This repository implements an empirical asset pricing study on asymmetric positive feedback trading (AFT) in the Chinese equity market, with a particular focus on small-cap stocks. The project constructs a novel AFT measure based on rolling regressions with realized volatility and incorporates it into a factor-based trading strategy.
The empirical analysis follows a standard research pipeline in asset pricing: data construction → factor estimation → portfolio sorting → statistical testing → strategy backtesting.
Positive feedback trading—buying after price increases and selling after price decreases—has been widely documented in financial markets. However, investor reactions to positive and negative past returns may be asymmetric, especially under varying volatility conditions.
This project aims to:
- Quantify asymmetric feedback trading intensity (AFT) at the stock level
- Examine whether AFT behaves like a priced characteristic
- Test whether AFT can be exploited in a predictive trading strategy
- Stock universe: CSI 500 constituents listed before 2003
- Frequency: Daily (returns, volatility), aggregated to monthly where appropriate
- Key variables:
- Daily stock returns
- Realized volatility (RV)
- Market capitalization (Size)
- Price-to-earnings ratio (Value)
- Amihud illiquidity measure (Liquidity)
All data are sourced from CSMAR and official index files.
For each stock, AFT is estimated using rolling regressions of the form:
[ r_t = \beta_0 + \beta_1 r_{t-1} + \beta_2 r_{t-1} \mathbb{I}(r_{t-1} > 0)
- \beta_3 RV_t + \beta_4 r_{t-1} \times RV_t
- \beta_5 r_{t-1} \mathbb{I}(r_{t-1} > 0) \times RV_t + \varepsilon_t ]
The coefficient (\beta_5) captures asymmetric feedback trading intensity.
Rolling windows of 12 months are used to obtain time-varying AFT measures.
Stocks are sorted monthly into quintiles based on:
- Market capitalization (Size)
- Price-to-earnings ratio (Value)
- Amihud illiquidity (Liquidity)
- Asymmetric feedback trading (AFT)
Value-weighted portfolio returns are computed, and factor spreads (High–Low or Low–High) are constructed.
Factor returns are evaluated using Newey–West adjusted t-tests to account for autocorrelation and heteroskedasticity.
A predictive trading strategy is implemented using:
- Rolling estimation of factor loadings
- Weekly return prediction
- Stock selection based on predicted returns
- Value-weighted portfolio construction
- Performance evaluation using cumulative returns, Sharpe ratio, and maximum drawdown
src/
├── 01_data_loading.R
├── 02_data_preprocessing.R
├── 03_volatility_measures.R
├── 04_AFT_estimation.R
├── 05_factor_construction.R
├── 06_factor_tests.R
├── 07_strategy_backtest.R