Help calculating expected earn

[Mr_J]

"'ve backtested it over the past 6 seasons and it has a win rate of 56.7% over that span. However, if I take the past 10 seasons then it only shows a win rate of 52.8%."

Don't assume you'll hit 56% going forward, you probally won't. What data did you use to form your hypothesis (system)? If you backtested over seperate data that's ok, but if you used the same data then your results are obviously biased.

You can't assume it'll hit 56% going forward.

You're probally fine betting 5% though if you heavily hit bonuses, line shop, low vig and free points. If you don't do these then definately lower the % you're betting.

I don't think you have any idea what you are getting into though. 5% units create huge variance, you have to be a real cool customer. If the money means anything to you then lower your %.

[DOTTT]

Well it basiaclly contians the following.

1. Public Opinion
2. Line Moves
3. Trend analysis


In all three I have anywhere from 1-10 stats (or data) inputs that I use. I'll be betting at Hollywood which offers -107 on all their bets, and I think I can handle the variance. Playing poker for the past 3 years helps a lot there.

I'm not sure what you mean though when you say

[ QUOTE ]
What data did you use to form your hypothesis (system)? If you backtested over seperate data that's ok, but if you used the same data then your results are obviously biased.


[/ QUOTE ]

One of the first things I set out to look for is a system that has been working for the past 3-5 years. I don't think historical data pre lets say 1990 is all that important, espcially not for my system since it uses a lot of variables that only recently have come about.

[Mr_J]

"Playing poker for the past 3 years helps a lot there."

The swings in sportsbetting are much, much bigger.

"I'll be betting at Hollywood which offers -107 on all their bets"

Very silly. In the longrun bonuses, low vig, free points, line shopping etc will probally be worth just as much as what your picks. You're leaving ALOT of money on the table by using just 1 book. The extra EV you make from the other stuff also acts as a buffer against overbetting.

"One of the first things I set out to look for is a system that has been working for the past 3-5 years."

Sounds like you used the same data, and if you did then your system isn't worth anything. When you test your system, you can't use the same data as you used to develop it. If you develop a system off the '03-'04 season, you can't include that data in your tests. So, what seasons did you use to develop your systems, and what seasons did you test over to see if the system worked?

[raccon]

Yes, $23,667 is the correct answer assuming that your every bet wins what's expected (impossible situation) and the expected profit is always added to the bankroll before making the next bet.

It's a bit difficult for me to explain this since english isn't my main language. But I'm sure me and whipsaw got the right answer.

This is how it goes:
1st bet
-------
Bet size: $10,000 * 5% = $500
Expected return: $500 * 1.909 * 0.56 = $534.52
Expected net profit: $34.52
(You EXPECT to win this bet 56 times out of 100, but with particular bet you can obviously lose the whole $500 or net $454.5 profit. What matters is that on average you net $34.52 with every bet)
Expected growth of the bankroll: $10,034.52 / $10,000.00 = 0,3452%

2nd bet
-------
Bet size: $10,034.52 * 5% = $501.73
Expected return: $501.73 * 1.909 * 0.56 = $536.37
Expected net profit: $34.64
Expected growth of the bankroll: $10,069.16 / $10,034.52 = 0.3452%

As you can see, on average your bankroll grows 0,3452% with every bet. Now it's easy to calculate how big the bankroll will be after 250 bets:
1.003452^250 = 2.366741
With your $10,000 bank, this will be $23,667.

Hope this clears things a bit.

[raccon]

With 46% win rate season (this happens even if you average 57% over long haul [thousands of bets]) your $10,000 bankroll is expected to drop to $2,200. If you can stand those kind of fluctuations, then it's ok.

[rvg72]

[ QUOTE ]

Sounds like you used the same data, and if you did then your system isn't worth anything. When you test your system, you can't use the same data as you used to develop it. If you develop a system off the '03-'04 season, you can't include that data in your tests. So, what seasons did you use to develop your systems, and what seasons did you test over to see if the system worked?

[/ QUOTE ]

That is a really good point... If you averaged 56.7% over 6 seasons and a 52.8% over 10 then assuming about the same number of bet games each year:

((52.8*10)-(56.7*6))/4 = 46.95% over the previous 4 seasons.

If you remove the season(s) that you built the system from the data then overall it is probably much closer to 50%...

This could be wrong and you may have found a great system but betting 5% per game would be a mistake IMO. Even a "real" 56% system stands about a 30% chance of dipping $10K over 250 games at $500/game.

rvg

[DOTTT]

I tested this over 6 seasons starting from 1999. The thing I love the most about is it's win rate has actually gotten better with every passing year. Even when I tested it for the past ten seasons it showed a some what similar trend. With only 1997 showing a decline from the year before.

[DOTTT]

.

[Mr_J]

But what seasons did you use to come up with the system?

With all due respect, I don't believe the 23k figure is correct. Since you're bet size is changing with bankroll, I belive this will change you're EV. However, this is grad school level stats. You can try simulating this in MATLAB or Excel to make it a bit easier.