本研究檢驗交易者持倉、投資人情緒、金融風險、能源庫存意外 (inventory surprise)、與貨幣供給成長率是否對於能源期貨的報酬率具有預測能力，並檢驗能源期貨市場的聰明錢假說。研究標的包括輕質原油、RBOB 汽油、天然氣及 ULSD 柴油與取暖油等商品的期貨價格，資料期間為 2011 年 1 月 1 日至 2026 年 4 月 30 日。由於 COT 資料為星期二持倉統計值，因此本文以週二至週三的日報酬率為預測目標，其餘變數則為星期二以前之最近期資料；因此本文分析 COT 持倉部位與次日期貨報酬率之間的領先資訊關係。本文的分析模型控制庫存意外、M2 成長率、經濟政策不確定性、股票與原油價格的隱含波動、金融壓力、散戶情緒及消費者信心。持倉資訊分別以 COT 淨部位比例（NetOI）、Wang 式部位情緒指標（WangSI）及部位變化（Flow）衡量，並比較五類交易者的差異。 本文發現，M2 成長率在多數基準模型中對於短期報酬率呈顯著負向影響，OVIX 則呈顯著正向影響；其餘情緒與風險指標未呈現一致的預測效果。在星期二持倉統計值方面，持倉部位變化（Flow）指標相對於 NetOI 與 WangSI 提供較多短期資訊；其中，生產者&商業避險者的持倉部位變化呈顯著正向影響，非可報告交易者的持倉部位變化呈顯著負向影響。GARCH-in-Mean 結果顯示，部分 COT 訊號與報酬或條件變異數具有顯著關係；但兩項條件均未在同一商品與同一訊號中同時成立，顯示五類交易者均不具有聰明錢效應。;This thesis examines whether trader positions， investor sentiment， financial risk， energy inventory surprise， and money growth are related to Tuesday-to-Wednesday returns in four NYMEX energy futures: Light Crude Oil Continuous， RBOB gasoline futures， natural gas futures， and ULSD diesel/heating-oil futures. The nominal sample period runs from January 1， 2011 to April 30， 2026. Except for the COT variables， which are measured using positions dated Tuesday， predictors are aligned to information available by Tuesday. Because Tuesday COT positions are normally released on Friday， the main COT estimates represent lead-lag information rather than a publicly executable Tuesday strategy. The models control for inventory surprise， M2 growth， economic policy uncertainty， equity- and oil-market implied volatility， financial stress， retail sentiment， and consumer confidence. NetOI， WangSI， and position change (Flow) are compared across five trader groups using fixed-effects panel regressions， commodity-specific regressions， and GARCH-in-Mean models. The results show that M2 growth is significantly negative in most baseline specifications， whereas OVIX is significantly positive; the remaining sentiment and risk indicators do not exhibit consistent effects. In the main analysis based on Tuesday position dates， Flow contains relatively more short-horizon information than NetOI and WangSI. Producer/merchant hedger Flow is significantly positive， while nonreportable Flow is significantly negative. Commodity-specific estimates indicate that COT relations vary across commodities， trader groups， and position measures. The GARCH-in-Mean results identify significant return or conditional-variance relations for some COT signals， but the two smart-money conditions never occur for the same signal in the same commodity. Therefore， none of the five trader groups supports the smart-money hypothesis.
