Our
tipping systems
International
Rugby Union Prediction
My
System By Stefan Yelas
My
International Rugby Prediction model uses a similar
method that Professor Stephen Clarke uses to predict
the Australian Rules Football results.
Each
team has a rating. The model subtracts the away team
rating from the sum of the home team rating and the
home ground advantage. The result is the predicted
margin. A positive predicted margin means the home
team is predicted to win and a negative predicted
margin means the away team is predicted to win.
After
a game is played, team ratings are updated using simple
exponential smoothing of the error. The model was
initialised and optimised using historical results
from all 566 matches between any of the 20 world cup
teams dating from 1996 (when Rugby Union became professional).
The data used were the team names, score margin and
which team was playing at home.
Initial
values for the ratings
The Zurich World Team Rankings, developed in 1998
by www.planetrugby.com were used as initial ratings
for the prediction models.
The
initialised model correctly predicted 78.8% of the
566 games and the average error in the predicted margins
was 13.9 points. Using the standard deviation of the
errors and inputting it to a function that returned
the normal cumulative distribution for this standard
deviation and specific mean (the predicted margin),
the probability that a team would lose or win can
be calculated.
Success
of the model
The
table below shows the predictions and actual results
for each match, as produced prior to each match in
the 2003 Rugby World Cup.
The
model performed extremely well in that the correct
winner was forecast in 46 out of 48 matches (96% correct).
Furthermore the average error was only 15 points.
The
model correctly predicted England to be the most likely
winner before the tournament had even begun. After
the first week of the competition (once the model
had been seen to work) a $50 pool was placed
with website www.betfair.com. Using the predicted
probabilities and margins, bets were placed on games
where a 20% or higher overlay was apparent. A fraction
of the pool equalling 40% of the probability of the
predicted margin range was bet. At the end of the
event the pool was $638.
Note
that even the semi final and final matches in which
the model predicted the incorrect winner resulted
in a significant contribution to this profit of $160
and $121 respectively. In both cases the model suggested
Australia would do better than the line indicated.
As these matches were late in the tournament, the
bets were a reasonable proportion of what was, by
then, a relatively sizeable pool.
Rugby
Union index page 
Rugby
Union Package
About
our experts
Testimonials
Refund
policy and disclaimer
Home|FAQs|Disclaimer|Contact us
©2007 Ozmium
Pty Ltd. All rights reserved
