Possible Problem with my Model?

[OMGRandyJackson]

So if you read my other thread you would know Ive got a model in the works. When back tested on the 09 and 10 season it showed profit with a 60-63% winning percentage on both years. By the end of the season the model had run the starting bankroll up about 85x for 09 and 90x for 10.

Since then we obviously believed the 85x and 90x were a bit unrealistic but on the other hand it is largely attributed to our betting strategy/rules.

Anyways we have discovered our first "bug". We found that our model was cheating and was using end of game ERA data instead of starting game ERA data. We have fixed this problem and the model now uses the pitchers era at the start of the game instead of from the end of the game.

This has altered our results significantly. When tested on 09 and 10, the bankroll now runs up about 10-12x the starting bankroll which honeslty does seem more realistic. But where Im worried is the win percentage has now dropped to about 51-52%.

Is a win percentage of 52% bad, if it is showing profit?

[jgilmartin]

So if you read my other thread you would know Ive got a model in the works. When back tested on the 09 and 10 season it showed profit with a 60-63% winning percentage on both years. By the end of the season the model had run the starting bankroll up about 85x for 09 and 90x for 10.

Since then we obviously believed the 85x and 90x were a bit unrealistic but on the other hand it is largely attributed to our betting strategy/rules.

Anyways we have discovered our first "bug". We found that our model was cheating and was using end of game ERA data instead of starting game ERA data. We have fixed this problem and the model now uses the pitchers era at the start of the game instead of from the end of the game.

This has altered our results significantly. When tested on 09 and 10, the bankroll now runs up about 10-12x the starting bankroll which honeslty does seem more realistic. But where Im worried is the win percentage has now dropped to about 51-52%.

Is a win percentage of 52% bad, if it is showing profit?

What are the average odds of your plays?

[OMGRandyJackson]

Of the top of my head I cant think of the average odd, but I can say if Im showing profit with only a 51-52% win ratio, then my average play would have to be over +100 odds?

[splash]

Are you betting moneylines? Cause a win% is pretty useless info without the bet odds.

[OMGRandyJackson]

Yes ml.

[jgilmartin]

Yes, as Splash said you are really going to need to know the odds. Is your backtesting based on betting the opener? You could use the line movement as a metric also.

[OMGRandyJackson]

We used closing lines when backtested.

[splash]

But the point is if the average line you bet was -200, then a win% of 52% sucks balls. However if the average line you bet was +200 then 52% is OMGAMAZING. So a win% by itself means nothing.

[Indecent]

I'd keep looking for more bugs.

[Data]

I'd keep looking for more bugs.

This reminds of PEBMAC which seems to be the most likely answer to the OP's question.

[lovesbaseball1]

possible problems with your model? using era is one of them.

[ferndog]

If you have a win % of 51-52 using ML & you made a profit then more than likely you are playing dogs. Of course playing the same units on each play.

[OMGRandyJackson]

But the point is if the average line you bet was -200, then a win% of 52% sucks balls. However if the average line you bet was +200 then 52% is OMGAMAZING. So a win% by itself means nothing.

Ya I got that now lol thx.

This reminds of PEBMAC which seems to be the most likely answer to the OP's question.

??

possible problems with your model? using era is one of them.

And why would you say that?

If you have a win % of 51-52 using ML & you made a profit then more than likely you are playing dogs. Of course playing the same units on each play.

Bet changes as bankroll increases/decreases.

[roasthawg]

Win percentage means very little unless you're playing only similar lines. 10x your starting bankroll for one sport over one season is extremely good... providing your betting strategy is sound.

[sharpcircle]

How do you model and know so little about statistics?

[curinator]

How do you model and know so little about statistics?

Lol, The main question from the OP leads us all to raise eyebrows...

[xyz]

Using end of the game ERAs, your model should do much better than 60-63%. Putting unearned runs aside, you can predict the winner using the end of game ERAs most of the time.

[jennahazeplays]

All systems will eventually fail, jusat my pessimistic theory lol!

[Marlin23]

So if you read my other thread you would know Ive got a model in the works. When back tested on the 09 and 10 season it showed profit with a 60-63% winning percentage on both years. By the end of the season the model had run the starting bankroll up about 85x for 09 and 90x for 10.

Since then we obviously believed the 85x and 90x were a bit unrealistic but on the other hand it is largely attributed to our betting strategy/rules.

Anyways we have discovered our first "bug". We found that our model was cheating and was using end of game ERA data instead of starting game ERA data. We have fixed this problem and the model now uses the pitchers era at the start of the game instead of from the end of the game.

This has altered our results significantly. When tested on 09 and 10, the bankroll now runs up about 10-12x the starting bankroll which honeslty does seem more realistic. But where Im worried is the win percentage has now dropped to about 51-52%.

Is a win percentage of 52% bad, if it is showing profit?

You think you can sort out the bug problem with the technique that you have shared over here. Have you tried it already. Is so, what are their effects to be seen.

[roasthawg]

Using end of the game ERAs, your model should do much better than 60-63%. Putting unearned runs aside, you can predict the winner using the end of game ERAs most of the time.

He was using end of game era's for the season to date... not for the one single contest.

[OMGRandyJackson]

How do you model and know so little about statistics?

Because I am not the doing the actual modeling durrrrrrrrrrrrrrrrrrrrrr

[Data]

??

Man, I am just relaying the answer, duhhhhhhhhhhhh.

[Marlin23]

You think you can sort out the bug problem with the technique that you have shared over here. Have you tried it already. Is so, what are their effects to be seen.

Apartments in Tucson