During my universities tug of war event, I had done a side betting project.
As I didn't know every one as well as I needed to, in order to estimate odds, I set up a LSMR in GSheet which computed the odds based on the bets placed with an upside of 5% for us. For the fun of it, we linked it to a virtual currency with price based on a bonding curve. The idea was to see if it is feasible and possible for casinos to make more money by changing the values of chips as they enter circulation.
This is rather attractive to the casinos as people are not gambling with things reading 100 just like the money.
Unfortunately, I was unable to test it to completion as I was handling the integration bee and my co-partner(and former roomate) was an _______ and _______ son of a _______ .
I think I should write a complete post on the research and the theoretical results I got.
Neat! Can you explain how you actually calculated odds and payouts? Was there a "house" or "bookie" that matched all bets? Or were payouts all coming from the other side?
I've seen a simple betting system where participants place bets on either side, and payouts are simply proportional to stake invested.
ex:
- Amy bets $10 on X
- Bob bets $20 on X
- Carl bets $30 on Y
- Daisy bets $30 on Y
If X wins, Amy gets $20 and Bob gets $40
If Y wins, Carl gets $15 and Daisy gets $15
This system works great for casual bets. It also does not require a "house" to match all bets. However, it has a cold start problem, where it is disadvantageous to the initial bettors, since the payout is initially 0, and the odds change over time so you can make better decisions if you jump in later. This works best when people want to bet on a specific side and don't care much about the odds.
The house was matching the bets but not entirely, to curtail the downside.
LSMR runs on the idea that the people know more about the game than we do and hence, have a better prediction of the odds. So after every bet, it changes the odds/payout offered.
If the system starts with no information(assumes 50-50) and assumes the size of average bet to be 15;
Amy is offered the simple 2:2 odds. If Amy bets on X, the system now offers 1.7:2.3 odds.
Say Bob bets on X. The system now offers 1.6:2.5 odds.
Say Carl bets Y, The system adjusts to 1:1 as Carl has shown as much confidence in team Y as much as Amy and Bob combined.
If Daisy bets on Y, the system offers 2.2:1.8 here onwards.
And so on.
Here the house loses 45 if Y wins and gets 35 if X wins.
The idea of the algorithm is to set odds such that given all betters are completely rational, risk averse, have no team preference and infinitely many(which some may argue is a bit too many) to make a set amount of profit for the house.
As the odds change, the agents are expected to bet on the slightly less likely to win team.
I was using Othman, Abraham, David M. Pennock, Daniel M. Reeves, and Tuomas Sandholm. “A Practical Liquidity-Sensitive Automated Market Maker.”
During my universities tug of war event, I had done a side betting project.
As I didn't know every one as well as I needed to, in order to estimate odds, I set up a LSMR in GSheet which computed the odds based on the bets placed with an upside of 5% for us. For the fun of it, we linked it to a virtual currency with price based on a bonding curve. The idea was to see if it is feasible and possible for casinos to make more money by changing the values of chips as they enter circulation.
This is rather attractive to the casinos as people are not gambling with things reading 100 just like the money.
Unfortunately, I was unable to test it to completion as I was handling the integration bee and my co-partner(and former roomate) was an _______ and _______ son of a _______ .
I think I should write a complete post on the research and the theoretical results I got.
Neat! Can you explain how you actually calculated odds and payouts? Was there a "house" or "bookie" that matched all bets? Or were payouts all coming from the other side?
I've seen a simple betting system where participants place bets on either side, and payouts are simply proportional to stake invested.
ex:
- Amy bets $10 on X
- Bob bets $20 on X
- Carl bets $30 on Y
- Daisy bets $30 on Y
If X wins, Amy gets $20 and Bob gets $40
If Y wins, Carl gets $15 and Daisy gets $15
This system works great for casual bets. It also does not require a "house" to match all bets. However, it has a cold start problem, where it is disadvantageous to the initial bettors, since the payout is initially 0, and the odds change over time so you can make better decisions if you jump in later. This works best when people want to bet on a specific side and don't care much about the odds.
The house was matching the bets but not entirely, to curtail the downside.
LSMR runs on the idea that the people know more about the game than we do and hence, have a better prediction of the odds. So after every bet, it changes the odds/payout offered.
If the system starts with no information(assumes 50-50) and assumes the size of average bet to be 15;
Amy is offered the simple 2:2 odds. If Amy bets on X, the system now offers 1.7:2.3 odds.
Say Bob bets on X. The system now offers 1.6:2.5 odds.
Say Carl bets Y, The system adjusts to 1:1 as Carl has shown as much confidence in team Y as much as Amy and Bob combined.
If Daisy bets on Y, the system offers 2.2:1.8 here onwards.
And so on.
Here the house loses 45 if Y wins and gets 35 if X wins.
The idea of the algorithm is to set odds such that given all betters are completely rational, risk averse, have no team preference and infinitely many(which some may argue is a bit too many) to make a set amount of profit for the house.
As the odds change, the agents are expected to bet on the slightly less likely to win team.
I was using Othman, Abraham, David M. Pennock, Daniel M. Reeves, and Tuomas Sandholm. “A Practical Liquidity-Sensitive Automated Market Maker.”
which is surprisingly readable.
Cool, I'll have to take a look at this paper. Everything I know about automated market makers comes from crypto in the last 4 years