close

Вход

Забыли?

вход по аккаунту

?

3055635.3056608

код для вставкиСкачать
Analysis of CSR activities Affecting Corporate Value Using
Machine Learning
Kanau Mitsuzuka, Feng Ling and Hayato Ohwada
Department of Industrial Administration, Faculty of Science and Technology
Tokyo University of Science, 2641 Yamazaki, Noda, Chiba, 278-8510, Japan
+8147-124-1501, +8147-124-1501, +8147-124-1501
7415618@ed.tus.ac.jp, fengl@rs.tus.ac.jp, ohwada@rs.tus.ac.jp
However, it is not clear what CSR activities actually affect profit
and financial performance. Therefore, many companies are still
seeking ways to tackle CSR activities [1]. In Japan, research that
verified the relevance between CSR activities and financial
performance has been carried out almost exclusive of some
research. This is because the definition and measurement of CSR
are difficult, and company data required for analysis has not yet
been released. However, in recent years, with the rise of social
interest in CSR, newspapers have independently investigated and
disclosed the CSR activities of individual companies, such as
Toyo Keizai, Inc. Analyzing the Toyo Keizai database, this study
aims to use machine learning to clarify the influence of CSR
activities on corporate value.
ABSTRACT
Corporate social responsibility (CSR) activities are attracting
attention in Japan. Interest has increased among researchers
investigating the potential link between CSR activity and financial
performance. In this study, we propose a method of finding CSR
activities related to corporate value using random forest, a
machine-learning technique. The definition of CSR is ambiguous,
so we used the International Organization for Standardization
(ISO) 26000 framework as the basic standard for investigating
companies, since it is a major guideline for CSR activities in
Japan. Following this framework, we used 36 CSR activities as
features. Results indicated that Occupational Health and Safety
Management System (OHSMS) is the most important activity in
corporate value classification. We then used association analysis
to clarify how CSR activities related to corporate value influence.
Results indicated that documentation of CSR policy is often
performed in parallel with other activities, whereas, stakeholder
engagement is not.
Some overseas studies on CSR activities and financial
performance have been conducted. Ahmed et al. [2] divided CSR
into five categories (labor practices, corporate governance.
environment, community, and workplace) and verified the
relationship between CSR and finance. Results indicated that
ROA tended to be higher for companies conducting these CSR
activities than for companies that were not. Chin-Wei-peng [3]
divided CSR into three categories (total, people, and product) and
verified the relationship between CSR and finance. Results
indicated that the people category has a positive effect on future
financial performance.
CCS Concepts
• Computing methodologies → Machine learning → Machine
learning approaches → Classification and regression trees.
Keywords
Corporate social responsibility; ISO 26000; corporate value;
machine learning; random forest; association rule learning.
However, these studies define CSR broadly and do not mention
specifically what kind of CSR activity affects financial
performance.
1. INTRODUCTION
Along with the development of information technology, research
on data mining that acquires new knowledge from large-scale data
has been conducted in various fields. However, information
obtained from the accumulated data is diverse and complicated.
Therefore, it is necessary to retrieve useful information from
databases of various formats when conventional statistical
analysis is difficult. Machine learning has attracted attention as a
technique for meeting this challenge.
Ozono [4] verified CSR relevance to financial performance using
individual data of the "Unique Questionnaire Survey on Corporate
Social Responsibility, 2005." However, since interpretation of
CSR differs for each company taking the questionnaire,
comparison among companies is not possible.
The present study uses the International Organization for
Standardization (ISO) 26000 framework as the basic standard for
investigating companies, since it is one of the major guidelines for
CSR activities in Japan. In doing so, it is possible to investigate
companies’ CSR activities in Japan and compare their efficacy.
Corporate social responsibility (CSR) activities are attracting
attention in Japan. Therefore, companies must actively manage
strategies that will lead to increasing profits from CSR activities.
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. Copyrights for
components of this work owned by others than ACM must be honored.
Abstracting with credit is permitted. To copy otherwise, or republish, to
post on servers or to redistribute to lists, requires prior specific permission
and/or a fee. Request permissions from Permissions@acm.org.
ICMLC 2017, February 24-26, 2017, Singapore, Singapore
© 2017 ACM. ISBN 978-1-4503-4817-1/17/02…$15.00
The purpose of this study is twofold. (1) We use a random forest
[5] to extract CSR activities related to corporate value. Random
forest can determine the importance of each feature after learning.
We focused on this function in our analysis. (2) We use
association analysis to clarify how CSR activities related to
corporate value influence each other [6] [7]. We examine the
mutual relevance of the top 10 CSR activities judged to be related
to corporate value using random forest.
DOI: http://dx.doi.org/10.1145/3055635.3056608
11
The greater the numeric value, the more actively the company is
engaged in CSR activities.
2. DATASETS
In this study, we used 36 CSR activities as features from the Toyo
Keizai database. (Table 1).
We used financial data from the Nikkei NEEDS CD-ROM. The
limitation of many previous studies is that they considered only a
short-term perspective. If companies conduct CSR activities, it is
insufficient to estimate the efficacy of CSR activities from this
perspective. In contrast, the present study focuses on the corporate
value model, which reflects corporate value estimation and
considers the relationship between CSR and financial performance
from a long-term perspective.
Table 1. 36 features (CSR activities)
Category
CSR activity
Establishment of Environment Department
Existence of Environmental Policy Document
Environment
Climate Change Initiatives
The Impact on Biodiversity by business activities
Policy on Anti-Corruption and Anti-Bribery Measures
Setting Compliance Department
Fair operating
CSR Procurement Measures
practices
Independent Internal Audit Department
Risk Management/Crisis Management System
Stakeholder Engagement
Community Philanthropy Department
involvement and Collaboration with NGO/NPO
development Community Investment Initiatives
Pro Bono Support
Establishment of Customer Service Department
Consumer
Establishment Departments Related to Product and
issues
Service Safety and Safety System
Establishment of Whistleblower Hotline(Internal)
Establishment of Whistleblower Hotline(External)
Human Rights
Policy on Anti-Discrimination, Human Rights, etc.
Human Right Due Diligence
Department for Managing Diversity of Human Resources
Re-Employment of Employee Who Resighned Due to
Labour
Pregnancy or Childbirth, etc.
practice
Occupational Health and Safty Management System
(OHSMS)
Employee Satisfaction Survey
Flexible Work Programs
Short-time Work System
Work life
Telecommuniting System
balace
Satellite Office(SO)
Discretionary Labor Systems
Establishment of CSR Department
Appointment of CSR Officer
Setting Materiality Levels
Corporate
Documentation of CSR Policy
governance
ISO26000 for Use
Disclosure of ESG Information
Dialogue with Investors and ESG Institutes
To calculate corporate value, we used the Ohlson Model [8],
which is based on the income approach. The income approach is
the most valid way to measure the value of a business or business
interest. The formulation of the Ohlson model is
∑
is corporate value; is equity at book value;
residual profit; and r is cost of equity.
is
We utilized company data to meet the following criteria: (1)
include both CSR data and financial data, and (2) include the
manufacturing industry from the Tokyo Stock Exchange’s first
section (food, fiber, chemistry, pulp and paper, medicine, rubber,
glass, steel, nonferrous metals, metal, machine, electronic goods,
transport equipment, precision equipment, wholesale, and other
products).
A total of 264 companies were obtained. We converted companies
into two labels according to corporate value. Arranged in
descending order of corporate value, the top 100 companies were
labeled "high," and the bottom 100 companies were labeled "low."
We removed the 64 companies in the middle as noise. Therefore,
we used 200 companies for analysis.
3. METHODOLOGY
3.1 Random Forest
We used a random forest classifier and labeled each sample state
("high" or "low") when it was learned. Random forest is an
ensemble learning method for classification. The forest includes
many decision trees, each of which is built using a bootstrap
sample of the dataset. The randomness used in random forest
gives good results and has been used in various studies [9] [10]
[11]. In addition, random forest can identify which features were
important in building a forest of trees; thus, we focused on this
function. In this process, random forest builds decision trees from
a sample drawn with replacement from the dataset, and each node
of trees is split by information gain for the categorical sample
state. However, random forest selects features randomly to make
its decision trees, and it is possible that some of the features could
be unused when it makes the forest. Thus, we should create
sufficient trees, because the importance of a CSR activity is
Table 2. Convert into numerals
content
Yes
Under consideration
Stakeholder Engagement
Other
No or No answer
(1)
The Ohlson model involves adding the discounted present value
of future residual profit to equity (at book value). This is in
accordance with the economy value that the shareholders expect.
This study uses the Ohlson model for calculating corporate value
for two reasons. (1) The Ohlson model is calculated from the
income approach (the method of estimating income based on
expected profits). This is suitable, since the aim of the present
study is to estimate CSR from a long-term perspective. (2) The
Ohlson model is calculated from accounting data. Thus, the result
is as good as market estimation, and this model is not affected by
changes in accounting policies
CSR activities are extracted by referring to the ISO 26000
framework as explained above. Because these contents are
categorical, it is necessary to convert each activity to a numerical
value for as data processing analysis. For example, CSR activity
"Stakeholder Engagement" is converted as follows (Table 2).
CSR activity
.
numeric value
3
2
1
0
12
to identify the most important relationships. Confidence
represents the probability (p(B|A)) of occurrence of event B when
event A occurs. Therefore, the closer this value is to 1, the
stronger the link between the antecedent and consequent of the
rule. Support represents the appearance frequency of the rule,
which is the probability (p(A ∧ B)) that A and B occur
simultaneously. If support is too low, the rule hardly occurs. In
rule extraction, minimum confidence and minimum support are
set, and only rules satisfying these conditions are extracted.
determined when it is used in the learning process. Therefore, we
created 2000 trees for classification.
3.2 CSR Activity Ranking by Random Forest
We use feature importance measured by random forest in this
process. We use a random forest classifier and assign a label to
each sample state.
Random-forest learning is affected by hyperparameters. If we
want the random forest to learn well, we should set the best
hyperparameters in each process. Therefore, our method uses a
grid search so that the random forest can learn with the best
hyperparameters
set
automatically
each
time.
The
hyperparameters and values that are set in the grid search are
listed in Table 3.
Under the antecedent (if) and a consequent (then) as CSR
activities, we extracted two kinds of association rules: positive
association rule and negative association rule.
A positive association rule is the relationship:
{CSR activity A = Yes ⇒ CSR activity B = Yes}.
Table 3. Hyperparameters and values
Hyperparameter
The maximum depth of tree
The minimum number of samples
reauired to split an internal node
The minimum number of samples
in newly created leaves
An example is presented in Figure 1:
Values(comma separated)
2, 3, 4, 5, 6
{CSR activity A = Yes ⇒ CSR activity B = Yes}
5, 10, 15, 20
After the ranking process, we focused on the top 10 ranked CSR
activities.
CSR
activity B
Figure 1. Positive association rule
In the above example, confidence of 0.8 means that 80% of the
companies that conduct CSR activity A also conduct CSR activity
B.
3.3 Random Forest Model Evaluation
We use accuracy (Eq. 2), precision (Eq. 3), recall (Eq. 4), and Fmeasure (Eq. 5).
accuracy =
confidence=.800
CSR
activity A
5, 10
A negative association rule is the relationship:
{CSR activity A = No ⇒ CSR activity B = No}.
(2)
An example is illustrated in Figure 2:
precision =
(3)
recall =
(4)
{CSR activity A = No ⇒ CSR activity B = No}
F-measure =
CSR
activity A
(5)
Figure 2. Negative association rule
here, TP is true positive; FP is false positive; FN is false negative;
and TN is true negative, as defined in Table 4.
In the above example, confidence of 0.8 means that 80% of the
companies that do not conduct CSR activities A also do not
conduct CSR activities B.
Table 4. Definitions of TP, FP, FN, and TN
Predicteed
high
low
CSR
activity B
confidence=.800
observed
high
low
TP
FP
FN
TN
4. RESULTS
This chapter describes our results. We classified corporate value
"high" or "low" using random forest. Classification results are
listed in Table 5.
In this study, five-fold cross validation was performed to obtain
classification prediction accuracy.
Table 5. Classification results
TP FP FN TN Accuracy Precision
84 17 16 83
0.835
0.832
3.4 Association Rules
After outputting CSR activity ranking by random forest, we focus
on the top 10 ranked CSR activities and apply association rules to
them. Association rule mining is an analytical technique that finds
strong relationships between events, known as association rules.
Association rules are if/then statements that help uncover
relationships between seemingly unrelated data in a relational
database or other information repository. An association rule has
two parts, an antecedent (if) and a consequent (then). A
consequent is an item that is found in combination with the
antecedent. Association rules are created by analyzing data for
frequent if/then patterns and using criteria support and confidence
Recall
0.840
f1
0.836
Using the random forest classifier, high classification prediction
accuracy was obtained. Therefore, the importance of each CSR
activity that contributes to this classification prediction model is
highly reliable.
Next, the top 10 CSR activities ranked by random forest are listed
in Table 6. Occupational Health and Safety Management System
(OHSMS) is the most important activity in corporate value
classification.
13
Table 6. Ranked CSR activities
5. CONCLUSION
In this study, with the aim of clarifying the relationship between
CSR activities and financial performance, we used random forest
classification to extract CSR activities related to corporate value.
We used 36 CSR activities from the Toyo Keizai database as
features. Results indicated that OHSMS ranked as most important
in corporate value classification with high accuracy.
Rank
CSR activity
importance
1 OHSMS
0.0953
2 Climate Change Initiatives
0.0872
3 Collaboration with NGO/NPO
0.0788
4 Stakeholder Engagement
0.0786
5 Dialogue with Investors
0.0604
6 Documentation of CSR Policy
0.0561
7 Human Right Due Diligence
0.0533
8 The Impact on Biodiversity
0.0523
9 Setting Materiality Levels
0.0408
10 Setting Compliance Department
0.0371
After that, we used association analysis to clarify how CSR
activities related to corporate value influence each other. Results
indicated that documentation of CSR policy is often performed in
parallel with other activities, whereas stakeholder engagement is
not.
For further study, if possible, we would like to add samples. In the
present study, we focused on only the manufacturing industry.
With the addition of samples, accuracy will become higher and we
can obtain more reliable results.
After outputting CSR activity ranking by random forest, we focus
on the top 10 ranked CSR activities and apply association rules to
them. Positive association rules among 10 CSR activities are
presented in Figure 3. The minimum support was set to 0.2, and
the minimum confidence was set to 0.8.
Setting Compliance
Department
[1] Kamei, Zentaro., Hitaro, Taku., 2015 Issues and Prospects
for CSR in Japan Analysis of Japan’s CSR Corporate Survey,
accessed: 18, January 2016
http://www.tokyofoundation.org/en/articles/2015/issues-andprospects-forcsr
OHSMS
.828
Setting Materiality
Levels
Climate Change
Initiatives
.874
.924
The Impact on
Biodiversity
6. REFERENCES
.834
.827
.815
.842
.969
.859
.880
Collaboration
with NGO/NPO
.946
.879
.827
Human Right
Due Diligence
.872
[2] Sarwar Uddin Ahmed, Md. Zahidul Islam, Ikramul Hasan:
Corporate Social Responsibility and Financial Performance
Linkage‐Evidence from the Banking Sector of Bangladesh,
J. Org. Management 1, 14-21, 2012.
.897
[3] Chih-Wei Peng and Yu-Cheng Chen, Corporate Social
Responsibility and Financial Performance: Does CEO
Compensation Really Matter?, Journal of Applied Finance &
Banking, vol. 5, no. 6, 51-67, 2015
.918
Stakeholder
Engagement
.897
.940
Dialogue with
Documentation
Investors
of CSR Policy .896
[4] Tavel, P. 2007. Modeling and Simulation Design. AK Peters
Ltd., Natick, MA.
Figure 3. Positive association rules among 10 CSR activities
[5] L. Breiman, "Random forests," Machine Learning, vol. 45,
2001, pp.5-32.
The numbers in the figure indicate confidence. As indicated in
Figure 3, documentation of CSR policy is often performed in
parallel with other activities.
[6] Zhang, C.,Zhang, S.:Association Rule Mining: models and
algorithms, Lecture Notes In Arti- ficial Intelligence,
Vol.2307, p.238, 2002
Negative association rules among 10 CSR activities are presented
in Figure 4. The minimum support was set to 0.2, and the
minimum confidence was set to 0.7.
Setting Compliance
Department
OHSMS
Setting Materiality
Levels
.900
.718
.862
.750
Human Right
Due Diligence
[8] Ohlson, J.A. “The theory of value and earnings and an
introduction to the Ball-Brown analysis” Contemporary
Accounting Research, Vol.8, No.2, pp1-19, 1991
Climate Change
Initiatives
.744
The Impact on
Biodiversity
[7] Hang, J., Pei, J., Yin, Y.: Mining frequent pattern without
candidate generation, Proc. of ACM SIGMOD,pp.1-12,
2000
.718
.795
Collaboration
with NGO/NPO
.765
.704
.738
.752
.726
Documentation
of CSR Policy
.713
[9] Y. Amit and D. Geman, "Shape quantization and recognition
with randomized trees," Neural Computation, Vol. 9, No. 7,
1997, pp. 1545-1588.
[10] J. Gall, A. Yao, N. Razavi, L. Gool and V. Lempitsky,
"Hough forests for object detection, tracking, and action
recognition," Pattern Analysis and Machine Intelligence, Vol.
33, No. 11, 2011, pp. 2188-2202.
.854
Stakeholder
Engagement
.950
.917
Dialogue with
Investors
[11] National Center for Biotechnology Information, "Home GEO DataSets," in GEO DataSets. [Online]. Available:
http://www.ncbi.nlm.nih.gov/gds. Accessed: Mar. 16, 2016.
Figure 4. Negative association rules among 10 CSR activities
Stakeholder engagement is not often performed in parallel with
other activities.
14
Документ
Категория
Без категории
Просмотров
2
Размер файла
796 Кб
Теги
3055635, 3056608
1/--страниц
Пожаловаться на содержимое документа