We have two parts for related works since our research focuses on the use of social network analysis on investment behavior: previous research on investment behaviors and previous research on social network analysis.
2.2 Related Research on Investment Behaviors
Prior studies on investment behaviors can be categorized into 6 categories based on the type of factors that drive investment behaviors.
1. Personal Opinions. Doran, Peterson and Wright (Doran et al., 2010) studied the role of personal opinions of finance professors on the efficiency of the stock market in the United States and found out that personal opinions do not affect investment behaviors. Rather, investment behaviors found in financial professors were largely driven by the same behavioral factor as amateur investors.
2. Investment Experience. Gorat et al., 2011 analyzed the differences in the investment behaviors of experienced and novice private equity firms and found out that novice firms tend to invest more slowly than experienced funds but the size and value of the funding size of novice firms tend to be larger.
3. Geographic Identities. Grinblatt et al., 2000 discovered that investment behaviors can be determined by the investors geographic identity: foreign investors in Finland tend to purchase past winning stocks and sell past losers. On the other hand, domestic investors sell past winning stocks and purchase losing stocks.
4. Online versus Offline Communities. Tan et al., in 2011 explored the roles played by online and offline communities and discovered that offline communities are more influential over investing behaviors. This is expected since offline communities often mean that interaction are offline and hence more likely in person, thus increasing the level of influence.
5. Psychology. Bakker et al., in 2010 on the other hand investigated into psychological factors that impact market evaluation and found out that trust and social influence affects the stability of investment markets.
6. Genetics. Barnea et al in 2010 investigated the relationship between genetics and investment behavior by studying the investment behaviors of identical and fraternal twins. They discovered that “a genetic factor” explains up to a third of twins investing behavior, though not long lasting.
•‧
2.3 Related Research on Social Network Analysis
There are numerous studies on social network analysis. More importantly in recent years we begin to see the marriage of social network analysis with management science, computer science and other fields, giving rise to what most of us term as “social computing” or “network science”.
Common social network analysis topics and its relevant techniques and applications are, but not limited to centrality analysis (Leskovec et al 2010), community detection (Girvan et al., 2010, Newman et al., 2006, Leskovec et al., 2007) link prediction label prediction (Gallagher et al., 2008, Kajdanowicz et al., 2010), information diffusion (Leskovec et al., 2007, Kempe et al., 2003 ) and team formation (Lappas et al., 2009, Kargae et al 2011).
Other related work includes statistical features of networks (Nowell et al., 2007, Newman 2011) such as information networks, collaboration networks, biological networks and social networks.
The similarities of the above applications is that the use of social network analysis techniques often improve the performance of the solution for the given problem domain. We often see the use of algorithms or similarity measures ranging from Common Neighbors, shortest paths, Katz, PageRank, Jaccard Coefficient, Adamic/Adar etc or its variants to help provide measures.
Link prediction is one of the most important topics in social network analysis. Link prediction seeks to predict the changes in terms of edges or nodes of social networks over time. Link prediction in social networks can be problematic: Nowell and Kleinberg performed extensive studies on link prediction in social networks and noted that there is no singular technique that can ensure the best performance. In fact, the techniques used shows limited performance. The techniques used for link prediction include PageRank (Page et al 1998), HITS (Kleinberg 1998), Adamic/Adar Adamic et al., 2001), Jaccard Coefficient, shortest paths etc. Moreover, Nowell and Kleinberg proposed that performance may be improved by taking into account of node-specific information. More recently, link prediction has been applied to datasets in popular social networks, which includes Twitter, Facebook and others as covered by Leskovec, Huttenlocher, Krause, Guestrin and Falousos (2007, 2010) and Fire, Puzis and Elovici in 2011. These studies include the prediction of positive and negative links to recommending friends on Facebook to using computationally efficient topologic features.
•‧
Guang, Zheng, Wen, Hong, Rose and Liu (Xiang et al., 2012) performed studies using the CrunchBase dataset and predicted company acquisition with factual and topic features using profiles and news articles on TechCrunch. Although they made use of a similar dataset as our work, their work did not make use of social relations as part of their feature set and focused on a different domain of mergers and acquisitions. In particular, they made use of node information, such as age of company, number of financing rounds and categories in addition to news articles related to mergers and acquisitions to build machine learning features. On the other hand, this research makes use of social relationships, represented by social network features to predict the act of investments; mergers and acquisitions are not covered in my work.
2.5 Link Prediction as a Model to Predict Investor Behavior
The originality of this paper is that we propose the use of social relationship as the main feature to predict if investments will occur. For example, given an Investor and a Company, can we predict if the Investor will invest in that particular Company just by understanding their social relationships? We believe that this will be a much easier approach for companies seeking investments since they are more likely to understand their social relations with potential investors.
We opted to use link prediction as a way to model investor behavior instead of other social network analysis methods due to the following reasons:
1. We find that link prediction suits our problem as it sought to predict new links within a social network as time progresses. This is very similar to how investors and start-up investing operate:
as time progresses, will new links (investments) occur between different pairs of investors and companies. Link prediction usually focuses on addition of link and do not take into account of removal of links, which suits our problem perfectly: we hardly see an investor pull off their investment after an investment is made into a company.
2. Link prediction allows us to input different characteristics of individual entities, which also reflects reality of investment behaviors and transactions: investors and companies both reflect
•‧
國
立立 政 治 大
㈻㊫學
•‧
N a tio na
l C h engchi U ni ve rs it y
readily reflected using network structures (such as “closeness” using shortest paths and similarity using Jaccard Coefficient) and investor/company information such as age and industries.
3. In addition and as pointed out by Kleinberg and Nowell, prediction models that uses only a singular metric (such as common neighbors only) yield less than satisfactory results; by taking into account different metrics (shortest paths, Jaccard Coefficient, common neighbors, adamic/adar, preferential attachment and number of shortest paths) we can derive a more complete perspective of the network we are dealing with.
•‧
3.1 Prosperity Taiwan Project Backgroud
My research work is part of a larger research project dubbed “Prosperity Taiwan”. The main goal of
“Prosperity Taiwan” is to create a set of IT-enabled policies and workflows that aids Taiwan’s economic transformation. My thesis represents the area of financing: helping companies gather investment and or helping investors find suitable companies to invest.
3.2 Taiwan’s Economic Strengths and Current Economic Landscape
As one of the Four Asian Tigers, Taiwan’s economic growth from 1960s to 1990s has been exceptional, maintaining at a growth rate excess of 7% per year. Taiwan’s economic strength stems from its manufacturing capabilities, especially in terms of information technology related products. Bulk of Taiwan’s business comes from building and manufacturing white-label products that is later branded with American or European brands. However, Taiwan’s greatest economic strength turns out to be its biggest liability in recent years due to changes in the global economic landscape.
The present global economic landscape presents several challenges for the Taiwanese economy: increase in cheaper manufacturing alternatives such as manufacturers from China or Brazil. This leads to increased price competition, severely cutting profit margins for Taiwanese manufacturers. There is also an increased premium placed on brand recognition and intellectual property; despite Taiwanese’s manufacturing prowess when it comes to building world class hardware such as iPhone, iPods and other globally recognized devices, Taiwanese manufacturers often receive up to only 3 to 5% of the profit share of each device sold. This is because the Taiwanese manufacturers do not own the brand and or intellectual property of the device.
In response to the above challenges, the Taiwanese government highlighted several industries poised for growth including medical services, bio-technology, green energy, culture/arts/creativity and high technology agricultural.