SBT’s OFFICIAL PLAYS
If you just want to see what plays I am on personally, just click on the SBT’s Plays link at the top of the page and you will see what games I personally bet every day. I select my plays by using a complex key generator/finder by cycling through 55 different combinations of the 6 projection methods to determine the most optimal keys there are. If you have the patience, the second half of this livestream featured the development of the Key Finder, which I call the Key Stripper, if you really want an under the hood look at how plays are selected: https://youtu.be/k7G2XWz8Pmo
The new and updated college basketball picks page can look very confusing, but I am here to explain it. The site is divided into 2 sections.
My college basketball model is more of a college basketball ecosystem of models – I run 5 different projection algorithms to make predictions with. In the image above these are numbered 1 to 6.
1 INDY METHOD – This method is the original prediction method I have been running since the start of the 2020-2021 season and was the only method featured on my site in the month of December 2020. INDY stands for Individual Monte Carlo. This approach takes the projected roster for each team before the game, the runs a monte carlo simulation by estimating the 5 players on the floor on each team each possession and calculating expectations based on the cumulative opponent adjusted and recent weighted stats of the 5 players on the floor. This method can be rather unstable in the early stages of the season before sample sizes have been built up, but the advantage to it is that stats are based on projected rosters, so any players who are not playing will not factor into the stats. It’s also how Daily Fantasy stats can be derived. I expect as the season goes along, for this method to become the strongest method of the 6 due to its granular nature.
2 TEAM METHOD – This method is also a monte carlo simulation, but it differs by simply using the cumulative team’s season to date opponent adjusted and recent weighted stats in the simulation. Individual stats are not factored in, so for example if a team has a major contributor go out with a season ending injury, that player’s statistics will still be factored into the team’s season statistical averages that are used in the simulation. The benefit of this approach is that it is more stable in the early stages of a season before a large enough sample of games have been built up.
3 BLENDER METHOD – The Blender Method is a Sports Betting Truth staple since the launch of the channel. A version of the Blender has appeared in every model in every sport I have displayed publicly since the debut of Sports Betting Truth. The Blender is rather simple – it is a regression based approach. In college basketball, I regress off a database going back 4 years(an entire recruiting cycle) filled with dozens of statistics that I have selected and deemed important. For team scores, totals and projected margins, the approach is a linear regression based approach. For win percentage predictions, it uses a logistic regression. The downside of this method and the ATM method is that the 4 year database it is regressing against has home advantages from 2017-2020 which hovered between 3.2 and 3.5 points for the home team baked in, which can inflate home advantages for 2021 contests as the 2021 home advantage as of the writing of this post is only 2.19 points. Therefore I have to make adjustments after the fact to account for this.
4 ATM METHOD – Sports Betting Truth regulars are also no stranger to the ATM Method, but I have been secretive in the past about the nature of the ATM Method. In this blog post, I will elaborate. The ATM Method goes hand in hand with the Blender Method. Both utilize the same four year database. Where the ATM Method differs from the Blender is that the ATM attempts to identify past matchups within the four year database that are statistically similar to the statistical profiles of the two teams that the ATM method is attempting to predict via K Means cluster analysis, and then uses hierarchal clustering and dimensionality reduction to determine where to cut the tree. This clustering process is a multi stage approach, with certain stats deemed “Tier 1” that are filtered for first as they are considered the most important, then “Tier 2” statistics which are less important but are used to help smooth the edges. For example, if the ATM were predicting a matchup between Gonzaga and Baylor(the current #1 and #2 teams at the time of this post), its going to filter out matchups between bottom feeding Northeast Conference teams from the database. Absent from the college basketball model but present in other sports, “Penny”, “Nickel”, “Dime”, “Quarter”, “Half” and “Dollar” are various designations of the K Means K Value.
5 SIMPLE METHOD – The Simple Method is exactly as it is named – simple. All it does is calculate Offensive/Defensive Efficiency and Tempo expectations based on each team’s OE, DE and Tempo vs. the National Average, factors in home advantage and comes up with a projected final score for both teams. Win percentages are calculated by a Log 5 Pythagorean Win Expectation equation plugging in efficiency expectations into the calculation.
6 COMBO METHOD – This method is simply an average of the 5 methods.
Those are the 6 methods that are used to project games. Each method has its own strength and weaknesses and only time will tell which methods are best and which are not.
7 – 5/6 AGREE – This column shows how many units have been won when 5 out of the 6 projection methods agree on a play.
8 – 6/6 AGREE – This column shows how many units have been won when all 6 projection methods agree on a play.
9 – UNITS WON – This column tracks how many cumulative units have been won across the various projection methods.
10 – KEY and KEY UNIT – This column shows the “keys” for each projection method. I have a video explaining what keys are: https://youtu.be/B9uZbx45xG0 but simply put, keys are a criteria that shows the optimal value/edge that a prediction has in order to bet it. For example, in the BLENDER column, the Key is 2.80 and the Key Unit is 8.40. This means that each projection in which the BLENDER margin differed 2.8 or more points from the spread, it showed a profit of 8.4 units. Keys attempt to find the “peak” units won and what the value/edge is for said peak.
The results section tracks 3 different layers – the current selected date, the last 14 days, and year to date.
The Plays Section is a lineup of all the games of the selected day that my projection system predicted or is predicting. If you select a past day you can see the game by game results, if you pick the current day you can see what the models are projecting for that day.
12 – Pre Game Score Projection – Rows 1 and 2, these rows/columns show the predicted scores for each team across the 6 different prediction methods, color coded by method.
13 – Margin and Total Edges vs. the Spread/Total – Rows 3 and 4, below the predicted scores show the edges against the spread/total for margin and totals. Row 3 shows the projected SPREAD margin(calculated as Away Score + Posted Away Spread – Home Score), so a negative value means the against the spread edge favors the home team, and a positive number favors the away team. This number is not simply Away Score – Home Score, but also factors in the spread. Row 4 shows the projected TOTAL edge(calculated as Away Score + Home Score – Posted Total) so a negative number means the UNDER is favored and a positive number means the OVER is favored. One thing to point out is the BLENDER TOTAL differs from the other 5 methods. Every other method the total is based on AWAY SCORE + HOME SCORE, but the BLENDER total projection regresses against a different set of coefficients than the BLENDER regresses against for scores, so BLENDER AWAY SCORE + BLENDER HOME SCORE will not completely equal BLENDER TOTAL PROJECTION.
14 – Pre Game Win % Projection – These rows/columns list the projected win % of each matchup. Rows 1/2 show the predicted win percentages for both teams. Rows 3 and 4 calculate the MONEY LINE EDGE. Money Line Edge is calculated as Projected Win % – Implied Moneyline Odds %. Example: If a team is a -200 favorite that means the money line is assuming they will win 66.7% of the time. If the pre game win projection shows them winning 75% of the time, their MONEY LINE EDGE = 75 – 66.7 = 8.3%.
15 – Plays or Results – If you select the current day(as illustrated above) you will see dots indicating which sides each prediction model is favoring. In the TOTAL columns, dots on the first row indicate an OVER play, dots on the bottom row indicate an UNDER play. If you select a past day, it will show the results of the plays each prediction method made. If no dot is in either row, then the model decided both sides were -EV. This usually only happens on moneylines.