Analysis of winning streaks in the NHL

The thesis is devoted to winning streaks in the National Hockey League. Analysis is implemented on the data from all regular-season games beginning in the 2013/14 season until 2018/19. Primary argument is that the investigation is only conducted on regular season statistics to avoid playoff bias. Inquiry revolves around whether a predictive and analytical methodology can accurately forecast the continuation or termination of a winning streak in the NHL. Descriptive statistics is used to find and estimate the factors affecting the continuation of an ongoing winning streak. Application of linear and logistic regression with Fixed effects, Mixed effects, Cross-random effects, and Discrete-time Survival analysis is used to discover possible statistical dependencies between variables and create the most precise predictive models. The results suggest that with a rising number of won matches in the time of streak provided by the team, the probability of losing the next game decreases. A statistically significant difference between the probabilities of losing the next match between long and short winning streaks has been proven by rejecting the hypothesis about the same probabilities over the number of consecutive matches won by the observed team. Models used in the thesis did not provide sufficient prediction rates to beat the odds of bookmakers with an accuracy rate of 56% by the most accurate model provided.