Do You Trust Me?
Imagine you are running from guards and someone you just met asks you to jump off a roof with them to escape?
"Do you trust me?"
Now, imagine a model you just built is telling you to bet the Dolphins?
"Do you trust me?"
After two weeks, we have a small sample size, yet amazing results: 17-4 overall (81%). I am most pleased with the Filtered results, 6-0 ATS and 7-4 Totals (overall includes everything I have posted to this site and filtered is a subset of that). These are what I call the "Filtered" picks, which I interpret as a pick worthy of a bet. To elaborate further, while building the model, some predictions are so close to 50% it is not practical to blindly follow the model for every game, especially when it is telling you it is a coin-flip game. There has to be some threshold for what is coin-flip versus trusted. Based on model accuracy and how many games were "Filtered", I chose a 10% range for ATS and 14% for Totals; practically meaning the 10% range around 50% (45-55%) and 14% (43-57%) would be a stay-away. Those games outside that range are my "Filtered" picks and during model build / testing those ranges helped build a 60% test set accuracy (2018 games were the test set, meaning blindly following the model of 2018 games yielded 60% results).
Regardless, its only two weeks in, and yet to be determined if the model is choosing the right side because I built a wonderful, amazing, cash-cow of a model woooooo, or.... it simply got lucky.
Interesting to note that the model usually selects away teams and unders, making me think if those tend to be undervalued in the market. The unders side of it is another factor to note, as the increase in holding penalties has theoretically caused more unders as it stifles offenses with 10 yard penalties.
Now onto the picks, and if you were worried about picking the 0-2 Dolphins, get ready for more concern, as four of the picks are 0-2 teams.
Interesting that the qualitative strategy of taking a 0-2 (backs against the walls, need it more, save their season) fits each of these narratives and in the Steelers case, even some talk that Rudolph is that bad and could surprise. Regardless, you can make a case for nearly any team, but the Dolphins?!
This week, the Top-5 match the Filtered, so consider these five bet-able.
1. Pit +6.5 (Pit @ SF -6.5) - Top ATS Pick
2. LAR -3 (LAR @ Cle +3)
3. Cin +6 (Cin @ Buf -6)
4. Car +3 (Car @ Ari -3) - NOTE: Line reflects that Cam is likely to be out
5. Mia +21.5 (Mia @ Dal -21.5)
Mia was close to being cut off, being just a shade over the 10% Filtered threshold. Here is the full breakdown.
NOTE: if the "Prediction" column is low, then it likes the away team. If high, then the home team. 0.500 would be a push.
Five "filtered" games this week, and curious that four of them match ATS games. Either the model is picking up on something in these games or that is a huge coincidence (or the model is wrong... but nah can't be that).
1. Mia @ Dal - Over 47 - Top Total Pick
2. LAR @ Cle - Under 49
3. Chi @ Was - Under 41.5
4. Pit @ SF - Under 44
5. Cin @ Buf - Over 44
So... Do you trust me?