Big Data,Efficient Markets,and the End of Daily Fantasy Sports As We Know It?

Introduction

Approximately 40 years ago, Rotisserie League Baseball was invented as an alternate means among friends to enjoy the game of baseball. ¹ While small at first, the game rapidly spread, allowing anyone who could read sports statistics in a newspaper the opportunity to be an armchair general manager, buying and selling players and trading players with other fantasy team owners, as well as an armchair manager in deciding which players to compete in a given day or week. All without the inconvenience of needing millions of dollars in start-up capital. As the game became more and more popular, it spread to other sports.

Fantasy football first gained wide acceptance in the 1980s. This was at the same time that the National Football League (NFL) was becoming more popular in the United States than Major League Baseball. ² While personal computing was becoming more entrenched in U.S. businesses during this time period, it was still a rarity to find in a residence, and Internet connections were still relatively rare outside the academic community. In addition to NFL popularity, many people liked that football games were (essentially) 1 day per week and as such, statistics only needed to be followed and updated once per week. Entire communities were even built around fantasy football.

With the introduction of the Internet and the relative ease of obtaining statistics, the games then began to explode. Databases were created and updated to assist fantasy sports players. Moreover, services were started to facilitate the leagues themselves; first on services such as CompuServe and AOL, and then to the wider Internet with the mid-1990s' standardization of the Internet in business and residences. Other ancillary services, such as scouting reports, new and “roto”-specific metrics, and draft advice were soon to follow. While sports such as baseball with its discrete plays have long lent themselves to easy analysis, ³ fan creativity and business opportunity then created fantasy sports opportunities in almost every sport imaginable: basketball, hockey, soccer/football, NASCAR, cricket, golf, and Formula 1 racing are just some of the more popular examples. Perhaps even one day there will be armchair provosts and deans creating fantasy research institutes—an interesting budget exercise. Fantasy sports were so popular by the mid-2000s that when the U.S. Congress created the Unlawful Internet Gambling Enforcement Act (UIGEA) in 2006, which was designed to remove the U.S. financial sector from payment processing related to Internet gaming, there was a special carve-out specifically exempting fantasy sports. ⁴

While one benefit of fantasy football over baseball was to reduce the need for onerous daily record-keeping by the fantasy league's commissioner, this is in direct contrast with the desires of “action junkies,” or players who need to constantly be in action. The daily fantasy sports (DFS) market provided competitions to those who wanted to compete more frequently than simply once per week. It also provided competition to those who did not have the time or desire to engage in a season-long competition. Furthermore, it allowed individuals to combine the outcomes of multiple sports, previously just a curiosity of proposition bets on Super Bowl Sunday. While 2007 saw the launch of Fantasy Sports Live, it was really in 2014 and 2015 that DFS, through its huge marketing budgets, became a mainstay of the market. It was estimated that a commercial for FanDuel or DraftKings ran every 90 seconds before the 2015 football season. ⁵

Games and Efficient Markets

Efficient market hypothesis (EMH) ⁶ posits that a market is efficient if the price of assets in a market reflects all available and relevant information about those assets. While traditionally used to explain pricing and develop theory about the capital markets, particularly the U.S. equity and debt markets, it has expanded to explain prices in other asset markets such as real estate, ⁷ commodity futures, ⁸ and currency exchange ⁹ among others. Moreover, regulators such as the Security and Exchange Commission strive to ensure that the public markets in the United States are fair and, as a result, as close to efficient as possible. In newer and unregulated markets, however, it is not uncommon for these markets to be inefficient, as market participants have unequal access to information and/or unequal ability to analyze, consume, and act on that information. When the markets are inefficient, it is theorized that there are abnormal economic profits to be made. ¹⁰ However, it is also theorized that these inefficiencies, such as arbitrage opportunities, will attract enough market participants such that the market will by nature become efficient. Market makers prefer efficient markets, as it leaves them without unnecessary risk or exposure.

While different than capital markets, many wagering opportunities (particularly in sports) are also viewed as markets. For example, in its simplest form, a sports book (market maker) generally tries to set a betting line on a football game not necessarily as a prediction of an outcome but rather to get the bettors (market participants) to wager an equal amount of money on both sides. In that case, the market “clears,” and regardless of the outcome of the event, the bookmaker has no adverse exposure and merely earns his or her commissions (the “vig” or the “juice”) for matching the market participants. As such, EMH has been tested in many gaming venues such as horse racing, ¹¹ professional football, ¹² soccer, ¹³ tennis, ¹⁴ golf, ¹⁵ cricket, ¹⁶ and ice hockey. ¹⁷ There has even been some crossover from the traditional capital markets to these sports markets, as Wall Street stalwart Cantor Fitzgerald spun off Cantor Gaming in the late 2000s to offer in-game betting services as well as traditional sports betting operations.

Fantasy sports can also be viewed as a market. In the large public markets, such as DFS, the market maker seeks to attract participants to compete in various contests (e.g., head-to-head, small- and large-group competitions with differing number of winners), where each participant sets a lineup of team members, generally restricted by the market maker placing a “salary” for that player based on perceived value, and giving each team entered a “salary cap.” This salary cap is what turns the DFS optimization problem from a reasonably simple one to one with a greater level of complexity. Big data and analytics are then used by market participants to deal with an additional constraint on the optimization problem.

One major difference between fantasy sports and equity investing, however, is the nature of the market. While each provides a series of trades for which a market maker extracts a fee, the nature of the financial markets is such that one can buy and hold a security or mutual fund that has some risk attached but in general has a positive expectation. There is no “buy-and-hold” strategy available in fantasy sports. Therefore, these markets more resemble the options derivative markets rather than a market where one owns an underlying asset such as a share of stock or a bond. Options markets, as derivatives, have been tested for efficiency many times and traditionally been lacking evidence of exploitable inefficiencies. ¹⁸

Analytics and DFS

It is no secret to the readers of this journal that fantasy sports players who use analytics have a distinct advantage over those who do not. Millions of dollars are spent developing and refining models to compete in the DFS markets, often to the gain of the tout sites such as RotoGrinders, RotoQL, and FantasyLabs, among others. Lineups based on backtested models are entered to see their validity in real time and are continuously refined and adjusted. The Sloan Sports Analytics Conference (SSAC) is but one example of a venue where people discuss models, although surely the participants' most valuable models are held back, similar to a financial conference. Since 2010, the SSAC has averaged just under one research article, poster, panel, or invited session each year dedicated to fantasy sports. The same can be said for journals and open access sites, for example, arXiv. ¹⁹ The market itself has begun to resemble the financial markets, with television shows, pundits, subscription websites, and print materials available constantly. SiriusXM satellite radio has a channel devoted to fantasy sports. It is rare that a mainstream sports programming channel does not have a fantasy columnist or section, much like a company has an investor relations department or a news program has a business segment.

When DFS is combined with analytics, the “sharp money” is separated from the “square money” with incredible speed. Winners and losers are decided every day, and when there is such a separation in skill levels, the markets will empty quickly. For example, in the first half of 2015, 91% of the profits in DFS baseball were earned by only 1.3% of the players. ²⁰ Contrast this to live poker tournaments, where analytics have also taken over but not as quickly as in online poker (more on this later), and the possibility of winning is not so remote as to eliminate the longshot dreamers. The phrases “All you need is a chip and a chair” and “any two cards” have helped to keep the hopes of many amateur poker players alive as they compete with the professionals. This aligns well with the quote from UNLV Center for Gaming Research Director David Schwartz, who says, “You can either take people's money or make them feel stupid, but you can't do both.”

Even with the salary cap and big data, variance will always play a part in each contest, as is the case in games such as poker. There is always the individual who will play well above or below expectations, the freak injury, or the unexpected weather condition that causes models to break down for a given evening or event. In the long run, variance of returns should even out and player returns should converge to the expected value. The number of trials required to get to the long run expected returns is beyond the scope of this article, but many players will submit dozens, hundreds, or even thousands of lineups ²¹ (p.95) in an effort to reach the long run more quickly. Figure 1 shows a screenshot from a DFS site showing that people even at the $1 level in head-to-head tournaments have submitted as many as 50 (not necessarily unique) lineups.

OPEN IN VIEWER

FIG. 1.

DFS head-to-head, September 25, 2018. DFS, daily fantasy sports.

Could this mean the end of DFS as we know it? Or, at a minimum, will “big data” and analytics lead us to a game structure that is an efficient market, increasingly dominated by those who are able to leverage these technologies? EMH posits that the markets move toward greater levels of efficiency as the search costs for information decrease. Let us examine several other games, contests, and markets and their popularity as analytics moved them toward weak, semistrong, and strong solutions.

Research Propositions

If DFS does indeed follow other markets, we should be able to test the solvability based on the complexity of the markets. As mentioned above, online poker is thought of as increasingly solved or solvable, but the size of the decision tree is directly related to the number of game states that need to be examined to develop optimal play. It is of course no surprise that the algorithms first tackle fixed-limit poker (where the decisions at each point are only bet, raise, call, or fold, and the amount that can be bet is a fixed amount) as opposed to no-limit poker where a player can bet any amount from the minimum to his/her entire stack of chips at any point. Early algorithms also first look at the heads-up game where there are only 2 players as opposed to a full poker table of 9 or 10 players, as the decision tree is exponentially greater even for 3 players rather than just 2. ^†

Similarly, the heads-up “cash games” for DFS are the ones that should be solved first. Indeed, in Barbarisi ²¹ (p.155), the leading market participants claim that it has already been solved for the National Basketball Association (NBA) for a couple of years at the highest levels. However, if logic follows that the decision trees are more difficult as the number of participants increases, then the tournaments with tens of thousands of players should be more difficult to solve. Thus:

Proposition 1: Heads-up DFS will be solved more easily than Tournament DFS

If the market leaves little opportunity for exploitation and only the market maker has a positive expectation, then the opportunity cost for participants will be too high and they will seek other opportunities for returns. However, people who play state lotteries, even with a rake approaching 50%, continue to play them even knowing the relative expected value. It follows that people will exhibit similar behaviors in DFS as they seek lottery-style “jackpot” payouts. This leads us to:

Proposition 2: Heads-up DFS markets will shrink relative to Tournament DFS markets

Moreover, as the information supporting the markets becomes even more publicly available and easily processable to those who choose to avail themselves of it, one will not be able to use any form of algorithm that is better for a particular day to beat the markets. Mistakes in pricing of athletes for a given day relative to the salary cap will be identifiable and exploitable by all and, hence, neither past performance nor any particular unknown analysis will allow one to consistently beat the markets. Therefore, we posit:

Proposition 3a: The DFS markets will be shown to be semistrong efficient

Proposition 3b: Although there will always be some form of private information, it will neither be frequent nor critical enough to matter and thus the DFS markets will act as strong-form efficient.

Discussion and Conclusion

What are documented above are some examples of market bubbles, greater fool theory, and smart money squeezing out square money. Such markets, although largely unavailable in 1981, were predicted by Rosen's ⁴¹ description of the economics of superstars. When the markets get too big and the stakes too high, an inevitable market contraction and, in some cases, market destruction can follow rather quickly, as seen in a quote from the famous poker player Amarillo Slim, who is said to have mused “You can shear a sheep many times, but you can only skin him once.” More recently, it has also been said that when you start getting advice from your hairdresser on a new moneymaking opportunity that you do not quite understand, then it is time to get out.

What will be the future of sports analytics as it relates to fantasy sports? The market is already seeing exit points from large facilitators such as Yahoo!, who exited Fantasy golf after the 2017 season. Leagues such as the NBA are also likely to demand payment for access to data as integrity fees after the U.S. Supreme Court in May 2018 struck down federal laws prohibiting sports gambling, and this may extend to fantasy sports of all kinds.

It would not be surprising to see top daily fantasy players shift their strategies to becoming providers of advice and data and similar tools. One might compare this to the gold mining strategy in the California gold rush, where few found gold but the ones who got rich were the ones supplying the miners with picks, axes, tents, and work pants, such as Levi Strauss. A similar path was observed in online poker, where top market participants (players) moved away from playing games, or waiting for games that never took place, to being actually market makers themselves (runitonce.eu), poker training sites (e.g., upswingpoker.com, advancedpokertraining.com, redchippoker.com), data miners (hhsender.com, hhsmithy.com), ancillary products to the software such as heads-up displays (PokerTracker, Holdem Manager, Jivaro), and other similar products. This happened when the places where the professionals used to make their money effectively dried up, particularly in “Heads-Up” matches. Professionals, acting economically rational, would refuse to play each other (“sit them out” or “quit them”) when they did not believe they had an advantage. This would result in dozens of heads-up tables being open with one person sitting, effectively either waiting for a less-skilled player to choose to play against them or having to decide to play each other, which rarely happened.

While not everyone has the means or the know-how to collect and analyze vast amounts of data related to fantasy sports, the proliferation of online “lineup optimizers,” coupled with “touts” selling their expertise and expected point projections, will naturally lead to a more efficient market. That is, the various inefficiencies in the games are reduced as the ability to predict the expected fantasy points scored on a given night improves through the use of more data and better analytical techniques. As was noted in Hunter et al., ¹⁹ the problem is a simple integer optimization problem at its core and everybody should be maximizing the same objective function and set of constraints once this occurs. At this point of inevitability, the games will be reduced to gambling on day-to-day stochasticity, or simply a game of chance.

Large market makers such as DraftKings and FanDuel are likely to try to protect their markets, and again may look to the online poker world for ideas as to how to make the casual customers stick around while not losing high-volume players to other pursuits. They have huge investments in their companies and will want to adapt as necessary to sustain their livelihood. They have already started to diversify into providing traditional sports book operations in places such as New Jersey. It is possible that they may try several ideas to reduce the advantage of professional players picking on the newcomers, a practice known in the online poker world as “bumhunting.” One can see even back, in Figure 1, for DFS that there are hundreds of matchups currently available, but the players are choosing not to play each other, even at the $1 level. They are seeking less skilled players and will even write scripts to identify when those “fish” are online and submitting lineups. Player anonymity, attempts to ban certain scripts and tools, and other mechanisms may be attempted to protect their markets. However, the market makers will need to attract a large volume of small-stakes players to make up for each large-stakes player who disappears as an active market participant.

Moreover, one cannot discount the asymmetry of payouts within the market. Although not necessarily related to big data, the payout structures from the large tournaments are so top-heavy that an individual who gets lucky and wins (“binks”) a tournament may find himself/herself well above his or her suggested return for a short term, while others who have achieved top 1% finishes but not a win may be unable to sustain as a participant. This could also be a factor in how the market evolves.

DFS markets may also be subject to the whims of regulators, which they can attempt to influence but cannot control. While the industry claims it is expressly legal and permitted due to the UIGEA carveout, it has been posited that this carveout was meant to keep the $20 season-long league participants from worry of running afoul of these laws as opposed to expressly permitting multi-$1000 daily participants in DFS. As states have come in to try to establish policies for their citizens, the leading DFS sites have exited markets in Washington, Nevada, Arizona, Hawaii, and seven other states. While not directly related to the presence or absence of big data, it is a concern for the market makers.

Given the zero-sum nature of fantasy sports, we posit that big data will likely send it back to the days where it (and poker) existed before there were online variants and huge data histories. There will be a small-stakes largely private market conducted among friends, but there will also be a tiny in size but high in stakes high-roller market where algorithms battle out for supremacy and the edges between them become so small that even the expected value of the best players cannot cover the costs of information gathering or the sanctioning fees (“the rake”) of the market makers.