Fantasy draft = fans choosing a team
Fans take turns selecting players before the season. Earlier selections are valuable because more players are still available. ESPN records the average draft position (ADP): where each player is typically chosen.
NBA research · Statistics & machine learning
Are NBA rookiesoverlooked?
Independent research · Python / statisticsOctober 2026
In fantasy basketball, fans build imaginary teams of real NBA players and earn points from their performances. I studied whether rookies—players in their first NBA season—were being picked later than their results justified.
See what the study foundWhat I found
No clear evidence that rookies were routinely overlooked.
Of 19 rookies with verified 2025 fantasy draft records, 9 finished the season ranked higher than where fans picked them. The sample did not show that rookies generally performed better than their fantasy selection positions suggested.
01 / The idea
Picked late. Performed well.
Imagine fans choose a player 100th, but he finishes 60th in season fantasy points. He performed better than his selection position suggests. That is the kind of overlooked value I was looking for. This is an example, not a study result.
NBA draft = pro teams hiring players
This is a separate draft. NBA teams select new players to join the league. I used that selection order as one clue to future performance.
Fantasy points = a score for performance
Real game statistics are converted into game points using ESPN’s scoring rules. More points mean more contribution to a fantasy team. This study compares totals for a whole season.
02 / My approach
Two questions, two comparisons.
I built a Python pipeline to bring together historical player data, make predictions and check the results. Predicting a strong player is only half the problem: a good fantasy pick also depends on when you can select him.
Can I predict a rookie’s first season?
Learn from 819 verified rookie seasons from 2010–2024. Then compare predictions with the results of all 59 players selected in the 2025 NBA draft.
Did fans overlook these players?
Compare season performance with where fans picked players on ESPN. Verified 2025 fantasy draft records were available for 19 of the 59 players. A broader analysis used 138 records across 10 draft classes.
03 / The findings
Some players exceeded expectations. A repeatable rule was harder to find.
9 / 19
9 of 19 finished above their selection position
Individual successes did not add up to clear evidence that rookies were generally overlooked. Fantasy draft records were missing for the other 40 players, so this finding covers only a small part of the class.
59
Predictions checked against a full class
The original model was closer to the real results than a simple prediction based only on NBA draft order. The difference was too uncertain to conclude that the extra information reliably helped.
0 / 96
No reliable pattern for finding overlooked players
Across 138 historical fantasy draft records, none of 96 tested relationships provided convincing evidence after accounting for how many comparisons I tried. Testing many ideas makes chance patterns easier to find.
See the exact model scores and uncertainty
For the 19-player comparison, actual points averaged 216.7 below the benchmark associated with their fantasy selection positions. The 95% uncertainty interval ran from 567.8 below to 119.4 above the benchmark. Because it includes zero, the study could not establish an average advantage or disadvantage.
A later experiment: adding college passing statistics
Adding college assists reduced average prediction error by 11.9 points, but the uncertainty interval still included no improvement (−26.3 to +2.4). I chose these inputs after seeing the 2025 results, so this was exploration, not an independent test.
| Model | Average error · points | Ranking agreement | When tested |
|---|---|---|---|
| Original model | 404.2 | 0.583 | Set before formal evaluation |
| Original NBA draft-order comparison | 447.4 | 0.550 | Set before formal evaluation |
| NBA draft order + college assists | 430.6 | 0.568 | Explored after seeing results |
| NBA draft order only | 442.5 | 0.550 | Explored after seeing results |
Average error (MAE) is how far a prediction misses by, ignoring whether it is too high or too low; lower is better. Rank correlation measures how closely predicted player order matches actual order, from −1 (opposite) to +1 (matching). All point values use the common season scale explained below.
04 / Explore the predictions
How close were the predictions?
Each dot is one player. Compare what a model predicted with what he actually produced. These are the two later, exploratory models; this chart tests performance predictions, not whether players were overlooked in fantasy drafts.
Points here are season totals calculated with ESPN’s fantasy scoring rules, then adjusted to a common 2024–25 scale so different seasons can be compared. They are not points scored on a basketball scoreboard, or raw fantasy totals.
College passing data availableCollege passing data unavailable
Above the dashed line: the player did better than predicted. Below it: worse than predicted.
12 / 59 players shown
Choose a player to see his result above, or select a column heading to sort. Error = predicted points minus actual points. A negative error means the player did better than predicted.
| 1 | 2,136 | 2,832 | -696 | |
| 4 | 1,414 | 2,700 | -1,285 | |
| 3 | 1,561 | 2,600 | -1,039 | |
| 13 | 822 | 2,340 | -1,518 | |
| 7 | 1,240 | 2,137 | -896 | |
| 42 | 262 | 1,918 | -1,655 | |
| 5 | 1,219 | 1,827 | -608 | |
| 2 | 1,793 | 1,765 | +28 | |
| 11 | 988 | 1,668 | -680 | |
| 34 | 351 | 1,558 | -1,207 | |
| 24 | 655 | 1,400 | -744 | |
| 9 | 1,022 | 1,343 | -321 |
05 / Research details
How the study was checked.
The fantasy draft records cover only some players. The later models were developed after seeing the results. The notes below document these limits, the data sources and corrections.
Open methods and audit notes
- Past rookie seasons used for training
- 8192010–2024 draft classes
- Players in the 2025 prediction test
- 59The entire 59-pick class
- Fantasy selection records
- 13810 classes · 2010–2025
Training and outcome coverage
The 2010–2024 source retains all 894 official picks. Models train on 819 verified first-season outcomes; 75 unknown outcomes remain missing and are excluded. Verified nonparticipants remain zero. All 59 players in the 2025 class have verified outcomes.
Inputs available before the fantasy season
The follow-up uses only NBA overall pick, transformed with its natural logarithm, and college assists per 40 minutes from the final season before the NBA draft. Imputation and standardization use training data only. In 2025, 47 college-assist values are observed and 12 are missing.
Frozen test and exploratory follow-up
The original model and predictions were frozen before formal test evaluation. A Phase 1 parser had accessed ancillary held-out cells; that incident is documented, and the cells were not used for model fitting. The later two-predictor choice was made after observing 2025 outcomes, so it is explicitly exploratory rather than a fresh untouched test.
Scoring and statistical uncertainty
Production models apply current ESPN points weights to historical box scores and normalize season totals to the fixed 2024–25 top-150 reference scale. Historical market comparisons separate ESPN’s older and newer scoring eras. The follow-up interval resamples players within one class; it does not measure uncertainty across future draft classes.
Fantasy prices and source corrections
The historical screen recovered 138 actual ESPN average picks across 10 classes, with substantial gaps and selective price coverage. The rank-to-points price benchmark is a surrogate, not a calibrated market expectation. ADPs near 140 form a draft-limit cluster; an ADP-below-139 sensitivity was reported. Points/categories source-format pooling was not independently verified. FantasyPros provider columns had an ordering structure and could not be verified as literal average picks, so their earlier literal-price comparisons were corrected and excluded from the historical price analysis.
Reproduction
Saved plans, dated sources and hash checks preserve each analysis. Completed reruns verify saved artifacts without refitting. No 2026 predictions are included.
Download results
59-player results, model metrics and methods.
Sources
Official NBA draft records, game statistics, college data and archived ESPN fantasy draft records.