by Chris Lardieri, Inside Sports, and Wells Fargo Investments
Nearly halfway through the fourth quarter of the 2019 Super Bowl, the New England Patriots and Los Angeles Rams were tied 3-3 in a game dominated by defense. Then, Patriots quarterback Tom Brady connected one last time with tight end Rob Gronkowsi on a 29-yard pass to the 2-yard line. Running back Sony Michel ran it in for a touchdown on the next play, and approximately 98 million Americans watching on television at home abruptly woke up. The Patriots went on to win 13-3 in the lowest-scoring Super Bowl in history. On the bright side, the Analytic Investors team correctly predicted the outcome—improving our overall record to a respectable 11-5 (69%) against the spread.
Thanks to a controversial non-call of obvious pass interference in last year’s NFC championship game, the 2019 season began with yet another rule change. The NFL Competition Committee approved the use of instant replay to review pass interference calls/non-calls for one year only. However, it was not without controversy as a mere 24% (24 out of 101) of such challenges were reversed by the NFL’s instant replay center in New York. In spite of this, and wide receiver Antonio Brown being released by both the Oakland Raiders and Patriots in a matter of weeks back in September, the NFL didn’t miss a beat.
The game remained as popular as ever. Its already-high television ratings increased, thanks to star quarterbacks both young (Lamar Jackson and Patrick Mahomes) and old (Drew Brees and Aaron Rodgers). The Cleveland Browns attempted to build on last year’s resurgence by trading for star wide receiver Odell Beckham Jr. but managed to underachieve in their typical fashion. The Oakland Raiders made one final run at the playoffs for their faithful in the Bay Area, but they ultimately fell short and packed their bags for their shiny new digs in Las Vegas. CBS commentator Tony Romo continued to show an uncanny ability to predict upcoming plays, earning him and/or his crystal ball a reportedly record-setting contract offer from ESPN. Finally, the surprising Tennessee Titans upset the defending champion Patriots in the wild card round of the playoffs, and football fans outside of New England rejoiced.
The NFL’s centennial season concludes in Miami on Sunday, February 2, 2020, when the San Francisco 49ers face the Kansas City Chiefs in Super Bowl LIV. This matchup features two historically successful franchises that haven’t hoisted the Lombardi Trophy in 25 and 50 years, respectively. While football fans, in general, are happy to see an AFC team other than the Patriots appearing in this game, it may actually also be a good sign for our prediction. More on that later.
At the core of this annual paper is a computation we developed way back when Tom Brady and Bill Belichick only had one Super Bowl ring to their names. We think of it as the merging of football and our quantitative approach to investment management, called “NFL alpha.” This version of alpha is the cumulative return on investment (ROI) for all 32 NFL teams’ performances relative to wagering market expectations for all regular-season games.
To illustrate, let’s assume a $100 “money line” wager on the Green Bay Packers to win before each of their 16 games. If the Packers win in a given week, one would collect the $100 wager plus an additional amount that’s a function of their win probability, as implied by the wagering odds that week. Should they lose, one would lose the $100. Once the season concludes, we then add up the winnings and compare that number with the $1,600 total amount wagered during the season ($100 per game for 16 games). Any amount above $1,600 would imply a positive NFL alpha, and anything less would indicate a negative one. In this case, one would have tallied just over $2,218 due to the Packers alpha of 38.7%—fifth-best in the NFL and a whopping 79.4% improvement over 2018 (see Table 1 on right). Could this be attributed to the Packers’ new head coach, Matt LaFleur? We’ll leave that to the sports analytics community to debate.
Can One Game Effect Alpha?
The process of writing this paper begins with an annual rite of passage: a number of Analytic team members submit their guesses on which teams will be at the top and bottom of our NFL alpha results. While the lowest-ranked team wasn’t much of a surprise, the top-ranked one was. The Miami Dolphins finished with a whopping 70.3% alpha, the highest for a team since 2009. As the underdog in all 16 of their games, the Dolphins were given plenty of opportunities to outperform expectations.
The Dolphins traded away a number of star players at the beginning of the season in hopes of ending up with the #1 pick in next year’s draft. Some experts coined this strategy “Tank for Tua”—a perceived effort to lose every game in order to take star Alabama quarterback Tua Tagovailoa. But as typical with the NFL, nothing ever seems to go as predicted. Tagovailoa suffered a season-ending hip injury, and the rag-tag Dolphins decided to rally behind rookie coach Brian Flores. After starting 0-7, the team upset the New York Jets, Indianapolis Colts, and Philadelphia Eagles to end up in the slightly positive alpha territory. But the best was yet to come.
In week 17, Miami went into Gillette Stadium to face the 17-point-favorite Patriots, who needed a win to secure a bye in the first round of the playoffs. Should be a no-brainer, right? Not for the resilient Dolphins. Journeyman quarterback Ryan Fitzpatrick threw a touchdown pass to Mike Gesicki with 24 seconds left in the game to give Miami a 27-24 victory, resulting in a massive surge in alpha (see Chart 1, below) and very likely contributing to New England’s early playoff exit the following week.
On the flip side, there are always the perennial alpha flatliners. This year, it was the Cincinnati Bengals. They lost star wide receiver A.J. Green to a fluke ankle injury in a July training camp practice, and it was all downhill from there. The team started the season 0-11 before upsetting the Jets (notice a trend?) and Browns, locking in a -69.6% alpha (see Chart 1, below) and the top pick in the 2020 draft. We’d like to add more color here, but this was simply a bad football team. Take heart, Bengals fans. Having the worst alpha means that your team will have the chance to draft your next franchise quarterback: Heisman Trophy and national championship winner Joe Burrow of LSU.
Let’s examine two more cases in which one game had a meaningful impact on teams’ alphas. The Atlanta Falcons started the season 1-7, and rumors were flying that head coach Dan Quinn was on the verge of being fired. After their bye week, the team had a tall task: going into New Orleans to face the heavily favored 7-1 Saints. In typical any-given-Sunday fashion, the Falcons upset the Saints 28-9, and their alpha jumped from -73.3% to 6.6% (see Chart 2, below). Then they upset the 49ers 2922 in week 14 on the way to recording the third-highest alpha of the season—39.4%—and managed to save Quinn’s job in the process.
The Los Angeles (NOT San Diego) Chargers had a similar turning point in their season. After defeating the Indianapolis Colts in week 1, the Chargers appeared to be the same playoff-caliber team they were last year. But it didn’t last. They went into Detroit and lost 13-10 to the hapless Lions, causing their +45.2% alpha to plummet to -27.4% (see Chart 2, below). The team stayed in negative alpha territory the rest of the season, thanks partly to the declining play of quarterback Philip Rivers. As the team prepared to move into the new Sofi Stadium with their cross-town rival Rams in 2020, rumors surfaced that the NFL may be realizing the error of its ways in relocating the Chargers from San Diego and would even consider moving the team to London in the future. No word yet on whether we’ll eventually be calculating alphas in pound sterling, but stay tuned.
“The Lottery Ticket”
Since 2004, the Analytic team has been managing low-volatility equity assets based on our published research, which demonstrates that low-risk equities tend to outperform higher-risk ones over time. As we’ve also outlined previously, there’s a similar trend in the world of football wagers. Specifically, bets placed on large favorites with lower payouts (in other words, less risky) typically outperform bets placed on heavy underdogs that pay out significantly more—often referred to as lottery tickets. In 2018, these higher-risk wagers outperformed lower-risk ones for the first time since the 2015 season.
This trend, albeit not as pronounced, continued in 2019. Low-risk wagers returned -0.7% while higher-risk, longshot bets returned 4.8% (see Chart 3, below)—an outperformance of 5.4% and consistent with the stories of the Dolphins and Falcons outlined above. A similar pattern occurred in the markets in 2019 as high-beta equities generally outperformed their low-beta counterparts.
The bulk of this paper has focused on the regular season. Now it’s time to focus on the playoffs and our predictions. When we first published this paper back in 2004, it was based on our finding that teams who outperformed expectations in one NFL season tended to underperform in the following season and vice versa. We found this same relationship also exists between the regular season and the postseason. As a result of this football form of mean reversion, the team with the lower alpha is typically undervalued when compared with its higher alpha opponent.
In the current postseason, this approach has resulted in a 4-6 record, breaking our streak of 15 consecutive years with at least a 50% postseason success rate. Two of the losses were by a mere half-point: The Buffalo Bills blew a 16-point lead to the Houston Texans and lost on a field goal in overtime, and the Seattle Seahawks failed to convert on a late 2-point conversion attempt in a loss to the Packers (see Table 2, below). Oh, that dreaded “hook!”
Despite the above, our historical average now stands at a respectable 59%.
A Second Predictive Approach: Computer-reading between the lines
As if one predictive approach weren’t enough, in 2019 we debuted a second one that’s based on machine learning (ML). Even though last year’s ML prediction incorrectly picked the Rams to win, we wanted to give this approach another chance. So, we’re again supplementing our traditional alpha model pick with a Super Bowl prediction courtesy of the machines. Readers may recall that our previous ML pick relied upon feature mining and gradient descent. Particularly astute readers may recall that the pick proved to be incorrect, in spite of encouraging historical performance (55% accuracy).
This year we took a turn from straight statistical ML and explored the value of natural language processing. What could we glean from the words of sportsbook “sharps” themselves? To accomplish this, we collected text from a group of professional handicappers that explains the rationale behind their picks. For each team in each matchup, we took all sentences justifying that particular pick and fed them into a deep, bidirectional linguistic neural network. These are a family of language models that have a powerful understanding of syntax and semantics. They exhibit superhuman performance on tasks ranging from document summarization to propositional logic—and they’re widely assumed to automatically generate the words spoken by the Joe Buck robot in the Fox booth. In our case, these cutting-edge tools simply allowed us to quantitatively measure the pairwise similarities of all arguments in favor of a particular pick.
Having created this distance matrix, we extracted the maximum eigenvalue—roughly, the magnitude of importance of the dominant argument being made in this collection of sentences—and compared it with the theoretical limit for this value. We interpreted this distance as the overall conviction held by the set of handicappers regarding the critical feature of the matchup that leads to the pick. Then, we picked the team having the more relatively dominant argument, conversely going against the team whose advantages weren’t widely apparent to our set of professional handicappers.
This approach yielded strong performance in these playoffs, with a 75% hit rate. For the Super Bowl, it favors the 49ers by 1 point. We’re assuming this machine pick is free of bias in spite of the fact that our hardware is from the Silicon Valley.
Super Bowl LIV
Unlike last year’s game, which had two single-digit alpha teams, Super Bowl LIV pits two teams in the top 10: the sixth-ranked San Francisco 49ers (37.8%) against the ninth-ranked Kansas City Chiefs (24.1%) (Table 3 on page 6). The 49ers also had the highest change in alpha from 2018, at 82.6%. This is primarily because the 49ers had a league-worst -44.7% alpha last year, thanks largely to quarterback Jimmy Garoppolo’s season-ending knee injury during a week 3 loss to—you guessed it—the Chiefs.
This year’s game features two evenly matched teams: the Chiefs’ high-flying offense led by superstar quarterback Mahomes going up against a stout 49ers defense anchored by rookie pass rusher Nick Bosa. It also features two offensive-guru head coaches with something to prove: young Kyle Shanahan, who was offensive coordinator of the Falcons during their epic collapse to the Patriots in the 2017 Super Bowl, and veteran Andy Reid, who’s seeking his first title in 21 seasons. The game is also reminiscent of a previous Super Bowl when the 49ers faced a young star quarterback in his second season as a starter. That was Super Bowl XIX in 1985, and they defeated Dan Marino’s Dolphins 38-16. Could history repeat itself 35 years later?
Our pick is the lower-alpha Chiefs to win the game by more than one point. Considering that our model has correctly predicted seven of the eight Super Bowls (88%) that have not involved the Patriots (dating back to 2004), we remain cautiously optimistic about this selection. And speaking of the Pats: This model does not have a prediction on what team Brady will be quarterbacking in 2020.
We want to help our clients build for successful outcomes, defend portfolios against uncertainty, and create long-term financial well-being. To discuss your investment needs, please contact them:
• To reach our U.S.-based investment professionals, contact your existing client relations director, or contact us at WFAMInstitutional@wellsfargo.com.
• To reach our U.S.-based intermediary sales professionals, contact your dedicated regional director, or call us at 1-888-877-9275.
• To reach our U.S.-based retirement professionals, contact Nathaniel Miles, head of Defined Contribution at Wells Fargo Asset Management, at firstname.lastname@example.org.
• To discuss environmental, social, and governance (ESG) investing solutions, contact Hannah Skeates, global head of ESG at Wells Fargo Asset Management, at email@example.com.
All investing involves risks, including the possible loss of principal. There can be no assurance that any investment strategy will be successful. Investments fluctuate with changes in the market and economic conditions and in different environments due to numerous factors, some of which may be unpredictable.
Each asset class has its own risk and return characteristics. Asset allocation and diversification do not ensure or guarantee better performance and cannot eliminate the risk of investment losses. Wells Fargo Asset Management (WFAM) is the trade name for certain investment advisory/management firms owned by Wells Fargo & Company.
These firms include but are not limited to Wells Capital Management Incorporated and Wells Fargo Funds Management, LLC. Certain products managed by WFAM entities are distributed by Wells Fargo Funds Distributor, LLC (a broker-dealer and Member FINRA). This material is for general informational and educational purposes only and is NOT intended to provide investment advice or a recommendation of any kind—including a recommendation for any specific investment, strategy, or plan.
PAR-2020-02962© 2020 Wells Fargo & Company. All rights reserved.