![author-image](https://ftnfantasy.com/wp-content/uploads/2024/04/Screen_Shot_2024-04-30_at_3.05.05_PM-removebg-preview.png)
In my initial primer on closing line value (CLV), I introduced exactly what CLV was, and how it could create an edge in NFL spread betting…if we could locate it in time. I’ve been spending some time digging through old betting lines with this in mind, hoping to dig up any clues and nuggets to help bettors be profitable.
Methodology
To assist me on my journey, I’ll be using classification trees using R, a coding language for stats and graphics. If you are interested in getting into any of this on your own, the packages I use are rpart and rpart.plot. If you’ve never seen what outputs for these trees look like, here is an example.
- The decimal at the top represents the examined probability of items in the tree node hitting what is being tested for.
- The “n” is simply the number of sample items in this tree node.
- The percentage tells us what percent of the sample is in this node (n/total).
- The bottom represents a split of the data based on a particular observation. In the example, that observation is o_spread (opening spread) < -1.2.
- As you move down the tree, going to the left means we have satisfied the condition, while going right means we have not satisfied the condition.
Initial Findings
I plugged 143 different variables into rpart with no additional modifications to the sample or settings. The objective was to locate sides that took on a CLV of -2 or less (more negative), as we established in Part 1 how valuable that movement could be. This was the result.
143 variables inserted, and the tree found just *one* to be significant when utilizing the entire sample of games: opening spread. A random side in the sample had roughly a 13% chance of snagging the desired CLV. If that side had an opening spread less than -1.2, those odds dropped to 9.4% (not desirable). However, if the opening spread was greater than or equal to -1.2, probability rose to 16%. If the opening spread was also less than 2.3, probability jumped to 26%. We can neatly summarize this as follows:
If -1.2 <= o_spread < 2.3, then there is a 26% chance that side will take on two points or more of spread value. From a practical standpoint, we are looking at sides ranging from -1 to +2.
Dead Zone
My initial thoughts were that this would give us a good launch point for mining CLV on NFL sides. However, a deeper investigation into this range presented some alarming findings.
o_spread |
# |
Cover |
% |
-1 |
39 |
18 |
46.2% |
-0.5 |
0 |
0 |
– |
0 |
78 |
45 |
57.7% |
0.5 |
2 |
1 |
50.0% |
1 |
62 |
31 |
50.0% |
1.5 |
28 |
14 |
50.0% |
2 |
23 |
13 |
56.5% |
Total |
232 |
122 |
52.6% |
Though our overall sample of sides had a cover rate of 59.3% when gaining two points or more of spread value, this smaller group covered at a much lower rate. In fact, it would be a break-even rate, with no discernible edge. There are subsets of this group that have been profitable, but we should not focus too much on them given the small samples.
It would appear that this is actually a dead-zone range for opening spreads and CLV. And it makes sense. These spreads are very low, and will ultimately not cross through a key number with the line movement. Most of the teams (120 of 122) that covered these lines won outright. In that case, the additional spread value does not have a ton of meaning.
Does this mean that we have no hope in mining useful CLV for NFL betting? Absolutely not. I’ll have some useful nuggets in the next installment of this series.