Research On The Time Varying Characteristics Of Silver Futures Market Based On The Beyesian Panel Smooth Transition Regression Models

The nonlinear relationship is common phenomenon in economic and financialproblem analysis. With the rapid development of nonlinear theory and non-linearmodels, the dominated linear paradigm analysis methods of modern economic theoryare challenged, and the consensus is that nonlinear models is better tools fordescribing complex economic phenomena and revealing the inherent law of economicoperation. Particularly, Panel smooth transition models, as a widely used researchmethod of regime switching models among the nonlinear models, are attracting moreand more attention from scholars because of their ability to reveal the possiblelyexisting nonlinear characteristics effectively and better decipt the cross-sectionalheterogeneity of panel data. A well-known advantages what it is convenient to reflectsmooth transition process in different mechanisms accurately and track the evolutionof the asymptotic behavior. In addition, the application of panel data which iscomposed of cross-sectional data and time-series data makes the model includeheterogeneity and commonality of economic variables, so the models are importanttools for multi-level analysis of social and economic phenomenon. However, thenonlinear least squares (NLS) algorithm, which is usually used for parameterestimation, may not reach the given accuracy, and it leads to the non-convergencetroubles. Moreover, it is difficult to test significance of the position and slopeparameter of model. Meanwhile, the theoretical hypotheses contradict with thepractical background of the changeable data generating behaviours and parameterdistributions of the economic variables. Furtherly, due to complex non-linear paneldata model construction frequently encounter high-dimension integral issues, it isdifficult to guarantee the reliability of the model parameter estimation.Bayesian panel data models with fixed-effect, Bayesian panel threshold modeland Bayesian panel smooth transition models are established in this paper to addressuncertain risks of the parameters in models. Based on the analysis of models statisticstructure and the selection of appropriate parameter prior, the inference of parameterposteriors have been done, then the Gibbs-MH hybrid sample algorithm is utilized toobtain the parameters’ distribution characteristics and its estimates. Monte Carlosimulation results show that the Bayesian panel smooth transition model canaccurately determine the conversion mechanism in the models and identify variables with non-linear characteristics at the same time, which provides the proof ofeffectiveness of the Bayesian MCMC sample algorithm design.The empirical researche’s panel data come from metal futures market (preciousmetals futures, gold and silver; non-ferrous metals futures, copper, aluminum, lead andzinc; ferrous metals futures, rebar). This paper regard the the gold futures price as thetransition variable, then constract the Bayesian panel smooth transition model toinvestigate the possiblely existing nonlinear relationship between silver futures marketand other non-precious metals futures markets, such as the copper futures, aluminumfutures, lead futures, zinc futures and rebar futures. The empirical results show thatthe nonlinear relationship between the silver futures market and the copper futures,aluminum futures, rebar futures market is significant; while the relationship betweensilver futures market and lead futures, zinc futures market is briefly linear. And theempirical results demonstrate the Bayesian panel smooth transition model is effective.