This paper aims to use the Bayesian Model Averaging (BMA) to identify the main cause of the global semiconductor industry cycle from a list of 75 U.S. macro and industry-level variables. By using the data from January 1997 to October 2012, we compare the out-of-sample forecasting performance accuracy BMA, ARMA and random walk models. The empirical results show that inventories are the most important variables in explaining the semiconductor industry cycle. Besides, BMA outperforms the other two models in terms of out-of-sample forecasts in both RMSE and DM tests. Nonetheless, BMA also provides better turning-point forecasts.