Freedom Odds Blog

Despite missing one week, i ended up in 62nd place out of all the thousands of people in the competition.  The success rate was 56.5%.  In this competition you choose five games against the spread each week of regular season NFL.  I struggled with the early weeks, not scoring that well really until about midseason things picked up.  I had a 60% result at one point but a few losses and a missed week (which counts as missing all five picks) brought that down.  Finally there was a strong finish the last four weeks from Jacqueline.  So a rather exciting season overall and I finished in the top 100 to boot!  Can’t wait until next year…

I’ve been imagineering a new program for a while now, and I finally coded it up and named it Jacqueline. The software is still in the development stage, so I haven’t posted it to the site yet, but I will be posting predictions. I also wrote a little utility program that goes along with Jacqueline, called Paper_Test, which analyzes the predictions made by Jacqueline and provides a summary report.

This is very helpful because Jacqueline produces what I call a confidence factor for each game, and some confidence factor values are good while others are not. By having Paper_Test collect statistics on Jacqueline’s predictions, I can easily see which confidence values to trust and which to reject.

Jacqueline has gotten off to a good start with a phenomenal (and lucky) 85% performance right off the starting blocks, and the results from Paper_Test show that Jacqueline is capable of performing at or above 70% success rate in NFL and NBA. Blink! Blink! Yes, those numbers are correct, based on two and a half weeks of NFL and NBA combined. I find them a bit difficult to believe myself as well, and I’m waiting for the winning streak to end, but it hasn’t yet. Until then we’ll just have to chalk it up to good source data and good programming with a little bit of luck mixed in (OK, a LOT of luck)!

Unfortunately Jacqueline is limited to only specific sports because of the availability of source data and the requirement of a name translation database, so for the moment Jacqueline is limited to NFL and NBA. I plan to put Jacqueline up on the site as soon as practical, so stay tuned. Until then I’ll be posting the predictions - good luck!

Haha, I have good news!  In the VegasInsiders NFL competition, I have a 60% success rate which is good enough to place me in 104th position out of all of the thousands of people who play there!  Not bad for a couple of computer programs.  The contest involves picking 5 games against the spread and then choosing a score total for Monday night football as the tiebreaker.  I play with the download, Predict Games and then if that doesn’t select 5 games I fill in with Predict.  So good news, I’m winning this year!

It has been a bleak year for my programs in NFL and NCAAFB, though this past week we posted an 11-9 winning result. This brings the overall average up to 41-41, now breaking even but not against the vig. Look for better results as the season goes foreward and the programs gain even better ratings estimates of the teams.

I just found and excellent source of information for use in consensus systems: The CBS Sports Expert Picks, located here:

http://www.sportsline.com/nfl/features/writers/expert/picks

If you take a look, you’ll see that some of the experts have very good success rates going back five years. I’ve included their picks in my consensus system for NFL and I’d suggest you do the same, it’s good data and worth the effort of recording it.

I have discovered a new source of computer predictions, the Wunderdog site. Here is the URL:

http://www.freeunderdog.com/CurrentMatchUps-NFL.php

There you will find computer predictions for all of the NFL games of the current and past seasons. The predictions include percent chance of winning and a predicted score, and therefore the predicted point spread. The only problem is they do not publish a record so we do not know how good the predictions are. It’s worth a look if you are interested in such things.

Someone in the Usenet newsgroup rec.gambling.sports noticed that most of my picks for the week were in favor of the underdog and commented on it in a post. This led to some thought that could be helpful in handicapping sports. It is all based on the fact that the betting public generally prefers to wager on the favorite team, not the underdog.

Let’s examine what the sports books might be doing in response to this fact. It is well known that the ideal goal of the sportsbook is to set the point spread such that half of the bettors bet on the favorite and the other half bet on the underdog. That way the sports book derives profit from the vigorish and the bettors just exhange money from one half to the other half.

Now let’s incorporate the observation that the betting public prefers the favorite. In order to equalize the wagers, the sports book must take this into account and adjust the point spread accordingly. That is, the sports book takes the genuinely predicted point spread and changes it from its optimal value to compensate for the public bias. Knowing that the sports books are doing this, the smart bettor can incorporate this bias into his/her wagering decision to achieve better results.

What it boils down to is that you should prefer to choose the underdog in your predictive analysis. To see if the program was doing this, I extracted the point spreads from my Betting_History.txt file which stores all my pretend wagering results. It turned out that 71% of the picks were for the underdog. So i think that what is happening is that the sports book point spread is skewed to one side, and the program sees value in this so it tends to prefer the underdog.

So we have the fortunate situation that the sports book point spread is modified to compensate for the mistaken behavior of the public, and we can take advantage of that to improve our prediction success rates. One strategy would be to select the underdog in borderline cases. Another strategy would be to always add one extra point to the underdog in your predicted point spread before comparing with the sports book point spread. Yet another strategy might be to simply always bet on the underdog - it might be profitable in the long run.

In conjunctoin with all this is the fact that the sports books adjust the point spread over time in an effort to accomplish this equalization. That suggests that the later odds, not the earlier odds would have the greatest underdog bias. So another strategy would be to wait until the odds settle down before placing your wager.

So, to summarize: the public prefers to wager on the favorite, the sports books adjust the point spread to compensate for this, and the net result is that the smart bettor will favor the underdog in making predictions. There you go, another little pearl of wisdom brought to you by Freedom Odds!

Just in case you haven’t read the article or the post in the forum about this, I thought I would outline my Consensus system for you here in my blog. It’s all based on this computer simulation that I wrote one energetic morning recently. The simulation created a group of pretend handicappers with a given percentage success ratio, and it then had them “vote” on the outcome of each game in a pretend football season. The votes were tallied up and success rates were recorded for the season, showing that if a group of X percentage handicappers “vote” on each game outcome, they will produce Y success rate.  The amazing result is that Y is greater than X, or the whole is greater than the sum of the parts.

To keep things honest and minimize statistical variance I made the program average the predictions over 100 seasons of NFL.  If the cappers are better than 50% (X > 0.5), then the result is greater than the capper prediction average (Y>X). This means that in handicapping sports games, two heads are better than one. Also three heads are better than two, and so on. The more good handicappers you combine with Consensus voting, the better your final success rate will be.

And it works the other way in reverse. If your cappers are all below 50%, then they will collectively produce a worse result than their average success rate. So the method is not foolproof, you have to make sure that all your handicappers are above 50% success rate for this to work.  Because of this observation, I have decided to change my pretend wagering strategy.  Instead of just using the results of my first program, I am using the combined results of the first program, the second program, the Sagarin predictions, and fading the public.

Time will tell whether this Consensus system works or not, but for now I am using it and at this time the early data is inconclusive, meaning a near 50% result.  We will see what happens, wish me luck!

My friend Stephen from Canada has come up with a very simple yet interesting system for wagering. His theory is based on the observation that sports books are very good at predicting the odds and point spreads of sporting events. They do amazing calculations including injuries, offensive/defensive match-ups, quarterback behaviors, weather, travel time lag, and who has the best criminal record (just kidding on that last one). In other words the sports books are the masters and we the wagering public are not so skilled.

Stephen’s idea is simple: why not use their own skills against them? Go to a site like www.vegasinsider.com and calculate the mathematical average of all the point spreads. We will consider this to be the true point spread. Now look at the various sports books and decide which one is offering a losing proposition based on the supposed fact that the averaged point spread is correct. Place your bet with that sports book and win!

My suggestion to Stephen was that if he were to do this, in paper testing or in actual wagers, then he might want to use the early odds rather than the later odds. This is because of the game that we all know the sports books play to optimize their profits. The ideal on any game for a sports book is to have half of the public betting on one team, half on the other team, and then they clean up on the vig. So as the wagers come in over time, the sports books adjust their odds to even things out. This is not conducive to accurate predictions, so we are in the bizarre situation that the sports books start out with accurate predictions and then they purposely make their predictions (the “line”) worse over time in response to the public consensus.

So in simpler terms, if you are going to do this then use the early odds, not the late odds.  That is, assuming the sports books are not doing some head-fake like anticipating average bettor’s behavior and pre-adjusting their early odds accordingly.  But let’s not get into second guessing and semantics like that.  Suffice to say that this simple system might just have some merit to it, so if you are looking for something new to track on paper or on your favorite pretend wagering site, you might want to give it a try.

Ah, the vig - or vigorish - without it how would the sports book make a profit? We all know that the sports books try to even out their wagers, with half of bettors on one side of the bet and half on the other. Then they clean up by charging the vig. But what does this mean to the average gambler? What success rate do we need to have to equalize that 10% vig? Some might say 60% since 50% is break-even and the sports book charges 10%, we have 50% + 10% = 60%, right? Wrong!

Actually it’s 52.38% and I’ll tell you why. It’s all based on an equation that I figured out, and I’m sure its in gambling textbooks everywhere. It goes like this:

P = N * B * (S * (2+V) / (1+V) - 1)

where

P is the total profit
N is the number of bets made
B is the bet ammount
S is the success fraction
V is the vigorish fraction

Now skim over this paragraph because we are going to do some math, and just about everyone hates math. Let’s set profit equal to zero, establishing the break-even point between the sportsbook and the gambler. Mathematically then, N and B divide into zero and disappear, leaving us with

0 = S * (2+V) / (1+V) - 1

We then pull the -1 over to the left hand side and divide by S, then take the inverse and what do we get? We get

S = (1 + V) / (2 + V)

That’s the formula for figuring out the success rate that is required to break even with a particular vig. Let’s say the vig is 10%, or 0.1, then we have S = 1.1 / 2.1 = 0.5238, or 52.38%.

Now let’s try another example. In the recent Houston Texans at Atlanta Falcons game one sports book charged a 25% vigorish for Houston -2.5. Now, I happen to have lost that bet so the vig didn’t matter, but what success percentage would I have needed to equalize that vig? S = 1.25 / 2.25 = 0.556, or 55.6% success rate. Even if I win I lose! Very few people are consistently above 55%, so in the long term its a losing propositon. I shouldn’t have taken that pretend bet at all.

This helps us out a great deal in deciding whether to take a bet or not. Let’s say you’re in the situation where you’ve chosen a team to defeat its opponent by more than the spread and you’re ready to bet. You go to your sports book and you find your bet, but there’s a different vigorish than 10% listed. Maybe its 15% or 20% or more. How can you determine if you should take that bet or not? Just use the formula S = (1 + V) / (2 + V) and you’ll come up with a success rate. Do you think you’re that accurate in the long term? Do you feel that good about this particular bet for some reason? It’s your decision, but now you have the math to back it up!