Core Insights - The report emphasizes the quantitative practices in the bond ETF era, highlighting the use of redemption lists to enhance transparency and the application of machine learning to predict which bonds are more likely to be purchased by ETFs [1][2] Group 1: Enhancing Transparency through Redemption Lists - Bond ETFs are not purely passively managed; they involve significant active sampling and timing operations. Typically, 90% of the portfolio comes from index and its alternative component bonds, with the remaining 10% sourced from non-component bonds to meet tracking error requirements [12][13] - The number of bonds in the credit bond ETF's benchmark market-making list exceeds 200, while the ETF holdings and redemption lists contain fewer than 200 and 70 bonds, respectively. This indicates a multi-layered selection process for bonds [12][13] - The rapid growth of bond ETF sizes can lead to liquidity disruptions in the primary market, affecting the cost of acquiring replacement bonds [15][24] Group 2: Machine Learning Predictions for Bond Selection - The expansion of bond ETF component bonds follows specific patterns, with a focus on selecting similar attribute bonds from the alternative pool. The redemption mechanism requires ETFs to disclose component bonds and ensure their liquidity [27][28] - The report notes that the weekly influx of new bonds into credit bond ETFs can range from a few dozen to nearly a hundred, reflecting the dynamic nature of the market [28][31] - A predictive model using LightGBM was developed to assess the likelihood of each bond being included in an ETF, with the model showing a correct prediction rate of 45% to 50% for the top 20 predicted bonds [36][38]