Analysis of stock selection strategy of multi-factorial model based on rotation factor and investor sentiment.

. With the wide application of advanced technologies including AI which with high precision and automation level, financial technology has shown a progressive trend, and quantitative trading strategies have bright prospect. Based on China's A-share market, this study constructs a quantitative strategy for stock selection by combining the industry rotation effect with a multi-factor model considering investor sentiment. Firstly, the strategy screens the target industries according to the rotation factor calculated by PER, and then determine the stocks by the multi-factor model formed by the factors after effective test. In this study, the effectiveness of the strategy is verified through six constructed back-test portfolios from both sides. According to the analysis, the return of the strategy considering the rotation effect and investor sentiment is higher than the market. The comparison of the impact of the rotation factor and the emotional factor shows that the change of macro and middle environment has a greater impact on stock prices than the micro level. Up to now, there are few stock selection strategies considering both rotation effect and investor sentiment. These results attempt to supplement this aspect and enrich the field of asset pricing.


Introduction
The concept of quantitative investment has a history of more than 30 years since it was put forward around 1900.Based on model construction, data screening and calculation, quantitative investment which relies on real data are more disciplined than traditional investment methods.On this basis, quantitative investment better overcomes the shortcomings of human nature, such as greed and fluke, and can also track market development timelier to grasp more opportunities.Quantitative investment capture financial products that are likely to profit by excavating laws from historical data continuously and combines with portfolio concepts to diversify investment risks.In addition, the rapid development of computer technology has been contributing the prosperity of quantitative investment, helping investors observe and analyse multi-dimensional massive data more efficiently and systematically as well as increasing the perspective of investment.Although quantitative investment in China started later, with the continuous updating and iteration of AI, big data, cloud computing and machine learning as well as the increasing popularity and recognition of quantitative trading strategies, quantitative investment is bound to become a trend and have broad development space.Referring to the previous literature, it is found that there are few studies on multi-factor models considering both industry rotation and investor sentiment.The model in this study is based on China's A-share market, adding mean reversion and irrational behaviour of investors, trying to seek a balance between law and anomaly, attempting to improve the explanation potency of the model and enrich the content of asset pricing.
The researches on industry rotation effect can be mainly divided into three aspects, namely whether it exists, what causes it, and how to apply it.Regarding the existence, Sorensen, Burke, and O'Neal constructed rotational strategies to demonstrate its effectiveness [1,2].Weiss proposed that there are two types of rotations in different industries, i.e., global and local [3].He confirmed the existence of the phenomenon of the Chinese stock market sector [4].Turning to the causes, most scholars have studied the correlation between economic cycles and industry rotation.Estrella and Chavuet have proved a close relationship between the stock market and economic cycles [5,6].After Merrill Lynch proposed the Merrill Lynch Clock Theory in 2004, it sparked numerous scholars and institutions to intend to use the investment clock theory to study industry investment portfolios, which thus derived the industry rotation theory.Zhang believes that industrial characteristics and correlations are the foundation that industry sector rotation can be achieved, while changes in investor expectations are the main influencing factors for industry sector rotation [7].In terms of the application, Sarwar studied the risk adjusted performance of US sector investment portfolios by using Alpha from the Fama French five factor model and sector rotation strategies [8].Zhao creates an industry rotation multi factor stock selection model by combining industry rotation strategy with a multi factor stock selection model to [9].Liu coined a rotation factor based on the mean regression of relative P/E ratio and brought out a stock selection strategy together with three indicators, i.e., book to market ratio, year-on-year growth rate of return on equity, and asset liability ratio [10].
Based on the complex of emotions, which are difficult to define, have a wide range of impact sources, and are abstract and hard to quantify, scholars have proposed numerous indicators to depict investors' emotions.These indicators include subjective indicators that directly communicate with investors through surveys (e.g., questionnaires, phone calls, and letters) to obtain emotional information, as well as market trading data such as trading volume, the first-day return on IPO and trading volume, which play a role in objective indicators of proxy investor sentiment [11][12][13].
For the researches of multi-factor model, after Ross got inspiration from Sharpe's CAPM model and proposed the famous APT model, Fama and French proposed the Fama-French three-factor model, establishing a milestone in the research process of multi-factor stock selection model [14,15].Piotroski used a multi-factor stock selection model by electing 9 financial indicators as judgment criteria and ranked them according to the scoring method to screen out the top-rated stocks, which turned out to be profitable [16].Fama and French found that profit and investment factors play an important role, thus they added them to the FF3 model, forming a more explanatory five-factor pricing model [17].Zhang et al. added machine learning algorithms to factor selection and prediction of stock price trend, including Elastic Net, Random Forest, GBDT and other models to screen factors, which obtained higher returns and lower risks compared with CSI 300 Index indexes [18].
This study constructs of a rotation factor based on relative PER to determine whether the industry is in a state of stagflation or hyperinflation.Since the stagflation industry will follow the law of mean regression, with a probability that stock prices will rise in the future, thus narrowing the stock pool by choosing the stagflation industry.Subsequently, candidate indicators are selected from five major categories of factors: valuation factors, growth factors, capital structure, technology factors, and sentiment factors.In this way, multi-factor models can be produced by synthesized indicators after effectiveness experience, to construct a scoring-based stock selection strategy.Based on rotational factors, this study not only considers valuation, growth, and capital structure, but also incorporates technical and emotional factors.Each major factor category has multiple indicators, forming a two round stock selection strategy.Moreover, this study constructed 6 back testing combinations for comparison, and tested the effectiveness of a multi-factor model incorporating rotational and emotional factors from both positive and negative perspectives.

Industry rotation and mean regression
Industry rotation refers to the phenomenon where stocks in the same industry rise and fall in a relatively short period of time, while stocks in different industries fluctuate in the long run.Mean reversion refers to the trend that social phenomenon such as stock prices and house prices, and natural phenomena, whether higher or lower than the value centre (or mean value), will return to the value centre with a high probability.Thaler and DeBondt found that stocks with a cycle of 3 to 5 years that is above or below the market average price will experience a reversal in performance [19].This means that after the stock price moves in one extreme direction, there is a phenomenon of mean regression.Furthermore, after returning to the mean, there is a tendency to move in the other extreme direction.The earliest research in China was conducted by Huang et al., who found that the Chinese stock market has a pulling force to pull back the average of excessively high or excessively low PER [20].Based on the above, the author believes that there is a close relationship between industry fluctuations and mean reversion, or to some extent, industry fluctuations are the process of mean reversion.

Multi-factorial stock selection model
For the stock selection strategy, the scoring method will not be affected by the data with large fluctuations, which will make the model relatively more stable.The basic idea of the scoring method is to arrange the order according to the size of factor values, assign different weights to each factor after arranging the order, and finally screen stocks based on the obtained weighted scores to construct the desired investment portfolio.This method is chosen in this study.However, this method also has shortcomings, such as causing a certain degree of subjectivity in the construction process of the investment portfolio.
The main principle of the Law of Return is to test the past return rate of the selected stock on each factor, obtain the equation after regression and the coefficient of each factor after regression, and then put the factor value of the next period into the equation obtained above, so that the return rate of the stock can be measured, and then select the stock to invest according to the predicted return rate.However, the disadvantage of the Law of Return is that the results are easily affected by data volatility.

Factors
Considering the dynamic changes in the macro environment and its impact on the industry, a relative PER is adopted to construct a rotational factor.The PER data are from 30 CITIC first tier industry indices, including the PER (TTM) and the Wind All A PER (TTM), sourced from the Wind database.The time interval is between January 1, 2011and December 31, 2019.
Here,   is the relative price to earnings ratio of the ith industry;   is the price to earnings ratio of the i-th industry; _ is the price to earnings ratio of all Ashares.The expressions are given as follows: Here,   is the minimum historical   ;   is the median historical   ;   is the maxi mum   ;   is t he quan tile valu e of  of the i-t h industr y, which is the rotation factor.One defines the median historical RPE i as a 50% percentile level, the lowest historical RPE i as a 0% percentile level, and the highest historical RPE i as a 100% percentile level.The quantile level of other RPE i is obtained by linear interpolation.If the RPE i is higher than the historical maximum, the quantile value is 1; if it is lower than the historical minimum, the quantile value is 0.

Selection candidate factors
As shown in (8) MktVal i is the total market value of stock i.Industry j,i is an industry dummy variable.If stock i belongs to industry j, the industry factor exposure is 1, otherwise it is 0. ε i is the factor value of market value divided by industry neutrality.Then IC analysis will be carried out.The three indicators with the highest IC mean are selected for each major category of factor.Afterwards, correlation analysis will be carried out.Perform correlation testing on the indicators that have undergone IC testing.Excluding indicators with absolute correlation values>0.75In order to further test the effectiveness of emotional factor, the indicators were divided into two groups: group 1 did not contain emotional factor, and group 2 contained emotional factor.Finally, one synthesises of effective factors.

Construction of backtesting combinations
In order to test the effectiveness of rotational factors and emotional factors, the following six backtesting combinations were constructed and compared with CSI 300 Index benchmark.The backtesting period is from January 2020 to December 2022.Backtesting combinations are given as follows.
• Combination 1: multiple-factor (Group 1) model with winner portfolio investment strategy; • Combination 2: multiple-factor (Group 2) model with winner portfolio investment strategy and the rotating factor; • Combination 3: multiple-factor (Group 1) model with loser portfolio investment strategy; • Combination 4: multiple-factor (Group 2) model with loser portfolio investment strategy; • Combination 5: multiple-factor (Group 1) model with loser portfolio investment strategy and the rotating factor; • Combination 6: multiple-factor (Group 2) model with loser portfolio investment strategy and the rotating factor; Before selecting stocks for each combination, first remove some obvious "junk stocks", and use the PER as the standard to exclude stocks with PER less than 0 or greater than 100 and stocks with ST markings.When selecting stocks, the equal weight scoring method is adopted, where the value of positive factors is positively proportional to their score, and the value of negative factors is negatively proportional to their score.One adds the scores obtained by all stocks on each factor equally to obtain the final total score for each stock.
In the winning combination, buy the 50 stocks with the highest score and sell the stocks that fell from the top 50, while in the losing combination, the opposite is true.Combination 2, 5, and 6 introduce a rotation factor, which selects stocks based on selecting 10 industries with smaller rotation factor values.

Results and discussion
The results of IC analyses, candidate factors after IC analysis and correlation analysis are shown in Table .2, Table .3 and Table .4, respectively.The results of correlation analysis are presented in Fig. 1.The results of correlation analysis are presented in Fig. 2, Fig. 3, Fig. 4, Fig. 5, Fig. 6 and Fig. 7, respectively.As shown in Table .5-6, results of back testing combinations are as follows.Based on the back testing results of the above winning combinations, it can be analysed.Firstly, all winner combinations outperform the market.Secondly, Combination 1, the annualized excess return of the multi factor stock selection model, which does not include rotational and emotional factors, reached 8.579%.Combination 2, the multi factor stock selection model after adding the rotation factor and emotion factor, increased the annualized excess return rate to 14.406%, nearly twice, and improved the Sharpe ratio, the ratio of Reno, the information ratio and other indicators.Based on the back testing results of the above loser combinations, it can be analysed that.Firstly, all four loser combinations underperformed the market, with the multifactor model that only added emotional factors having the lowest returns.Secondly, the addition of rotation factors to combinations 5 and 6 resulted in improvements compared to combinations 3 and 4.Among them, the improvement of combination 5 is significant, increasing the annualized return from -13.898% in combination three to -10.199%, and reducing losses by over 25%, indicating that the rotation factor also has an optimization effect on loser portfolios.Compared to Combination 3, the addition of emotional factors in Combination 4 resulted in a decrease of nearly 2.8% in annualized returns.In Combination 6, compared to Combination 3, the addition of rotational factors based on emotional factors resulted in an increase of 3.337% in annualized returns, indicating that the impact of industry rotation on stock prices is greater than the impact of investor sentiment on stock prices.

Conclusion
In summary, this study first takes 30 CITIC primary industries as the research object, constructs the rotation factor as an indicator based on the industry rotation mean regression theory, and combines it with a multi-factor model incorporating emotional factors to construct six multi-factor stock selection strategies.Through the comparison and analysis of the back testing results, the following conclusions can be drawn.
Firstly, the addition of rotational and emotional factors to the multi-factor stock selection model can improve investment returns.Besides, the rotation factor still has an optimization effect on the investment strategy of loser portfolios.Secondly, the impact of industry rotation on stock prices is greater than the impact of investor sentiment on stock prices, indicating that changes in the macro and micro environment have a greater impact on stock price fluctuations than at the micro level.Overall, there are industry rotation and mean regression effects in China's stock market.Investors can first select industries, then combine multiple-factor models to select stocks, and finally obtain returns from the ultra large market.There are also shortcomings in this study.When considering industry rotation and mean regression effects, the construction of the rotation factor is relatively simple, and only the P/E ratio is considered to construct the valuation interval.When constructing a scoring and stock selection strategy, for stability reasons, equal weight scoring was selected without considering the differences in actual weights of each factor.In the future, more indicators reflecting industry rotation and mean regression effects can be added, as well as optimization of stock scoring and screening methods.
Table.1, this study selects indicators from five categories of factors, namely, valuation factor, growth factor, capital structure factor, technical factor, and emotional factor.The data time interval is selected as the monthly data from January 2011 to December 2019, which is sourced from Rice Quant.

Table 2 .
IC analysis of candidate factors.

Table 3 .
Candidate factors after IC analysis.

Table 4 .
Candidate factors after correlation analysis.

Table 5 .
indexes of winner combinations.

Table 6 .
indexes of loser combinations.