[Waterstpub87]

I've built many losing models over the years. Everyone has. Save yourself some money, and avoid obvious losing situations.

1. You don't consider things that move lines- If you are running a system where you bet on a team without looking at injuries, adjusting lineups, or considering their starting pitcher in baseball, you will lose. This mostly applies to chase systems, or systems based on "trends" like performance after a major loss, or on the third Tuesday in June. With chase systems, you can never make +EV bets by combining -EV bets.

2. You only scratch the surface with analysis - Several years ago, I managed to have a +15 unit season in NFL and a +25 unit season in NBA. I took the average score a team had, divided it by the league average, and multiplied the result by the other team's points allowed to get estimated score. This is the worst experience that someone could have. I gave it all back and more next year, not knowing why my "system" failed. Its obvious to see why it failed, it was no good to begin with. If you system is based on something simple, with simple inputs and formulas, you are going to lose. It is not that easy to beat sports betting.

3. If, and when you backtest, you have skewed results- Another system that I built a few years ago, lost small amount most months, but had a single 40 unit month. Of course, I focused on that 40 unit month, and ignored the others. Look for you system to win relatively consistent, and do not become biased by occasional outperformance.

4. You bet the same team/fade the same team every night- Sometimes, you will weight certain statistics too highly, and this will cause you to overemphasize certain teams. If you find yourself day after day betting the Minnesota Grizzles, or betting against the Miami Marlins, something in your system is wrong. You can use this to find place where your model is incorrect. Or, where parts of your spreadsheets are accidentally values and not formulas.

5. If you think that every line is off crazily, your wrong, your going to lose- If you run your daily analysis, and compare it to the point spread, and find that every line is off 3 pts, and some are more then 7, you don't stand a chance. Understand that if a line was off 7 points on an NBA game, you would be betting an approximate -350 favorite at even money. If bookmakers routinely offered those odds, they wouldn't exist. Even 5 points is approximately -200, which would be like a 16% edge. The lines are not off this much, if you think they are, you are wrong and your system is wrong.

6. If your model looks good for one type of bet, but doesn't work for another, you are probably going to lose- If your model only works for totals after 300 bets, but not for spreads/moneylines, you likely don't have a good model.

7. When you backtest, you don't beat the closing line- You should test your system against openers. This will give you a good idea if you are making good bets. It would be difficult to win consistently once the lines have been hammered into place. Test your system against openers. If you are consistently beating the vig free line, you likely have a winning model. If you are not, you model is likely bad, unless you know something that the world does not.

That is all I can think of for now. Save yourself some money and avoid these mistakes.

[Mayberry]

If you built several losing systems and theyre not chase systems, why not fade the worst one? System doesnt need to give you winners to be a system.

[Waterstpub87]

If you built several losing systems and theyre not chase systems, why not fade the worst one? System doesnt need to give you winners to be a system.

This is a question that is asked often.

Losing system does not mean picks less than 47%, meaning you can fade and win. Losing means picks 50%, meaning you lose long term because of the juice. Over the long run, everyone will hit roughly 50% and lose due to the juice. Your losing system hits 50% and you lose.

[evo34]

Some good stuff in there, but I couldn't disagree with #6 more. It's very possible to have a model that predicts overall scoring rates without doing a great job at predicting relative team strength. This is particularly likely if your model is using elements that affect both teams (officiating, weather, park dimensions, etc.). A model does not have to be all-encompassing or perfect to be useful.

My thoughts on the general topic of system building:

1. Don't be satisfied with small samples, no matter how much you want to be able to predict rare events. Some events (NFL playoffs) are too rare to observe many valid trends. So don't force the issue.

2. In adjusting system parameters, don't be afraid to trim down outliers that are unprofitable. If your system is using closing lines, and you see that it fails whenever the open to close line move is > 1.5 either direction, remove those games. It makes logical sense that a situational system is is going to break down in games with unusual pre-game events. Apply this type of logic to all your parameters -- does reducing the window for this param actually make sense, or I am over-fitting?

3. Be concerned with margin of victory for your bets. Running up the score doesn't allow you to win more money on individual bets, but it tells you a lot more about the quality of your system than a bunch of 1-point covers do.

4. Don't become obsessed with matching the market (I.e., I disagree with #5 in the OP). If your system shows a 10% theoretical edge, but you only realize 2% of it (over a large sample of bets), that's great. You know that your edge projection is overly optimistic, but also that it reflects true value. That's all you really need is to be good enough to know in what direction the line is off -- and that the magnitude is enough to comfortably cover the vig. Knowing your precise edge down to the tenth of a percent would be awesome, but it is virtually impossible to know and any time you spend thinking you do know it with that level of precision is going to be more dangerous than helpful.

5. The market is changing monthly. Don't think that a high-volume system that had a horrible year last year (after crushing it for nine years) is likely to bounce back. Odds are people are looking at the same thing you are and the inefficiency is gone for good. How far is appropriate to look back? It depends on the complexity of the system and how much proprietary info. involved. All I know is that less than three years of data is meaningless to me, when creating a system. But also anything more than 5-6 years old is probably worthless. There are no magic numbers, but being acutely aware of the annual ROI of your systems and where it is trending is critical for success.

[Waterstpub87]

Some good stuff in there, but I couldn't disagree with #6 more. It's very possible to have a model that predicts overall scoring rates without doing a great job at predicting relative team strength. This is particularly likely if your model is using elements that affect both teams (officiating, weather, park dimensions, etc.). A model does not have to be all-encompassing or perfect to be useful.

My thoughts on the general topic of system building:

1. Don't be satisfied with small samples, no matter how much you want to be able to predict rare events. Some events (NFL playoffs) are too rare to observe many valid trends. So don't force the issue.

2. In adjusting system parameters, don't be afraid to trim down outliers that are unprofitable. If your system is using closing lines, and you see that it fails whenever the open to close line move is > 1.5 either direction, remove those games. It makes logical sense that a situational system is is going to break down in games with unusual pre-game events. Apply this type of logic to all your parameters -- does reducing the window for this param actually make sense, or I am over-fitting?

3. Be concerned with margin of victory for your bets. Running up the score doesn't allow you to win more money on individual bets, but it tells you a lot more about the quality of your system than a bunch of 1-point covers do.

4. Don't become obsessed with matching the market (I.e., I disagree with #5 in the OP). If your system shows a 10% theoretical edge, but you only realize 2% of it (over a large sample of bets), that's great. You know that your edge projection is overly optimistic, but also that it reflects true value. That's all you really need is to be good enough to know in what direction the line is off -- and that the magnitude is enough to comfortably cover the vig. Knowing your precise edge down to the tenth of a percent would be awesome, but it is virtually impossible to know and any time you spend thinking you do know it with that level of precision is going to be more dangerous than helpful.

5. The market is changing monthly. Don't think that a high-volume system that had a horrible year last year (after crushing it for nine years) is likely to bounce back. Odds are people are looking at the same thing you are and the inefficiency is gone for good. How far is appropriate to look back? It depends on the complexity of the system and how much proprietary info. involved. All I know is that less than three years of data is meaningless to me, when creating a system. But also anything more than 5-6 years old is probably worthless. There are no magic numbers, but being acutely aware of the annual ROI of your systems and where it is trending is critical for success.

I think we are on the same page evo. You can have an uncertain edge, even though it is large, it can certainly be small no problem. My issue is more with having every game with a 10% edge in a given day, which is ridiculous common sense wise. Not being able to get your edge estimate down to 4 decimal places no problem.

[Miz]

good post. nice job. Beating the closing line is a huge indicator.

Having built many losing models and several winning ones also, I'd say that not being a pussy about your bet size when you know you've done your homework and backtesting is something to work on also. When you have a winning model, it won't remain a winning model forever, so seize the day.

[Gene Paulson]

One thing you'll forgot is the test sample, which is 2, 500 min 10, 000 best. Not many sytems ever pass those cir. 1/500!

[agendaman]

500 factorial wow

[Meansteam]

If you have a user name at sbr forums your system doesn't work.

[BrickJames]

You know when you get broke

[Rich Boy]

Why try and beat the system in the first place, much easier to use professional opinions and lean on square books. Virtually no way you can lose and requires no actual "handicapping"

[Waterstpub87]

Why try and beat the system in the first place, much easier to use professional opinions and lean on square books. Virtually no way you can lose and requires no actual "handicapping"

You would need to assess if "professional opinions" were actual winning plays. This would you require you to assess at least 1000 historical plays. You have no way to get this easily and if they provided it, you have no way to know if this is accurate.

Leaning on square books is possible, but beating them on off market numbers consistently will lead to getting limited and then you will no longer be able to profit. Building a good system would allow you to continue profit and bet large amounts in more liquid markets.

[Hu$tle]

I dont think you have a long lifespan beating online bookie line shooting.

I could see it for Large cash amounts in vegas max a few weeks nowadays...unless u know something i dont ???

[Rich Boy]

You would need to assess if "professional opinions" were actual winning plays. This would you require you to assess at least 1000 historical plays. You have no way to get this easily and if they provided it, you have no way to know if this is accurate.

Leaning on square books is possible, but beating them on off market numbers consistently will lead to getting limited and then you will no longer be able to profit. Building a good system would allow you to continue profit and bet large amounts in more liquid markets.

I agree however that is really the only way to guarantee a profit long term. Any system that works now is unlikely to work in the future as markets sharpen and your edge is slowly grinded away.

By professional opinion I mean sharp books (Pinny, 5dimes, etc) with high volume and the most efficient lines.

To work around the limiting issue is not that difficult either.

The hard part with systems for me personally is when do you know that it no longer produces edges? When you start to lose or when the markets disagrees with your plays and on average you no longer beat the closer?

In smaller markets sometimes I get stubborn and dont really care what the closing line is or weather the market agrees with me or not, other than the fact that I could have gotten a better line. I specialize in NHL totals and I dont regard them to be very efficient with relatively low volume. At what point do I say my system works or doesnt work and its time to readjust?

[Waterstpub87]

I agree however that is really the only way to guarantee a profit long term. Any system that works now is unlikely to work in the future as markets sharpen and your edge is slowly grinded away.

By professional opinion I mean sharp books (Pinny, 5dimes, etc) with high volume and the most efficient lines.

To work around the limiting issue is not that difficult either.

The hard part with systems for me personally is when do you know that it no longer produces edges? When you start to lose or when the markets disagrees with your plays and on average you no longer beat the closer?

In smaller markets sometimes I get stubborn and dont really care what the closing line is or weather the market agrees with me or not, other than the fact that I could have gotten a better line. I specialize in NHL totals and I dont regard them to be very efficient with relatively low volume. At what point do I say my system works or doesnt work and its time to readjust?

I disagree with your first part. If you model is fundamentally good at valuing games, I don't believe it is unlikely to continue to work in the future. There is a limit on how sharp markets can be.

Your system will no longer produces edges when:
1. Over the course of several hundred plays, you do not beat the closing line
2. In game results/small markets, over the course of several hundred plays you no longer have a statistical edge

I would argue that if you have a fundamentally sound model, it would stop working when you are no longer able to find games to bet on as the market has sharpened to the point that you don't have an advantage. If it suddenly starts to lose consistently, perhaps it was never actually good to begin with.