These situations can occur when you have given COV or EWSTATS a small number of effective samples, increasing the effects of machine imprecision and causing small negative eigenvalues in the covariance matrix. If you are not familiar with the types of errors present when using computers to perform floating point calculations, see the Related Solution listed at the bottom of the page.
With EWSTATS, this is typically caused by using a small WindowLength input or by having a few observations compared to the number of securities being used. With COV, this means that you need more observations in your data to counter the effects of this round-off error.
In the meantime, one possible alternative is to use a larger value for WindowLength or to use more samples in your data, which is a more numerically sound suggestion for forming the covariance matrix.
Another alternative includes removing the error checking for this type of situation in the FRONTCON or PORTCON functions. The first method is recommended, however if you want to do this you can comment out the lines which look like:
if any( eig(ECov) < 0 )
error('Covariance matrix must be positive semi-definite.');
end
These lines are found in the beginning of the FRONTCON and PORTCON functions.
For some customers who are unable to add more samples, the following File Exchange submission might be useful:
Best Answer