[The final entry on this post can be accessed here]
On the 28th of August I began recording Sky Sports Pundit and ex-Premier League Footballer Paul Merson’s predictions for EPL fixtures. These predictions are usually published by Sky Sports the day before EPL matches. The first three entries that followed his predictions can be found here , here and here. Given that Paul will make 380 predictions over the course of the Premier League season (210/380 to date), we have a rare opportunity to analyse the accuracy of a football pundit and Sky Sports football expert who systematically predicts.
After a busy Christmas period, 70 more fixtures have past and Merse has now predicted the results for 210 matches. He has successfully predicted the right score line 20 times, the right result 85 times and been incorrect on 105 occasions. As per the past entries Merson's predications are compared to the output of a random number generator. Merse's predictions have improved over the Christmas period and he no longer gets more wrong than right. Paul's pie chart is below and he has widened the gap between his estimates of match outcomes from that of randomness.
Also included below is Merson's Premier League Table and how he believed the league would look in light of his estimations. Merson has successfully predicted the standing of the first two teams and has accurately estimated the points West Ham have collected. There does however appear to be tentative evidence of what behavioural economists would refer to as a misattribution effect - their seems to be excessive optimism in relation to the performance of certain clubs such as Chelsea, Manchester City, Arsenal and Liverpool and excessive pessimism associated with Southampton, Newcastle, Tottenham and Burnley. For instance Merse thinks that by this stage Burnley would have a goal difference of minus 35 (!) and that Chelsea would have remained unbeaten and only have conceded 7 goals. It may be the case that Merse is misattributing desirable outcomes (big wins for Arsenal or Chelsea for example) and overestimating their likelihood in comparison to neutral events.