The Different Risk Levels of Fantasy Positions in Achieving their Projected Season Score -- A Historical Analysis

I've been excited to share with you this updated new study.

It's about your relative payoff from drafting each different fantasy position.

It's about the fantasy "return on investment", so to speak: Which position (QB/RB/WR/TE) has been most dependable on hitting the mark?

Why care about Positional Risk?

Like pretty much all of you, I care a lot about what sequence to take positions in a draft. 
Do we go Zero-RB? Late-QB?

I think it's important to understand risk better-- and to take any lessons we can to assist our draft strategy.
And this is especially true for the early draft, where de-risking is more important than pure upside.

Yes, these results feed into TapThatDraft valuations, whenever you use the default "BEER+" baseline selection.

At the end of the analysis, I'll get to describing "Risk-Adjusted Valuation" of each position.
But as always, I also hope you'll find little nuggets of insight along the way.

Methodology -- New and Updated

Projections. After the 2025 season, I dug up ACTUAL individual player pre-season point projections, from the last 10 years.

Therefore, this study is representing pre-season expectations from the best information analysts could have had, at the time of drafting.

We're not just assuming "the previous season's RB5 should’ve start the next season expecting to be RB5 again".
Lots of things change around players during the off-season: different teams, different coaches, different QBs, new team-mates, an of course injuries.

These pre-season point projections are essential, because they incorporate the specific scenarios of each player, as was presented at the time.

Outcomes vs. Expectations. With projections in hand, we can look at actual plots of RESULTS (fantasy points) versus ESTIMATES (pre-season projections). I think they're really interesting: they reveal the real amount of scatter, and each position has slight differences.

Variance vs. Projections. The next question was "How does VARIANCE change as a function of player score projections?" I fit variance to a continuous curve dependent on score projections, which is not perfect but shows a consistent trend: higher uncertainty for higher projections.

VORP. Finally, it's important to define a "replacement player" for each position, to get the right scaling against projections. Intuitively: the "risk" of a player can't measured be against a worst case of ZERO points, when subs are allowed. This points-per-game replacement amount is going to be used to calculate VORP. Conveniently, the same concept feeds right into the next part about Sharpe Ratio. For this study, I am assuming replacement points from: QB16 (meant to estimate streaming of QBs above a QB25 equivalent), RB40, WR48, and TE20.

Sharpe Ratio. Explained below, the information in the above steps feeds into calculations of Sharpe Ratios for each data point. Taking the dependence of these on pre-season projections, we arrive at a way to quantify the relative Risk-adjusted value of pre-season projections.

Resulting Scatter Plots

Let's get our heads wrapped around what the resulting data looks like. Here's an example scatter-plots for QB:

The above chart shows Adjusted Seasonal Fantasy Point outcomes versus season point expectations.

This graph is already interesting itself, because scatterplots tell you how dependable the projections are. If there was more of a straight line— instead of a cloud of scatter— then you'd know we're playing a predictable game in fantasy football. That is, with a straight-line result, you could assume decisions can be made like clock-work. Actually, QB looks more clear-cut than the other positions. But obviously we see scatter-- That means our hobby is vey much a game of statistics, and it justifies the rest of the approach.

The next question is “Are higher-projected players more predictable, or less predictable?” We're caring about the changing accuracy of predictions, as you move from left to right across the graph. For an example, let's look at the variances for WRs of the last 10 years:

As you can see, the expected variance increases gradually, almost linearly, with the WR projection. (Though clearly with much noise.)

In particular: We really care a lot more about the scatter of the higher-forecast players. We care less for the lower ones.

Intuitively: When we draft a top-tier player, we're looking for certainty. We always know there will be players popping up from lower ranks (maybe on waivers), but draft priorities are supposed to be about getting the guys we can trust.

As it turns out, for all positions, higher-projected players always tend to be have LARGER error. As shown in the above example graph.

The Sharpe Ratio is Legit for Addressing the Impact of Variance on “Value”

It's a shame that you've probably heard of the Sharpe Ratio (wikipedia) ONLY if you work with economics, insurance, etc. My opinion: too many articles try to fudge their way around by concocting new "clever" metrics, inventions that sound like “subtract the floor from the ceiling, and divide by the average, and multiply by the standard deviation…"— or whatever. All kinds of made-up math. And very few readers will be able to judge if those ideas got applied the wrong way. To avoid fluffy conclusions (like those sometimes are) I hope in the future more studies will consider to use the Sharpe Ratio calculation, as a way to help them get their point across.

The Sharpe ratio is an equation for making a simple estimate of risk-adjusted returns on investment. It works by deducting an amount for "Risk-free returns", and then dividing by standard deviation. You can find more correct assessments in a given situation, but this is a simple estimate.

Be aware: the trickiest part of utilizing the Sharpe Ratio is usually about defining what is a risk-free rate. In our case, I will a convenient assumption: The risk-free return is the same as the production of the "replacement" player. (By assumption, that player is always available at no additional cost.) I described above which rank player I'm assuming for each position. (In a VBD calculation, these are the players you would assume as your point "baseline".)

Take a look at how this equation transforms the initial scatter-plot, by combining the variance-dependence and the replacement value:

As mentioned, it’s convenient for us that the Sharpe ratio calculation intersects the x-axis where VORP does.

Results of Sharpe Ratio Trends for the 4 Positions: QB, RB, WR, and TE

The important part is that this gives us a trend to directly transform "point projections" to the “Risk-adjusted Expected Value”. This is what we wanted.

And we also wanted to see: Each position has a unique trend, like the one shown above, and it depends on scoring settings (std. or ppr).

One quality check seems necessary though: Is a straight linear justified for the different positions?

I chose to observe a moving-average trend, for each of the 4 scatterplots, to judge whether linear fitting made sense. Here’s what I see:

Two conclusions stand out.

First, it is generally acceptable to assume a linear fit, in scaling projections to risk-adjusted value.

Secondly, there is one glaring exception: Higher end Running backs have lagged significantly behind in their Return on Investment.

This is apparent, despite the fact that RBs projected for <230 points per season fit the linear trend.

I should note, for the purposes of applying the approximation for RBs, that removing the higher-end “outlier” points does not significantly impact the slope, for converting points to risk-adjusted value. Therefore, it is reasonable to assume the coefficient can be applied without worry that it represents a “low end” estimate.

In TapThatDraft’s BEER+ baselining routine, I do not apply this extra deduction to the high-end RBs, even if I should. I observe that RBs already take a hit in risk adjustment, as described further below.

Zooming in on the Biggest RB Busts

There’s no clear consensus on whether the high-end RBs were a result of an unlikely succession of very bad luck— versus whether higher-end RBs attract a more risky scenario.

Without trying to conclude, I will instead just label the individual points with the player and year that they represent:

Again, despite the significance of these busts in the most crucial range of consideration, removing them from the data-set does not significantly shift the slope of the best-fit line. Apparently the large number of other RB instances already follow a lower-slope (higher risk per return) conversion from projections to risk-adjusted returns.

So which positions are best / worst, after risk-adjustment?

Running backs. As just described, RBs take a hit from risk adjustment, and it’s not just because of the elite busts. I want to be clear about my bias: this is not a conclusion that I wanted or liked. I've been a "Robust-RB" guy, personally, in my standard leagues. But the data makes it very clear: Higher ranked RBs deserve some discount in your draft assessment, in PPR leagues. If I apply a conservative estimate, their VBD valuation should receive something like a 10-15% discount.

Like I said, not what I wanted to hear, to confirm my bias! But very often I lean on a comment from a follower that always makes me laugh: "It feels weird and I don't like it, but I have to trust the data". It's become something of a motto!

Tight Ends. Meanwhile, who benefits? Does anyone position gain from risk-adjustment? Once again the result is against my intuition: It's usually TEs who get the slight bump. But in contrast to the case for RBs, this effect has not been so prominent for PPR leagues. Anyway... what this means is, if you see high-ranked TEs, they've historically been more likely to get closer to meeting expectations-- rather than bring the risk of a low floor. Again, not a conclusion I expected. After all, I'm a guy who thought it was super smart, last year, to take LaPorta early, and then oh heck why not grab Pitts too when he fell to me. So this one kind of hurts.

Wide Receivers also get a mild bump that appears more relevant in Standard leagues. It seems that counting "Receptions" introduces additional variance (even though net scores are higher).

Quarterbacks? The apparent scatter, shown in the earlier picture above, seemed at a glance to imply QBs are random. But applying the method to focus on the highest-ranked QBs, there is a reasonable average-risk to selecting the top tier QBs. So it turns out that QBs are "normal" regarding risk-adjusted draft value.

TL;DR RBs appear to deserve a downgrade most of the time, and sometimes TEs deserve a bump according to historical trends. All positions get at least a slight adjustment, but the degree of risk-adjustment depends on settings (standard vs. PPR)

One skeptical question you might ask: Don't the analysts who make projections take this positional risk into account? If they did, that would negate the use of applying it. However, their numbers don't appear like they discount differently across positions (instead the numbers seem uniformly high), and I don't even think they should (because their focus should usually be ordering within a position). So it still makes sense for the rest of us to apply risk-adjustment according to our league settings.

TapThatDraft

Finally, yes. Of course, these results are automatically integrated into the TapThatDraft draft sheets that I created and freely opened up to everyone! The tool modulates risk adjustment automatically, according to which PPR scoring setting you selected. Accessing a sheet URL is as quick as this: [LINK]

Specifically, the risk-adjustment to player valuation is part of my "BEER+" default valuation (article link), and is one of 5 key improvements to VBD (simple image view here, or Reddit post link here, see section 4) I made to help you prioritize players smarter. You wouldn't notice these upgrades right away-- they're tucked behind the high simplicity of display, which should help you act quicker in your draft-- especially when you're feeling "in the fray"!

So, if you're wondering "Do I use Hero-RB? Late-QB? Zero-RB?" etc. , then the guidance is clear. You react to your specific situation in the draft, taking the best player available to you at the time. If a certain position is riskier, then they are automatically discounted appropriately in value, by the list ordering. When your valuation is automatic like this, you can focus on secondary effects, like finding ADP deals or balancing your roster.

Thanks, and good luck!

/Subvertadown