Building the Ultimate Draft List -- Part 1

This article is part of our MLB Observations series.

Last year, I switched up my draft prep in one crucial respect. Instead of building my own cheat sheets by hand, I outsourced the process to a combination of the National Fantasy Baseball Championship (NFBC) Average Draft Position (ADP) and an average between two well-known high-quality projections algorithms, Steamer and Derek Carty's The Bat. I weighted the NFBC ADP -- comprising dozens of recent drafts on which informed players wagered their hard-earned money -- at 60 percent and Steamer and The Bat at 20 percent each. 

The idea was to mine the best information available by combining the vast knowledge contained in the high-stakes market with the most predictive performance metrics.  The market handicaps who's slated for closer roles and how healthy players looked late in the season, while the projections systems aren't mislead by recency bias, buzzy players or cosmetic results not commensurate with underlying skills.

The 60/40 split was somewhat arbitrary, but I got the rough idea from high-stakes professional bettor Rufus Peabody who uses his own metrics to estimate the relative strength of NFL teams and combines them with the market number (point spread) to make his bets. As rigorous as he is with his own process, Rufus still gave the market number more weight. 

But where the light bulb went on was after I read an article about a horse-racing bettor who had a moderately successful algorithm that only took off when he combined it with publicly available market data: 

A breakthrough came when Benter hit on the idea of incorporating a data set hiding in plain sight: the Jockey Club's publicly available betting odds. Building his own set of odds from scratch had been profitable, but he found that using the public odds as a starting point and refining them with his proprietary algorithm was dramatically more profitable. He considered the move his single most important innovation, and in the 1990-91 season, he said, he won about $3 million.

Accordingly, I ran the 60/40 split last year to generate my draft list, tinkered with it to account for only my strongest leans and assembled my teams. While a couple were duds (perhaps not coincidentally the ones where I departed most from my new methodology), this one finished 12th overall in the NFBC online (more than 2,100 teams) and this one cashed in the NFBC Main Event. 

Given last year's results, I reprised the method for 2020, but modified it slightly. I gave more weight (two-thirds) to the NFBC values and only one sixth to each of the algorithms. I did this initially because I mis-remembered last year's split as 70/30 and wanted to make sure the algos had enough of an effect on my rankings to differentiate them from ADP and afterward, when I realized it was actually 60/40, out of inertia. I could still switch them, but 60/40 was just a guess last year, and the algo-based rankings aren't tailored specifically to the NFBC game with its overall contest element that prevents you from being too light on any category out of the draft, e.g, the projections-based rankings don't account for the scarcity of stolen bases. That three and a third percent probably doesn't make a huge difference, and for now I'm content to leave it, but full disclosure, it's not exactly the same as last year's, and I might change it later in draft season. 

I also re-calculated the NFBC hitter/pitcher/closer/catcher splits to generate dollar values. Last year, I just went off my 2017 numbers, using a 66/34 hitter/pitcher split, with $293 of a hypothetical $3,120 ($260 * 12 teams) budget spent on closers and $145 on catchers. But given the way starting pitching has been pushed up more than ever and saves have declined in the NFBC, I decided to update it. 

Accordingly, I used recent NFBC ADP, setting the No. 1 player, Ronald Acuna, at $45 and the number 276 (23 starters * 12 teams) player (Carter Kieboom) at roughly $1, and made the total add up to $3,120 by multiplying $45 (and each successive draft slot value) by .98586. (As you might imagine, it took some trial and error to get the parameters to line up.) I went with $45 for the No. 1 overall pick instead of $52, or whatever Acuna fetches in the NFBC auctions, because auctions are a different ecosystem, and the top overall valued player by Steamer and The Bat, according to my formula, was closer to $45. In any event, the starting number wasn't likely to sway the overall distribution much. 

Here's what it looks like -- keep in mind this is recent NFBC ADP only leading up to my draft on Friday, February 21st. 

When you add up the implicit NFBC expenditures by position, it looks like this: 

The first thing I noticed was that pitching -- already expensive in the NFBC three years ago relative to the old 70/30 industry-league standard -- was an even bigger portion of the budget, coming to $1.104.36 out of the $3,120 for a 64.6%/35.4% split. Moreover, relief pitching that went for $293 in 2017 dropped to only $210.73. That means an even bigger portion of total resources ($893.63) is going toward starting pitching:

Knowing these splits is necessary for converting the algorithmic projections into dollar values -- and ultimately rankings. The projections can tell you how hitters are valued in each category relative to other hitters, but to compare starting pitchers to relievers to hitters, you need to know how much of the total budget to apportion to each. 

In Part 2 - I'll dive into the Steamer and The Bat projections for hitters and pitchers, how and why I separated relievers and catchers from the rest and how I integrated them back into an overall dollar-value based ranking. Then I'll add the NFBC ADP and aggregate it back into the master list.