A multilevel model for repeated cross-sectional data with stochastic volatility

In this paper we propose a multilevel approach for the analysis of repeated cross-sectional data that exhibit volatility effects. We treat individuals as clustered within time-points so that the dynamics over time is modelled at the second level. Items sold in auction present a structure like that of repeated cross-sectional surveys since different goods are sold at different time-points. For prices of artworks, as well as for other assets (financial, insurance, etc.) the hypothesis of constant volatility appears unreasonable. In this work we combine a multilevel model with autoregressive random effects and a stochastic volatility model in order to account for the kurtosis and the volatility pattern of prices. We apply the model to Tribal art auction prices and show improvement over existing proposals both in terms of fit and forecasting