The Fundamental DFS Misunderstanding

Over the last couple weeks, I’ve examined the current DFS debacle and some reader perspectives. And I keep getting hammered with a fundamental misunderstanding of the “sport.” Sure, any given day of DFS is wildly unpredictable. The same is true of baseball. Does that mean there’s an incongruity between DFS and the need for large sample sizes we espouse here on FanGraphs? No.

At its heart, DFS is a process-driven lottery. As with any lottery – or for that matter, a baseball game – winning on a single day involves a lot of statistical noise. Where DFS differs from your typical state-run money pit is that the right process can create a positive expected value (EV). Theoretically. The wrong process still has a negative EV, but that doesn’t mean you can’t win.

Here’s where the misunderstanding comes into play. Some people say it’s impossible to offer meaningful advice about a single slate of games. The teeny tiny sample size of one game ensures chaos. That’s true-ish, but it’s never been about one game. You’re thinking too small. Succeeding at DFS is about a full season of games. The outcome of one contest is close to random. The outcome of 250 contests reflects talent level. If you have a positive EV process, you’ll probably make money over a full season.

Daily fantasy really isn’t much different than your typical daily roto league. You pay for your team, you draft it, then you set a lineup for 180 straight days. Every one of those days you accrue stats. You have to be better than your fellows with some frequency in order to win the league.

With DFS, you’re simply paid with cash rather than cumulative roto points. You also have the option to skip contests, draft new lineups for every time slot, and explore a new range of strategies and tactics. But it’s still deceptively similar to a roto league. You’re trying to grind out a winning season, not a winning day.

Of course, nothing is simple. Players rarely use the same process over the entire season. For example, I used an aggressive, multi-entry technique last April. I lost a bunch of money. When I reverted to my previous one or two lineup approach built around bargains, I made over $100 in the final five months. I’m not shopping for a yacht, but it’s a sign that the process is probably reliable.

The quality of a process depends on both expected production and the picks of all other players. I had a lot of near misses this year. They came because my lineups had too much overlap with other owners. At least I think that was the problem. It’s hard to tell. When I rostered Joey Votto’s three-home run night, so did 25 percent of the field. I think 50 percent of the field rostered Bryce Harper’s three-home run day (it was a small contest).

In this case, my good process was nullified by its growing popularity. When I participated in 2014, not that many people were using the simple principles that I follow. Sometime over the winter, other venues got a LOT better at offering DFS advice. Bad for me. Good and bad for our readers.

In August and September, I started reaching for more obscure picks with mixed results. It’s something I’ll try more frequently next year. As the average user gets smarter, the large contests (GPPs) have become more about anticipating who will and won’t be popular relative to their actual value. So there are two important angles to analyze: expected versus actual value and popularity.

Ideally, the game would be about picking the lineup with the highest expected value. In actuality, there are more variables to balance. This is partly why algorithms have an advantage. Their process allows them to perform tertiary analysis on expected ownership rates. Perhaps that’s the next frontier for The Daily Grind.