How confident are we?
TipBetPro is a player based statistical model used to predict the outcome of NRL, AFL and shortly cricket matches (first match is England v Sri Lanka in May).
The model is based on the principle of “value betting” – which in essence means it is an approach to gambling for long term profit. The concept of long term is an important one, as from match to match and week to week the model will not always get it right – but over the longer term, and especially a full season we are reasonably confident that the model will make a profit.
So how does it work?
Simply put the model (based on a range of factors) determines the expected result, expressed as a margin – for example in Round 6 we expected the Warriors to win by 8 points.
The margin then drives the confidence we have in the result – unsurprisingly the higher the margin the more confident we are. We were 58% confident the Warriors would win based on the margin of 8 and conversely 42% confident that Manly would pull off an upset.
Now if the odds offered the bookmakers multiplied by the confidence is greater than 1.10 we generally recommend a bet. However, we do NOT recommend bets if the probability is less than 35%.
So for Warriors v Sea Eagles:
Warriors odds 1.38 x 58% = 0.80, so way under the 1.10 threshold – no bet
Sea Eagles odds 3.05 x 42% = 1.28, so passes the value test of 1.10 and is above minimum probability of 35%, so is a recommended bet.
This basically means we think the odds offered by the bookmakers do NOT reflect the probability/confidence we have in the Sea Eagles winning and over the long term if we continued betting on such games we win!
The purpose of this week’s article is to explore how we set the confidence levels, as its key to determining value and in turn what we recommend to you lot.
Confidence levels
In order to determine confidence levels you need data – so based on the factors we believe influence a game of NRL we went back and calculated the expected result for the past 5 seasons of NRL games. We then compare these with the actual results.
Taking the 2015 NRL season as an example we have the following results:
So in 2015 where the margin was low we had around a 50% success rate – which is what you want as it shows your model is well correlated with reality! As the margin increases so does the amount we get right, with margins over 25 nearly all coming in.
But there are some anomalies. There were 4 games where we predicted a margin of either 22 or 24 points and they all lost! All teams were at home, and all were favourites (remember in NRL favourites lose a lot):
Can you remember these games? What was the reason? Interestingly Cronulla appear twice!
Now in the graph you will see Excel has kindly fitted a line of best fit to the data – it effectively says the confidence is 50% plus 1.06% for every extra point of margin. So for a game where we predict the margin will be 10, we would be 60.6% confident we’d get it right.
Now this isn’t quite what we use as we have more data than this but it’s very close. There is an argument that as the margin increases, say above 20 points we should be more confident than we are. But games like the 4 above are the main reason we aren’t that confident. If we remove these results you can see the line shifts up a fair bit.
So what is the answer? Well until we get more data we’re taking a cautious approach and it seems to working so far with the season to date return on head to head bets a healthy 50% return.
Interestingly these are the games this year we have “stuffed” up, where the margin was expected to be 12 or more points but we got it wrong. Unsurprisingly our great friends the Rabbitohs appear more than once!
As always please let us know what you think. What do you use? Help us help you!
GUEST WRITERS WANTED – Do you have a view on anything NRL related? Maybe a response to this article or something else. We are happy to post here* and make you famous (well sort of).
To see our betting recommendations, including our Bet of The Week, and to read more about our approach, and to compare our performance with your own, please visit www.tipbetpro.com or follow us (abuse us) on Twitter @tipbetpronrl.
*Subject to editors discretion
