201901 Hokkaido Conference Proceedings (SS)

Table 1: Shows the level of opinions about consumer purchasing decisions through online media. Consumer purchasing decisions through x S.D. Level of online media agreement 1. You use a device such as a personal 3.99 1.078 High computer. (PC, Notebook), mobile phones, tablets to buy products through online media. 2. You use social networks such as 3.94 1.049 High Facebook, Instagram, Twitter to buy products through online media. 3. You buy products online because the 3.73 1.066 High products are diverse and rare in the market. 4. You buy products online because the 3.69 1.096 High value of the price compared to buying through other channels. 4.02 1.023 High 5. You buy products through online media because it is convenient and fast. 3.68 1.085 High 6. You buy products through online media because it is the best choice. 3.77 1.060 High 7. You buy products online because it corresponds to the current lifestyle. 3.79 1.221 High 8. You can order online through 24 hours. Total 3.8250 0.87446 High The table 1 shows the level of opinions about consumer purchasing decisions through online media. When considering all variables, the opinions were at a high level. They can be sorted from highest to lowest as following; Buying products through online media because it is convenient and fast, with an average of 4.02, followed by use a device such as a personal computer. (PC, Notebook), mobile phones, tablets to buy products through online media, with an average of 3.99, using social networks such as Facebook, Instagram, Twitter to buy products through online media with an average of 3.94, allowing to order online 24 hours with an average of 3.79, buying products online because it corresponds to the current lifestyle with an average of 3.77, buying products online because the products are diverse and rare in the market with an average of 3.73, buying products online because the value of the price compared to buying through other channels with an average of 3.69, buying products through online media because it is the best choice with an average of 3.68, respectively. 248

Table 2: shows the opinions on the role of online marketing promotion. Online marketing promotion x S.D. Level of agreement 1. Advertising 3.7571 0.75652 High 2. Public relations 3.8651 0.77460 High 3. Personal selling 3.6952 0.79954 High 4. Direct marketing 3.6881 0.82193 High 5. Promotion 3.7595 0.85555 High The table 2 shows the opinions on the role of online marketing promotion. When considering all variables, the opinions were at a high level. The highest to lowest can be sorted as follows: public relations had an average of 3.8651, followed by promotion had an average of 3.7595, advertising had an average of 3.7571, personal selling had an average of 3.6952, and direct marketing had an average of 3.6881, respectively. Hypothesis testing by multiple regression analysis using independent variables (X): advertising, public relations, personal selling, direct marketing, sales promotion. The dependent variable (Y) is the purchase decision through the online media of the consumer. Table 3: shows the results of multiple regression analysis in case the dependent variable is consumer purchasing decision via online media. Role of online Unstandardized Standardized t Sig. marketing Coefficients Coefficients promotion B Std. Error Beta (Constant) 0.188 0.230 0.817 0.000** Advertising 0.252 0.079 0.218 3.181 0.002** Public relations 0.275 0.090 0.244 3.049 0.003** Personal selling 0.228 0.089 0.208 2.545 0.012* Direct marketing 0.112 0.088 0.105 1.268 0.206 Promotion 0.100 0.069 0.097 1.451 0.148 Adjusted R2 0.553 1.964 Durbin - Watson Note **Correlation is significant at the 0.01 level * Correlation is significant at the 0.05 level The table 3 shows the results of multiple regression analysis using the Enter method in case the dependent variable is the consumer purchasing decision via online media. It was found that the role of online marketing promotion can forecast online consumers decision to purchase via online media with the percentage of 55.3, and by Adjusted R2, it can be found that the public 249

relations (������ = 0.27, P < 0.01), advertising (������= 0.25, P < 0.01), personal selling (������= 0.22, P < 0.05) It can be explained that the three elements which are public relations, advertising and personal selling influences the purchase decision through online media, and direct marketing and promotion cannot predict consumer purchasing decisions through online media. Discussion The study found that the role of online marketing promotion can predict the consumer decision to purchase through online media. Advertising and public relations had a significant level of 0.01, and personal selling had a significant level of 0.05. It was consistent with Jian et al. (2018), stated that the use of online media to promote marketing improved brand awareness and boosted sales, but should be used online to advertise and promote right to the target consumer. Businesses should have a long-term plan to promote their marketing through online media at any given time to enhance their competitive advantage and long-term performance. In addition, research by Li, Timon C. Du. (2017) found that communicating with consumers through online media should include the use of micro-blogs to send information through short messages for a clearer and more effective communication to consumers. Thus, Businesses can use micro-blogs to send short messages via Facebook, Line, Twitter to help in advertising, promoting, and person selling to be clear and trustworthy, as well as increase brand awareness and increase sales in the future. Recommendation This study investigated the role of online marketing promotion that influences consumer purchasing decisions through online media which can be presented as follows; 1. Online merchants should focus on advertising through online media, such as advertising through various online media such as Mobile ad, Banner, Social media, with the clear and interesting content, to send to consumer through online media. Public relations, such as the introduction of micro-blogs, used to send short messages via Facebook, Twitter, and Twitter to provide clear and engaging information by allowing consumers to engage in online publicity or interactive communications. In case of personal selling, salespeople should continually introduce and persuade consumers, and they should be able to answer questions or solve problems quickly and clearly so that they can continue to purchase products and services through online media. 2. Online merchants should focus on building credibility by registering electronic commerce to get the symbol of DBD registered, which will make the shop tangible, can be followed or tracked. When the problem occurs, consumers will be relieved because it is registered with the Department of Business Development, Ministry of Commerce. It will make the shop look credible, professional, and most importantly, consumers who come to buy goods through online media are feeling more confident in the safety of the store. 250

3.2 Acknowledgments This study was supported by the Faculty of Business and Information Technology, Rajamangala University of Technology Suvarnabhumi in all aspects to conduct this research until it was successful. In addition, the researcher would like to take this opportunity to thank the respondents who answers the questionnaires as well. 4. References Chonnikarn Julmakorn (2013). Factors Influencing Internet Purchasing Behavior of Undergraduate Students, Faculty of Science, Burapha University. Graduate School, Burapha University Economic Intelligence Center (EIC), Siam Commercial Bank. (2561). Expect Thai will raise to 4.7 in 2022. Retrieved on 19 September 2018, from https://www.tcijthai.com/news/2018/3/scoop/7828 Hair, J., Black, W., Babin, B., Anderson, R., & Tatham, R. (2006). Multivariate data analysis (6th ed.). Uppersaddle River, N.J.: Pearson Prentice Hall. Jian, et al. (2018). Optimizing online recurring promotions for dual-channel retailers: Segmented markets with multiple objectives. European Journal of Operational Research, (267), 612 – 627. Kallaya Wanichbancha. (2006). Statistics Principles. Bangkok: Department of Statistics, Faculty of Commerce and Accountancy Chulalongkorn University. Kotler, Philip and Armstrong, Gary. Principles of Marketing. Eleventh Edition, New Jersey: Prentice – Hall International, Inc., 2006. Li, Timon C. Du. (2017). Maximizing micro-blog influence in online promotion. Expert Systems with Applications, (70), 52 – 66. Nantaporn Huaikaew et al. (2016). Factors related to the decision to use the online reservation service of consumers in Phra Nakhon Si Ayutthaya Province. Rajamangala University of Technology Suvarnabhumi Nattakit Wantamay. (2014). Marketing Communication. Bangkok: Kasetsart University Press. Professors of Faculty of Marketing Faculty of Business Administration Rajamangala University of Technology Thanyaburi (2016). Marketing Principles. Bangkok: Triple group. Ltd Supaporn Polnikorn (2005). Consumers' Behavior. Bangkok: Holistic Publishing Surangkana Wayupap.(2018). Internet user behavior in Thailand in 2018. Retrieved on 20 September 2018, from https://www.etda.or.th/content/etda-reveals-thailand-internet-user-profile-2018.html Time Chuesathapanasiri. (2015). Online media with role in determining social news issues. Retrieved on 15 September 2016 fromhttps://blog.infoquest.com/iqmedialink/news_onlinemedia/ 251

ISSSM-0352 Potential of Ecotourism Destinations in Suphan Buri Province Sunantha Khamnuanthong *, Phannarai Paiboon Faculty of Business Administration and Information Technology, Rajamangala University of Technology Suvarnabhumi, Thailand * E-mail: [email protected] Abstract Province under the purposes of the studying of overall physical condition of the tourism destinations, the tourists’ behavior in the ecotourism destinations, and the potential of the ecotourism destinations in Suphan Buri Province. The prospective result of this research is expected as the guidelines for the promotion of sustainability development of tourism industry in Suphan Buri Province. This research was conducted under the survey method on 400 tourists visiting the ecotourism destinations in Suphan Buri Province, selected by the purposive sampling and the quota sampling method. The data collected from the distributed questionnaire was analyzed under the statistic methods of percentage, standard deviation, and weighting score equation which was applied for the evaluation of the potential of tourism destinations. Results of this research revealed the tourists’ behavior in 8 tourism destinations (Bueng Chawaak Chalerm Phrakiat, Phu Toey National Park, Hoop Khao Wong Reservior, Huay Tha Dua Reservoir, Ground Lizards Conservation Village, Phu Haang Naak Natural Rock Garden, Kra Seow Dam, and Kra Seow Dam Beach) considering the key 7 factors (attractions, accessibility, safety measures, amenities, activities, ancillary services, and the community involvement) as follows : Bueng Chawak Chalerm Phrakiat, Kra Seow Dam ,and Phu Toey National Park – high potential in attractions for tourists and variety of activities, Hoop Khao Wong Reservior and Ground Lizards Conservation Village- high potential in community involvement, Huay Tha Dua Reservoir – low potential in attractions for tourists, convenience in accessibility, amenities, and the community involvement, Phu Haang Naak Natural Rock Garden and Kra Seow Dam Beach – moderate potential in all expects together with low potential in ancillary services. Keywords: Tourism, Potential, Ecotourism. 1. Background The tourism industry of Thailand in the year 2017 was in the growing situation considering the increasing of foreign tourists which was 30.50-34.15 million or 2.8-4.8% in comparison to those in 2016. Receipt/revenue from this sector was 9.30-9.38 billion Baht or 7.0-8.0% in comparison to those in 2016 (Office of Tourism and Sports, Tourism Authority of Thailand (TAT), 2560: online). 252

Suphan Buri is one among 25 provinces of the central Thailand, located in the western part of the Kingdom and northwest of Bangkok, with the geographical characteristic of the agricultural community on the Taa Chine River Basin and forests. Suphan Buri Provinces is consisted of 10 districts, 110 sub-districts, and 1,007 villages ( Strategic Plan of Suphan Buri Province, 2017 : online) which houses more than 107 tourists’ attractions such as Paa Lay Lai Wora Wiharn Temple, Banharn-Jaamsai Observatory Tower, Mangorn Sawan Park, Don Chedi Royal Monument, Phai Roang Wua, Temple, Taap Kradaan Temple, Sam Chook’ Buffalo Village, 100 Years Market, Ground Lizard Conservation Village, Agricultural Promotion and Development Centre, Natural Rock Garden, Phu Haang Naak natural Rock Garden, Khao Thiem Temple, Kra Seow Dam, Phu Toey National Park, Kra Seow Dam Beach, Hoop Khao Wong Reservior, and Nong Ratchawat Archeological Site ( Namelist of the district’s attraction site, 2017 : online) The Strategic Plan of Suphan Buri Province for 2014-2017 was consisted of 5 areas as follows: 1) the enhancement of agricultural potential in connection to the industrial commercial sectors for the domestic consumption and the export, 2) the tourism development to reach the international standards aligning with the sustainable natural resources and environmental development, 3) the elevation of life quality and safety measures, 4) the promotion of education and sports towards the international standard, and 5) the development of public administration and public services. Result of the analysis of the Strategic Plan of Suphan Buri Province for 2014-2017 (Development Plan of Suphan Buri Province for 2014-2017) revealed the weakness of Suphan Buri Province in tourism sector in these following aspects: the lack of solidarity and unity of the entrepreneurs, the promotion of local products, and the distribution of the supportive infrastructures and ancillary services. Those are considered as the negative factors to the 2nd strategy for the “tourism development to reach the international standards aligning with the sustainable natural resources and environmental development”. Therefore, under the purposes of the studying of overall physical condition of the tourism destinations, the tourists’ behavior in the ecotourism destinations, and the potential of the ecotourism destinations in Suphan Buri Province, the researcher had decided to conduct research for the potential of tourism destinations in Suphan Buri Province under the expected result which would lead to the guidelines for the promotion of sustainable development of tourism industry and the economy of Suphan Buri Province. Literature Review For this study, Potential of Ecotourism Destinations in Suphan Buri Province. The researcher studied and researched data from various sources such as textbooks, research reports, papers by studying concepts, theories and research papers. This can be summarized as follows: Related concepts and theories 253

Typologies of Tourists’ Behaviors and the Tourism Development Cohen Erik (1993) classified the types of tourists’ behavior as: 1) the leisure, 2) the drifter,3) the explorer,4) the the discoverer, and 5) the local explorer. Whereas Swarbrook and Horner (1999) classified the types of tourists/travelers as :1) the sea travelers, 2) the friendship travelers, 3) the naturalist travelers, 4) the explorer, 5) the discoverer,6) the family travelers, and 7) the traditional travelers. Kosith Panpiamrath (2010) noted about the meaning of tourists’ destination development as the utilizing of existing resources which could be classified into 3 types as follows: the well-known area, the area with international potential, and the area with provincial potential. Whereas the Tourism and Education Commission of the Senate of Thailand noted the meaning of the development of tourists’ destination as the management of each area in accordance to its unique characteristic and the existing condition under the maintaining of its educational and environmental value. McIntosh Ritchies and Goeldner (1989) noted about the stakeholders in the tourism development which are: 1) the tourists (travelers looking for experiences and pleasures), 2) the entrepreneurs providing tourists’ services and goods, 3) the local government expecting the benefits from the tourist towards the local economy, and 4) the local people expecting the jobs opportunity from the cultural tourism sector. Whereas Sirinthip Phanmakkawan (1999) stated the problems of the eco-cultural tourism of Maha Sarakham Province which requires the effectively management due to the lack of related governments agencies and the insufficiency of budget. Kittisak Klinmuenwai (2001) suggested the tourism development guidelines as follows :1) the provincial’s development of natural and cultural attraction sites by utilizing its unique characteristic for the increasing of its economic value, 2) the enhancement of mutual cooperation between the public and private sectors for the development of the sites, 3) the concerned agencies’ enhancement of the community’s knowledge on the management of community tourism, 4) the promotion of community tourism by the local transportation such as horse chariot or bicycle, and 5) the creation of community tourism learning center /museum. 254

Ecotourism Elizabeth Boo (1991) defined the meaning of ecotourism as “travel to natural areas that conserves the environment by the provision of financial sponsorship from revenue obtained and by the improvement of the local people’s well-being as well as the enhancement of their awareness towards the natural conservation\". Whereas Western (1993) had another meaning of the ecotourism as “the responsible traveling to natural sites with the conservation of the sites and the development of community and local citizen’s quality of life. In conclusion, “ecotourism” could be defined as “the travelling to unique natural sites with awareness of the natural conservation aligning with the creation of learning environment under the local community involvement for their well being. Therefore, the key concept of the ecotourism is the management of tourism aligning with the development of people, community, and society by the utilization of the area’s existing potential under the community involvement and the integration of local context and its unique characteristic. (Thailand Institute of Scientific and Technological Research, 2007 in Ramkhamhaeng University, 2015). This is aligned with the concept of ecotourism in the research paper of Naowarat Plainoi (2002) on the studying of profession skills of the individual and group of workforce in the tourism industry in the upper northern region of Thailand (Nan, Chiang Mai, Chiang Rai, and Mae Hongson) where the development of desirable tourism is considering the key success factors. Factors for the Analysis of Potential of the Tourism Destinations Buhalis (2000) mentioned about the thinking framework for the analysis of tourists requirements regarding the development of tourism destinations which was consisted of 6 aspects (6As) as follows: (1) Attractions: was considered the most important factor for the tourism destinations development since it could effectively persuade the tourists to their desirable destination. The attractions destinations was classified into 4 types: natural, cultural, ethnical, and entertainment. For this aspect, Ajake, Anim Obongha (2016) said that, based on these findings, it is recommended that pursuance of all aspects of development in peripheral areas to improve destination and attraction sites thereby encouraging tourists from other parts of the world is essential. Also, public and private stakeholders should work together through recognizing destination marketing organizations to create promotional packages that possess the likelihood of presenting an image that can stimulate tourists' interest to visit the destinations. (2) Accessibility (3) Amenities (4) Available packages. (5) Activities. (6) Ancillary services. Those fact was also mentioned in Araya Inkotchasarn (2011) about the potential of Wat Klang Ku Wiang Floating Market which considered moderate due to the requirement for the improvement of services, community involvement, and the safety measures. 255

The community involvement and the safety measures were considered as another key factors for the consideration of the tourism destinations’ potential. Chartweerawat Prakhongjitmun (2012) mentioned about the success of development and promotion of the Koii Kee Walking Street which was the result of the effective collaboration between the local people and local government agencies, using the unique characteristic of the market with traditional wooden shop houses located on the bank of Mae Klong River. Research Concepts This research is aimed on the study of the general data and behavior of the tourists and the overall condition of the ecotourism area which is the independent variation of the potential of each area. In addition, this research is also aimed on the study of the tourism area’s potential considering the key 7 factors as follows: attractions, accessibility, safety measures, amenities, activities, ancillary services, and the community involvement The conceptual framework for the study is as follows. Tourists Data 7 Factors for the potential of the tourism destinations Condition of the tourism -Attractions tourism area -Accessibility -Safety measures -Amenities -Activities -Ancillary services -Community involvement Figure 1 Conceptual Framework Research Objectives 1. To study the general condition of tourist sites and tourist behavior of ecotourism sites in Suphanburi area. 2. To study the potential of ecotourism in Suphanburi area. 2. Methods Population was 2,021,529 tourists visiting the ecotourism destinations in Suphan Buri Province in 2016 (National Statistical Office, 2017: online). Number of the sample population determined by Taro Yamane’s formula was 398 and 2 substitutions (total of 400). The sampling methods were 1) in accordance to the research’s objectives, the purposive sampling method was applied 256

for the selection of the visitors from 8 aforementioned ecotourism destinations, and 2) the quota sampling was applied for the 400 sample population (50 sample for each ecotourism area). Tool for this research was the questionnaire sturctured in accordance to the result of literature review on academic articles, theses, research reports, online sources, and textbooks. Structure of the questionnaire was consisted of 2 parts which were : part 1 general data and behavior of the tourists ( collected data was analyzed by the statistic method of percentage) ,and part 2 questionnaire on the tourists’ opinion towards the 7 factors of the potential of ecotourism destinations in Suphan Buri Province (collected data was analyzed by the statistic method of average, standard deviation, and weighting score equation) for the consideration of the potential of the destinations. In addition, there was the comparison of the destinations’ potential with the designated criteria in order to determine the current condition and situation of the area which was classified in 5 levels: 1- very low, 2- low, 3-moderate, 4- high, 5- very high. 3. Results 3.1 Research Results Part 1: the condition of the 8 ecotourism destinations were as follows: (1) Bueng Chawaak Chalerm Phrakiat the natural wetland covering the area of 2,700 Rai (1,070 acres) located at 64 km away from Muang Supah Buri District and connect to Hankha District of Chai Naat Province and Derm Bang Nag Buat District of Supah Buri Province. This destinations is comprised of aquamarine creatures display zone, zoo, and the local plants and vegetable garden in commemorative to the King. (2) Phu Toey National Park – this only national park of Suphan Buri Province located in the area of Daan Chaang District is known as the greenery forest with number of sheer rocks and precipitous mountains. The highest peak is known as “Yod Khao Tewada (peak of the deity)”. There are number of nteresting places such as the Merkus Pine forest, Ta Poen Klee Waterfall, Phaa Yai Waterfall, Tard Yai Waterfall, Phu Krathing Waterfall, Nakhee Cave, Mee Nhoi Cave, Hoi Rayaa cave, and Phaa Yai Cave. (3) Hoop Khao Wong Reservior (Suphan Buri’s Paang Oong)- the latest tourists’ destination known as the natural site with fresh air which was officially opened to the public in December,2015. (4) Huay Tha Dua Reservoir- due to the beauty of its location surrounded by the lush mountains, this reservoir was one among the most well-known destinations among tourists. (5) Ground Lizards Conservation Village- a new ecotourism destination known as the study center for the life of ground lizards and the prototype village of the sufficiency economy philosophy. (6) Phu Haang Naak Natural Rock Garden- the rock garden which houses a million years prehistoric rocks and petrified woods that visitors can enjoy drawing their imagination and interpretation of its shape, is considered for the carefully conservation as the “Natural Heritage of Suphan Buri”. (7) Kra Seow Dam – this largest rolled-earthfill dam in Thailand is also used as the large aquamarine farm in Suphan Buri Province. (8) Kra Seow Dam Beach – this freshwater beach appearing in 257

the dry season blessed with the white sand is known as another new attraction site of Suphan Buri Province. Part 2: The analysis results on the general data and the tourists' behavior. Analysis results shows the data regarding the tourists and their behaviors as follows: Bueng Chawak Chalerm Phrakiat 70% of tourists are male at the age of 46-60 with the domicile of birth in Suphan Buri, whereas 40 % are from another provinces such as Singh Buri, Chai Naat, Nakhon Phathom. Most of the tourists are accompanied by family members by their own vehicles. Phu Toey National Park - 70% of the tourists are male, 40% are at the age of 15-35 and another 40% is 46-60. 60% of their domicile of birth are Bangkok. Most of the tourists are accompanied by family members with 80% usage their own vehicles. Hoop Khao Wong Reservior there is the equivalent number of the male and female tourists with 60% of age at 31-45. 60% of their domicile of birth are Suphan Buri 50% of the tourists are accompanied by friends with 70% usage of their own vehicles. Huay Thaa Dua Reservior 80% of the tourists are male at the age of 31-45 with 90% of the domicile of birth in Suphan Buri. 60% of the tourists are accompanied by friends with 86% usage their own vehicles. Lizard Conservation Village 62% of the tourists are female at the age of 31-45 with 68% of the domicile of birth in Suphan Buri. 70% of the tourists are accompanied by family members with 90% usage their own vehicles. Phu Haang Naak Natural Rock Garden - there is the equivalent number of the male and female tourists. More than 50% of their age are 15-30 with 70% of the domicile of birth in Bangkok. Most of the tourists are accompanied by friends and family members by the usage of their own vehicles. Kra Seow Dam- there is the equivalent number of the male and female tourists, with the age of 15-45 and 72% of the domicile of birth in Suphan Buri whereas those in Bangkok. and 90% of both groups use of their own vehicles. Kra Seow Dam Beach – 68% of tourists are female with the 60% of age at 15-35, whereas 30% are at the age of 31-40. 84% of their domicile of birth are Suphan Buri and those in Bangkok. 70%of the tourists are accompanied by friends with the usage of their own vehicles. Part 3: analysis results on the ecotourism destinations considering 7 factors (table 2) Table 2 Mean of the potential level of the 8 ecotourism destinations. Factors Bueng Chawak Chalerm Phrakiat Phu Toey National Park Hoop Khao Wong Reservior Huay Thaa Dua Reservior Ground Lizard Conservation Village Phu Hang Nak Rock Garden Kra Seow Dam Kra Seow Dam Beach 1. Attractions 3.96 4.32 258

Mean high Very high 3.84 2.54 3.07 2.99 3.41 3.29 Potential level high low moderate moderate high moderate 2. Accessibility Mean 3.56 2.56 3.27 2.48 3.06 3.30 3.39 3.40 Potential level high moderate moderate low moderate moderate moderate moderate 3.Safety measures 3.90 3.04 3.84 2.62 2.21 3.23 3.23 2.61 Mean high moderate high moderate low moderate moderate moderate Potential level 4. Amenities 3.77 2.99 3.27 2.31 2.45 3.23 3.43 3.00 Mean high moderate moderate low low moderate moderate moderate Potential level 5. Activities 3.65 3.65 3.36 2.56 3.18 3.14 3.51 3.19 Mean high high moderate low moderate moderate high moderate Potential level 6.Ancillary 3.61 3.53 3.02 2.43 2.54 2.44 3.11 2.69 Services high high moderate low low low moderate moderate Mean Potential level 2.84 2.54 3.89 2.40 3.85 2.94 3.04 2.69 7. Community moderate low high low high moderate moderate moderate Involvement Mean Potential level Analysis results from the table 2 shows the tourists’ opinion for the ranking of the ecotourism destinations in accordance to the level of its potential as follows: Bueng Chawak Chalerm Phrakiat- ranked as the highest potential ecotourism destination regarding the factors of attractions, amenities, and ancillary services which are aligning with the mean of 3.96,3.90, 3.77,3.65,3.61 and 3.56 respectively, whereas the community involvement is considered as the moderate at the mean of 2.84. Phu Toey National Park - ranked as the highest and high potential ecotourism destination regarding the factors of attractions, activities, and ancillary services which are aligning with the mean of 4.32, 3.65 and 3.53 and 3.56 respectively. Safety measures, amenities, accessibilities, are consider moderate at the mean of 3.04,2.99 and 2.56 respectively ,whereas the community involvement is considered as the low level at the mean of 2.54. Hoop Khao Wong Reservior- ranked as the high potential in community involvement and attractions with the mean of 3.89 and 3.84, moderate potential in activities, safety measures, accessibilities, and ancillary services at the mean of 3.38,3.36,3.27 and 3.24 respectively, whereas the amenities is considered as low potential at the mean of 3.02. Huay Tha Dua 259

Reservior- overall potential and factors are ranked low by the mean of attractions, ancillary services, amenities, accessibilities, community involvement, and activities, at 2.56, 2.54, 2.48, 2.43, 2.40 and 2.31 respectively, whereas the safety measures is considered as the moderate level at the mean of 3.62.Ground Lizards Conservation Village- ranked as the high potential destinations at the mean of 3.85 regarding the community involvement ,and moderate in attractions, accessibilities, activities which are at the mean of 3.18, 3.07 and 3.06 respectively. Whereas the ancillary services, amenities, and safety measures are considered as low potential at the mean of 2.54, 2.45 and 2.21 respectively. Phu Hang Naak Natural Rock Garden- overall potential and factors are ranked high in activities, safety measures, amenities, accessibilities, and community’ s involvement which are at the mean of 3.30, 3.23, 3.23, 3.14, 2.99 and 2.94, whereas the ancillary services is considered as low potential at the mean of 2.44. Kra Seow Dam - ranked as the high potential in activities, and attractions at the mean of 3.51 and 3.41, whereas the safety measures, amenities, accessibilities, and community involvement are considered moderate. Kra Seow Dam Beach - overall potential and factors are ranked moderate in attractions, activities, safety measures, amenities, accessibilities, and community involvement at the mean of 3.40, 3.29, 3.00, 3.19, 2.69, 2.69 and 2.61 respectively. Discussion According to the aforementioned results, the discussions of this research are as follows: 1. Bueng Chawak Chalerm Phrakiat, Kra Seow Dam ,and Phu Toey National Park are considered as the high potential and most popular ecotourism destinations regarding the factors of activities which is results of the continually development (Bueng Chawak Chalerm Phrakiat), its most beautiful natural ambient of Suphan Buri Province (Phu Toey National Park) and the its location as the most relaxing destination (Kra Seow Dam).This is in accordance to the research of Ajake, Anim Obongha (2016) who noted that, Based on these findings, it is recommended that pursuance of all aspects of development in peripheral areas to improve destination and attraction sites thereby encouraging tourists from other parts of the world is essential. Also, public and private stakeholders should work together through recognizing destination marketing organizations to create promotional packages that possess the likelihood of presenting an image that can stimulate tourists' interest to visit the study areas. 2. Hoop Khao Wong and Ground Lizards Conservation Village are considered as the high potential and most popular ecotourism destinations regarding the factors of community involvement. This is in accordance to the research of Chartweerawat Prakongjitman (2012) which noted that the key success factor of the development of Koii Kee Walking Street as the famous tourists’ destination is the result of the effective collaboration between the local people and local government agencies, using the unique characteristic of the market with traditional wooden shophouses located on the bank of Mae Klong River. 260

3. Huay Tha Dua Reservior is considered as the low potential ecotourism destination regarding the factors of attractions, accessibilities, amenities, ancillary services, and community involvement. This is in accordance to the research of Sirinthip Phanmakkawarn (2009) which noted about the problems of the eco-cultural tourism of Maha Sarakham Province which requires the effectively management due to the lack of related governments agencies and the insufficiency of budget. The communities with insufficient knowledge and experiences have to operate the community-based tourism activities on their own. Therefore, for the sustainable development of the community tourism, there should be a supportive efforts by the government for the development of the tourism system management, the enhancement of local people’s international language skills, the creation of new tourism activities, the conservation of existing cultures, and the cleanliness and safety measures of attraction sites. 4. Phu Haang Naak Natural Rock Garden and Kra Seow Dam Beach is considered as the moderate potential ecotourism destination in all factors of attractions, which is in accordance to the research of Kittisak Klinmuenwai (2011) which suggested the tourism development guidelines as follows :1) the provincial’s development of natural and cultural attraction sites by utilizing its unique characteristic for the increasing of its economic value, 2) the enhancement of mutual cooperation between the public and private sectors for the development of the sites, 3) the concerned agencies’ enhancement of the community’s knowledge on the management of community tourism, 4) the promotion of community tourism by the local transportation such as horse chariot or bicycle, and 5) the creation of community tourism learning center /museum. However, since the ancillary services of these two destinations are considered low, therefore, there is another suggestions for the improvement of the services such as the provision of ATM and mobile phone relay stations. This is in accordance to the research of Araya Inkotchasarn (2011) about the potential of Wat Klang Ku Wiang Floating Market which considered moderate due to the requirement for the improvement of services, community involvement, and the safety measures. The suggestions for the improvement for this area to the sustainable ecotourism site was the increasing of shops and food stalls, provision of local guides, promotion of the site via media, persuasion of local people’s involvement, and the elevating of the safety measures. Recommendation Huay Tha Dua Reservior – for the improvement of this ecotourism destination, it is highly suggested that there should be the efforts for the promotion and development of the attractions as well as the creation of its unique characteristics together with the maintaining of its own values. Together with the conservation of the values of place, the cleanliness must be the priority for the mentioned development as well. Phu Toey National Park and Hoop Khao Wong Reservior- for the improvement of these ecotourism destinations especially for the conveniences of accessibility, 261

it is highly suggested that the route to these destinations should be equipped with road and direction signs providing the accuracy information for the accessibility to the destinations. Ground Lizards Conservation Village - for the improvement of these ecotourism destinations especially for the safety measures against the criminal actions, it is highly suggested that there should be the criminal-warning signs together with the provision of the active security guards and the security station located in the area which is easy to observe and access. In addition to those measures, there should be the training for concerned personnel regarding the safety measures for the provision of tourists’ assistances in emergency situations, as well as the enhancement of cooperation with related agencies for the safeguarding of the tourists’ safety. 3.2 Acknowledgments The researcher would like to extend my sincerest gratitude to the kind support of the chief executive officers of Rajamangala University of Technology Suvarnnabhumi, as well as the kind cooperation of the valued tourists who provided times and information as the sample population of the research on the “Study of the Potential of Ecotourism Destinations in Suphan Buri Province”, which is currently completed successfully. 4. References Araya Inkotchasarn. (2011). Study for the Potential of Wat Klang Ku Wiang for the Ecotourism Destination. Chartweerawat Prakhongjitmun. (2012). Guidelines for the Development of Koii Kee Walking Street, Na Muang Sub-district, Muang District, Ratcha Buri Province. Development Plan of Suphan Buri Province of the Year 2014-2018- Commission of Information. (2017). Retrieved on July 26th, 2017 from www.oic.go.th/ Kiitsak Klinmuenwai. (2011). Guidelines for the development of community tourism in Lam Pang Province. Commission of Tourism and Education of the Senate. Tourism Management. Retrieved on July 26th, 2017, from www.senate.go.th. Kosith Panpiamrath. Natural Resources and Rural Development. Bangkok. Office of the National Council for Economic and Social Development Board, 2010 Namelist of the Tourists’ Attractions by the District. (2017). Retrieved on July 26th, 2017 from www.suphan.biz/ Naowarat Plainoi. (2002). Working Skills and the Conditions for the Development of Workforces in Local Tourism Industry: the Study in the Civic Groups of the Upper Northern Thailand Area. Office of Tourism and Sports, Suphan Buri Province. (2017). Retrieved on July 26th, 2017 from https://suphanburi.mots.go.th/more_graph.php Office of National Statistic. (2017). Domestic Tourism. Retrieved on July 26th, 2017 from http://statbbi.nso.go.th/staticreport/page/sector/th/17.aspx 262

Ramkhamhaeng University. (2005). Research Project for the Strategy Development of the Upper Area of Bhumibhol Dam for the Promotion of Conservative Tourism. Strategic Development Plan of Suphan Buri Provincial Administration Organization. (2017). Retrieved on July 26th, 2017 from www.suphan.go.th/content-4.html.suphantour.htm Sirinthip Phanmakkhawarn. (2009). The development and Management of Eco-cultural Tourism Destinations in Maha Sarakham Province. Ajake, Anim Obongha (2016).Tourism marketing strategies performance: evidence from the development of peripheral areas in Cross River State, Nigeria.(Online) from http://dx.doi.org/10.1007/s10708-015-9643-5 Buhalis, D. (2000). Marketing the competitive destination in the future. Tourism Management Vol. 21 No.1. pp. 97-116. Cohen Erik. “Hilltribes, Island and open-ended prostitution.” Thai tourism. 30 (1993): 48 Elizabeth Boo. (1991). Meaning of Ecotourism. Retrieved on August 20th, 2016 from www.msu.ac.th Mcintosh, R. W., Goeldner, C.R. Tourism : Principles, Practices, Philosophies.New York : john Wiley & Sons: 1989. Swarbrooke, J. and Horner, S. (1999) The green tourist - myth or reality. In Swarbrooke, J. and Horner, S. (eds) Consumer Behavior in Tourism. Butterworth-Heinemann, Oxford: 198-208. Western, D. (1993) .Defining Ecotourism. In: Lindberg, K. and Hawkins, D. (eds) Ecotourism: a guide for planners and managers. The Ecotourism Society, North Bennington, VT, pp. 7-11. 263

Economics/ Finance Thursday, January 24, 2019 08:45-10:15 Lailic Session Chair: Prof. Jörg Paetzold ISSSM-0390 Can Commuting Subsidies Increase Mobility? Evidence from a Regression Kink Design Jörg Paetzold︱Salzburg University ISSSM-0257 Uncertainty and the Declining Business Dynamics Puzzle De-Chih Liu︱National Taipei University ISSSM-0311 Housing Market Deleveraging Under Heterogeneous Household Preference: Case of Korea Joonhyuk Song︱Hankuk University of Foreign Studies ISSSM-0231 Using Deep Learning Neural Networks and Candlestick Chart Representation to Predict Stock Market Rosdyana Mangir Irawan Kusuma︱Yuan Ze University Trang-Thi Ho︱National Taiwan University of Science and Technology Wei-Chun Kao︱Omniscient Cloud Technology Yu-Yen Ou︱Yuan Ze University Kai-Lung Hua︱National Taiwan University of Science and Technology ISSSM-0377 Difference between Conventional, Transitioning and Organic Rice Yield: Case Study in the South of Thailand Kanjana Kwanmuang︱Office of Agricutural Economics 264

ISSSM-0390 Can Commuting Subsidies Increase Mobility? Evidence from a Regression Kink Design Jörg Paetzol Department of Economics and Social Sciences, University of Salzburg, Austria E-mail: [email protected] Abstract I exploit a kink in the benefit scheme of a large commuter tax break to study the effect of subsi- dizing commuting costs on the commuting distance of employees. My results show a significant increase in commuting distances exactly at the income level where the subsidy becomes more generous. I test the robustness of this finding by using variation in the location of the benefit kink over time. My results show that commuting subsidies can indeed influence commuting decisions. This finding contributes to discussions about the efficacy of such subsidies, which often are justified on the grounds of making workers more mobile. Keywords: Public policy, commuting subsidy, commuting behaviour, taxation, administrative data JEL codes: H21, H24, J22, J38, R23, R41 1. Introduction This paper aims to study the effect of commuting subsidies on the mobility of employees. Specif- ically, it investigates whether commuting subsidies can increase the willingness of employees to commute further. Such commuting subsidies exist in many countries, sometimes included in gen- eral work-related deductions (e.g., France or Italy), designed as a single allowance for commuters (e.g., Germany, Netherlands, Denmark, Switzerland), or come in the form of tax-free benefits paid by the employer (e.g. in the U.S.) (see Potter et al. 2006 for an overview). In general, these sub- sidies are seen to incentivize wage earners to accept jobs that are more distant from their homes. A body of theoretical literature shows how such commuting subsidies should be designed in order to reach an efficient level of job search and commuting (e.g. Wrede 2001, Richter 2006, Borck and Wrede 2008, 2009). While there exists a larger number of empirical studies that examine the relationship between wages and commuting (see, e.g., Van Ommeren and Guti´errez-i-Puigarnau 2011, Mulalic et al. 2013, Guglielminetti et al. 2015), evidence of the effect of commuter tax breaks on commuting behaviour is relatively scarce. An exception is Agrawal and Hoyt (2017) who show how U.S. state income tax differentials distort commuting behaviour, but they do not analyze com- muting tax subsidies specifically. Boehm (2013) examines whether workers become more likely to switch job or move house when commuting subsidies are significantly reduced. Relatedly, Heuer- mann et al. 265

(2017) investigate to which extent firms compensate workers for their commuting expenses in response to a large reduction in the German commuting tax subsidy. Overall, the only limited amount of empirical research on the efficacy of such commuting tax subsidies seems surprising, given the non-negligible sums in tax refunds many countries spend on these subsidies. This paper provides novel graphical evidence of the effect of commuter tax subsidies on individuals’ commuting distances to work. I investigate the effect of such subsidies by exploiting a kink in the benefit scheme of the Austrian commuter tax break. In essence, the design of the Austrian commuter tax break implies lower effective costs of commuting for individuals with income above the first income tax bracket, other things being equal. I show that sorting of male wage earners around the first income tax bracket is minimal, and use a Regression Kink Design to uncover the effect of the commuter tax break on distance travelled to work. My results show a significant kink in the relationship between income and commuting distance exactly where the commuting subsidy becomes more generous. I find the commuting distance to discontinuously increase above the first income tax bracket, paralleling the rise of the commuting subsidy. I test the robustness of this result by using variation in the location of the first tax bracket threshold over time. Interestingly, the discontinuity between income and commuting distance shifts with the location of the first tax bracket. This works in favour of my hypothesis, suggesting that the commuter tax break has indeed an effect on the commuting distance of employees. Negative placebo tests with fake tax bracket thresholds corroborate this result. Furthermore, I find that the tendency to longer commutes above the first tax bracket threshold is also reflected in a higher take-up rate for the commuter tax break. Overall, my results indicate that commuting subsidies can influence commuting decisions. Specifi- cally, I find that a more generous compensation of commuting costs increases commuting distance. This is an important finding when discussing the efficacy of such subsidies, since they are often justified on the grounds of encouraging workers to increase their job search radius and to commute further for better job matches. 2. Institutional Setting To identify the effect of commuting subsidies on commuting distances, I exploit a kink in the benefit scheme of the Austrian commuter tax break. For wage earners above the first income tax bracket, the commuter tax break comes in the form of an allowance which reduces taxable income. Importantly, for such commuters above the first income tax bracket, the commuter tax break increases with commuting distance (see left side of Table 1). In contrast, for taxpayers with income below the first tax bracket, there exists only a flat commuter tax credit which does not increase with commuting distance. Thus, the commuter tax break creates a discontinuous budget set in terms of commuting expenses, where the extra cost of commuting an additional 266

kilometre to work is different for wage earners below and above the first tax bracket. Put differently, being above the first tax bracket results in a higher commuter tax break for a given commuting distance, other things being equal. This higher commuter tax break for taxpayers above the first income tax bracket should induce individuals to accept longer commutes. Table 1: Commuter tax break in the Austrian tax code (in EUR) Income above ftrst tax bracket Income below ftrst tax bracket (Pendlerpauschale) (Pendlerzuschlag) Commuting public transport public transport not applicable Distance available not available 2–20 km - 372 130 20–40 km 1,476 130 40–60 km 696 2,568 130 1,356 >60 km 2,016 3,672 130 Figure 2 illustrates the design of the commuter tax break by plotting the (after-tax) cash-value of the allowance versus gross income relative to the first income tax bracket. As depicted by the figure, for commuters with income below the first income tax bracket (indicated by the dashed vertical line), the cash-value of the commuter tax break is flat and EUR 130. In contrast, for commuters who earn above the first tax bracket, the cash-value of the commuter tax break increases with both income and driving distance. For all commuting distances, the compensation when being above the first tax bracket is higher (or at least the same) than compared to being below the first bracket. For instance when I suppose that an individual commutes 25km (which is the average distance of a claimant of the commuter tax break) and earns EUR 9,500, the commuter tax break would equal EUR 130. However, if the same individual would earn EUR 11,500 instead and thus be above the first income tax bracket, the commuter tax break would amount to around EUR 500. As a consequence, the effective costs of commuting differ for the two individuals and thus, I expect those above the first tax bracket to accept longer commutes. It is important to note that previous research has found Austrian taxpayers to misreport their driving distances in order to receive a higher commuter tax break (see Winner and Paetzold 2016). This non-compliance does not invalidate my research design used here, because the data I use to measure actual (i.e. ’true’) commuting distance is unrelated to the distance taxpayers report for tax purposes (see Section 3 for details). Furthermore, the possibility to misreport does not eliminate the fact that the commuter tax break increases substantially with commuting distance, since most cheaters game ’only’ at the margin of one of the distance bracket thresholds (i.e. the 20, 40, 60km thresholds; see also Winner and Paetzold 2016). If anything, the non-compliance turns the commuter tax break into a rather continuous function of distance, in 267

contrast to the step function intended by the tax law. This is also supported by the data, where I observe the amount of the commuter tax break to increase quite continuously with commuting distance (see Appendix Figure ??). 3. Data and Summary Statistics For the empirical analysis I use the Austrian Social Security Database (ASSD), a linked employee- employer dataset. Using the ASSD, I are able to link employees to their workplace, which allows to observe both the residence and workplace location of the individual at the zip-code level. I then use a route planner (as is commonly used in navigation devices) to calculate the commuting distance between the centroids of these two zip-codes, which will be my (outcome) variable of interest. This procedure for calculating travel distances to work has been used before to uncover behavioural responses of Austrian commuters (see, e.g., Paetzold and Winner 2016, Frimmel et al. 2017), and I will build on this approach. Importantly, the administrative purpose of the ASSD is to collect social insurance contributions of Austrian employees and their employers, which makes this data completely unrelated to any reporting regarding tax matters, such as the commuter tax break. Thus, by using the workplace location recorded by the ASSD to measure commuting distances, misreporting in my variable of interest (i.e. distance-to-work) should be negligible. I restrict my analysis along some dimensions. First, I exclude employees with more than one employer, since it is unclear which commuting distance I should assign to those workers. Sec- 268

ond, I follow previous studies and focus on workers who commute less than 100km and work for companies with less than 3,000 employees (see, e.g., Guglielminetti et al. 2015, Heuermann et al. 2017).4 Workers who live more than 100km away from their workplace are most likely weekly commuters, which have not been eligible for the commuter tax break during the time of my study (see Einkommenssteuergesetz 1988). Furthermore, the ASSD provides no clear provision on whether the employer identifier is used for a company or for single establishments of a (larger) company. Sometimes only the headquarter of a company with several establishments is recorded in the ASSD. This would inflate my variable of interest (i.e., commuting distances), since I assign such employees a much greater distance (i.e., to the headquarters) than where they actually work (i.e., the local establishment). Finally, I remove companies where zip-code information is missing, or which are only listed with P.O. boxes. In sum, the final sample consists of approximately 8 million person-year observations, stemming from 2,385,560 unique individuals. Descriptive statistics are given in Table 2. In the first column I present descriptive statistics for the whole sample. In the third column, I provide descriptives for taxpayers around the first income tax bracket, defined as having gross income within EUR 5,000 on each side of the tax bracket. For the whole sample, I find that taxpayers are on average 40 years old, 46 percent are female, and 21 percent work part-time. The mean gross income is EUR 23,695, and more than 32 percent of all taxpayers file for the commuter tax break. The average commuting distance to work for all taxpayers is around 17km. Looking at taxpayers around the first tax bracket, I find a mean age of 38 years, 73 percent women, and the share working part-time being 44 percent. The mean gross income is EUR 12,061, and around 26 percent file for the commuter tax break. The average travel distance to work is ca. 15km. The latter restriction leaves my findings qualitatively unaffected. Results available upon request. 269

4. Identification Strategy To identify changes in the commuting distance in response to the asymmetric incentives below and above the first income tax bracket, I use a Regression Kink Design (RKD). Hence, I exploit the discontinuity in treatment intensity (i.e. in the cash-value of the commuter tax break) as a function of a continuous assignment variable (i.e. gross income). In other words, I use the change in the slope of the treatment function at the tax bracket to identify changes in the willingness to commute longer. The identification assumption is that unobserved determinants of the commuting distance evolve smoothly around the first tax bracket. Following standard practice (see, e.g., Card et al. 2015a), this identifying assumption can be evaluated by analysing whether there is i) a manipulation of the assignment variable, and ii) a discontinuity in covariates at the first tax bracket. To assess i), I inspect gross income distributions around the first tax bracket for women and men separately (see Panel A of Figure 2). The female income distribution (series in diamonds) shows a clearly visible hump around the first tax bracket, suggesting potential behavioural responses of women to it. In contrast, there is no evidence for manipulation of the assignment variable in the case of male wage earners, indicated by the lack of bunching in the male income distribution (series in dots).5 Using other methods to detect manipulation of the assignment variable confirm this null result for male wage earners (e.g., employing a McCrary-test does not detect a break in the density at the first tax bracket, results available upon request). Hence, I proceed with using male wage earners only in my analysis.6 Second, I evaluate whether observable characteristics of male wage earners are smooth around the first tax bracket. In the spirit of Card et al. (2015a), I construct a ’covariate index’ - the predicted commuter distance of wage earners using underlying personal characteristics as regressors. Thus, I build a vector of predetermined covariates including information on age, education, working part-time, years of job tenure, and region. Panel B of Figure 2 plots the mean value of the covariate index against EUR 100 bins of gross income. The index moves reasonably smooth across the first tax 270

20000bracket (the vertical line), with no evidence of discontinuities in these variables. 17Finally, it should be noted that the identification strategy does not allow to fully isolate the commuting decision from the residential mobility decision. However, I find that a change in workplace is much more prevalent in Austria than a change in residency, in line with previous research (see, e.g., Guglielminetti et al. 2015). Specifically, I find that only 5% in my sample change residency during the period of my study, compared to 27% who change the workplace. In addition, I exploit a shift in the location of the benefit kink and focus on job movers and new entrants to the labour market in order to isolate the commuting decision (see Figure 4). It turns out that the influence of residency changes does not alter my main findings, which considerably facilitates the interpretation of my analysis. 5. Empirical Results Figure 3 plots the actually observed commuting distance of wage earners against EUR 100 bins of gross income. Please note that I plot the entire population of wage earners irrespective of whether a person receives the commuter tax break or not, since take-up might be endogenous.9 Two things are worth noting in Figure 3. First, I see that commuting distance increases with income, which is in line with previous empirical evidence (see, e.g., Mulalic et al. 2014, Guglielminetti et al. 2015). Second, the figure shows a clear change of slope occurring exactly at the point where the commuter tax break becomes more generous. The solid lines represent best-fit linear regressions estimated on the microdata separately for observations above and below the kink. The hypothesis that the two slopes are equal is rejected with a p-value smaller 271

than 0.01. Thus, the commuting distance increases discontinuously above the kink, mirroring the rise in the cash value of the commuter tax break. In Table 3, I present regression results for the regression-kink coefficient (change in the slope) with and without demographic control variables and for different bandwidths. I estimate the effect of the commuter tax break on the commuting distance using the following linear model: Yit = α + βRit + δ1(Rit > 0) ∗ Rit + γXit + ηt + sit (1) where i denotes the individual and t the year. Rit is (binned) gross income before deducting the commuter tax break, the parameter δ measures the treatment effect, the change in the slope at the first tax bracket. Xit are a set of control variables, ηt are year fixed effects and sit is the error term. To account for correlation in the error term at a level higher than the individual I cluster my standard errors at income groups of EUR 100 (Bertrand et al. 2004). First, I find that when including control variables in my regression, the estimated size of the kink remains very similar and is always statistically significant. This reassures us that individuals to the left and to the right of the first tax bracket are indeed very comparable regarding their characteristics. Second, I find that the size of the kink at zero somewhat decreases with increasing bandwidth. However, even when using a very large bandwidth of 5000, the kink remains to be highly statistically significant. Given that only few people qualify for a commuter tax break large enough to create a discontinuous budget set in commuting expenses several thousands EUR above the first tax bracket (e.g., only 10% of all commuters receive the 60km+ commuter tax break), the smaller kink for large bandwidths seems not surprising. 272

15 16 17 18To substantiate the claim that it is the discontinuity in the benefit scheme of the commuter tax break which induces people to accept longer commutes, I use variation in the location of the first income tax bracket over time (see, e.g., Landais 2015). In 2009, the location of the first tax bracket and hence, the location of the discontinuity in the generosity of the commuter tax break was shifted by EUR 1,000 to EUR 11,000.10 Thus, I can study whether the change in the location of the first tax bracket corresponds with a change in the location of the break in commuting distances. Since only job movers and new entrants to the labour market may respond to this new location of the first tax bracket, I focus on this subgroup. Figure 4 plots the relationship between the commuting distance and gross income for all taxpayers from 2005-2008 (empty circles), and for all job movers and new entrants from 2009-2011 (full circles), respectively. There is a clear kink in this relationship for the 2005-2008 group at the location of the 2005-2008 income tax bracket (bin 0). Interestingly, this kink disappears for the 2009-2011 group, but a new kink appears at the location of the shifted 2009-2011 income tax bracket (bin 10). Furthermore, in the interval between the 2005-2008 and 2009-2011 tax bracket, there is a change in slope in the relationship between the commuting distance and gross income. The fact that the location of the change in slope shifts with the location of the first tax bracket is supportive of an effect of the commuter tax break on commuting distances. 273

6. Conclusion Many OECD countries provide tax subsidies for commuting. These subsidies are usually justified by assuming that financial incentives to commute will make workers more willing to take up jobs that are further away from their homes. While this seems theoretically well understood, it has been difficult so far to provide direct empirical evidence of it. In this paper, a kink in the benefit scheme of the Austrian commuter tax break is exploited to study its effect on commuting distances. Applying a Regression Kink Design I observe commuting distances to discontinuously increase above the benefit kink where the commuting subsidy becomes more generous. Using variation in the location of the benefit kink over time I find that the willingness to commute further shifts in accordance of the kink. In sum, my results indicate that commuting subsidies can affect commuting distances. This presents an important finding when discussing the efficacy of such subsidies, which are often justified on the grounds of encouraging workers to increase their job search radius and to make them more mobile. In light of the non-negligible amounts many governments spend on such commuting subsidies, this finding also carries important policy implications. 7. References Agrawal, D. and W. Hoyt. 2017. ‘Commuting and Taxes: Theory, Empirics and Welfare Implica- tions’, Economic Journal forthcoming. 274

Bertrand, M., Duflo, E. and S. Mullainathan. 2004. ‘How much should we trust differences-in- differences estimates?’, The Quarterly Journal of Economics 119(1), 249–275. Boehm, M. 2013. ‘Concentration versus re-matching? Evidence about the locational effects of commuting costs’, CEP Discussion Paper 1207, 1–44. Borck, R. and M. Wrede. 2008. ‘Commuting subsidies with two transport modes’, Journal of Urban Economics 63, 841–848. Borck, R. and M. Wrede. 2009. ‘Subsidies for intracity and intercity commuting?’, Journal of Urban Economics 66, 25–32. Brueckner, J.K. 2005. ‘Transport subsidies, system choice, and urban sprawl?’, Journal of Regional Science and Urban Economics 35, 715–733. Bundesministerium fu¨r Finanzen. ‘Bundesrecht konsolidiert: Einkommensteuergesetz 1988 §16, aktuelle Fassung’. url: https://www.ris.bka.gv.at/Dokumente/Bundesnormen/NOR40174029/NOR40174029.pdf Card, D., Lee, D.S., Pei, Z. and A. Weber. 2015a. ‘Inference on Causal Effects in a Generalized Regression Kink Design’, Econometrica 83(6), 2453–2483. Card, D., Johnson, A., Leung, P., Mas, A. and Z. Pei. 2015b. ‘The Effect of Unemployment Benefits on the Duration of Unemployment Insurance Receipt: New Evidence from a Regression Kink Design in Missouri, 2003-2013’, American Economic Review 105(5), 126– 130. Chetty, R., Friedman, J.N., Olsen, T. and L. Pistaferri. 2011. ‘Adjustment costs, firm responses, and micro vs. macro labor supply elasticities: Evidence from Danish tax records’, Quarterly Journal of Economics 126, 749–804. Frimmel, W., Halla, M. and J. Paetzold. 2017. ‘The Intergenerational Causal Effect of Tax Evasion: Evidence from the Commuter Tax Allowance in Austria’, Working Paper, 1–55. Guglielminetti, E., Lalive, R., Ruh, P. and E. Wasmer. 2015. ‘Spatial Search Strategies of Job Seekers and the Role of Unemployment Insurance’, Working Paper, 1–84. Heuermann, D., Assmann, F., Vom Berge, P. and F. Freund. 2017. ‘The distributional effect of commuting subsidies - Evidence from geo-referenced data and a large-scale policy reform’, Regional Science and Urban Economics 67, 11–24. Landais, C. 2015. ‘Assessing the Welfare Effects of Unemployment Benefits Using the Regression Kink Design’, American Economic Journal: Economic Policy 7(4), 243-278. Mulalic, I., van Ommeren, J. and N. Pilegaard. 2013. ‘Wages and commuting: quasi-natural experiments’ evidence from firms that relocate’, Economic Journal 124, 1086–1105. aetzold, J. and H. Winner. 2016. ‘Taking the high road? Compliance with commuter tax allowances and the role of evasion spillovers’, Journal of Public Economics 143, 1–14. Potter, S., Enoch, T., Black, C. and B. Ubbels. 2006. ‘Tax treatment of employer commuting support: an international review’, Transport Reviews 26, 221-237. Richter, W.F. 2006. ‘Efficiency effects of tax deductions for work-related expenses’, International Tax and Public Finance 13, 685-699. 275

Saez, E., Slemrod, J. and S.H. Giertz. 2012. ‘The elasticity of taxable income with respect to marginal tax rates: a critical review’, Journal of Economic Literature 50 (1), 3-50. Umweltbundesamt. ‘Umweltsch¨adliche Subventionen in Deutschland (2014) §37’. url https://www.umweltbundesamt.de/publikationen/umweltschaedliche-subventionen-in- deutschland-2014 Van Ommeren, J. and E. Guti´errez-i-Puigarnau. 2011. ‘Are workers with a long commute less productive? An empirical analysis of Absenteeism’, Regional Science and Urban Economics 41, 1–8. Wrede, M. 2001. ‘Should Commuting Expenses Be Tax Deductible? A Welfare Analysis?’, Journal of Urban Economics 49, 80–99. 276

ISSSM-0257 Uncertainty and the Declining Business Dynamics Puzzle De-Chih Liu Department of Economics, National Taipei University E-mail: [email protected] Abstract Some studies pointed out that the pace of business dynamism, as measured by job reallocation, in the United States. has declined after 2000. This study argues that uncertainty is potential an important factor in the evolution of job reallocation. In particular, we explore what degree could the job creation uncertainty and job destruction uncertainty respectively interpret the declining business dynamics. Moreover, we will explore whether the degree of interpretation will differ by firm age. This study will also assess the relative contribution of five uncertainty indices in the job reallocation process. Finally, this study argues that different regime switching feature of job flows uncertainty among firm age. Keywords: Uncertainty; job creation: job destruction 277

ISSSM-0311 Housing Market Deleveraging Under Heterogeneous Household Preference: Case of Korea Joonhyuk Song Hankuk University of Foreign Studies E-mail: [email protected] Abstract Korea's household debt has increased rapidly since the mid-2000s, and as a result, many researchers are concerned that there is not much time left to contain it from hampering real economic activities, such as consumption, investment, etc. In this paper, we examine the dynamic effect of deleveraging, which is a phenomenon in which household debt rapidly declines. For this, we first construct a dynamic general equilibrium model (DSGE) model that includes heterogeneous households and housing production, and analyze the macroeconomic effects of deleveraging. This study analyzes the assumption that deleveraging can occur when a housing market shock or a currency shock occurs, as the household debt of Korea is increasingly related to the housing market and the sensitivity of interest rates of households and other economic entities has increased. As a result, in the case of LTV 70% and the baseline model that does not respond to monetary policy in the case of deleveraging, real GDP and consumption deteriorated. In addition, the impact of the currency shock on the relative price of housing was stronger than that of the housing market shock in the short term. In order to see how the LTV ratio and the housing price adjustment affect the ripple effect of deleveraging, we conducted a hypothetical simulation and found that if the LTV ratio is low (50%), And the impact of the currency shock on the economy are mitigated compared to the baseline model. In the case of considering the housing price in the monetary rule, even if the housing market shock occurs, it does not affect the relative price of the housing investment or the house, and the effect on the real economy is somewhat eased. Finally, assuming both the strengthening of the LTV ratio and the adjustment of the monetary policy, the negative impact of the housing market shock on real GDP, inflation and consumption is relatively moderate. This implies that the risk of deleveraging can be controlled to some degree by performing macro-prudential policy through LTV ratio management and post crisis management through monetary policy. Keywords: Heterogenous Preference, Housing market, DSGE JEL Classification: E31, E52 278

ISSSM-0231 Using Deep Learning Neural Networks and Candlestick Chart Representation to Predict Stock Market Rosdyana Mangir Irawan Kusuma a, Trang-Thi Ho b, Wei-Chun Kao c, Yu-Yen Ou a, Kai-Lung Hua b a Department of Computer Science and Engineering, Yuan Ze University, Taiwan Roc b Department of Computer Science and Engineering, National Taiwan University of Science and Technology, Taiwan Roc c Omniscient Cloud Technology E-mail: [email protected] Abstract Stock market prediction is still a challenging problem because there are many factors effect to the stock market price such as company news and performance, industry performance, investor sentiment, social media sentiment and economic factors. This work explores the predictability in the stock market using Deep Convolutional Network and candlestick charts. The outcome is utilized to design a decision support framework that can be used by traders to provide suggested indications of future stock price direction. We perform this work using various types of neural networks like convolutional neural network, residual network and visual geometry group network. From stock market historical data, we converted it to candlestick charts. Finally, these candlestick charts will be feed as input for training a Convolutional Neural Network model. This Convolutional Neural Network model will help us to analyze the patterns inside the candlestick chart and predict the future movements of stock market. The effectiveness of our method is evaluated in stock market prediction with a promising results 92.2 % and 92.1 % accuracy for Taiwan and Indonesian stock market dataset respectively. The constructed model have been implemented as a web-based system freely available at http://140.138.155.216/deepcandle/ for predicting stock market using candlestick chart and deep learning neural networks. Keywords: Stock Market Prediction, Convolutional Neural Network, Residual Network, Candlestick Chart. 1. Introduction The stock market is something that cannot be separated from modern human life. The Investment in stock market is a natural thing done by people around the world. They set aside their income to try their luck by investing in stock market to generate more profit. Traders are more likely to buy a stock whose value is expected to increase in the future. On the other hand, traders are likely to refrain from buying a stock whose value is expected to fall in the future. Therefore, an accurate prediction for the trends in the stock market prices in order to maximize capital gain and minimize 279

loss is urgent demand. Besides, stock market prediction is still a challenging problem because there are many factors effect to the stock market price such as company news and performance, industry performance, investor sentiment, social media sentiment and economic factors. According to Fama’s efficient market hypothesis argued that it is impossible for investors to get advantage by buying underrated stocks or selling stocks for exaggerated price[9]. Therefore, the investor just has only one way to obtain higher profits is by chance or purchasing riskier investments. With the current technological advances, machine learning is a breakthrough in aspects of human life today and deep neural network has shown potential in many research fields. In this research, we apply different types of machine learning algorithms to enhance our performance result for stock market prediction using convolutional neural network, residual network, virtual geometry group network, k-nearest neighborhood and random forest. Dataset format in machine learning can be different. Many kind of dataset format such as text sequence, image, audio, video, from 1D (one dimension) to 3D (three dimension) can be applicable for machine learning. Taken as an example, the image is used not only as input for image classification, but also as an input to predict a condition. We take the example of Google DeepMind’s research in Alpha Go[4]. Recently, they are successfully get a lot of attention in the research field. By using the image as their input, where the image represents a Go game board, which later this image dataset is used to predict the next step of the opponent in the Go game. On the other occasion, from historical data of stock market converted into audio wavelength using deep convolutional wave net architecture can be applied to forecast the stock market movement[2]. Our proposed method in this work is using the represented candlestick charts of Taiwan and Indonesian stock markets to predict the price movement. We utilized three trading period times to analyze the correlation between those period times with the stock market movement. Our proposed candlestick chart will represent the sequence of time series with and without the daily volume stock data. The experiments in this work conduct two kind of image sizes (i.e. 50 and 20 dimension) for candlestick chart to analyze the correlation of hidden pattern in various image size. Thereafter our dataset will be feed as input for several learning algorithms of random forest and k-nearest neighborhood as traditional machine learning, CNN, residual network and VGG network as our modern machine learning. The goal is to analyze the correlation of some parameters such as period time, image size, feature set with the movement of stock market to check whether it will be going up or going down in the next day. 2. Related Work There are many researchers have been started to develop the computational tool for the stock market prediction. In 1990, Schneburg conducted a study using data from a randomly selected German stock market, then using the back-propagation method for their machine learning architecture [13]. To our knowledge, stock market data consist of open price data, close price data, high price data, low price data and volume of the daily movement activity. In addition, to use the 280

historical time series data from the stock market, some researchers in this field of stock market predictions began to penetrate the method of sentiment analysis to predict and analyze movements in the stock market. J. Bollen reported the sentiment analysis method by taking data from one of the famous microblogging site Twitter to predict the Dow Jones Industrial Average (DJIA) stock market movements[1]. There are more studies on stock market predictions; they use the input data not only by using elements of historical time series data, but by also processing the data into other different forms. (Borovykh, Bohte et al.) tried to use the deep convolutional wave net architecture method to perform analysis and prediction using data from S & P500 and CBOE [2]. We also found some related works using candlestick charts in their research. (do Prado, Ferneda et al. 2013) used the candlestick chart to learn the pattern contained in Brazilian stock market by using sixteen candlestick patterns[3]. (Tsai and Quan 2014) utilized the candle- stick chart to combine with seven different wavelet-based textures to analyze the candlestick chart[15]. While, (Hu, Hu et al. 2017) used the candlestick chart to build a decision-making system in stock market investment. They used the convolutional encoder to learn the patterns contained in the candlestick chart[5] while (Patel, Shah et al. 2015) used ten technical parameters from stock trading data for their input data and compare four prediction models, Artificial Neural Network (ANN), Support Vector Machine (SVM), random forest and nave-Bayes[11]. Traditional machine learning like Random Forest has been applied to predict the stock market with a good result. (Khaidem, Saha et al. 2016) combine the Random Forest with technical indicator such as Relative Strength Index (RSI) shown a good performance[7]. Adding more feature set can be one of the way to enrich your dataset and enhance the result of classification. According to (Zhang, Zhang et al. 2018) input data is not only from historical stock trading data, a financial news and users sentiments from social media can be correlated to predict the movement in stock market[16]. Different from most of existing studies that only consider stock trading data, news events or sentiments in their models, our proposed method utilized a representation of candlestick chart images to analyze and predict the movement of stock market with a novel to compare modern and traditional neural network. 3. Dataset 3.1 Data Collection Getting the right data in the right format is very important in machine learning because it will help our learning system go to right way and achieve a good result. We trained and evaluated our model on two different stock markets, i.e. Taiwan and Indonesia. We collected 50 company stock markets for Taiwan and 10 company stock markets for Indonesia based on their growth in technical analysis as a top stock market in both countries. In this data collection, we use the application program interface (API) service from Yahoo! Finance to get historical time series data for each stock market. From the period that we have been set in the following Table 1, we certainly get 281

some periods of trading day, starting from Monday until Friday is the period of trading day. Segregation of data based on predetermined time for data training and data testing is important, while some studies make mistakes by scrambling data; this is certainly fatal because of the data, which we use, is time-series. Table 1: The period time of our dataset, separated between the training, testing and independent data. Stock Data Training Data Testing Data Independent Data Start End Start End Start End TW50 2000/01/01 2016/12/31 2017/01/01 2018/06/14 2017/01/01 2018/06/14 ID10 2000/01/01 2016/12/31 2017/01/01 2018/06/14 2017/01/01 2018/06/14 3.2 Data Preprocessing From historical time series data, we converted it into candlestick chart using library Matplotlib[6]. To analyze the correlation between different period times with the stock market movement, we divided the data used to create candlestick chart based on three period times such as 5 trading days data, 10 trading days data and 20 trading days data. Besides the period time, we also di- vided our candlestick chart with and without volume indicator. Adding a volume indicator into candlestick chart is one of our approaches to find out correlation between enrich candlestick chart information and prediction result. 4. Methodology The architecture of our proposed method is shown in Figure 1. The first, we collected the data from stock market historical data using Yahoo! Finance API. After that, we applied the sliding window technique to generate the period data before using computer graphic technique to generate the candlestick chart images. Finally, our candlestick charts are feed as input into some deep learning neural networks model to find the best model for stock market prediction, and the outputs will be binary class to indicate the stock price will going up or down in the near future. 4.1 Candlestick Chart Candlestick chart is a style of financial chart used to describe the price movements for a given period of time. Candlestick chart is named a Japanese candlestick chart which has been developed by Japanese rice trader- Munehisa Hooma [10]. Each candlestick typically shows one day of trading data, thus a month chart may show the 20 trading days as 20 candlestick charts. Candlestick chart is like a combination of line-chart and a bar-chart. While each bar represents four important components of information for trading day such as the open, the close, the low and high price. Candlesticks usually are composed of 3 components, such as upper shadow, lower shadow and real body. If the opening price is higher than the closing price, then the real body 282

will filled in red color. Otherwise, the real body will be filler in green color. The upper and a lower shadow represent the high and low price ranges within a specified time period. However, not all candlesticks have a shadow. Candlestick chart is a visual assistance to make a decision in stock exchange. Based on candlestick chart, a trader will be easier to understand the relationship between the high and low as well the open and close. Therefore, the trader can identify the trends of stock market for a specific time frame [8]. The candlestick is called bullish candlestick when the close is greater than the open. Otherwise it is called bearish candlestick. Figure 2 and Figure 3 describe our candlestick chart representation in different period time and size with volume and without volume respectively. Figure 1: Our methodology design. Figure 2: Proposed candlestick chart without volume indicator in different period time and size. Figure 3: Proposed candlestick chart with volume indicator in different period time and size. 283

Input4.2 Learning Algorithm Conv2D-32In this work we will use some Deep Learning Networks (DLN) based on Convolutional Neural maxR-epoLoUlingNetwork to perform our classification on stock market prediction. Besides the DLN, we also Conv2D-48apply some traditional Machine Learning (ML) algorithms to compare with DLN. Those maxR-epoLoUlingtraditional Machine Learning algorithms are Random Forest and K-Nearest Neighbors algorithms. Dropout Conv2D-644.2.1 Convolutional Neural Network maxR-epoLoUling Conv2D-96Table 2: Our proposed CNN architecture. maxR-epoLoUling Convolutional neural network (CNN) is a feed-forward artificial neural networks which includes Dropoutinput layer, output layer and one or more hidden layers. The hidden layers of CNN typically Flattenconsist of pooling layers, convolution layers and full connected layers. It is similar to ordinary Dense-256Neural Networks (NN) made up of a set of neurons with learnable weights and bias. The difference Dropoutis Convolutional layers use a convolution operation to the input then transfer the result to the next Dense-2layer. This operation allows the forward function more efficient to implement with much fewer parameters. As shown in Table 2 ,our CNN model architecture consist of 4 layers of convolutional 2d, 4 layers of max pooling 2d, and 3 dropouts. 4.2.2 Residual Network It is an artificial neural network developed by He in 2015 [4]. It uses skip connections or short- cut to jump over some layers. The key of residual network architecture is the residual block, which allows information to be passed directly through. Therefore, the back propagated error signals is reduced or removed. This allows to train a deeper network with hundreds of layers. And this vastly increased depth led to significant performance achieves. 4.2.3 VGG Network The VGG network architecture was introduced by Simonyan and Zisserman[14]. It is named VGG because this architecture is from VGG group, Oxford. This network is characterized by its simplicity, using only 3x3 convolutional layers stacked on top of each other in increasing depth. Reducing volume size is handled by max pooling. Two fully connected layers, each with 4096 nodes are then followed by a softmax classifier. The 16 and 19 stand for the number of weight layers in the network. Unfortunately, there are two major drawbacks with VGGNet. First, it is 284

painfully slow to train and the second the network architecture weights themselves are quite large. 4.2.4 Random Forest Random Forest classifier is a classifier with consist of many decision trees and adopted the technique of random decision forest prioritizes predictive performance by using multiple learning algorithms (ensemble learning). In general, Decision trees are a learning methods used in data search technique. The method used by the idea of combining the” bagging” idea or called” Bootstrap Aggregating” (reduce variance) and the random selection of features in the training sets (classification and regression tree). The difference between Random Forest algorithm and the decision tree algorithm is that in Random Forest, the processes of finding the root node and splitting the feature nodes will run randomly. We applied our random forest algorithm from a machine learning python library called skicit-learn[12]. 4.2.5 K-Nearest Neighbors K-Nearest Neighbors (KNN) is a classifier with based on the Lazy learning and Instance-based (IBk) learning algorithms (selection K based value based on model evaluation method or cross validation). Further, Lazy learning is a learning method with the purposed to store training data and enables the training data is used when there is a query request is made (waits until it is given a test) by the system. Similarity measure applied to the KNN with the aim to compare every new case with available cases (training data) that has been previously saved. KNN adopted a supervised learning approach by utilizing the labeled data and this learning model of the algorithm can be used for classification and regression predictive problems. We also using skicit-learn python library for our KNN classifier. Furthermore, we used a K- D Tree algorithm in our KNN to perform prediction with default parameter from scikit-learn library. 4.3 Performance Evaluation There are some statistics measures of the performance evaluation to evaluate the result of all the classifiers by measuring the sensitivity (true positive rate or recall), specificity (true negative rate), accuracy and Matthew’s correlation coefficient (MCC). In general, TP is true positive or correctly identified, FP is false positive or incorrectly identified, TN is true negative or correctly rejected and FN is false negative or incorrectly rejected. Formulated as follows: 285

5. Experimental Results and Discussion In this section, we perform classification based on some traditional and modern machine learning algorithms (random forest, kNN, residual network, VGG, CNN) and then evaluate the performance of our best classification algorithm compared to three state-of-the-art methods [7, 11, 16] 5.1 Classification for Taiwan 50 Dataset Table 3: Summary result of Taiwan 50 with their best classifier for each trading days and image dimension. Classifier Period Dimension Sensitivity Specitivity Accuracy MCC CNN 5 50 83.2 83.8 83.5 0.67 with volume CNN 10 50 88.6 87.3 88.0 0.758 CNN 20 50 91.6 91.3 91.5 0.827 CNN 5 20 83.9 82.7 83.3 0.666 Random Forest 10 20 87.0 88.3 87.6 0.751 CNN 20 20 90.8 90.2 90.6 0.808 CNN 5 50 83.6 85.1 84.4 0.687 without volume CNN 10 50 89.2 88.1 88.7 0.773 CNN 20 50 93.3 90.7 92.2 0.84 CNN 5 20 84.8 83.0 83.9 0.678 CNN 10 20 88.0 88.2 88.1 0.761 CNN 20 20 81.7 91.4 91.0 0.817 From all experiments about Taiwan 50, we conclude a summary result with and without volume indicator for different trading days period and image dimension result. Table 3 shows that CNN in 20 trading days period with 50-dimension image and volume indicator is better than the others with 91.5% accuracy. In addition, without volume indicator for Taiwan 50, CNN in 20 trading days period with 50 dimension performs better than the others with 92.2% accuracy. From the 286

result of both of those experiments, it indicates that the method using CNN model with longer trading days period without volume indicator can achieve the best result for Taiwan 50 dataset. 5.2 Classification for Indonesia 10 Dataset Table 4: Summary result of Indonesia 10 with their best classifier for each trading days and image dimension. Classifier Period Dimension Sensitivity Specitivity Accuracy MCC ResNet50 5 50 80.7 85.4 83.1 0.661 with volume ResNet50 10 50 88.6 88.4 88.5 0.77 CNN 20 50 90.0 90.1 90.0 0.798 ResNet50 5 20 78.8 82.3 80.6 0.612 CNN 10 20 83.3 85.4 84.3 0.686 CNN 20 20 89.1 84.6 87.1 0.738 ResNet50 5 50 79.1 87.9 83.3 0.671 without volume CNN 10 50 87.5 86.6 87.1 0.74 CNN 20 50 92.1 92.1 92.1 0.837 CNN 5 20 83.4 82.4 82.9 0.658 CNN 10 20 85.4 85.6 85.5 0.708 VGG16 20 20 91.5 89.7 90.7 0.808 From all experiment results with Indonesia 10 dataset, we conclude a summary result with and without volume indicator in Table 4 respectively. It shows that the CNN method with 20 trading days period in 50 dimension using volume indicator show the best result with 90.0% accuracy. While the CNN method in 20 trading days period with 20-dimension image with- out using the volume indicator performs better result with 92.1% accuracy. It indicates that the method using CNN model with longer trading days period without volume indicator can achieve the best result for Indonesia 10 dataset. 5.3 Independent Testing Result 287

Measuring our model result not only used performance evaluation. We also performed an independent test to see that our proposed method is reasonable. During this independent test, we used two index stock exchange data from each country. Yuanta/P-shares Taiwan Top 50 ETF represented independent data test for our Taiwan50, whereas Jakarta Composite Index is our independent data set test for Indonesia10. Both of the stock exchange data are taken from 1st January, 2017 until 14th June 2018. Table 5 shows our independent test result for Taiwan50 using volume indicator and without using volume indicator respectively. The independent test result for Indonesia10 using and without using volume indicator are shown in Table 6 respectively. As shown in Tables 5 and 6, our CNN with 20 trading days period and 50- dimension image get best result for both independent test. 5.4 Comparison To further evaluate the effectiveness of our predictive model, we also compare our result with the other related works. The first comparison is between our proposed method with Khaidem’s work[7], they used three different stock market datasets with different trading period time. Samsung, General Electric and Apple are their stock market data with one, two and three months of trading period respectively. We applied our proposed model in their datasets to compare our 288

prediction performance with their result. The comparison result for Samsung, Apple, and GE stock market shown in Table 7 respectively. Based on these comparison results, it revealed that our performance results outperformed the prediction results from Khaidem’s work[7]. Table 7: Comparison result with Khaidem. Khaidem, Saha et al. Samsung Name Trading ACC Precision Recall Specificity Period Khaidemm 1 Month 86.8 88.1 87.0 0.865 Our 1 Month 87.5 88.0 87.0 0.891 Khaidem 2 Month 90.6 91.0 92.5 0.88 Our 2 Month 94.2 94.0 94.0 0.862 Khaidem 3 Month 93.9 92.4 95.0 0.926 Our 3 Month 94.5 94.0 95.0 0.882 Khaidem, Saha et al. Apple Khaidem 1 Month 88.2 89.2 90.7 0.848 Our 1 Month 89.6 90.0 90.0 0.863 Khaidem 2 Month 93.0 94.1 93.8 0.919 Our 2 Month 93.6 94.0 94.0 0.877 Khaidem 3 Month 94.5 94.5 96.1 0.923 Our 3 Month 95.6 96.0 96.1 0.885 Khaidem, Saha et al. GE Khaidem 1 Month 84.7 85.5 87.6 0.809 Our 1 Month 90.2 90.0 90.0 0.86 Khaidem 2 Month 90.8 91.3 93.0 0.876 Our 2 Month 97.8 98.0 98.0 0.993 Khaidem 3 Month 92.5 93.1 94.5 0.895 Our 3 Month 97.4 98.0 98.0 0.983 Table 8: Comparison result with Patel. S&P BSE SENSEX NIFTY 50 ACC F-Measure ACC F-Measure Patel 89.84 0.9026 89.52 0.8935 Our 97.2 0.97 93.4 0.93 Reliance Industry Infosys Patel 92.22 0.9234 90.01 0.9017 Our 93.9 0.94 93.9 0.94 Second comparison is between our proposed method with J. Patel’s work[11]. They utilized four 289

different stock market datasets from India stock exchange. In this comparison, we followed their dataset using Nifty50, S7P BSE Sensex, Reliance Industry and Infosys stock market datasets. Accuracy and F-measure were used for their performance evaluation. As comparison results shown in Table 8, Our proposed model yielded 97.2 %, 93.9 %, 93.4 % and 93.9 % for accuracy with S7P BSE Sensex, Reliance Industry, Nifty50 and Infosys stock market datasets respectively. It indicated that our proposed method is superior to Patel work [11]. Table 9: Comparison result with Zhang. Hong Kong - Zhang Accuracy MCC Zhang 61.7 0.331 Our 92.6 0.846 Last comparison is between our proposed method with Zhang’s method[16]. Their dataset composition is similar with us. They are using thirteen Hong Kong stock market, whereas we used fifty Taiwan stock market datasets and ten Indonesia stock market datasets. Their methodology is combine sentiment analysis on social media and finance news. As shown in Table 9, Our proposed method achieved 92 % significantly outperforms Zhang method [16]. 6. Conclusions and Future Works In this study, we present a new method for stock market prediction using 2 stock market datasets including 50 company stock markets for Taiwan50 datasets and 10 company stock market for Indonesian datasets. The first, we employ the sliding window technique to generate the period data. To find out correlation between enrich candlestick chart information and stock market prediction performance, we utilized the computer graphic technique to generate the candlestick chart images for stock market data. Finally, an CNN learning algorithm is employed to build our prediction for stock market. We found that the model using long-term trading days period with CNN learning algorithm achieves the highest performance of sensitivity, specificity, accuracy, and MCC. It is proved that Convolutional neural network can find the hidden pattern inside the candlestick chart images to forecast the movement of specific stock market in the future. Adding the indicator such as volume in candlestick chart not really help the algorithms increase finding the hidden pattern. The comparison experiments indicated that our proposed method provide highly accurate forecast for other datasets compare to the other existing methods. Patel used trading data from Reliance Industries, Infosys Ltd., CNX Nifty and S & P Bombay Stock Exchange BSE Sensex during 10 years with accuracy in the range of 89 % - 92 % while we achieved accuracy in the range of 93 % - 97 %. Khaidem method achieved the accuracy in the range of 86 % - 94 % using three trading 290

data from Samsung, GE and Apple while we achieved in the range of 87 % - 97 %. Zhang utilized 13 different companies in Hong Kong stock exchange with accuracy 61 %. Meanwhile, our method achieved 92 % for accuracy. For the future works we want to extend our work being able to predict the percentage change on the price movements. For the convenience of experimental scientists, we developed a user- friendly webserver for predicting stock market using our final model Available at http://140.138.155.216/deepcandle/, DeepCandle is a system through which users can easily predicting stock market in the near future. Users only need to input the target date, and our models will process them and return the prediction result of stock market movement on that target date. The provided web interface is constructed such that users can easily access its functions and comfortably use it without a deep understanding of computing. 7. References [1] J. Bollen, H. Mao, and X. Zeng. Twitter mood predicts the stock market. Journal of computational science, 2(1):1–8, 2011. [2] A. Borovykh, S. Bohte, and C. W. Oosterlee. Dilated convolutional neural networks for time series forecasting. [3] H. A. do Prado, E. Ferneda, L. C. Morais, A. J. Luiz, and E. Matsura. On the effectiveness of candlestick chart analysis for the brazilian stock market. Procedia Computer Science, 22:1136–1145, 2013. [4] K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770– 778, 2016. [5] G. Hu, Y. Hu, K. Yang, Z. Yu, F. Sung, Z. Zhang, F. Xie, J. Liu, N. Robertson, T. Hospedales, et al. Deep stock representation learning: From candlestick charts to investment decisions. arXiv preprint arXiv:1709.03803, 2017. [6] J. D. Hunter. Matplotlib: A 2d graphics environment. Computing in science & engineer- ing, 9(3):90–95, 2007. [7] L. Khaidem, S. Saha, and S. R. Dey. Predicting the direction of stock market prices using random forest. arXiv preprint arXiv:1605.00003, 2016. [8] T.-H. Lu, Y.-M. Shiu, and T.-C. Liu. Profitable candlestick trading strategiesthe evidence from a new perspective. Review of Financial Economics, 21(2):63–68, 2012. [9] B. G. Malkiel and E. F. Fama. Efficient capital markets: A review of theory and empirical work. The journal of Finance, 25(2):383–417, 1970. [10] G. L. Morris. Candlestick Charting Explained: Timeless Techniques for Trading Stocks and Futures: Timeless Techniques for Trading stocks and Sutures. McGraw Hill Profes- sional, 2006. [11] J. Patel, S. Shah, P. Thakkar, and K. Kotecha. Predicting stock and stock price index 291

movement using trend deterministic data preparation and machine learning techniques. Expert Systems with Applications, 42(1):259–268, 2015. [12] F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, et al. Scikit-learn: Machine learning in python. Journal of machine learning research, 12(Oct):2825–2830, 2011. [13] E. Scho¨neburg. Stock price prediction using neural networks: A project report. Neuro- computing, 2(1):17–27, 1990. [14] K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014. [15] C.-F. Tsai and Z.-Y. Quan. Stock prediction by searching for similarities in candlestick charts. ACM Transactions on Management Information Systems (TMIS), 5(2):9, 2014. [16] X. Zhang, Y. Zhang, S. Wang, Y. Yao, B. Fang, and S. Y. Philip. Improving stock market prediction via heterogeneous information fusion. Knowledge-Based Systems, 143:236– 247, 2018. 292

ISSSM-0377 Difference between Conventional, Transitioning and Organic Rice Yield: Case Study in the South of Thailand Kanjana Kwanmuang Office of agricultural economics zone 8, Office of agricultural economics, Thailand E-mail: [email protected] 1. Background In Thailand, organic agriculture is now under the interest of government policy, which “National Organic Agricultural Development Plan 2017-2021” has been implemented. This plan aims to increase both organic area and the number of organic product farmers as well as enhancing market channels and the organics standard (Office of Agricultural Economics, 2017). Organic farming in Thailand is increasing by 20.97% from 2014 to 2015 with the exported values of 1,817.80 million baht. Especially, more than half or 59.07% of whole organic farming is rice farming (Earth Net Foundation and Green Net Cooperative, 2016). Even though organic farming particularly rice farm is increasing, its ratio of organic rice planted is only 0.27% compared with producing conventional rice ones, and its production is less than consumer’s demand. A small number of farmers changed into organic farms, it is more severe in the southern part of Thailand where the main production are perennial crops like rubber and palm oil. There are many constraints of changing to organic farm, for instance, time for transitioning, farmers attitude, lack of organic and market knowledge (Piyachan et al.,2009), organic standard (Lukrak et al., 2013), organic farming requires labor-intensive operation (Antos et al.,2009). Moreover, the important concern is yield. There was debate about yield gap between organic and conventional farming. USDA data and researchers revealed that organic yields significantly lower than with conventional farming. However, this study tries to fill the gap of literature by identifying the yield deference in each stages of transitioning from conventional to organic rice farm. the questions of this study are that, is its organic rice yield worth for changing? and how difference on yield each stage of changing to organic farming?. The purpose is to contribute to the understanding of features of changing on rice yield by each stage, conventional, transitioning and organic rice, in particularly southern part of Thailand. 2. Methods This study was conducted in mainly provinces for rice production in southern, Nakorn Sri Thammarat and Phattalung where organic rice farms have been practiced and started since last 3 years. A survey was conducted among organic, transitioning and conventional rice farmers by an in-deep interview with the head of those farms’ households. 391 sample of rice farms were selected by statistical sampling. These 391 farm households consist of (1) 92 farms household who started produce organic farm for one year or in the year 2017, it represented as the first year of transitioning. (2) 60 farms household who started 293

producing organic farm for 2 years or in the year 2016, it represented as the second year of transition. (3) 64 farms household who started produce organic farm for 3 years or in the year 2015, or produced organic rice, it represented as organic rice farms. And (4) 175 farms household who produce conventional rice. The yield, production and cost of seasonal rice were colleted. In the first phase of this research, a simple T-test method was employed to estimate the yield difference between organic, transitioning and conventional rice farmers. 3. Results The result reveals that the yield of rice significantly sharply dropped when changing from conventional to organic rice on the first year of transition, or decreased from 479.73 to 305.96 kg/rai. However, the yield continues increasing from the first year of transitioning until totally changing to organic one. But only 3 years of organic produced cannot reach the yield of conventional rice farm. While, the price of organic rice is higher than the conventional one. However, some of production was keep for self-consumption. Yield (kg. per rai) Selling Price (baht/kg.) 600 479.73 15 12.83 12.37 400 305.96 324.23 358.99 200 10 8.94 10.66 0 5 Note: 1 rai = 0.16 hectares 0 conventional first year of second year organic rice rice transitioning of farms transitioning This comparison result may be a reason of hesitating of farmers to start organic farming, in order to increase the number of farmers, the strategies of supporting especially first-year transition are required. Keywords: organic farming, transitioning farmer, T-test estimation 4. References Antos, Florence Ivy M. & Park, Timothy A. & Escalante, Cesar L., 2009. \"The Impact of Labor Constraints on the Farm Performance,\" 2009 Annual Meeting, January 31-February 3, 2009, Atlanta, Georgia 46821, Southern Agricultural Economics Association. Earth Net Foundation and Green Net Cooperative (2016). Thailand organic farming situation [online]. Retrieved from http://www.greennet.or.th/article/411. Lukrak, N., Athinuwat, D., & Sindecharak T. (2013). Problems and Barriers in Changing to Organic Vegetable Production of Ratchaburi farmers Who Qualified in the Organic Farming Development Project. Thai Journal of Science and Technology. 2, 2 (2556): 125-133. (in Thai) 294

Office of Agricultural Economics (2017). National Organic Agricultural Development Plan 2560-2564 B.E. Office of agricultural Economics, Ministry of Agriculture and Cooperatives, Thailand. Piyachan, N., Suthumchainutata, C., Klaiprasit, P., &Ratthananupong, K. (2009). Oraganic fruit plantation Problem and threat; Case Study: in Rayong Chantaburi and Trat Province [online]. Retrieved from http://romphruekj.krirk.ac.th/books/2552/2/04.pdf. 295

Education (4) Thursday, January 24, 2019 10:30-12:00 Lailic Session Chair: Prof. Teng-Huang YU ISSSM-0343 Foreign Language Classroom Anxiety in Vietnamese Students Studying in English Class in the University in Taiwan Teng-Huang Yu︱Takming University of Science and Technology ISSSM-0320 The Estimation of Single Mediation Model in Variables with Measurement Errors Jia-Ren Tsai︱Fu Jen Catholic University Ching-Yun Yu︱University of Taipei ISSSM-0336 Evaluation of Satisfaction Toward Admission to Further Education of the Undergraduate Students in the Faculty of Education and Liberal Arts at Nakhonratchasima College, Thailand Saruda Chaisuwan︱Nakhonratchasima College Chalinee Paladprom︱Nakhonratchasima College Siriphon Suwanrangsi︱Nakhonratchasima College Thiraphat Witchayaphong︱Nakhonratchasima College Titima Panyong︱Nakhonratchasima College Kanthima Sawangwong︱Nakhonratchasima College 296

ISSSM-0339 Factors Affecting Decision to Study and Satisfaction Toward Admission to Further Education in Master Degree in Education of Nakhonratchasima College, Thailand Saruda Chaisuwan︱Nakhonratchasima College Wiralphat Wongwatkasem︱Nakhonratchasima College Chutima Prompouri︱Nakhonratchasima College Adisorn Bansong︱Nakhonratchasima College Naiyana Chumchong︱Nakhonratchasima College ISSSM-0346 The Relationship Between Characteristic of Teacher and Administrator and Effectiveness of Administration Schools under Local Administrative Organizations in the Lower Northern, Thailand Saruda Chaisuwan︱Nakhonratchasima College Wattana Jantanupan︱Northern College Suphawadi Jeckjuntuek︱Director Verification Center 3 Phenpphan Saengnet︱North-Chiang Mai University Nipon Saengnet︱Northern College Thinnakit Chaisuwan︱Northern College Bunsum Inkong-gnam︱Metheewudthikorn School Venus Manmungsil︱High Thai tension Master LTD 297