Abstract
We give three methods for estimating Markov transition matrices when observed state probabilities are not all either zeros or ones and a simulation–based comparison of the performance of the estimators. Best performing was the least squares estimator with the constraint that elements be non-negative. Another estimation method resembles a method proposed for analyzing fuzzy time series that is also a generalization of the traditional zero–one Markov estimation technique. This estimator performs well for small sample sizes but is not consistent. The third estimator, the least squares estimator, does not always perform better with increasing sample size. Surprisingly, there is a sample size for which discarding an observation selected at random improves the performance of the least squares estimator
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call