Sector rotation multi-factor stock selection strategy based on six boards

: This paper proposes an investment strategy based on a regression linear model to predict sector rotation and a multi-factor scoring model to select stocks to achieve excess returns. The strategy uses January 2006 - December 2020 as the model construction period and January 2021 - June 2021 as the model validation and backtesting period. Using Shenyin & wanguo's industry classification as a standard, the cyclicality of industries is classified by β value and the existence of sector rotation is verified using the difference in rate of return between cyclical and non-cyclical industry. The factors associated with the annualized rate of return are selected and the fitted formula for the annualized rate of return is obtained by regressing linear equation and neural network multidimensional correlation to select the dominant board. Valid factors for each board were selected for fitting, and the factors were assigned weights for scoring, resulting in 32 stocks. The backtest results significantly outperformed the CSI 300 Index, and the model performance is stable and can achieve better returns.


Introduction
As the "barometer" of the stock market, stocks have the function of optimizing resource allocation and value discovery. Its return rate is affected by the market liquidity (Xie, C., et al., 2007), the "media effect" of financial news topics (Long, W., et al., 2019), the financialization level of enterprises (Xu, N.,et al., 2014) and other factors. However, its own profitability, creditworthiness, operating capacity, as well as the dividend payout status and growth prospects, are the most decisive factors. Over the years, in each round of a typical bear-bull market, there are always several industries or boards that take turns to lead the market and have a greater impact on other boards. The phenomenon of sector rotation in the stock market has been widely discussed. Domestic and international researchers have explored the law of sector rotation in terms of economic cycles, investment cycles, monetary cycles, capital expenditure and so on. Can sector rotation be combined with multiple factors affecting stock returns to accurately determine investment opportunities and capture strong rising sectors through sector return regression prediction models and multifactor scoring models to achieve excess returns? This paper analyses the phenomenon of sector rotation and innovatively blends the six boards of Cycle upstream, Cycle midstream, Cycle downstream, Macrofinance, Consumption, Growth with two data processing models, neural network and multiple linear regression, to select dominant boards and dominant stocks in that board. The stock selection model after two rounds of screening can better achieve the investment objectives and obtain higher returns. As the domestic stock market development system becomes more mature and market-oriented, sector allocation will gradually become the focus of investment guidance. At present, the existing literature on stock selection strategies is relatively single, with methods based on fundamental analysis and technical analysis, and a relatively small proportion of stock selection from a sectoral perspective. This paper expects a marginal contribution in the following two areas. First, summarize the law of development and change of various industries and make quantitative prediction, innovatively link the sector rotation with the factor scoring model that affects the stock yield. A multi-factor stock selection model based on six boards is constructed to find advantageous industries and advantageous stocks, which is not only a demonstration of sector rotation and an attempt at a stock selection strategy. Second, from the perspective of the practical significance, the quantitative investment scheme constructed in this paper, can effectively screen the industry stock pool, and make the forecast and stock selection within the advantage board, narrowing investors' investment scope and helping them to reduce investment risks and increase investment returns.

Sector rotation
The core of the sector rotation lies in to firmly hold the strong opportunity of each industry board, and make investment transformation on this basis to maximize the investment return and income. A review of domestic and international research on sector rotation mainly focuses on the explanation of the causes and the identification of the phenomenon. The existing literature has used multiple cycles as the background and framework for analysis of the causes of sector rotation. However, the economic cycle, as the overall reflection of economic dynamism and prosperity, is the basis of all cycles and contains the most comprehensive news of economic operation (Su, M. and Y. Lu, 2011). Therefore, the existing research have mainly used the economic cycle as the entry point for exploring the pattern of sector rotation. Peng, H. and X. Liu (2016) applied the correlation rule Aprior algorithm to explore the phenomenon of sector rotation, and believed that the emergence of this phenomenon comes from two aspects. Firstly, due to the inherent growth cycle of the industry itself, different industries have differences in demand income elasticity, demand price elasticity and cost composition. Secondly, it is due to the influence of the macroeconomic level, the implementation and adjustment of fiscal and monetary policies have brought different opportunities to various industries. Chen, M. and F. Cao (2005) found that the formation of sector rotation is related to the behavioral characteristics of investors in the stock market. Due to the incomplete rationality of investors, they are vulnerable to policy expectations, which will lead to the phenomenon of "rising and falling together" in a group of stocks. Regarding the identification of sector rotation phenomenon, He, C. (2001) used the relative return cr index to give the quantitative indicators to reflect the intensity of the sector rotation phenomenon. Shang, Y. and W. Xu (2020) used the two indicators of output gap and inflation to divide China's economic cycle from 2008 to 2018, and empirically find that different industries have different sensitivity to economic cycle fluctuations, completing the identification and verification of the rotation phenomenon of the industry. The strategy proposed in this paper is based on the law of sector rotation, which simply means that we hope to capture the better performing boards and eliminate the underperforming ones in the future, based on the "ebb and flow" of board. It is an investment strategy that takes advantage of the differential allocation of markets, complemented by the subjective initiative of investors.

Multi-factor stock selection
Multi-factor stock selection is a selection strategy with high popularity and application. The basic idea is to use the method of mathematical statistics to find out the indicators most closely related to the rate of return, construct a portfolio of stocks based on these indicators and judge the future returns of that portfolio against the broad market index. Compared to the single-factor model focusing on the relationship between the risk asset return and the market risk quantity (Sharpe, W., 1977), namely the capital asset pricing model (CAPM), the multi-factor model can test the correlation between different factors, assign different weights to the factors, eliminate redundant factors, and repeat the test method of the singlefactor combination, so as to obtain a better investment strategy than the single-factor strategy.
The key to the multi-factor stock selection model lies in the selection of factors and the model setting (Li, W. and J. Li, 2018). Partha (2005) used the top 20% of stocks in terms of PB ratio as the initial stock pool, and selected nine indicators from three dimensions of growth ability, profitability and robustness to score stocks, using the scoring results as the basis for portfolio construction. Du, H. and S. Ping (2005) selected data from eight points in time from mid-2000 to late-2003 to prove that the double low stock selection strategy combining low PS and low PE can obtain excess returns, which confirmed the combination of growth and value characteristics of the Chinese stock market. In terms of model setting, it is mainly divided into two categories: statistical models represented by Logistic regression, and machine learning models represented by random forest, neural network, and SVM (Zhang, H., et al., 2020). In the new economic situation, the complexity and uncertainty of the financial market are increasing, and the category and number of effective factors may also be replaced, which puts forward new requirements for factor screening and model construction. How to accurately identify the factors that generate excess returns and construct robust multi-factor models is an urgent problem to be solved (Wang, X., et al., 2023).

Industry board division
The starting point of the sector rotation must be the industry. This paper uses Shenyin & wanguo's industry classification and exclude the national defense industry. The β value can reflect the sensitivity of individual stocks to systemic risks. Referring to relevant literature, after calculating the β value of 27 first-level industries from January 2006 to December 2020 through Flush iFinD, the cyclicality of the industry is divided. In this paper, 1 is used as the threshold. If the β value exceeds the threshold, it indicates that the asset or portfolio is more sensitive to market changes. When the market changes, assets or portfolio will generate β times earnings or losses, identifying this type of industry as cyclical. On the contrary, if the β value is less than the threshold, it means that the asset or portfolio is not sensitive to the market fluctuations, and such industry is identified as a non-cyclical industry. In the cyclical industries, there are Cycle upstream, Cycle midstream and Cycle downstream based on industry chains. In the noncyclical industries, they are divided into Macrofinance, Consumption and Growth boards according to their economic characteristics.

Test of the existence and effectiveness of sector rotation
After the division of the industry, the paper introduces the key variable ∆Ri, which is the difference between the return rate of cyclical and non-cyclical boards. By obtaining the relevant data in Flush iFinD and processing it, the following conclusions are drawn: ∆ Ri has a uniform distribution of positive and negative values and expects to be close to 0, indicating that there is certain alternating fluctuation phenomenon between industries; the standard deviation of ∆Ri is large, indicating that the alternating fluctuation phenomenon is significant. The sector rotation phenomenon has been tested.

Factor screening
On the basis of establishing sector rotation, in order to select strong boards, this paper refers to existing research results and combines knowledge of finance and accounting, among 40 factors in the following seven dimensions, using data from January 2006 to December 2020, to determine the correlation level between each factor and the annualized return through SPSS, and finally identifies seven high correlation factors with a correlation of 0.7 or more.

Factor fitting
Linear regression equation is a statistical analysis method to determine the interdependent quantitative relationship between two or more variables. In this paper, the seven factors selected as indicators are fitted to annualized returns by normalizing the importance of the factors, and multiple linear regression equations are used to determine the return fitting formulae for each board to predict its development. Set the model as follows, where ε is the random error term. Z α α X α X . . . α X ε i=1,...,n Through SPSS analysis, the correlation level between each factor and the annualized return can be judged, and the formula related to the annualized return is preliminarily obtained:    Table 6, the two boards with the highest correlation with the Cycle downstream are the Cycle upstream and Consumption. Therefore, this paper will conduct stock selection analysis in 14 industries within these three boards.

Primary screening and elimination of factors
Screening 467 stocks in the three main boards of Cycle downstream, Cycle upstream and Consumption of CSI 300. The IC and IR analysis of the factors are conducted based on the performance of the 40 factors in the above seven dimensions from January 2021 to June 2021. The Pearson matrix is also used to eliminate redundant factors and 17 factors are finally selected.
ons for each stock factor scoring model  Table 6: Importance of the independent variables of the As can be seen from Table 9, the three stock selection models fit well and pass the tests for each indicator, identifying them as stock selection scoring formulas.  Table 8: Importance of independent variable of Table 9: Summary of three board models

Backtesting analysis
To verify the effectiveness of the strategy, this paper uses MindGo to backtest the returns of stock selection based on this strategy from January 2021 to June 2021. Compared with the returns of the CSI 300, the returns of the three boards show a significant improvement over the benchmark returns of the broader market, proving the effectiveness of the stock selection strategy.

Risk control
Conduct position management. The position is maintained at around 20%-80% under normal circumstances, using mainly stocks in the equity pool, leaving a certain amount of assets to wait for better investment opportunities. The portfolio is adjusted every 20 trading days and at the beginning of each position adjustment cycle, a composite score of stocks in the pool is calculated to select stocks based on the factor scoring approach described above. When a stock's score ranking is higher than the position adjustment ranking, the stock is sold and the top ranking stock is selected for purchase based on the desired position. When the stock in the position does not meet the sell condition, if the stock position is within the range of the individual stock position, it is marked as "Continue to hold"; if the stock position is greater than the maximum value of the individual stock position range, it is marked as "Reduce the position to the ideal position"; if the stock position is less than the minimum value of the individual stock position range and is selected by the stock selection strategy, it is marked as "Increase the position to the ideal position". At the start of the next 20 trading days, the portfolio will be reconstructed as described above and held for 20 trading days and so on until the end of the day. Extreme public opinion risk control. Public opinion risk refers to the negative information, false information and rumors that government departments and enterprises may face from the society or the network when engaged in social management and economic activities. When some extreme public opinion appears, the stock will be greatly affected, and the strategy constructed in this paper may may have errors as a result. The strategy will suspend its operation of buying and selling stocks for four days to deal with extreme public opinion risk when there is an abnormal short-term stock price fluctuation, enlarged volume and significantly negative excess return for the corresponding stock.

Conclusion
This paper combines the sector rotation phenomenon with multiple factors affecting the annualized return rate of stocks, and constructs the sector rotation multi-factor stock selection strategy based on six boards. The study results show that. First, sector rotation phenomenon exists, and confirms the linkage effect between the boards. Second, considering from the seven dimensions, like risk response ability, profitability and growth ability can more comprehensively select the factors that reflect the characteristics of the stock. Third, the performance of the stock selection strategy constructed in this paper throughout the backtesting period was significantly better than that of the CSI 300 in terms of win rate, return and annualized return, indicating that this strategy is conducive to leading investors to step into the rhythm of the market's sectoral ups and downs and select superior stocks from which to reduce investment risks and increase investment returns, thereby promoting the professionalization of investors' behaviour and the healthy development of China's capital economy market. In order to improve the effectiveness of this strategy, strengthening the dynamic position management and improving the ability of signal collection will be the research focus in the next stage.