After careful analysis, we believe that it is appropriate to use the ARIMA (0,1,0) model (Random Walk Model) to forecast the
average goals scored per game in a NHL season.
This data was modelled using statistical methods such as linear regression, exponential smoothing, and ARIMA models. The
analysis indicated that and ARIMA (4,1,0) model had the best predictive results, based on several statistical measures. However,
trying to interpret this model was extremely difficult. In the end, it was decided that although the data set was a fair size,
it was not large enough to completely eliminate the possibility of modelling statistical noise. It seemed reasonable that the
AR (4) component of the model was arising due to statistical. Therefore, the conclusion is that the ARIMA (0,1,0), the "random
walk" or "naive forecast," is the best model to use to predict the future average number of goals scored per game in a season.
It was somewhat disappointing that the number of goals scored per season could not be explained more clearly. Upon reflection,
it seems that, similar to the stock market, trends or correlations between number of goals and other variables can only be
determined after they have occurred, and rarely last for an extended period of time.