The Efficiency of Pinnacle.com's Opening Prices compared to bet365
Posted 25th August 2017
In July 2016 I published an article on Pinnacle.com discussing the efficiency of Pinnacle's closing prices on the home-draw-away football match betting market, and specifically whether we could use them to calculate a value expectation. The hypothesis formulated was as follows: expected value is given by the ratio of the price you bet to Pinnacle's closing price with margin removed. Hence, if you placed a bet at odds of 2.5 and Pinnacle closed at 1.94 or 2.00 with its margin removed your expected value would be given by 2.5/2.00 or 25%.
Typically however, we don't know in which direction the market will move, hence we can't know whether the odds we bet will prove to be longer or shorter than Pinnacle's closing price. Consequently, it would be nice to know whether we could apply a similar hypothesis at the time the bet is placed, i.e. the ratio of the odds we place to Pinnacle's odds at the same time (with margin removed) will provide a useful guide to our expected value. This is just another way of asking whether Pinnacle's pre-closing odds are as efficient or accurate.
We already know from my original article published on Pinnacle that their opening odds are a pretty decent reflection of 'true' result probabilities. Using a metric known as Shannon Entropy which measures the uncertainly associated with a random variable there appeared to be almost no difference in market efficiency between Pinnacle's opening and closing prices. Similarly, comparing expected outcome probabilities (as implied by the betting odds) with their actual result frequencies showed almost no difference. So let's now take a look at an opening odds dataset and see if we can use it to find betting value in a football match betting market.
Using a dataset of opening Pinnacle prices for 147,629 football matches from 11th August 2007 to 15th February 2017, I used my logarithm function method to calculate the fair odds with Pinnacle's margin removed. [To learn more about how to apply that method, as well as similar methods for removing a bookmaker's margin, see my updated Wisdom of the Crowd betting document.] Alongside this dataset were opening prices from the bookmaker bet365, the world's most popular bookmaker and one less inclined than other UK, Irish and European brands to limit or close accounts of winning punters. From these data I could calculate a ratio of opening bet365 price to opening Pinnacle price with margin removed, to estimate expected value. Removing matches where odds from both bookmakers were not available, matches for where it was obvious the data were erroneous (for example underround books) and odds with calculated ratios greater than 1.20 and less than 0.80 left 396,419 betting odds ratios.
The first chart below shows the relationship between expected and actual returns. There is a strong 1:1 correlation between the two.
The table below compares actual returns to average expected returns whilst the second chart shows the profit time series betting all bet365 opening prices where expected return is greater than 102%. Average expected return for this sample of 28,748 bets is 106.3%, whilst actual return was 107.4%
|Average expected return||Actual return||Number of bets|
The astute amongst you might have spotted a problem with this analysis, with a clue being that the performance looks far too good to be true. The problem rests with the relative time of opening prices. Unlike closing prices, which by their nature are the prices published at kick-off, bookmakers open their markets at different times. Hence, we are not necessarily comparing like with like. If a later bookmaker opens hours or days later than an earlier bookmaker, the price of the earlier bookmaker may very well have changed by the time the later ones opens their market.
Sure enough, analysing a set of 50 matches and 150 betting prices, bet365 were typically found to open their market before Pinnacle, doing so for 48 of those games. For 35 of those matches, bet365's odds had changed from opening prices by the time Pinnacle opened their market. For these 150 betting prices, the ratio of opening bet365 price to opening Pinnacle price (with margin removed) ranged from 1.60 to 0.49 with a standard deviation of 0.13 and 29 of them over 1.00. By contrast, the ratio of the bet365 price at the time Pinnacle opened their market to Pinnacle's opening price ranged from a much narrower 1.23 to 0.75, standard deviation 0.07 and with just 14 over 1.00. Evidently, there is some betting away of extreme prices at bet365 before Pinnacle open their market. This will be a consequence of a number of factors, including bet3365 managing their liabilities and following other bookmakers, Pinnacle choosing to watch the market for longer than bet365, and Pinnacle's odds setters adopting a more efficient pricing model than that of bet365.
Without a much larger dataset of bet365 prices at the moment Pinnacle open their market we cannot be absolutely sure that betting a bet365 price that is higher than an opening Pinnacle price with its margin removed will be offering profitable expectation. It is, however, extremely probable that such value will exist. We already know from an earlier analysis of Football-Data's freely available betting odds that such betting value exists at the time I collect those odds (Friday afternoon for weekend matches and Tuesday morning for midweek matches). This value forms the basis of my Wisdom of the Crowd betting system and weekly picks. Although the majority of these odds are not opening prices for Pinnacle, they are typically those a long time prior to closing. The chart below shows the relationship of expected to actual return betting those bet365 prices and using the equivalent Pinnacle price as the predictor of value expectation.
We can also show that using bet365's opening match odds does not provide a reliable predictor of value expectation. This time using the ratio of Pinnacle's opening price to bet365 opening price (with margin removed) as a measure of the expected return, and calculating actual returns from betting Pinnacle's opening odds, there is no correlation at all between the two.
Finally, by removing the extreme odds ratios (i.e. those above 1.20 and below 0.80) from the analysis, we have already gone some way to removing those bet365 opening odds which contained the largest errors and as a consequence would have been mostly likely to change before Pinnacle had opened their market. Indeed, this was the primary rationale for removing them from the data analysis in the first place.
Conclusion: We have demonstrated the likelihood that even Pinnacle's opening prices, whilst not as efficient as their closing ones, are nonetheless efficient enough for the purposes of using them as a reasonable measure of assessing 'true' match result probabilities. Furthermore we have demonstrated that there is likely to be sufficient profitable expectation available even at one bookmaker like bet365 which is also known for being more tolerant of winning customers than the majority of UK, Irish and European brands. It is also clear that there is a lot of inefficiency in bet365's opening prices which if theoretically knowable would provide for a very significant profit expectation indeed. Of course, knowing how to find that inefficiency before Pinnacle's opening prices highlight it is the difficult bit, and by the time Pinnacle do open their markets, much of that inefficiency will have been bet away. By contrast, Pinnacle are the benchmark for opening price accuracy in comparison to other brands. This will be on account of delaying opening their markets until much later than others and an embracement of efficient market principles which allow them to tolerate, indeed actively encourage, winning customers to continue to play with their odds.