Stock Price Prediction Using Statistical and Machine Learning Techniques (Part II)

Although decision trees are among the most interpretable models, they exhibit highly variable behavior. Consider a single training dataset randomly split into two parts. Each part trains a decision tree, yielding two distinct models. When both models are fitted, they produce different outputs. Due to this behavior, decision trees are associated with high variance. Ensemble methods such as Bagging or Boosting help reduce the variance of any learner. Several decision trees constructed in parallel form the base learners of the Bagging technique. These learners are trained on bootstrapped (i.e., sampled with replacement) data. The final prediction output is the average of all learners’ outputs. 3.3.2 Boosting In Boosting, all trees are built sequentially so that each subsequent tree reduces the errors made by the previous one. Each tree learns from its predecessor and updates the errors; thus, all subsequent trees learn from updated error versions. Base learners with high bias—termed weak learners—possess predictive power only slightly better than random guessing. Each weak learner contributes meaningful information toward prediction, and combining many such weak learners yields a strong learner. The final strong learner exhibits both low bias and low variance. In contrast to Bagging techniques like Random Forest, Boosting models such as XGBoost employ trees with relatively few splits. Such shallow trees are highly interpretable due to their minimal depth. Key parameters—including the number of iterations (or trees), the learning rate for gradient boosting, and tree depth—can be optimized using validation techniques such as k-fold cross-validation. Overfitting arises when too many trees are used; therefore, careful selection of the stopping criterion for boosting is essential. 3.4 Hybrid Model The Hybrid Model integrates the statistical Kalman filter with the machine learning model XGBoost. This hybrid approach improves accuracy and consistency in stock price prediction across diverse datasets. When the variable of interest cannot be measured directly, Kalman filters optimally estimate it using available indirect measurements. For such indirect measurements, 14 parameters can be employed. Through feature extraction, the XGBoost algorithm selects the top five parameters—ranked by weight—among these 14. The top five parameters vary by stock. These five parameters are fed into the Kalman filter to forecast future stock values and generate predictions. Each model was tested on a test set containing 400 days of stock data, with predictions required for several days ahead. Testing was conducted using two distinct datasets: one comprising price movements only, and another incorporating both price movements and market sentiment. Price movements include open price, high price, low price, closing price, trading volume, dividend yield, and volume-related metrics. Market sentiment refers to investors’ collective psychological attitude toward a specific security.

Breaking the Information Cocoon: U.S. Equity Investment Finance Series (I)

• Berkshire Hathaway Annual Letters to Shareholders (full original texts from 1977 to present) — https://www.berkshirehathaway.com/letters/letters.html • Oaktree Capital / Howard Marks’ Memos — https://www.oaktreecapital.com/insights/memos • Bridgewater Associates / Ray Dalio’s Principles and Observations — https://www.bridgewater.com/research-and-insights Sell-Side Research Aggregators (free access to abstracts and partial full texts): • StockAnalysis.com (company financials + analyst rating summaries) — https://stockanalysis.com • TradingView Research — https://www.tradingview.com/research Macroeconomic and Financial Data: • FRED (Federal Reserve Bank of St. Louis — vast macroeconomic database) — https://fred.stlouisfed.org • BLS (U.S. Bureau of Labor Statistics) — https://www.bls.gov • BEA (U.S. Bureau of Economic Analysis) — https://www.bea.gov • CME (Chicago Mercantile Exchange) — CME Group’s blog and investment education resources Financial Media (practice-oriented): • Financial Times — https://www.ft.com • Wall Street Journal — https://www.wsj.com • Barron’s — https://www.barrons.com • The Economist — https://www.economist.com • Caixin (in-depth Chinese-language financial investigations) — https://www.caixin.com

Calumet, Inc. (NASDAQ: CLMT) Financial Estimates

Oct 24, 2026

TradingView displays financial estimates and analyst data for Calumet Specialty Products Partners, an energy minerals company listed on NASDAQ as CLMT.

TradingView

2026-03-30

Adams Asset Advisors LLC Holds $102.46M Stake in Calumet, Inc.

2026-03-30

Calumet (NASDAQ:CLMT) Given New $60.00 Price Target at HC Wainwright

2026-03-31

H.C. Wainwright Raises Calumet Stock Price Target to $60 from $33

2026-04-02

Calumet (NASDAQ:CLMT) Director Daniel Sajkowski Sells 3,310 Shares

2026-04-05

Calumet, Inc. Receives Consensus 'Hold' Rating from Brokerages