2017 : Simulating the Final Series After Week 1 of the Finals

And then there were six ...

With the Elimination and Qualifying Finals now completed, our simulations involve generating predictions for the 6 remaining games by applying the methodology as described in this blog.

In the chart below we record the proportion of the 10,000 simulations in which a particular team went out in a given week of the Finals.

The weekend's results were, according to the simulations, particularly beneficial for Richmond, who are now almost equal-favourites with Adelaide for the Flag. On a team-by-team basis, the changes in Flag chances were as follows:

  • Geelong -9%
  • Port Adelaide -6%
  • Essendon -2%
  • GWS -2%
  • West Coast +2%
  • Adelaide +2%
  • Sydney +5%
  •  Richmond +10%

The probabilities of going out in a particular week are shown in the heatmap below.

If you look closely at the methodology I've been using for these simulations, it includes perturbing the offensive and defensive ratings of both teams independently in every game based on random draws from a Normal distribution with a mean of 24 points. These perturbations add more variability to the simulated results relative to that which is incorporated in the one-week ahead forecasts used in my weekly tipping. Most importantly, the additional variability tends to move probabilities close to 50% for both teams (see this blog post to understand why this happens).

There's ample empirical rationale for perturbing the ratings in this way when projecting the remainder of the home and away season, but I've not assessed the efficacy of this practice when applied to the finals series. 

So, for the charts below I've re-run the simulations without perturbations.

Doing this shrinks the estimated Flag chances of the teams with small probabilities (viz Geelong, GWS and West Coast) and grows the estimated chances of the remaining teams (viz Adelaide, Richmond and Sydney - though for the latter only slightly).

As a heatmap we have.

So, being conservative, we might conclude that the Flag chances of each team lie in the following ranges:

  • Adelaide 33-37%
  • Richmond 29-33%
  • Sydney 15-16%
  • GWS 8-11%
  • Geelong 5-8%
  • West Coast 2-4%

As a last output, let's look at the underlying team-versus-team probabilities that fall out of the two simulation approaches. The first set of probabilities relate to the simulations we looked at first in this blog and you can see that, as I noted, they tend to be nearer 50% than the second set of probabilities, which come when we remove the additional perturbations.

Interestingly, what both sets of probabilities suggest is that the more even-matched Grand Finals would have Adelaide facing Richmond, Sydney facing Richmond, or Geelong facing GWS.

A Sydney or Adelaide versus West Coast Grand Final would be the most mismatched.