We investigate price discovery, volatility spillovers and impacts of speculation in the dairy sector. Examining the time series properties of cheese cash and implied futures price we find that the unit root hypothesis is strongly rejected for cash prices, while unit roots cannot be rejected for nearby futures prices in the framework that carefully controls for rollovers. To explain this result, we built a model that illustrates the time series properties of the nearby futures price series for a futures contract written on a second-order stationary cash series and identified the mean-reverting nonlinear dynamics that will occur at rollovers. Given the time series properties of the cash and futures series we propose an error-correction model using spreads between cash and the second nearby futures instead of the cointegration vector. To account for volatility dynamics we propose an extension of the BEKK variance model that we refer to as GARCH-MEX. That model does not restrict the sign of the additional regressors on the conditional variances, and can easily insure positive-definiteness of the conditional covariance matrix. We find that the flow of information in the mean model is predominantly from futures to cash, while volatility spillovers are bidirectional. It is possible that cash prices that include unfilled bid/offers react differently to increases in volatility in futures prices than sales cash prices, indicating that liquidity in the cash market is reduced with increase in conditional volatility of the futures price. Utilizing GARCH-MEX model we find strong evidence against the speculation on realized volatility of futures prices, bid-ask spread and magnitude of slippage.