|
This study employs fractional cointegration analysis method to investigate the appropriate form of gold-silver parity in both cashand futures markets. The identified parity relationships are then incorporated into the corresponding error correction models to forecast variations of gold-silver spread. Thereafter,spread tradingsimulations are conducted to examine the information value revealed by the parity relationships. It is found that the gold-silver parity contains a time-varying risk premium and is a slow adjustment-longmemory process as analyzed using a two-stage procedure similar to Engle & Granger''s(1987). Since ADF tests are less powerful than GPHtests in detecting general mean-reverting relationships, the gold-silver parity could not be identified by more rigorous cointegrationanalysis. On the other hand, futures gold-silver spreads lead cash spreads in terms of information reflection. And the cash and futures spreads are cointegrated. The 5-step ahead forecast results of spreaderror correction models embodying the parity relationships outperformsthe nodels without the parity informations. But the opposite is true for 1-step ahead forecast. Considering the long memory property of the gold-silver parity, the forecast findings are justified. In addition, the efficiency of gold and silver marklet as a whole is questionable since abnormal profit could be made based on the simulation results of spread trading.
|