Quantitative Models and Factor Construction Quantitative Models and Construction Methods - **Model Name**: Extreme Style Segmentation Model - **Model Construction Idea**: The model identifies extreme style periods by analyzing the excess returns of specific styles relative to the CSI All Share Index over rolling periods of 10 to 60 trading days. [7][8] - **Model Construction Process**: 1. Define extreme style periods as those where a specific style achieves an excess return of 15% or more relative to the CSI All Share Index. 2. For overlapping periods with the same start or end date, retain the period with the highest excess return. 3. For adjacent periods with overlapping dates, merge them by taking the earliest start date and the latest end date. [7][8] - **Model Evaluation**: The model effectively captures periods of extreme style dominance, providing insights into market behavior during such times. [7][8] - **Model Name**: Market Sentiment and Style Sentiment Indicators - **Model Construction Idea**: This model uses market-wide and style-specific sentiment indicators to predict the end of extreme style periods. [13][14] - **Model Construction Process**: 1. Market sentiment is measured using turnover rate, trading volume, and the William's variable dispersion index. 2. Style sentiment is assessed using metrics such as return dispersion, crowding (measured by price deviation weighted by trading volume), and turnover rate of free-float market capitalization. 3. Compare sentiment changes at the start and end of extreme style periods to identify patterns. [13][14][17] - **Model Evaluation**: The model provides a robust framework for understanding the role of sentiment in driving and ending extreme style periods. [13][14] Model Backtesting Results - **Extreme Style Segmentation Model**: - Example: Growth style (represented by the ChiNext Index) achieved an excess return of 102.71% over 315 days from November 28, 2012, to October 9, 2013. [10] - Example: TMT style (CITIC) achieved an excess return of 30.56% over 111 days from November 6, 2019, to February 25, 2020. [10] - **Market Sentiment and Style Sentiment Indicators**: - Example: During the extreme growth style period from November 18, 2019, to February 21, 2020, the William's variable dispersion index rose from 0.346 to 0.924, indicating heightened market activity. [20][21] - Example: During the extreme TMT style period from December 28, 2022, to April 12, 2023, style crowding reached a historical high before declining below 90%, signaling the end of the extreme period. [35][40] --- Quantitative Factors and Construction Methods - **Factor Name**: Style Crowding Factor - **Factor Construction Idea**: Measures the degree of crowding within a style based on price deviation, trading volume, and other metrics. [28][29] - **Factor Construction Process**: 1. Calculate price deviation as the weighted average distance of component stock prices from their 30-day moving averages. 2. Compute the proportion of trading volume during uptrends over the past 40 days. 3. Combine these metrics with equal weights to form a composite crowding factor. [28][29] - **Factor Evaluation**: The factor effectively identifies periods of excessive concentration within a style, which often precede reversals. [28][29] - **Factor Name**: Return Dispersion Factor - **Factor Construction Idea**: Captures the degree of return variability among component stocks within a style. [28][29] - **Factor Construction Process**: 1. Calculate the standard deviation of returns for component stocks over the past 40 trading days. 2. Normalize the dispersion values to percentile ranks over the past six months. [28][29] - **Factor Evaluation**: High return dispersion often signals the end of extreme style periods, as observed in historical data. [28][29] Factor Backtesting Results - **Style Crowding Factor**: - Example: During the extreme TMT style period from January 6, 2025, to February 21, 2025, the crowding factor peaked at 0.85 before declining below 0.90, signaling the end of the period. [46][47] - **Return Dispersion Factor**: - Example: During the extreme growth style period from March 15, 2021, to August 4, 2021, return dispersion reached the 99.6th percentile, indicating heightened variability among component stocks. [30][34] --- Summary of Key Insights - Extreme style periods are identified using excess returns relative to the CSI All Share Index, with additional insights provided by sentiment and crowding indicators. [7][8][13] - Backtesting results highlight the effectiveness of these models and factors in capturing market dynamics during extreme style periods. [10][20][35] - Style crowding and return dispersion factors are particularly useful in predicting the end of extreme style periods, as evidenced by historical data. [28][30][46]