scientific article; zbMATH DE number 4022364

From MaRDI portal
Publication:3765029





Bayesian inference is presented on parameters of the trend \[ y_ t=\alpha \exp (u_ t)/(1+\delta e^{-\gamma t}),\quad \alpha,\gamma,\delta >0, \] where \(u_ t\) are independent and normally \((N(0,\sigma^ 2))\) distributed. It is proved, that in the case of a priori distributions of the type: \(g(\beta,\gamma,\delta,\sigma)\propto \sigma^{-1}g(\gamma,\delta)\) (where \(\beta =\ell n \alpha)\), the Bayesian analysis requires only double and single numerical integrations. Since the Jeffreys' rule gives an a priori distribution, belonging to the mentioned above class, there are also presented some empirical examples with distributions \(g(\gamma,\delta)=1/\gamma \cdot \delta\), \(g(\gamma,\delta)=1\) and g(\(\gamma\),\(\delta)\) resulting from this rule. The results obtained by means of Bayesian estimation are compared with the results produced by the maximum likelihood method.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3765029)