LATEST PROJECTS
Research 01
Evaluating the risk of the Chinese borrowers’ default in peer-to-peer lending- using the Logistic regression model for ordinal response variables
The interpersonal lending has appeared since the historical record. The key difference between
the new P2P and previous interpersonal lending is that the borrows and lenders no longer need to
meet each other before the transition completed. Using data from a P2P lending platform in China,
this article explores the P2P load characteristics, evaluate the risk of the Chinese borrowers’ default.
The article finds that the ordinal categories of the number of late repayment days and the ordinal
logistic regression is much more precise than the binary variable and the binary logistic regression
when solving this problem. By using the ordinal logistic regression, the article finds that older male
and divorced or widowed borrowers who have children and have ever repaid late, with lower
education level, lower monthly earning and larger amount of load money are more likely to repay
late. The P2P lending platform in China must find ways to attract younger, married borrowers with
higher education level, higher monthly earning and great borrowing history.
Research 02
The analysis of NEEQ in China
The year 2015 has witnessed the booming of NEEQ (National Equities Exchange and
Quotations). The advantages of listing in the NEEQ, such as the policy support and the improvement
of stock liquidity have stimulated the enthusiasm of enterprises in a large extent. Driven by different
parties, the NEEQ has obtained the remarkable results. The number of the listed companies and
market capitalization volume are experiencing the most rapid growth since the NEEQ has been
established.
Research 03
The Reform of the Humanities-sciences Division in 1999
Humanities-Sciences division is under the system of the national college entrance examination. When high schools organize students, they have humanities and sciences two different types of specific classes with two teaching methods. Students can choose to join in the humanities or sciences classes according to their interests. The humanities classes mainly study humanities and social courses, while the science classes mainly study nature science. High school humanities sciences division is a form of education system that meets the needs of the national college entrance examination. This report mainly focuses on the reform of humanities-sciences division after the Great Culture Revolution, especially the reform in 1999 which had made the historical breakthrough. After this round of the reform, both the number of the students who attended the national college entrance examination and admitted rate has increased by a large extend. The report mainly describes the background of the policy, the pre- and post-policy, the implementation process and the experimental unit of the policy, the related data, the current research results of the humanities-sciences division at home and board and the conclusion.
Research 04
Cryptocurrency using the Statistical Learning
The goal of this report is to ascertain with what accuracy the direction of cryptocurrency price in USD can be predicted. By collecting the data of top 20 ranking cryptocurrencies from Apr. 28th 2013 to June 5th 2018 together, the report uses the statistical learning methods to predict the price trend and price of cryptocurrency next day. The best method to predict the next day trend of cryptocurrency is quadratic discriminant analysis (QDA), which produces the probability that the prediction is correct to be 54.13%. The best method to predict the next day price of cryptocurrency is random forests model, which produces the MSE to be 14566.2.
Research 05
Artificial Intelligence and Machine Learning in Financial Service - Fraud Detection in Cryptocurrency Exchanges Reported Volume
Artificial intelligence and machine learning can be used widely in financial service, especially in
surveillance and fraud detection. In March 2019, the news of fraud trade volume in cryptocurrency
exchange has make a big attention. According to the LDA Model and Decision Tree Model, the
trading size, bid ask spread, the standard deviation of bid ask spread and trading size are the most
important index to detect the fraud of Bitcoin exchange. The Bitwise Asset Management even
called for the issuers to field applications for bitcoin or bitcoin future ETFs to avoid the fraud of
volume and better detection. The Bitcoin ETF or the index created by the variables listed below
in my machine learning method can intend to provide direct exposure to bitcoin, priced off the
equivalent of a crypto consolidated tape, while custody assets at a regulated, insured, third-party
custodian.
