StatisticsNZ / demest

Bayesian demographic estimation and forecasting.
Other
21 stars 6 forks source link

Numerical problems with small scales in DLM models #7

Open johnrbryant opened 6 years ago

johnrbryant commented 6 years ago

When the scale terms in DLM models are very small (ie less than 0.001), the results seem to be quite sensitive to different values for the scale - eg 0.0001 gives different results from 0.00001. In theory this shouldn't happen. Are there numerical problems?