2017 has been a rough year for tipping, even tougher than 2016, which itself was a wake-up call after the lovely, predictable Hawthorn dynasty.
Sadly, none of the models managed to outperform the bookies in tipping winners head-to-head during the Home & Away season, nor in terms of probabilistic Bits.
| Tips | Bits | MAE | Correct | |
|---|---|---|---|---|
| Punters
4 8 4 7 7 6 3 5 5 7 4 4 3 7 5 5 7 6 6 9 8 7 5 |
132 | 22.1970 | 66.7% | |
| PlusSixOne
5 8 4 7 6 6 3 5 6 7 3 3 2 8 5 7 7 5 6 5 8 7 6 |
129 | 17.9055 | 28.89 | 65.2% |
5 9 4 6 7 6 3 5 4 7 4 3 2 6 5 5 6 7 7 8 6 6 7 |
128 | 17.5804 | 64.6% | |
3 8 5 6 7 7 4 5 4 7 4 1 3 6 4 6 8 7 7 8 6 6 6 |
128 | 16.3202 | 28.87 | 64.6% |
5 7 5 6 6 6 4 5 7 7 3 2 3 7 3 5 7 5 7 7 7 7 7 |
128 | 15.4157 | 30.11 | 64.6% |
| Aggregate
5 8 4 6 7 6 3 5 5 7 3 3 2 6 4 5 7 6 7 8 7 6 5 |
125 | 20.1305 | 28.92 | 63.1% |
5 7 5 6 6 6 3 5 5 7 3 3 3 6 4 6 7 6 6 7 7 6 6 |
125 | 16.5200 | 29.03 | 63.1% |
4 7 4 6 7 6 3 5 5 7 4 3 2 6 3 5 6 6 7 8 8 6 5 |
123 | 18.6520 | 28.73 | 62.1% |
5 8 5 5 5 6 3 5 5 8 4 2 2 7 3 6 7 6 6 7 6 5 6 |
122 | 17.0455 | 29.79 | 61.6% |
A particularly rough year for Squiggle’s in-house tipping algorithm, which came in dead last and didn’t deserve to be any higher. But very credible performances elsewhere, especially from Plus Six One (for the second year in a row), and it’s worth noting The Arc‘s chart-topping Mean Average Error in a year when the line-ball games didn’t fall their way.
I was curious this year to see what kind of performance the Aggregate would have, where it represents a simple average of everyone else’s tips. Would there be some kind of wisdom of the crowds effect, where it could outperform most of the individual models that provided its inputs? Well, kind of: It landed mid-table in terms of tips and MAE, but with more Bits than all models. So there could be something there.