| Invented at Bell Lab in 1947, the semiconductor is becoming the key to today's technology progress. The purposes of this paper are to examine the causes of the semiconductor industry cycle and to develop a modern quantitative method to better forecast the cycle. We first survey industry trade articles on the semiconductor industry cycle issue and then collect and examine the rules of thumb used to predict the semiconductor cycle by professional forecasters. A 12-variable Vector Autoregression (VAR) model is constructed to explore the dynamics among the macroeconomic and industry-level variables. The results from impulse responses and variance decompositions of the VAR model are shown and discussed. Finally, a comparison of the forecasts from three different forecasting methods, the Kalman Filter, the Standardized VAR and the Random Walk methods, is presented. We find that the Standardized VAR forecasting method outperforms the others in terms of both Minimum Mean Squared Forecast Errors and Minimum Standard Deviation of Forecast Errors criteria. |
