Preliminary Analysis

As part of this analysis, we will be looking at Linear Regression, Auto-regressive, Simple Exponential, Holt's Double Exponential and ARIMA models and compare these to each other to determine the most suitable model to forecast with. The graph below is our preliminary time series sequence graph for the total number of goals scored per season from 1946 - 1998.

Observing the graph, we see that for the first twenty years there is a negligible increase in total goals per year as the league was static in terms of the way the game was played and the number of teams in the NHL. During the expansion years from 1967 to 1979, we see an increasing trend to the number of goals scored. In this time period, as teams were added (15 in total), the number of games played per team increased from 70 games in the 1966 season to 80 in the 1979 season. The remaining years from 1980 to 1998, we see a fluctuation of goals scored, which warrants further analysis. The significant dip in the 1994 season was caused by the strike that resulted in 36 regular season games being forfeited.

The first logical transformation was to compute the number of goals per game. This eliminates the effects of the independent variables teams, games played, and the effect of the strike-shortened season. This was done so that other explanatory variates may be identified in each year.