This paper proposes a factor model for interval-valued panel data. The authors exploit that the first r largest eigenvalues of the sample covariance matrix divided by N (i.e., \(\frac{\langle{s'_{Y},s_{Y}}\rangle_{K}}{NT}\) ) of interval-valued response variables are Op(1), while the rest are op(1). Then the eigenvalue ratio-type estimators of the number of factors are proposed. Under certain conditions, the proposed estimators are all proven to be consistent. Moreover, the estimators of interval-valued factors and the loadings can be obtained by the principal components method. Monte Carlo simulation studies show that the proposed estimators have the desired finite sample properties. A real example is analysed for illustrations.
