During the off season there have been two aspects of the Carlo Monty AFL model that have been under review and development. The first has been the initial examination of weather data and the second has been the refinement of the weighted shot difference forward estimate model.
With respect to the weather model data those (few) followers that are interested in this aspect of game day variation and the potential impact on the score differences etc resulting in a game of AFL I am afraid you
will have to wait for more details. What I can say that I have done is I have had a good look at the Bureau of Meteorology data available - which is exensive to say the least. I have then started downloading aspects that I thought would be relevant and worthy of modelling in more depth – air temperature, wind speed (and direction), rainfall (previous 24 hours), humidity and solar radiation.
will have to wait for more details. What I can say that I have done is I have had a good look at the Bureau of Meteorology data available - which is exensive to say the least. I have then started downloading aspects that I thought would be relevant and worthy of modelling in more depth – air temperature, wind speed (and direction), rainfall (previous 24 hours), humidity and solar radiation.
I have settled on examining airport data as the most reliable and readily available data for game day conditions. Airport data across the different states also offers a reasonable comparison. I have started with an examination of the Melbourne, Essendon and Avalon data. I was intending to expand to Moorabbin, Sydney, Brisbane, Adelaide and Perth with the initial focus on the Melbourne airport data which also includes solar radiation data (at one minute intervals for the past decade). This last phrase
should offer the insight into the current issue that I have – being buried in data and more specifically being able to extract from the massive dataset the pre game and game specific data.
should offer the insight into the current issue that I have – being buried in data and more specifically being able to extract from the massive dataset the pre game and game specific data.
The downloading of a decade’s worth of data (in one minute intervals) is the easy part. Managing
those files is beyond Excels capability and I must resort to Access and enhancing my data extraction macro skills...plodding along. I had hoped to have meaningful data in the spreadsheet by now but it will be a race to the start of the season I feel.
those files is beyond Excels capability and I must resort to Access and enhancing my data extraction macro skills...plodding along. I had hoped to have meaningful data in the spreadsheet by now but it will be a race to the start of the season I feel.
Weighted shot difference prediction model
The other aspect of the Carlo Monte spreadsheet which is allowing me to develop my skills of the Vlookup function is my simplistic weighted shot difference forward prediction estimate model. This takes weighted averages of the previous 20 rounds worth of shots for – shots against and compares them to the 22 round
future prediction. The model was developed on the historical patterns of the Hawthorn data and then applied to the collection of teams. The Hawthorn data was chosen as it has a range of wooden spoons and premierships (and they are the team I follow). I am still fine tuning the simple model but using some simple cut offs of >+5 or <-5 to predict grand finalists and wooden spoon contenders seems sensible.
future prediction. The model was developed on the historical patterns of the Hawthorn data and then applied to the collection of teams. The Hawthorn data was chosen as it has a range of wooden spoons and premierships (and they are the team I follow). I am still fine tuning the simple model but using some simple cut offs of >+5 or <-5 to predict grand finalists and wooden spoon contenders seems sensible.
I am also using the “going into the current round” weighted shot difference and the end of year prediction to graphically highlight the relative strengths of the competing teams each round.
The end of year predictions is one way of forming a season end predicted ladder:
Based on this thinking we are in for a repeat of this year’s Grand Finalists; North and Port have a
realistic crack at making the Grand Final; the battle for the wooden spoon will be keenly fought as will the final two spots for the 8.
realistic crack at making the Grand Final; the battle for the wooden spoon will be keenly fought as will the final two spots for the 8.
The alternative season end predicted ladder is via the monte carlo modelling of all of the individual games during the season and adjusting the pre game prediction inputs based on the previous results. By running the full season a thousand times gives a spread of possibilities which I have chosen to rank via predicted games
won (average, min and maximum):
won (average, min and maximum):
This method seriously boosts Richmond’s chances and dramatically puts Collingwood into contention for the wooden spoon. Both aspects I feel are intuitively wrong. What I will say (and I am reiterating many earlier blog posts) is that the monte carlo model prediction method round by round I am really only confident of for 6 games into the future and the end of year predictions contain just too much variation – up to a point. If I view the predicted shot difference as a form guide going into the season and the monte carlo model above as an indication of the relative ease of a teams fixture (smoothed to an extent) then I am happy to use both as commentary aids during the year. I have managed to sort out some macros to run the end of year predictions overnight after the completion of each home and away round.
By taking a weighting of both ladders and re ranking
The three teams that move around a lot are Collingwood, Port Adelaide and Richmond. How each of these
fair against each other will have a large impact on the make up of the ladder and the respective fortunes of the three. The differences from each model also warrants a closer examination of each of these teams.
fair against each other will have a large impact on the make up of the ladder and the respective fortunes of the three. The differences from each model also warrants a closer examination of each of these teams.
So starting with Collingwood who I am predicting will finish in the bottom eight: on first examination the Pies have a reasonable fixture with a number of games against lowly sides from 2014 and a lot of time playing
on the big grounds. The first few rounds could be good – a win against St Kilda should be pencilled in; who knows what Essendon team will front up in round 4. A first up loss to Brisbane (at the Gabba) followed by a loss to Adelaide and a poor performance against the Saints would see the papers in a frenzy going into the two “blockbuster” games against the Dons an Blues. By the time Collingwood play the Blues again if they do not have 9 or 10 victories then finals is out of the question (Possible Wins by then: St Kilda, Melbourne x2, Brisbane, GWS and Bulldogs... puts a lot of emphasis on the games against Essendon, Richmond and West Coast.
on the big grounds. The first few rounds could be good – a win against St Kilda should be pencilled in; who knows what Essendon team will front up in round 4. A first up loss to Brisbane (at the Gabba) followed by a loss to Adelaide and a poor performance against the Saints would see the papers in a frenzy going into the two “blockbuster” games against the Dons an Blues. By the time Collingwood play the Blues again if they do not have 9 or 10 victories then finals is out of the question (Possible Wins by then: St Kilda, Melbourne x2, Brisbane, GWS and Bulldogs... puts a lot of emphasis on the games against Essendon, Richmond and West Coast.
What is of concern is the combined dramatic drop in scoring shots for and increased scoring shots against.
The defence has been erratic over the last few seasons and the attack has dramatically dipped in recent times. This combined trend is raising many red flags and compared to the other teams in the competition I struggle to see the Pies giving their supports much to cheer about in 2015; without well targeted recruitment or a change in coaching philosophy – neither of which I am not seeing.
The defence has been erratic over the last few seasons and the attack has dramatically dipped in recent times. This combined trend is raising many red flags and compared to the other teams in the competition I struggle to see the Pies giving their supports much to cheer about in 2015; without well targeted recruitment or a change in coaching philosophy – neither of which I am not seeing.
Going into round one I (and the bookies) are predicting Brisbane to defeat Collingwood at the moment.
If Collingwood have two or fewer wins by the end of round 6 a top eight finish will seem out of reach to me.
If Collingwood have two or fewer wins by the end of round 6 a top eight finish will seem out of reach to me.
Richmond
The Tigers graphical trends suggest to me that they are more middle of the road than their finals appearance suggests, however their fixture suggests that by the middle of the season they could realistically be well on
their way to another finals appearance in 2015:
their way to another finals appearance in 2015:
The early game against North should give a good indication of how both of these teams are travelling even though it is an “away game” for both teams perhaps to the detriment of the Roos? If the Tigers are on 8 or 9
wins by the middle of the year then a top four spot beckons; the lower the win total is less than 8 the lower on the ladder they will definitely finish. At the moment I am predicting that Richmond will not defeat a team that finishes above them at season end.
wins by the middle of the year then a top four spot beckons; the lower the win total is less than 8 the lower on the ladder they will definitely finish. At the moment I am predicting that Richmond will not defeat a team that finishes above them at season end.
Port Adelaide
The first eight rounds will define Port’s season – home games against last year’s grand finalists, an away game to the Roos and Brisbane and a potentially ladder defining game against the Tigers are all tantalizing. The turnstiles will be spinning for the AFL I predict. The tough fixture is counter balanced by the positive shot trend graph. I will be watching the pre season form of Port closely. At the moment I am predicting Fremantle to win the opening round.
Despite their good attack and defence I am predicting they will finish 9th – they will have a good percentage and could be in the finals with a superior percentage. Four games against the Hawks and Swans I think tip the scales.
In the next blog I shall examine the cellar dwellers and the top teams.
