Do We Have Time for a Story?
When I first started playing daily fantasy sports, I came at it from an algorithm-only perspective. DFS seemed pretty straightforward to me: For every sport, all you had to do was create a projection system, run it through an optimizer, and have the computer pick your winning lineup every day. It took me a couple of months and losing streaks to figure out that there’s a little more to it than that.
My first instinct was to go through the projections themselves and see if there was anything I was missing. Sure, in theory you should be doing that anyway — but, by the beard of Zeus, losing money really gives you a new level of focus when it comes to getting your numbers right.
Also, once you start to pick similar players over multiple weeks who underperform your expectations, it’s much easier to dive into the data and ask specific questions about what you might be getting wrong.
If you notice a pattern, you can backtest your general rules (perhaps using our free Trends tool), see if you can refine them, and get your models to work better.
Beyond the Projections
It took me even longer to accept that there are factors beyond projected points that really do make a difference for roster construction, such as Consistency and Upside.
Consistency, Upside, and other premium exclusive metrics are accessible via our free Ratings tool.
In most sports, you want high-floor, low-variance players for cash games and high-ceiling, high-variance players for guaranteed prize pools. Your approaches to Consistency and Upside vary drastically dependent on the format — and those factors have nothing to do with projected points.
In PGA, you have additional soft factors: Line movement, course fit, course history, etc. The best DFS players are able to consider such factors in the abstract and properly weight them. Sure, some of them might be already baked into projections, but long-term averages don’t necessarily say whether these factors hurt or help more in cash games or GPPs.
The Algorithmic Bias
I was slow to incorporate these factors because of a classic bias of the algorithm-only approach: It was just too hard to move off of a lineup that I thought would score the most points over the long run.
Sure, Consistency might be important in the abstract, but how much pure value would I be willing to trade off for the sake of a semi-nebulous soft factor? None of the analysis I read could answer that question really well either. So I did what bad players do: I stuck my head in the sand and told myself that I was making great plays, that I was merely the unfortunate victim of variance, and that everything would work out over the long run.
Even if I were correct on that note (I wasn’t), this is not the way to improve your decision-making process. Sure, it sucks to admit that you might not be as good at DFS as you thought that you were, but that’s a required step if you actually want to get better.
Hard Facts for Soft Factors
Of course, not all of the burden should be placed on individuals to improve. The DFS community as a whole also needs to get better. Here are two specific ways in which we can all improve our approach to soft factors:
- First, there could be more attempts to quantify some of these soft factors.
- Second, after we’ve quantified them, we should weight these factors relative to each other, i.e., we should show how much production/value it’s reasonable to forgo when increasing the Consistency of a cash game lineup.
And since FantasyLabs takes seriously the imperative of Mahatma Gandhi to “be the change that you wish to see in the [DFS] world,” for the next couple of weeks I’m going to explore the topic of soft factors and see if I can make some changes.
Be The Change
Each week, I’m going to take a non-projection factor, try to put some numbers to it, and explain how it interacts with projections. I’ll also do a little bit of hypothesizing about how it would be utilized in roster constructions for cash games vs. GPPs.
And, for the grand finale, I’ll outline A) the process I’ve been using for figuring out how to evaluate all of these factors explicitly and B) the findings from that process.
This will probably be one of the slower series I’ve done, but I think that it will fill a troublesome gap in DFS analysis in general, to say nothing of PGA-specific analysis.
