Inference Discussion
I am a die-hard football fan and a big follower of the English Premier League. I developed this football prediction model in Microsoft Excel which helps me to predict or infer the probability of goals in a match, or to infer whether a team will win, lose, or if a game will end in a draw. Most of my friends – who are also football fans – like betting on the English premier league, and they sometimes ask me to tell which team I think will win, or how many goals a team will score. My friends’ requests made me do some research online, and I was able to find an article that guided me on how to build a prediction model in Excel.
I use my model to infer goals that will be scored in a match, and sometimes I use my inferences to advise my friends on bets that they can place; I am not a big fan of betting. My model basically uses historical data to make inferences. What I do is that I take the number of goals scored and goals conceded in the first half of an EPL season and feed it into Microsoft excel. I have formulas in my spreadsheet that use the data to calculate the defensive strengths and the attacking strengths for each team. The attacking and defensive strengths are then converted into goal expectation figures (using formula) which are then fed into a Poisson distribution, and then the Poisson distribution formulas work out the probability of every score, and the probability of a team winning, losing or a match ending in a draw. My spreadsheet also infers whether a match will end in a home or an away win, whether both teams will score and whether only one team will score.
For example, if Arsenal scores 36 goals in 19 matches and concedes 29 goals, then the average goals scored by the team will be 36/19, which will be equal to 1.89. The average goals conceded will be 29/19, which will equal 1.53. To calculate Arsenal’s attacking strength, I divide 1.89 by the average number of goals scored by all teams. The defensive strength is calculated by dividing 1.53 by the average goals scored by all teams in all the 19 matches. Once all these figures have been calculated for all teams, they are then fed into a Poisson distribution to calculate the probability of different scores.
As games progress, I input new scores into my model because outdated data can give inaccurate results. My spreadsheet is automated, so all I do is to capture scores for the home and away teams, and I leave every other thing to Microsoft excel. My predictive model has proved to be beneficial in most of the cases, but results are not always accurate. The accuracy of my model is not one hundred percent because of certain reasons. For example, it predicts or infer future results using past data. Something that happened over six months with different players and under different weather conditions may not help us accurately infer what will happen next. Managers and coaches also come and go, but my model does not consider these things. Another limitation of this model is that it only uses past results to infer future scores. Past goals may not give an accurate picture of a team because there are instances when a dominant team loses to a weaker team. The probability of zero and the probability of draws may also be underestimated by the Poisson distribution, a problem that can be corrected using a zero-inflated Poisson.
