# NBA Rookie Fantasy Value

Fantasy basketball lets fans build virtual teams from real NBA players and earn points from their game performances. A rookie is a player in his first NBA season. This study asks whether rookies performed better than their fantasy selection positions suggested.

ADP means **average draft position**: the average place where fans select a player in fantasy drafts. A fantasy “price” here means that selection position, not money. This is separate from the **NBA draft**, where professional teams select new players. NBA draft order is an input for predicting performance; fantasy draft position is used to assess value.

**The available data did not establish average ESPN rookie underpricing or a reliable rule for finding overlooked players before the fantasy season.**

In 2025, **9 of 19 rookies with verified ESPN fantasy ADPs** finished above their fantasy draft position in season fantasy-point rankings. Their average return relative to the ADP benchmark was **−216.7 adjusted fantasy points**, with a 95% interval of **[−567.8, +119.4]**. That interval includes zero. The result does not establish that rookies are systematically cheap or expensive. [Preseason ADP archive](https://web.archive.org/web/20251009173943id_/https://hashtagbasketball.com/fantasy-basketball-adp)

## Data and results

The Python pipeline retained all official draft picks. **819 verified first-season outcomes from 2010–2024** entered training; 75 unknown outcomes stayed missing. The 2025 prediction test covered **all 59 picks**, including four verified nonparticipants with zero production. Verified fantasy prices were available for only 19 of those 59 players.

| Test | Result |
|---|---|
| Original frozen model | Mean absolute error **404.2 adjusted FP**, versus **447.4** for the draft-position baseline. The paired error-difference interval **[−105.8, +23.6]** includes zero. |
| Historical pre-draft traits | **138 actual ESPN prices across 10 classes**. None of **96 trait comparisons** remained significant after multiple-testing correction; minimum BH q-value **0.240**. |
| Later two-predictor model | NBA draft position and college assists per 40 minutes: error **430.6**, versus **442.5** for its pick-only baseline. Difference interval **[−26.3, +2.4]** includes zero. Features were selected after seeing 2025 results, so this test is exploratory. |

The original model specification and predictions were frozen before the authorized 2025 outcome collection. This does not remove the data-access incident disclosed below. Later feature exploration did not change the original saved test.

“Adjusted FP” expresses season totals on a common 2024–25 scale. It is not the raw points a player earned. The ADP-to-production curve is a comparison benchmark, not a validated market forecast. Intervals resample players within one class; they do not establish accuracy across future classes.

## Limits and corrections

Historical ESPN prices are selective, and the 2025 price test covers a small subset. Draft counts, values near the apparent pick-140 limit, and pooling across league formats remain unverified. ESPN's scoring change required separate legacy and modern price analyses. There is no matched-veteran comparison. [Historical ESPN prices](https://web.archive.org/web/20171016112921id_/http://games.espn.com:80/fba/livedraftresults) · [ESPN scoring change](https://web.archive.org/web/20210504001028id_/https://www.espn.com/fantasy/basketball/story/_/id/30296896/espn-fantasy-new-default-points-league-scoring-explained)

Earlier FantasyPros provider columns were treated as average picks. Their integer ordering structure was later identified; literal average-pick semantics remain unverified. Those price conversions were withdrawn from price-based inference. Original reports and the correction record were retained.

An early recruiting-table parser accessed and discarded ancillary NBA career cells. Those values were not exported, analyzed or used for fitting. The incident remains an audited QA failure; selective parsing and regression checks were added.

---

# NBA 新秀的梦幻篮球价值

梦幻篮球让球迷挑选真实的 NBA 球员组成虚拟球队，再按球员的比赛表现获得积分。新秀指进入 NBA 第一个赛季的球员。这项研究考察：新秀的赛季表现，是否优于玩家选中他们的位置？

ADP 是 **Average Draft Position（平均选秀顺位）**，表示玩家平均在第几位选中一名球员。文中的梦幻“价格”指这个选择位置，并不是金钱。它与职业球队招募球员的 **NBA 选秀**不同：NBA 选秀顺序用于预测球员表现，梦幻选秀位置则用于判断这个选择是否划算。

**现有数据未能证明 ESPN 整体低估了新秀，也未找到可靠的方法，在梦幻赛季开始前识别被忽略的球员。**

2025 年有真实 ESPN 梦幻选秀 ADP 的 **19 名新秀中，9 人的赛季梦幻积分排名优于 ADP**。相对 ADP 积分基准的平均回报为 **−216.7 调整后积分**，95% 区间为 **[−567.8, +119.4]**。区间包含零，因此无法确认新秀整体偏便宜或偏贵。[季前 ADP 存档](https://web.archive.org/web/20251009173943id_/https://hashtagbasketball.com/fantasy-basketball-adp)

## 数据与结果

Python 流程保留了官方选秀名单中的全部球员。训练使用 2010–2024 年 **819 个已核实的首季结果**，75 个未知结果保持缺失。2025 年预测测试覆盖 **全部 59 名球员**，包括四名已核实未出场、积分为零的球员。其中只有 19 人有已核实的梦幻选秀价格。

| 测试 | 结果 |
|---|---|
| 原始锁定模型 | 平均绝对误差 **404.2 调整后积分**，选秀顺位基线为 **447.4**。配对误差差异区间 **[−105.8, +23.6]** 包含零。 |
| 历史选秀前特征 | 收集到 **10 届新秀的 138 个真实 ESPN 价格**。**96 项特征比较**在多重检验校正后均未显著，最小 BH q 值为 **0.240**。 |
| 后续双特征模型 | NBA 选秀顺位与大学每 40 分钟助攻数：误差 **430.6**，同一模型形式的顺位基线为 **442.5**。差异区间 **[−26.3, +2.4]** 包含零。特征在看过 2025 年结果后选出，因此属于探索分析。 |

原始模型设定和预测在获授权收集 2025 年结果之前已锁定，但这不能消除下述数据访问事件。后续特征探索没有改动保存的原始测试。

“调整后积分”把赛季总积分换算到共同的 2024–25 年尺度，并非球员实际所得积分。ADP 到积分的换算只是比较基准，尚未验证为市场预期。区间来自对单届球员的重抽样，不能证明未来各届的预测表现。

## 限制与更正

历史 ESPN 价格覆盖不完整，2025 年价格测试也只包含少数球员。底层选秀次数、接近第 140 顺位的数值是否受选秀长度影响，以及是否混合不同联赛格式，均未核实。ESPN 积分规则变更前后的价格分析分别进行。研究没有同价格老将对照。[历史 ESPN 价格](https://web.archive.org/web/20171016112921id_/http://games.espn.com:80/fba/livedraftresults) · [积分规则变更](https://web.archive.org/web/20210504001028id_/https://www.espn.com/fantasy/basketball/story/_/id/30296896/espn-fantasy-new-default-points-league-scoring-explained)

早期曾把 FantasyPros 平台列当作平均选秀顺位。后续发现其整数排序结构，但其是否代表实际平均选秀价格仍未核实，因此撤回相关价格换算结论。原报告与更正记录均已保留。

早期招生排名表解析器曾读取并丢弃附带的 NBA 生涯单元格。这些值未被导出、分析或用于训练。该事件仍保留为质量审核失败记录，后续改为只解析指定单元格，并加入回归检查。