CN116823487A - ESG evaluation system investment decision-making system - Google Patents

ESG evaluation system investment decision-making system Download PDF

Info

Publication number
CN116823487A
CN116823487A CN202310769091.4A CN202310769091A CN116823487A CN 116823487 A CN116823487 A CN 116823487A CN 202310769091 A CN202310769091 A CN 202310769091A CN 116823487 A CN116823487 A CN 116823487A
Authority
CN
China
Prior art keywords
esg
investment
year
feature
strategy
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310769091.4A
Other languages
Chinese (zh)
Inventor
易蓉
周煜寰
宋靖
叶子薇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangxi University of Finance and Economics
Original Assignee
Jiangxi University of Finance and Economics
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jiangxi University of Finance and Economics filed Critical Jiangxi University of Finance and Economics
Priority to CN202310769091.4A priority Critical patent/CN116823487A/en
Priority to LU504769A priority patent/LU504769B1/en
Publication of CN116823487A publication Critical patent/CN116823487A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/27—Regression, e.g. linear or logistic regression
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G06N5/045—Explanation of inference; Explainable artificial intelligence [XAI]; Interpretable artificial intelligence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0282—Rating or review of business operators or products
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/06—Asset management; Financial planning or analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Development Economics (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Economics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Game Theory and Decision Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Tourism & Hospitality (AREA)
  • Computational Linguistics (AREA)
  • Human Resources & Organizations (AREA)
  • Medical Informatics (AREA)
  • Health & Medical Sciences (AREA)
  • Educational Administration (AREA)
  • Technology Law (AREA)
  • Operations Research (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Train Traffic Observation, Control, And Security (AREA)
  • Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
  • Warehouses Or Storage Devices (AREA)

Abstract

The application discloses an ESG evaluation system investment decision system, which belongs to the financial field, and evaluates ESG scores of enterprises from the aspects of environment, society and management, and brings ESG factors into an investment model; meanwhile, by adopting the introduction of an AI algorithm and XAI interpretation and adopting a nonlinear non-preset self-adaptive algorithm for evaluation, the existing linear weight evaluation method is overcome, and an investor can be helped to more accurately identify ESG companies which consider financial and social benefits; testing the optimal threshold position of investment screening by a data inflection point method, and optimizing the configuration of investment combinations; the ESG investment strategy is developed, so that investors gradually turn to comprehensively consider the environment, society and management performance of enterprises from the traditional financial indexes. The investment mode is more focused on sustainable development of enterprises, so that not only can return on investment be obtained, but also the sustainable development of society can be promoted, and the aim of considering financial and social benefits is fulfilled.

Description

ESG evaluation system investment decision-making system
Technical Field
The application relates to the financial field, in particular to an ESG evaluation system investment decision system.
Background
As the concept of ESG investment has spread, more and more mainstream investment institutions began to pay attention to and incorporate the ESG factors into the process of decision making, investment and evaluation, since the 21 st century. The guiding role of the ESG concept in terms of investment is gradually increasing, which many enterprises and even government agencies take as important principles for evaluating investment risk. Through the establishment and popularization of ESG indexes, the contribution of enterprises in the aspects of promoting economic sustainable development and fulfilling social responsibility can be effectively evaluated, and important references are provided for investment and financing; and the method can also promote enterprises to achieve maximization of own benefits from single pursuit to maximization of social value.
The traditional ESG scoring system is used for reference; the ESG score is typically evaluated and calculated by a third party entity and the scoring system flow includes: 1. collecting data: the evaluation entity may collect data and information regarding the performance of the enterprise ESG from a number of sources including official documents, media reports, information disclosed by the enterprise disclosure, and the like. 2. Screening indexes: the evaluation facility may screen out key metrics related to the ESG performance of the enterprise, such as carbon emissions, employee welfare, board management, etc., from the large amount of data collected. 3. Weight index: the assessment mechanism can distribute weights of different indexes according to factors such as importance degrees of the indexes, industry characteristics and the like. 4. Calculating the score: the evaluation mechanism calculates corresponding scores according to the performance of the enterprise on each index, and integrates the scores to obtain a final ESG score. As can be seen, the conventional ESG scoring system generally adopts the delfei method, which is highly subjective, and has questionable and improved effect on ESG investment guidance.
Therefore, the application provides an ESG evaluation system investment decision system.
Disclosure of Invention
1. Technical problem to be solved
The application aims to provide an ESG evaluation method based on AI, XAI and Knee and an investment decision system, which evaluate ESG scores of enterprises from the aspects of environment, society and management. Incorporating the ESG factors into an investment model; meanwhile, by adopting the introduction of an AI algorithm and XAI interpretation and adopting a nonlinear non-preset self-adaptive algorithm for evaluation, the existing linear weight evaluation method is overcome, and an investor can be helped to more accurately identify ESG companies which consider financial and social benefits; testing the optimal threshold position of investment screening by a data inflection point method, and optimizing the configuration of investment combinations; the ESG investment strategy is developed, so that investors gradually turn to comprehensively consider the environment, society and management performance of enterprises from the traditional financial indexes. The investment mode is more focused on sustainable development of enterprises, so that not only can return on investment be obtained, but also the sustainable development of society can be promoted, and the aim of considering financial and social benefits is fulfilled.
2. Technical proposal
In order to solve the problems, the application adopts the following technical scheme.
An ESG evaluation system investment decision system comprising the steps of:
s1, collecting the existing ESG scoring index system and historical data;
s2, processing a missing value;
s3, ESG investment score analysis;
s4, obtaining the score of each feature;
s5, adopting Knee inflection point analysis;
s6, constructing an ESG factor investment strategy;
s7, strategy comparison;
in the step S1, the collected data sources are from existing domestic and foreign scoring institutions, and the listed companies are main data sources and analysis objects.
Preferably, the step S2 is to perform missing value processing by adopting a zero padding mode, and the target variables of the ESG score index system are respectively set to be the current year rate of return, the next year rate of return and the third year rate of return; to investigate whether the ESG factor has an immediate effect or a delayed effect on the investment value of the enterprise.
Preferably, the step S3 adopts an XGBoost algorithm in an AI algorithm to analyze ESG investment scores, input variables are secondary indexes and tertiary indexes respectively, and target variables are stock annual yields of marketing companies in the current year, the next year and the third year respectively. And (5) carrying out an AI algorithm to detect the contribution degree of each ESG factor to the investment yield.
Preferably, the S4 obtains the score of each feature by calling the feature_importance_attribute of the XAI, and the higher the score is, the greater the contribution of the feature to the output is, which can be used for screening out the most important feature or performing operations such as feature dimension reduction. And visualizing the importance scores of the features using a plot_importance function.
Preferably, the S5 uses Knee inflection analysis to analyze trend changes based on the ESG investment prediction sequence, identify important locations for data pattern conversion, and explore maximization of the balance gain between ESG responsibility and investment gain.
Preferably, the method comprises: and S6, constructing an ESG factor investment strategy: considering that part of stocks in China limit the qualification of the coupon, the strategy does not consider the emptying mechanism, the predicted value is higher than the Knee inflection point to buy, the stock is not opened and lower than the Knee inflection point value, the fund weight is equal, the bin rotation period is one year, and the bin is replaced in the beginning of each year.
Preferably, the step S7 performs a policy comparison, and performs a performance comparison with a broad base standard policy of the hun-deep 300 index.
3. Advantageous effects
Compared with the prior art, the application has the advantages that:
the application evaluates the ESG score of enterprises from the aspects of environment, society and management, and brings ESG factors into an investment model; meanwhile, by adopting the introduction of an AI algorithm and XAI interpretation and adopting a nonlinear non-preset self-adaptive algorithm for evaluation, the existing linear weight evaluation method is overcome, and an investor can be helped to more accurately identify ESG companies which consider financial and social benefits; testing the optimal threshold position of investment screening by a data inflection point method, and optimizing the configuration of investment combinations; the ESG investment strategy is developed, so that investors gradually turn to comprehensively consider the environment, society and management performance of enterprises from the traditional financial indexes. The investment mode is more focused on sustainable development of enterprises, so that not only can return on investment be obtained, but also the sustainable development of society can be promoted, and the aim of considering financial and social benefits is fulfilled.
Drawings
FIG. 1 is a diagram of pre-ESG profitability data in accordance with the present application;
FIG. 2 is a diagram of the yield data after ESG operation according to the present application;
FIG. 3 is a current annual rate corner map of the present application;
FIG. 4 is a next year yield corner plot of the present application;
FIG. 5 is a third year yield corner plot of the present application;
FIG. 6 is a graph of the next year rate of return quantized investment in accordance with the present application;
FIG. 7 is a diagram of the current annual rate of return quantized investment in accordance with the present application;
FIG. 8 is a graph showing the return comparison of annual yield quantified investments in the present application.
FIG. 9 is a flow chart of a data system of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application; it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments, and that all other embodiments obtained by persons of ordinary skill in the art without making creative efforts based on the embodiments in the present application are within the protection scope of the present application.
As shown in fig. 1 to 9, the present application is mainly performed in the following 7 steps.
Step S1, collecting the existing ESG scoring index system and historical data.
The traditional ESG scoring system is used for reference; the ESG score is typically evaluated and calculated by a third party entity and the scoring system flow includes: 1) Collecting data: the evaluation entity may collect data and information regarding the performance of the enterprise ESG from a number of sources including official documents, media reports, information disclosed by the enterprise disclosure, and the like. 2) Screening indexes: the evaluation facility may screen out key metrics related to the ESG performance of the enterprise, such as carbon emissions, employee welfare, board management, etc., from the large amount of data collected. 3) Weight index: the assessment mechanism can distribute weights of different indexes according to factors such as importance degrees of the indexes, industry characteristics and the like. 4) Calculating the score: the evaluation mechanism calculates corresponding scores according to the performance of the enterprise on each index, and integrates the scores to obtain a final ESG score. As can be seen, the conventional ESG scoring system generally adopts the delfei method, which is highly subjective, and has questionable and improved effect on ESG investment guidance.
And S2, processing missing values.
Carrying out missing value processing in a zero padding mode, wherein target variables of an ESG scoring index system are respectively set to be current year rate of return, next year rate of return and third year rate of return; to investigate whether the ESG factor has an immediate effect or a delayed effect on the investment value of the enterprise.
Step S3, ESG investment score analysis
The application adopts AI algorithm to construct scoring model. Because the traditional ESG scoring model has higher subjectivity and cannot reflect nonlinear real-change financial activities by adopting linear weights, the defects of traditional scoring can be eliminated by scoring by adopting an AI algorithm with self-adaptability, expandability and interpretability. Taking the AI algorithm process of XGBoost model as an example, the artificial intelligence algorithm of ESG will be scored from the AI algorithm and XAI.
XGBoost is a number of improvements over traditional GBDT, mainly: firstly, performing second-order expansion on a loss function by using a Taylor formula; secondly, adding regularization term to the loss function to control the complexity of the model, so as to prevent overfitting; third, parallelism is supported. The XGBoost core idea is to add CART tree as weak learner, the weak learner learns a new function to fit the residuals of all the previous trees, and the final output is the accumulated sum of all the tree prediction results. The XGBoost algorithm is a machine learning algorithm based on a gradient-lifted decision tree (Gradient Boosting Decision Tree, GBDT) that implements a series of optimizations based on the decision tree algorithm. In practice, it has proven to be more efficient and faster than other decision tree-based algorithms (including random forests and AdaBoost, etc.).
A training dataset in which feature vectors representing the ith sample represent the true labels of the ith sample. Our goal is to learn a function so that.
The following is described:
a loss function is defined, representing the gap between the real label and the predicted label, and common loss functions include a square loss function, a cross entropy loss function, and the like. The mathematical description of the objective function of XGBoost is expressed as follows:
wherein n is the number of samples, x i For the ith sample, y i Andthe true value and the predicted value of the ith sample are respectively corresponding to the real value and the predicted value of the ith sample; k is the tree number of CART tree, f k Represents the kth CART; />As a loss function, Ω (f k ) Is an L1 or L2 regularization term.
XGBoost trains the t-th tree at each split, and the objective function of the t-th tree is:
XGBoost performs second-order Taylor expansion on the loss function so as to approximate the loss function, and the objective function of the t-th tree is approximately as follows:
wherein g i 、h i Is the first and second derivatives of the i-th sample loss function L. Omega (f) t ) Regularization term for the t-th tree may be defined as:
f t represents the T-th tree, T is the number of leaf child nodes, and w j For the weight on the j-th leaf node where the sample point x is located, γ and λ are regularization parameters.
In XGBoost, each tree structure is split into two parts: a structural part q and a weight part w, the formula is:
f k (x)＝w q(x) ,w∈R T ,q:R d →{1,2,…,T}
wherein f k (x) The T dimension vector corresponds to the scores on the T leaf nodes; q is a mapping that maps sample points to a leaf node. The tree can be completely determined as long as q and w are determined. w is the weight vector of the leaf nodes, T is the number of leaf nodes, and q is the function mapping samples x to the leaf nodes.
Will f t And omega is substituted into the approximate objective function, and a constant term is ignored, so that the following can be obtained:
the above formula is simplified to be as follows by step 3.3:
starting from the leaf node, all the leaf nodes are accumulated, and the above formula is equivalent to:
the t-th CART tree is a certain determined structure, the above formula is a unitary quadratic function about leaf nodes, and after minimized optimization, the minimum values of the minimum point and the objective function are obtained by solving the following steps:
minimum point:
minimum value:
wherein, the liquid crystal display device comprises a liquid crystal display device,G j representing the sum of the first derivatives of all input samples mapped to the jth leaf node, and similarly, H j Representing the sum of the second derivatives.
And constructing a tree structure by using a greedy algorithm, and selecting a tree with the smallest structure score.
AI aims to solve the problem of black box models, i.e. it is difficult to understand and explain the decision making process of machine learning models. As XAI (interpretable artificial intelligence) studies have progressed, a number of interpretable methods have been developed that allow users to better understand the decision making process of a model and thus better understand the predicted outcome of the model. Common global interpretation methods include SHAP, LIME, PDP, etc., and common local interpretation methods include LIME, counterfactual Explanation, etc. The variable importance in XAI is adopted to judge the variable nonlinear attribute of the contribution of each factor of ESG scoring system to the income.
Step S4 obtains a score for each feature.
And S41, calculating the increment degree of the model prediction error after the argument is disturbed by adopting Perchange disturbance or shuffle, and taking the increment degree as an importance judgment index of the feature. There are two modes: permulation disturbance and shuffle disturbance. The basic steps are as follows: 1. and (3) subtracting the estimated parameters of the two models to obtain a difference value, 4, sequentially repeating the steps 2 and 3, wherein the difference value and 5 represent the importance of the feature, and the difference value is large, so that the influence of the independent variable disturbance on the model prediction precision is larger.
Step S42, PDP part dependency graph. PDP (Partial Dependence Plots) on the basis of training the original data set, researching the relation between the dependent variable and the characteristic, and for each sample, the other characteristic values are unchanged, while the characteristic performs traversal disturbance of the characteristic values, so as to calculate the average value of model prediction. PDP function expression: . Wherein: representing the ith feature of the kth sample of the training set. PDP brief steps:
1) And training a model.
2) The feature is specified. For each sample, performing value traversing disturbance on a specified feature, for example, the specified feature is X1, traversing the value range of the specified feature to be [ a, b ], and performing disturbance with the step length being h.
3) Independent variable disturbance. And (3) predicting each disturbed sample through a model to obtain a corresponding predicted value without changing other characteristic variables.
4) For each feature value, an average value of the predicted values of all samples is calculated as the PDP value of the feature value.
5) And (3) calculating all reference values corresponding to the X1 value range given in the step (3).
6) PDP partial dependency graph: and drawing a PDP part dependent graph, wherein the X axis represents the value range of the designated feature, and the Y axis represents the corresponding PDP value.
If step 4) does not average, then there is no reference value, but is called single condition expectation (ICE). The ICE curve is also another model based on independent variable disturbance, namely the corresponding model predicted value change after the specified characteristic value is changed, which is beneficial to exploring individual differences and identifying subgroups and interactions between model inputs.
The PDP partial dependency graph can help us understand the relationship between model predictions and features. For example, the degree of influence of features on model predictions, and interactions between features, can be observed through the PDP map. The PDP map can also be used to verify whether the model is reasonable and to find outliers or missing values in the data. S is S
And S5, adopting Knee inflection point analysis.
And adopting a knee method to judge the investment threshold. Inflection point detection is a technique for determining the location of an inflection point in a data sequence. In general, an inflection point refers to a point at which a sudden change occurs in a data sequence, which may be a change in trend or the occurrence of an abnormal point. Inflection point detection can help us identify these change points in the data, thereby better understanding the trends and characteristics of the data. Common inflection point detection methods include slope change, knee point detection, elbow method, etc. The application adopts a Knee point detection method.
The formula of the method of steps S51, knee point detection is as follows:
1) Ordering the model predicted revenue data set to obtain an ordered sequence
2) Calculating the sum of the distances of each data point from the whole data sequence, namely, taking each data point as an inflection point, calculating the sum of the distances before and after the point, and adding the sum of the two distances:
where D (i) represents the sum of distances when the i-th data point is taken as the inflection point.
3) And drawing a relation graph of the sum of the distances and the data points, and finding out the point with the maximum change rate of the sum of the distances, namely the inflection point position.
4) If multiple inflection points locations need to be determined, the sequence preceding the inflection point locations may be removed and the above steps repeated.
The basic flow of the method of step S52, knee point detection.
1) Collecting data: data was collected that needed Knee point detection.
2) Drawing a curve: the data is plotted as a graph.
3) Calculating the slope: the slope between each point on the curve and the previous point is calculated.
4) Calculating the distance: the distance from each point to the starting point of the curve is calculated.
5) Calculating a Knee point: and carrying out weighted average on the distance and the slope to obtain a Knee point value of each point.
6) Selecting a Knee point: the point with the largest Knee point value is selected as the Knee point.
7) Segmentation curve: the curve is split into two parts at the Knee point.
8) Analysis data: analyzing the two segmented portions to obtain further insight
In step S53, it should be noted that Knee has a certain requirement on the shape and number of data sequences, and is applicable to data sequences with a single inflection point or a small number of inflection points. The application only needs to separate investment and non-investment objects, and is suitable for selecting the Knee inflection point method.
And S6, constructing an ESG factor investment strategy.
And constructing a quantization strategy, buying a predicted value higher than a Knee inflection point, not opening a bin lower than the Knee inflection point value, and waiting for fund weight, wherein the bin rotation period is one year, and the bin is replaced in the beginning of each year.
And S7, strategy comparison.
Example 1: the application adopts data of social responsibility reports 2010 to 2020 of the company on the Internet and a professional evaluation system of the social responsibility report of the company on the Internet, and finally forms a total of 102 indexes which relate to 5 primary indexes, 13 secondary indexes and 37 tertiary indexes. The data processing and verification process is realized through python 3.9.
And S3, decomposing by utilizing system methods such as AI, XAI, knee and the like to obtain an ESG evaluation method and a method of an investment decision system. The method specifically comprises the following steps:
in the implementation process, XGboost, XGBoost in the AI algorithm is a powerful machine learning algorithm, so that the method has the advantages of high efficiency, expandability, accuracy, interpretability and the like, and is widely applied to practical application. The parameters of XGBoost are relatively large, including the depth of the tree, the minimum number of samples per leaf node, the learning rate, regularization parameters, etc. Therefore, tuning takes a lot of time and effort. After parameter optimization, searching optimal parameters of the models, wherein the optimal parameters of each model are as follows:
TABLE 1 XGBoost Algorithm optimization parameters for various inputs
Step S32, in the model training process, searching the optimal parameters of the model in the mode of grid pattern parameters, setting the parameters of the model to optimize the model effect, comparing the predicted value of the model with the effect of the true value on the training set, and evaluating the effect on R 2 Effects on RMSE, MAE, fitting results are shown in the table below:
table 2 each model goodness-of-fit parameters
From R 2 In the above, the model input by the secondary index is uniformly lower than the three-level index input model, the three-level index has higher decision coefficient and better fitting effect, and the model result of the decision coefficient shows that the secondary index decision coefficient is too low to be suitable for investment decision reference, and the three-level index is larger than 0.15, so that the model has a certain investment reference effect. Among the three-level indexes, the current annual yield is the coefficient of the target variable on the RMSE and the MAE, the fitting effect is relatively the best, the third-year model is the second-year model, and the next-year model is the first-time model. The demonstration results support that the ESG factors have an immediate and delayed effect to some extent.
The model conclusion shows that the scoring by adopting a given weight mode has a certain subjectivity, so that the primary and secondary indexes have little effect on investment reference. The three-level index is not weighted, but is an original unweighted index, and shows a more obvious investment reference effect in XGBoost. When the three-level index of ESG is used for investment reference, the decision coefficient is about 0.2, and the ESG has certain model interpretation capability, but is not strong. Thus, we choose the ESG investment pattern of the screening class to exclude or select investment targets based on certain criteria. Specifically, enterprises or industries that are poor in performance or have significant disputes in terms of environment, society, administration are screened out according to certain exclusion criteria, and those that are good in terms of ESG are selected. Not only is favorable for reducing the risk of ESG and improving the return on investment, but also is favorable for promoting enterprises to improve the ESG performance and promote sustainable development.
In step S4, the XGBoost algorithm automatically calculates the importance of each feature in the training process, and the importance scores of the features can be used in aspects of feature selection, feature engineering and the like. In XGBoost, a feature_importance_attribute is called to obtain a score of each feature, and the higher the score is, the larger the contribution of the feature to output is, so that the feature can be used for screening out the most important feature or performing operations such as feature dimension reduction. Meanwhile, XGBoost also provides a plot_importance function, so that the importance scores of the features can be visualized, and the importance of each feature can be displayed more intuitively.
Step S41, selecting the first 10 factors with larger influence on the annual income score as shown in figure 1,
as can be seen from the feature importance results, in the first ten important variables of the three-year yield model, all sub-indexes in the return are used as important indexes ranked at the top. There are sub-indicators in employee responsibility and environmental management in the current and next years. None of the sub-indices in social responsibility appear in the first ten important variables. The research target is mainly ESG investment effect analysis, and three secondary indexes of profit, debt and return in a professional evaluation system of a social responsibility report of a marketing company are all financial indexes, so that the method has important importance on annual yield and covers the effect of ESG, and therefore, the analysis is carried out after the three secondary indexes are removed.
In step S42, the research objective is mainly ESG investment effect analysis, and three secondary indexes of profit, debt repayment and return in the professional evaluation system of the social responsibility report of the marketing company are all financial indexes, which have important roles on annual yield, so that the effect of the ESG is covered, and therefore, after the three secondary indexes are removed, the analysis is performed.
Table 3 each model fitting goodness parameters:
in the current year Next year of the year Third year
R 2 0.11 0.10 0.11
MSE 0.15 0.16 0.15
MAE 0.28 0.29 0.28
The yield is shown in figure 2. The first five important variables of the three models are the total sum ratio of the obtained tax to the profit, the average income of staff, the development expenditure of products, the public benefit donation amount and the number of technical innovation projects.
Step S5, performing Knee inflection point analysis as an example, analyzing trend changes based on the ESG investment prediction sequence, identifying important positions of data mode conversion, and exploring the maximization of balanced benefits between ESG responsibility and investment benefits.
In the implementation process, the specific steps include:
as shown in FIG. 3, the x-axis where the current year yield inflection point is located is-0.2;
as shown in fig. 4, the x-axis where the next year yield inflection point is located is-0.003;
as shown in FIG. 5, the x-axis where the third year rate of return inflection point is located is-0.042.
And S6, constructing an investment strategy based on ESG factors, buying with a predicted value higher than a Knee inflection point, not opening bins and waiting for fund weights lower than the Knee inflection point, wherein the bin rotation period is one year, and the bins are changed in the beginning of each year. The strategy avoids using future data, each model will run predictions of corresponding year samples each year, such as the third year revenue prediction model in 2015, early 2015 binning, XGBoost model samples are 2012 samples. Operating environment: and (5) a width-gathering platform.
The next year yield quantified investment return diagram is shown in fig. 6;
the current annual yield quantitative investment return diagram is shown in figure 7;
annual yield quantitative investment return comparison chart is shown in figure 8.
And S7, investment strategy analysis. The ESG scoring system under the simple linear weighting and AI algorithm has self-adaptability, and can reflect market variation in investment guidance. The threshold value of investment or not is divided through a Knee inflection point, and positive and negative values are adopted to divide the investment or not more simply, wherein the threshold value is divided through the Knee inflection point, and the threshold value has more excellent bottom logic.
The above description is only of the preferred embodiments of the present application; the scope of the application is not limited in this respect. Any person skilled in the art, within the technical scope of the present disclosure, may apply to the present application, and the technical solution and the improvement thereof are all covered by the protection scope of the present application.

Claims (7)

1. An ESG evaluation system investment decision system, which is characterized by comprising the following steps:
s1, collecting the existing ESG scoring index system and historical data;
s2, processing a missing value;
s3, ESG investment score analysis;
s4, obtaining the score of each feature;
s5, adopting Knee inflection point analysis;
s6, constructing an ESG factor investment strategy;
s7, strategy comparison;
in the step S1, the collected data sources are from existing domestic and foreign scoring institutions, and the listed companies are main data sources and analysis objects.
2. The ESG evaluation system investment decision system of claim 1 wherein: s2, carrying out missing value processing in a zero padding mode, wherein target variables of an ESG scoring index system are respectively set to be current year rate of return, next year rate of return and third year rate of return; to investigate whether the ESG factor has an immediate effect or a delayed effect on the investment value of the enterprise.
3. The ESG evaluation system investment decision system of claim 1 wherein: and S3, performing ESG investment score analysis by adopting an XGBoost algorithm in an AI algorithm, wherein input variables are secondary indexes and tertiary indexes respectively, and target variables are stock year yields of market companies in the current year, the next year and the third year respectively. And (5) carrying out an AI algorithm to detect the contribution degree of each ESG factor to the investment yield.
4. The ESG evaluation system investment decision system of claim 1 wherein: and S4, the feature_importance_attribute of the XAI is called to acquire the score of each feature, wherein the higher the score is, the larger the contribution of the feature to the output is, and the feature can be used for screening out the most important feature or performing operations such as feature dimension reduction. And visualizing the importance scores of the features using a plot_importance function.
5. The ESG evaluation system investment decision system of claim 1 wherein: and S5, analyzing trend change based on the ESG investment prediction sequence by adopting Knee inflection point analysis, identifying important positions of data mode conversion, and exploring the maximization of balanced benefits between ESG responsibility and investment benefits.
6. The ESG evaluation system investment decision system of claim 1 wherein: and S6, constructing an ESG factor investment strategy: considering that part of stocks in China limit the qualification of the coupon, the strategy does not consider the emptying mechanism, the predicted value is higher than the Knee inflection point to buy, the stock is not opened and lower than the Knee inflection point value, the fund weight is equal, the bin rotation period is one year, and the bin is replaced in the beginning of each year.
7. The ESG evaluation system investment decision system of claim 1 wherein: and S7, performing strategy comparison and performance comparison with a broad-base standard strategy of the Shanghai-Shen 300 index.
CN202310769091.4A 2023-06-27 2023-06-27 ESG evaluation system investment decision-making system Pending CN116823487A (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202310769091.4A CN116823487A (en) 2023-06-27 2023-06-27 ESG evaluation system investment decision-making system
LU504769A LU504769B1 (en) 2023-06-27 2023-07-19 Ein esg-bewertungssystem für investitionsentscheidungen

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310769091.4A CN116823487A (en) 2023-06-27 2023-06-27 ESG evaluation system investment decision-making system

Publications (1)

Publication Number Publication Date
CN116823487A true CN116823487A (en) 2023-09-29

Family

ID=88114097

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310769091.4A Pending CN116823487A (en) 2023-06-27 2023-06-27 ESG evaluation system investment decision-making system

Country Status (2)

Country Link
CN (1) CN116823487A (en)
LU (1) LU504769B1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117422334A (en) * 2023-10-27 2024-01-19 国网北京市电力公司 Multi-level panoramic carbon efficiency analysis method and system based on multi-energy data

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117422334A (en) * 2023-10-27 2024-01-19 国网北京市电力公司 Multi-level panoramic carbon efficiency analysis method and system based on multi-energy data

Also Published As

Publication number Publication date
LU504769B1 (en) 2024-01-22

Similar Documents

Publication Publication Date Title
CN110110981B (en) Credit rating default probability measure and risk early warning method
Chen et al. An intelligent pattern recognition model for supporting investment decisions in stock market
Brauers et al. MULTIMOORA for the EU Member States updated with fuzzy number theory
Chuang Application of hybrid case-based reasoning for enhanced performance in bankruptcy prediction
Deng et al. Inter-company comparison using modified TOPSIS with objective weights
Gao The use of machine learning combined with data mining technology in financial risk prevention
Smith et al. Predicting firm-level bankruptcy in the Spanish economy using extreme gradient boosting
Özdağoğlu et al. Evaluating financial performance with grey relational analysis: an application of manufacturing companies listed on borsa Istanbul
CN116823487A (en) ESG evaluation system investment decision-making system
Yang et al. Evaluate the sustainable reuse strategy of the corporate financial management based on the big data model
Yang et al. Fortify the investment performance of crude oil market by integrating sentiment analysis and an interval-based trading strategy
Yang et al. How environmental regulation influences the green economy efficiency of resource-based cities: An empirical study from China
KR20180104967A (en) Investment Value Index and Model
Li et al. A comparison between linear regression, lasso regression, decision tree, XGBoost, and RNN for asset price strategies
Liu et al. A comparative study of environmental information disclosure between banks in net-zero banking alliance and China
Sun Management Research of Big Data Technology in Financial Decision-Making of Enterprise Cloud Accounting
Kaisar et al. Explainable Machine Learning Models for Credit Risk Analysis: A Survey
Huang Research on strategy of HS300 index based on random forest
Nawani et al. A Quantitative approach to create a Hybrid stock filtering platform
Abdul Salam et al. Machine learning algorithms on price and rent predictions in real estate: a systematic literature review
Adakh et al. Comparison of Some Data Mining Models in Forecast of Performance of Banks Accepted in Tehran Stock Exchange Market
Vinogradova et al. Software for data analysis and prediction of bankruptcy of organizations in the fuel and energy complex
Yang et al. An efficient loss function and deep learning approach for ranking stock returns in the absence of prior knowledge
Alburshaid et al. Forecasting Methods for the Real Estate Market: A Review
Zhang et al. Digital transformation by firms and the cleanliness of China's export products

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination