Introducing cricodds: Match Odds and Analysis

[viriya]

Introducing cricodds: the first step in my plan for a comprehensive hub for live match odds and detailed analysis.

The tool gives you match odds based on historical scenarios for limited over matches, with detail on where the numbers come from, and input options and graphs for better customization.

To illustrate with an example, say you're following an ODI match where India is chasing, 4 down and need 80 runs in 10 overs. Given those inputs, cricodds gives preliminary odds of 50% based on similar historical scenarios (given in the table).


The column chart shows wins and losses by required run rate for similar match scenarios historically, and the bubble chart plots all historical scenarios over time by required run rate. Browsing the bubble chart tells you that chases such as these tended to end as losses more often in earlier ODIs (I've circled the relevant red dots).

To get better odds, you can customize the date range of historical matches to consider. Once you do that by setting the start date to 2005-08-10 (based on insight gleamed from the bubble chart - hover on the dots to get dates), the odds update to 54.44%:


The next step to tweak the odds further is to input the teams involved. India is chasing vs Bangladesh, and since India's current team rating is significantly higher than Bangladesh, setting the teams as inputs increases the win odds even further to 62.69%:


The team ratings adjustment is shown below the adjusted win odds as 8.25% so you can compare with the raw odds. Check this example here. The site now shows 63.19% for the win odds since it just picked up a few more recent similar ODIs that I just updated.

Currently it works for ODIs and T20s, with T20s using T20I and FT20 history for more data. The odds are calculated by using a simple ratio of the number of similar match scenarios that ended in wins and the total number of similar match scenarios.

Let me know what you guys think. These odds are intentionally somewhat simplistic, because a black box solution (say using machine learning methods) might give better odds, but will add a lot more complexity to the numbers.

Thoughts, ideas and criticisms welcome!

 

[Athlai]

Why don't we keep it on the downlow, check to see how it compares with real odds over the course of a year. And if it looks legit, throw money at it.

 

[viriya]

Why don't we keep it on the downlow, check to see how it compares with real odds over the course of a year. And if it looks legit, throw money at it.

I've been using a souped up version of this to bet for 1+ years now. That factor model works pretty well, so I thought of opening it up for those who want to make their own odds.

 

[Zinzan]

Why don't we keep it on the downlow, check to see how it compares with real odds over the course of a year. And if it looks legit, throw money at it.

You don't think bookies worth their salt use their own historical analysis programs along with several other analysis tools already? Not knocking what viriya is doing, it actually looks cool, but I'd be surprised if it's cutting edge.

 

[Athlai]

You don't think bookies worth their salt use their own historical analysis programs along with several other analysis tools already? Not knocking what viriya is doing, it actually looks cool, but I'd be surprised if it's cutting edge.

Well yeah but any kind of advantage you can gain on when odds are very wrong can be taken advantage of easier.

 

[viriya]

You don't think bookies worth their salt use their own historical analysis programs along with several other analysis tools already? Not knocking what viriya is doing, it actually looks cool, but I'd be surprised if it's cutting edge.

This by itself is definitely not cutting edge. Other factors like home/away, momentum, batting/bowling ratings, ground adjustments, rule changes are all valid for a proper model - this is more of a first foray that everyone can play with.

 

[viriya]

Reddit post: https://www.reddit.com/r/Cricket/comments/54na74/introducing_cricodds_match_odds_analysis/

 

[Burgey]

Introducing cricodds: legitimizing illegal betting and match fixing in India since 2016.

 

[viriya]

Introducing cricodds: legitimizing illegal betting and match fixing in India since 2016.

hey now - it's legal in UK.

 

[the big bambino]

England will win a world cup one day using this data.

 

[weldone]

Fantastic...The batting team, bowling team filters aren't useful for me...But knowing the exact averages for different scenarios is just superb...saves me so much data-mining effort (that I was doing currently)

Love you so much viriya

 

[viriya]

Fantastic...The batting team, bowling team filters aren't useful for me...But knowing the exact averages for different scenarios is just superb...saves me so much data-mining effort (that I was doing currently)

Love you so much viriya

Good to hear. Let me know what you think can be improved. Why don't you find the batting/bowling team filters useful?

 

[weldone]

Oi viriya but there is multiple counting of the same match when I calculate a scenario.

 

[weldone]

Good to hear. Let me know what you think can be improved. Why don't you find the batting/bowling team filters useful?

Because more than the teams, the chances are affected by the exact batsmen who are playing, the exact batsmen who are to come next (i.e. batting order) and which bowler has how many overs left. In fact, you may find a way to link this with your player current ratings. But that's way too much work I guess.

 

[viriya]

Oi viriya but there is multiple counting of the same match when I calculate a scenario.

Yes, this is by design. This is partly to avoid insufficient data issues. I can see the argument for restricting it to unique matches only - probably a good option to add.

 

[viriya]

Because more than the teams, the chances are affected by the exact batsmen who are playing, the exact batsmen who are to come next (i.e. batting order) and which bowler has how many overs left. In fact, you may find a way to link this with your player current ratings. But that's way too much work I guess.

Yes, this would add much more complexity. I've toyed with using cumulative batting and bowling ratings of the teams with good results - but that would only work if I source live scorecards and pick up the players dynamically (manually adding up the ratings as an input would be painful).

 

[weldone]

Yes, this is by design. This is partly to avoid insufficient data issues. I can see the argument for restricting it to unique matches only - probably a good option to add.

yea should be unique matches only for one query (can't do anything about sample size issue when the issue is genuine)

..should take the closest scenario from each match (i.e. where the number of overs matches exactly)

 

[weldone]

'No result' should be omitted from the %age calculation

And tie should be counted as 0.5 win.

Right now, win %age is (=win/total matches), which slightly understates the number.

 

[viriya]

'No result' should be omitted from the %age calculation

And tie should be counted as 0.5 win.

Right now, win %age is (=win/total matches), which slightly understates the number.

This one is debatable (Tie and NRs are classified the same btw). For betting purposes having a 0.5 win gets you nowhere since if you bet for a win you still lose 100% of your bet on a NR/Tie so I thought just looking at raw wins/matches made more sense.

 

[viriya]

yea should be unique matches only for one query (can't do anything about sample size issue when the issue is genuine)

..should take the closest scenario from each match (i.e. where the number of overs matches exactly)

This sounds like a good idea, probably will improve odds with big enough similar scenarios already and worsen odds for those with limited data. But the odds calculated when there is limited data is questionable anyway so probably a worthwhile trade.