Searching for Value in Week 4
As I mentioned last week, I'll be ranking the "top expected-value money-line plays of the week" going forward, which is what I focused on a few years ago. (I'll still highlight the picks that meet the ELDORADO threshold or come close, as I've done the last few weeks.) This year, when ranking the top money-value plays of the week, I am also going to show how those plays stack up relative to others this season and to thousands of games over many seasons.
Let's take our successful Seattle (+245) pick in Week 1 as an example. Not only did the Seahawks carry the highest expected value of the week (and so far this season) as a money-line play at +245, they were in the 99.7th percentile in terms of expected value over several thousand potential picks in recent NFL history.
Last week, the top three potential plays were Washington (+240), which lost 24-8; Tennessee (+115), which won 24-22; and New England (+135), which lost 37-26. When set in context of the last few years, Washington fell in the 96th percentile, Tennessee in the 95th percentile, and New England in the 94th percentile.
Those are all pretty high percentiles, but they were not high enough to qualify. That cutoff essentially falls at the 99th percentile, which means that for every 50 games played (and therefore 100 teams to potentially play), we can expect approximately one to clear the threshold .
In Week 2, Dallas (+270) actually cleared this mark, coming in at the 99.5th percentile for expected value when measured against all potential money-line plays for the last few years. And they won. Houston (+270) was in the 98th percentile and lost, and Seattle fell in the 96th percentile and lost.
Here are this season's top 10 plays so far:
1. SEA (+240) vs. DEN - Week 1 - 99.7th percentile - WON 17-16
2. DAL (+270) vs. CIN - Week 2 - 99.5 - WON 20-17
***99th percentile threshold***
3. HOU (+390) at DEN - Week 2 - 98.2 - LOST 19-6
4. WAS (+240) vs. PHI - Week 3 - 96.1 - LOST 24-8
5. SEA (+310) at SFO - Week 2 - 95.9 - LOST 27-7
6. HOU (+283) vs. IND - Week 1 - 95.6 - TIED 20-20
7. HOU (+205) vs. LAC - Week 4 - 95.5 - TBD
8. CHI (+265) vs. SFO - Week 1 - 95.0 - WON 19-10
9. TEN (+115) vs. LVR - Week 3 - 94.7 - WON 24-22
10. NYJ (+275) vs. BAL - Week 1 - 94.1 - LOST 24-9
And here are this week's top 5 money-line expected value plays:
1. HOU (+205) vs. LAC - 95.5th percentile
2. SEA (+180) at DET - 93.3
3. NEP (+375) at GBP - 92.1
4. CHI (+152) at NYG - 88.0
5. PIT (-159) vs. NYJ - 87.2
When reviewing these, remember that the system does not necessarily think these teams are likely to win their games; it thinks they're some degree more likely to do so than the prevailing money line implies. (The exception to this is Pittsburgh in Week 4, the rare favorite to the make the list; the system basically thinks they should be heavier favorites. Still, their expected-value percentile is not very high.)
That top 10 list above looks pretty good — 4 wins, 4 losses, a push, and one game not yet played — which, when you're betting on modest to heavy underdogs, is a superb outcome. At 100 units per play, that's +890 on the wins (over and above what you initially risked) and -400 on the losses, netting +490 on 800 risked, good for a 61% ROI.
Nice as that sounds, if we leave this week's post latching onto any "thresholds," I would point again to the 99th percentile that I mentioned above, which basically aligns with where I initially pinned the ELDORADO cutoff. The 99th percentile and above holds up well to several years of back-testing.
(In other words, this is not a standardized test where falling in the 80th or 90th percentile might feel good. The back-tested ELDORADO threshold is really high — the 99th percentile and above — so don't be fooled when you see numbers in the high 80s or low 90s, even if they make a certain week's top five.)
If anything else, I would point to the 94th or 95th percentile as the threshold for "honorable mention" considerations. Prior to this year, it basically broke even via back-testing, which means that betting them would have cost time and/or stress for essentially no return. (But hopefully some entertainment, which is really the point here.) I'll keep these weekly rankings, percentiles, and thresholds in focus going forward.
Enjoy Week 4!
Searching for Value in Week 3
After a solid win with Seattle +245 in Week 1, there were no official ELDORADO selections in Week 2. The three money-line picks that flirted with hitting the qualifying thresholds — Dallas, Seattle, and Houston — did not see their lines move enough to make the cut-off.
Dallas needed to hit +300 versus Cincinnati; the best I saw was +285 in the middle of last week and +270 on Sunday. Seattle needed to make it all the way to +400 at San Francisco; the best I saw was +365 last week, before it came down to the low 300s (e.g., +310). And Houston needed to be +415 or better at Denver; the best I saw was +400 in the middle of last week, and then +390 over the weekend. (They basically all moved in the wrong direction.)
Even though none of these became qualifying picks, I had some free plays, so I actually dabbled (equally) on all three. My historical analyses suggest that this is foolish, and that if a play doesn't meet the necessary expected-value thresholds, you should stay away. But I did it anyway (for the entertainment, I suppose), and because it was free.
This trio of "honorable mention" picks went 1-2, which when you're betting fairly heavy underdogs on the money line, makes for a winning day (even if the bets weren't free). Dallas won at +270; Seattle got blown out by the 49ers (I am in the camp that thinks the 49ers are better off with Jimmy Garoppolo); and Houston was actually beating Denver into the beginning of the fourth quarter, before losing 16-9.
Unfortunately — and as I warned would happen — there are no qualifying picks in Week 3, and there aren't really any "almost" or "honorable mention" picks, either. The top three considerations for the week would be Washington +240 versus Philadelphia (needs +315 to officially qualify), Tennessee +115 versus Las Vegas (needs +175), and New England +135 versus Baltimore (needs +210).
If we end up with too many "officially pick-less" weeks, it might be worth bringing back the "top five expected value plays" construct that we used in 2018, at the very least to keep this interesting, and so we can track their performance week to week.
For example — and I realize this sounds convenient to say — but the teams with the highest expected money-line value in each of the first two weeks were Seattle (+245) in Week 1 and Dallas (+270) in Week 2, both of which won. At current levels, Washington (+240) has the highest expected money-line value in Week 3, though it's a lot less than where Seattle and Dallas landed.
So let's keep on monitoring, and I'll think about putting out money-line rankings in future weeks (a top three or five sort of thing), which will probably be more fun. I'll also take a look at the futures board some time in the next few weeks to see if there's any numerical value lurking in teams' odds to win their division, conference, or the Super Bowl. Enjoy Week 3!
Searching for Value in Week 2
Year-to-Date: 1-0 (+245); ARZ, ATL, DAL, GBP, TEN all lost in Week 1 (good for our win-total under bets)
As luck — or good ELDORADO models? — would have it, Week 1 was a resounding success. Geno Smith and the Seahawks upset Russell Wilson and the Broncos in Seattle. I actually ended up getting in at +245, which made the play even more attractive than the +230 consensus when I posted the article. (It remained at +220 at several books, and I even saw a +200 out there; down in those ranges, I would have stayed away.)
And while season-long bets aren't worth tracking on a weekly basis, it's worth pointing out that all five of ELDORADO's win-total "unders" went the way we wanted in Week 1. All five teams lost, which is good.
There are three possible money-line plays in Week 2. As always (or almost always), they are modest to heavy underdogs. Dallas is +285 at home against Cincinnati; if that goes to +300 or higher, it qualifies. Dak Prescott is out, and Cooper Rush is starting, meaning you will have to make a subjective call. Do you believe the Cowboys have much more than a 26% chance to win the game, as the current line implies?
Our friend Geno Smith is back in the mix, as the Seahawks are +365 underdogs (21.5% implied probability to win) at San Francisco; If that moves to +400 or greater, it also qualifies. And finally, Houston is a +400 underdog at Denver (20% implied probability to win); the model likes the Texans at +415 or better. As you can likely tell, the models think Seattle is undervalued and Denver is overvalued by oddsmakers thus far.
Note that I am usually quoting the best underdog money line I see. Depending on where and when you look, the lines you see could be different. The important question is whether the aforementioned thresholds are met. (And as always, none of this necessarily means the model thinks the pick will win; it simply means the model thinks the team has "enough" of a better chance than the money line implies.)
Searching for Value in Week 1
The only Week 1 play that (barely) made the new Searching for Value cut is Seattle, which is a +220 or +230 home underdog against Denver. The +230 implies that Seattle has a 30% chance to win. The system doesn't think Seattle is especially likely to win; it thinks they have a better chance than that +230 line implies. (It's also important to note that the system likes Seattle at +230 or higher but not at +220 or lower.)
Skeptics might also argue that the ELDORADO system hasn't adequately accounted for the Russell Wilson trade. Do you believe Geno Smith gives the Seahawks significantly more than a 30% chance to win at home? Between line variance and who's at quarterback, this week's Searching for Value "pick" is a perfect example of how a "data-based" system is still unavoidably subjective — ultimately up to you and your gut.
I also plan to track five (small) win-total plays that I made a few days ago. They're all "unders" (how boring), and they grew out of a "model" (too strong a word?) that I originally started working on (and didn't revisit for several years) after taking a deep-dive into the topic for an ESPN Chalk article that I wrote four years ago.
Win totals and their corresponding odds (how much you win based on how much your risk) vary a lot between sportsbooks, so you really have to be discerning, especially if you're trying to identify expected value. (The 17-game regular season also makes historical analysis a bitch, and while 2022 win totals are set in the context of 17 games, it does make betting "unders" feel a little weirder.) Anyway, here's what I did:
- Arizona under 8.5 wins (-110) and 7.5 wins (+150)
- Atlanta under 4.5 wins (+115)
- Dallas under 9.5 wins (+135)
- Green Bay under 10.5 wins (+135) and 11.5 wins (-120)
- Tennessee under 8.5 wins (+120)
In a typical season, the model thinks that about three of these will hit; it does not imply a clean sweep. So as with pretty much everything related to wagering on the NFL (or anything else), single events and choices — a game here or a game there, a missed or made field goal, a decision to make a play or not make a play — will materially swing results. But at the very least, it will keep me (us?) a bit more engaged.
It's been about three-and-half years since I've posted with some consistency. Since then, and at risk of sounding cliché, "I got married, had a baby, and got a dog" — all of which has been amazing but leaves no time for ELDORADO posts, especially when the original raison d'être for this blog was to tackle topics for which the data was so hard to come by that no one had ever given the topic at hand its proper due before.
While I won't be taking any of those vintage ELDORADO deep-dives in the near future, I have been pondering how I might go about dabbling again, this time in a simpler but perhaps more fun and reader-relevant way. One of the topics that's lurked in the back of my mind for much of those three-and-a-half years was the "Searching for Value: NFL Picks and Futures" column I wrote during the 2018 NFL season.
Beyond being a weekly read that (some?) people (might have?) enjoyed, the column's pick system did extremely well that season, going 44-34-3 against the spread (+770, 9.5% ROI) and 26-45 on the money line (+1914, 24% ROI), as almost every money line pick was a sizable underdog. I wanted to see if there was anything to the money line piece of this, so I downloaded and back-tested many years of historical data.
Unsurprisingly, and as I repeatedly warned in 2018, there was not — the exact "Searching for Value" system used in 2018 (and close variations thereof) would have lost money in several other seasons. This validated my assumption that the 2018 column's results were more a matter of small-sample, single-season luck, that the bottom eventually falls out, and that "Joe the bettor" can't beat Vegas over the long term.
Still, I had the data, so I kept adapting and testing, and I ultimately built my own system. The fruits of those analyses are what I will offer you on a weekly basis during the 2022 NFL season in the form of money-line picks and periodic looks at NFL futures. As in the past, this is all based on "expected value" — trying to identify situations where you think a team has a higher probability to do something than the line implies.
Given the thresholds the system has defined, we could only be looking at a handful of qualifying picks per year. That is admittedly not a good way to launch a "weekly" column, but I don't want to throw picks out there just to fill space. That also means that the system's historical pick universe is not that large, so one outcome per season can really swing results, both for that season and for the system's record overall.
Finally, as objective and rules-based as I hoped to make this, some degree of subjectivity is impossible to avoid. For example, if the system supports a certain play but that team's star quarterback just got injured, do you buck the system and decide not to do it? (Some NFL prediction models out there do attempt to account for quarterback quality, but I tested gambling applications for those, and they were not pretty.)