This blog post aims to explain the results of MAGOS open test that took place earlier this year. We will try to stick to the very basics, without drifting away into more technical terms, and keep it brief. A more detailed analysis, as well as the future projections can be found in our bluepaper.
To understand the results more clearly, its recommended to be familiar with the basic terms of betting/wagering, such as “unit size”, “bankroll”, “ROI”, etc
When and where?
Testing started in Februry 2017, and lasted for 120 days. A sport betting network “FansUnite” was used as a testing platform. Roughly speaking, the conditions of the test can be described as :
- You are supplied with 250 “units”. This is your total bankroll
- You can bet from 1 to 5 units per event, not more (0.4–2% of the bankroll)
- Go ahead and try to make a profit
In a nutshell, that was it. However, to take full advantage of this testing period, we introduced some additional goals for our model :
- We tried to lower the bet amount as much as possible, avoiding the risks of high wagers
- We focused at betting on events that were perceived as “coinflips” (offered for even odds by the markets)
Before looking at the final results
When you look into the results of a model, interpretation is the key. It might sound counterintuitive, but the overall accuracy is not reall an important factor. Its much more important to look into the other factors : overall profit, ROI, and avg unit size.
“Winrate” looks very important on paper, but when it comes to forecasting models, especially applied to sports betting, it does not tell you much. Theoretically you can simply place wagers on the massive 90%+ favorites, achieving 85% winrate, and losing money in the process of doing it.
Profit and ROI, on the other hand, will tell you a real story. If a model is able to generate profit over a decent sample, it safe to assume that it has an edge.
Finally, the results:
- Profit of +153 means that a bankroll of 250 ETH would’ve resulted in a profit of 153 ETH after 4 months of betting
- ROI of 28% means that for 1 invested ETH, we are getting 0.28 ETH of profit on average, over a large sample
Were our additional goals achived?
- We managed keep the average odds of a bet near 2.00 (50%). This means that we were mainly forecasting events that were considered to be 50–50 by the markets, and reached a winrate of 66.7%.
- We kept AVG bet pretty low — at 1.72, thats 0.688% of a total 250 unit “bankroll”
How did we fair against the other handicappers?
Lets take a look at E-Sports leaderboard with following filters:
- Period : All Time
- Minimum picks : 100 — we are only interested in comparing large samples of results
- Priority stat : ROI %
Looking at the top 6 of Handicappers leaderboard — MAGOS completely crushed the competion: reaching 27.9% Roi, almost doubling what second best contenter was able to achieve, while maintaining the lowest AVG bet in the process (and therefore, lowering the amount of wagering risks).
In the next blog, we will brienfly jump back into the past, taking a look at some older performance gems from MAGOS collection.