panel data approach with high-dimensional predictors and factor structures. Journal of the American Statistical Association, 112, 1182-1198.
# NOT RUN {
fit <- PDMIFCLUST(data5X,data5Y,2,c(2,2,2),20,0.5)
# }
Run the code above in your browser using DataLab