Wisdom of crowds

From MMA Tycoon Help
Jump to: navigation, search

The Concept

The Wisdom of Crowds term was coined by James Surowiecki in 2004. Essentially, it states that many people giving totally independent predictions can be extremely accurate. The key point is that you must give your prediction independently of anyone else, or the system fails because you have be influenced by others. This is a concept that's always fascinated me, having heard about it many years ago.

What are you trying to do?

I've wondered for a few years now if I could do something with WOC to predict MMA results. Having come up with a system I'm happy with, I want people to predict the % chance of each fighter winning, then we'll compare that to bookmaker odds and see which fights we think are good betting opportunities. I then want to either sell that information to bookmakers (and pay the best predictors a cut of that income as a reward), or charge users a small fee to get the "best bets" ahead of the event.

Of course, it won't work for every fight but hopefully over time, we will get net positive results. There's only one way to find out! :)

How to enter predictions

There are two sliders; prediction and confidence.

  • Prediction: Fairly self explanatory. What percentage chance do you think either fighter has of winning? If the fight happened 100 times, how many would each fighter win? Maybe you wanna set this to 100% for one fighter but in reality, no fight should be 100%. Also, if you set it to 100% win for fighter A and fighter B wins, you're going to get rated down heavily in your predicting ability (read on for more info).
  • Confidence: How confident are you in your prediction? You could also view this value as "predictability". If you don't think the fight is very predictable or you don't feel you know that much about these fighters then potentially it might be better if you don't predict that fight at all OR you can just give a low confidence value.
  • Don't leave a fight as 50:50 and set a confidence - that will just get deleted as we can't really work with that as a prediction.

Not entering a prediction

Importantly, you don't have to predict every fight, just predict the ones you think you know something about. If you don't want to enter a prediction, just leave the confidence at 0% and that fight will not be logged. Once you enter a prediction though, you can't currently un-submit it.

Calculations

With your raw data, hopefully this is where I earn my money by coming up with some clever formulae. I will be weighting good predictors' data more highly, as well as (over time), building up a database of which fighters / types of fighter are more easily predictable.

Thank you!

Last of all I want to say thanks to everyone for entering data and taking part in this little experiment. I find it fascinating and hopefully you will too, as we move forward. Also, I fundamentally need to earn more from Tycoon to justify it as a full time job; hopefully this will provide both an extra income and also extra exposure (if it does work I expect it will gain significant interest and publicity for the site).


Stats

Event List Page

  • Method A: Based on a $10 bet on all our picks (raw or weighted). For whole events, the profit (or loss), is the total net return we'd get if we place a $10 bet on each of our picks.
  • Method B: Instead of placing a $10 bet on each of our picks, we look at which bets offer the best potential returns, based on the odds and our predictions. We then bet the same total amount ($10 x number of fights) but weighted so that more money is bet on fights with high potential returns. This is more high risk / high reward.
  • Method C: In some instances, betting against our picks will provide a higher average return than going with our pick. To give an extreme example, if the bookies think fighter A will win 90% of the time but we only think they'll win 55%, we're much better off betting on the 45% chance (opponent), with very high returns, compared to the actual pick and getting low returns, even if they do win. This method will likely result in more incorrect predictions overall but should see higher returns over a long period of time.
  • ROI: Return on investment. Income/Bets, displayed as a percentage.

Event Page - Viewing Predictions and Placing Bets

Viewing Predictions

We finish predictions and reveal the results 7 hours before the first fight, to give people plenty of time to place bets. Viewing predictions is currently free but will eventually cost 30 days VIP / 1 credit, per event.

Earning Credits

You can also earn credits by successfully predicting results, to view the results for free. I don't want to reveal exactly what criteria you have to meet to earn credits (as then people would adjust their predictions to earn credits not to make the system work as best as possible) but I try and make it as fair as possible.

How To Place Bets

We recommend you use the unit system for placing bets. We also display our old system ($10 average bets) just for building up statistics and a better understanding of how it's working.

What is Unit Betting?

We look at each fight and compare the odds given by the bookmaker compared to our predicted odds. The higher the difference, the higher the value for us and the larger bet we suggest placing. It's then up to you to convert "units" into a cash equivalent. 1 unit is usually said to be 1% of your bankroll. So if you have $100 in your betting bankroll, 1 unit = $1.

With this system we do not recommend betting on all fights. We only suggest bets for fights with sufficient value. We also suggest parlays when appropriate, with a fixed bet of 2 units per parlay.

Bookmaker Odds

The initial plan was to display the average of odds available by all bookmakers. There tends to be an average difference of around 1.7% between the average and the best possible odds offered by all the bookmakers. One might think this therefore produced a minimal difference in profits but that's not the case! After 15 events, using the best possible odds saw an increase in profits of $80 (roughly a 27% increase in profit)!

So why the massive increase? Well, first of all, you are getting a ~2% improvement in odds on all your income over time, not your profit. Whereas the profit was $280ish after 15 events, the income was $780. However, more importantly, a 2% increase in win chance according to the bookmakers does not equal a 2% increase in income. going from 50% win chance to 45% win chance is actually a 10% increase in profit. The less chance the fighter is given of winning, the more pronounced the difference. For example, going from a 20% chance of winning to a 15% chance of winning is a 26% increase in profit.

So with that said, I've begun to use best possible odds and will include a link to the relevant bookmaker once I have time to program that in!


Past Predictions Changed?

I never change any of the predictions made and never delete any data. My job is to try and make the most profit from the data going forward. With that in mind, I sometimes try out different formulae and see if they would have made us more profit on past events. If the new equation works better, I implement it and run the whole WOC from the beginning with the new equations, then using the new equations for future predictions. This will alter previous picks and results (winnings / losses etc), so you aren't going mad.

Redos

  • 5th Jan 2015: Edited formulae to downvote people who just give 100% (or close to 100%) picks for all their picks.

Results: Improved straight & weighted pick results. Slightly improved results on value bets. People who did this will have a worse WOC ranking, therefore my hope is that this will force people to start doing it properly, which will result in more accurate predictions in the future.

  • 23rd Jan 2015: I spotted a bug whereby we were giving people credit (to their WOC rating) for fights that ended up as draws / were cancelled. I removed this credit and unfortunately it saw a small dip in profits for the weighted predictions as we have two big win fights that were 49/51% picks that were affected by this.

See Also