The article discusses a method to estimate covariance matrices of different groups when sample sizes are limited. By combining sample covariance matrices in a linear way, the method reduces estimation errors, especially when the true covariance matrices are similar. The researchers developed a technique to find the best coefficients for this combination, which minimizes errors. They used a special matrix to estimate the normalized covariance matrix accurately. The method can also help in choosing parameters for estimating multiple covariance matrices. The researchers tested the method using simulations and real stock data for portfolio optimization.