Newhavensoft forum

**trollerboy** · Newbie

Hello

how do we know that odds wizard on average gives fair odds?

How can it be tested? Has anyone ever done back testing with historical
results and odds? Any experiences?

Sascha

**admin** · Administrator

Odds Wizard uses advanced statistical methods to compute performance-based odds. These odds have neither betting volume bias made up by bookies, nor they have profit margin. Their values in decimal notation are equal to the inverse probabilities (1/p). Odds Wizard's tool "Performance analysis" shows how the actual 1X2 distribution over the certain period corresponds to the distribution obtained from computed odds. In case of absolutely correct fair odds the above distributions should be identical. Therefore, computed odds may pretend to be fair odds if they produce as little gap between these distributions as possible.

**trollerboy** · Newbie

Hello, thanks for your answer.

Does that mean that in the "Performance analysis" the value for Success rate Total should be as close to 100 % as possible? Or should it be as high as possible? Is 120 a good value?

**ColdFish** · Newbie

100 means the odds reflect the outcomes exactly, which is actually unlikely. 102 that the model is good, 120 that the model is doing considerably better. In general though, scores that are much above 110, would almost certainly belong to a league with very few teams, of which about half dominate in playing ability.

-----
ColdFish

**admin** · Administrator

Actual and computed 1X2 distributions are shown in columns 3 and 5, respectively (both 'Relative rate, %'). The less difference between those values, the greater odds 'fairness' can be claimed.

The last column 'Success rate' indicates the extent of success in statistical forecasting of 1, X, and 2. Exactly, it is equal to the ratio:
(Sum of those computed probabilities 1,X,2 that are justified by the actual result of an event) / (Sum of computed probabilities 1,X,2) * 100%. Therefore the 'Total' value in the last column could be at most 300% if the program knew the future surely (100%+100%+100%). In the latter case, distribution of probabilities for each (1,X,2) event would be (1,0,0), (0,1,0), or (0,0,1). And the odds 'fairness' would be of course perfect, as columns 3 and 5 would be identical.

What we see in a reality, was well described by ColdFish in a previous post. Relatively high value of 'Total' success rate (over 110%) indicates a good predictability of a given league with a big difference in teams' strength.