Deep learning for modeling the collection rate for third-party buyers

This study evaluates a wide range of machine learning techniques such as deep learning, boosting, and support vector regression to predict the collection rate of more than 65,000 defaulted consumer credits from the telecommunications sector that were bought by a German third-party company. Weighted performance measures were defined based on the value of exposure at default for comparing collection rate models. The approach proposed in this paper is useful for a third-party company in managing the risk of a portfolio of defaulted credit that it purchases. The main finding is that one of the machine learning models we investigate, the deep learning model, performs significantly better out-of-sample than all other methods that can be used by an acquirer of defaulted credits based on weighted-performance measures. By using unweighted performance measures, deep learning and boosting perform similarly. Moreover, we find that using a training set with a larger proportion of the dataset does not improve prediction accuracy significantly when deep learning is used. The general conclusion is that deep learning is a potentially performance-enhancing tool for credit risk management.     

Asymmetric effects of housing wealth on college enrolment

We investigate the impact of housing wealth, credit availability and financial distress on college enrolment decisions. We find that housing wealth is negatively related to enrolment in public schools and positively related to enrolment in private schools. This evidence suggests that, on average, students substituted away from private schools towards public institutions during the recent financial crisis.     

Structural holes and hedge fund return comovement: evidence from network-connected stock hedge funds in China

Using data from a new hedge fund database, we examine the impact of social networks on the return comovement of stock hedge funds in China. We use structural holes in the college alumni networks of managers to measure the managers’ social network positions. We perform an empirical analysis on a sample of 3,012 hedge fund products in China from 2010 to 2017. We find that greater structural holes are associated with higher return comovement. The positive impact of the structural holes on return comovement is not affected by market cycles, a manager's major in college, or his or her abilities.     

The impact on stock prices of deferral and elimination of internal control audit requirement for small firms

This paper examines the cumulative market reaction to the events related to deferral of internal control audit requirement under the Sarbanes-Oxley Act of 2002 and its elimination under the Dodd-Frank Act of 2010 for nonaccelerated filers (small firms). We document that small firms experienced negative cumulative abnormal returns around these events; and the differences between the cumulative abnormal returns for small firms and the two control groups (accelerated and large accelerated filers) were negative and significant at the 1% level. These results support the notion that market participants value the reliability of financial information irrespective of the firm size. Within the small firms, we find no firm characteristic significantly explains the market reaction to the events considered. That is, all small firms lost market value in reaction to the events that delayed and eliminated their internal control audit requirement.     

Data vs. information: Using clustering techniques to enhance stock returns forecasting

This paper explores the use of clustering models of stocks to improve both (a) the prediction of stock prices and (b) the returns of trading algorithms.We cluster stocks using k-means and several alternative distance metrics, using as features quarterly financial ratios, prices and daily returns. Then, for each cluster, we train ARIMA and LSTM forecasting models to predict the daily price of each stock in the cluster. Finally, we employ the clustering-empowered forecasting models to analyze the returns of different trading algorithms.We obtain three key results: (i) LSTM models outperform ARIMA and benchmark models, obtaining positive investment returns in several scenarios; (ii) forecasting is improved by using the additional information provided by the clustering methods, therefore selecting relevant data is an important preprocessing task in the forecasting process; (iii) using information from the whole sample of stocks deteriorates the forecasting ability of LSTM models.These results have been validated using data of 240 companies of the Russell 3000 index spanning 2017 to 2022, training and testing with different subperiods.     

Comparative study of deep learning models for analyzing online restaurant reviews in the era of the COVID-19 pandemic

Online reviews remain important during the COVID-19 pandemic as they help customers make safe dining decisions. To help restaurants better understand customers’ needs and sustain their business under current circumstance, this study extracts restaurant features that are cared for by customers in current circumstance. This study also introduces deep learning methods to examine customers’ opinions about restaurant features and to detect reviews with mismatched ratings. By analyzing 112,412 restaurant reviews posted during January-June 2020 on Yelp.com, four frequently mentioned restaurant features (e.g., service, food, place, and experience) along with their associated sentiment scores were identified. Findings also show that deep learning algorithms (i.e., Bidirectional LSTM and Simple Embedding + Average Pooling) outperform traditional machine learning algorithms in sentiment classification and review rating prediction. This study strengthens the extant literature by empirically analyzing restaurant reviews posted during the COVID-19 pandemic and discovering suitable deep learning algorithms for different text mining tasks.     

A neural network based price sensitive recommender model to predict customer choices based on price effect

The impact of price and price changes should not be ignored while designing algorithms for predicting customer choice. Consumer preferences should be modeled with consideration of price effects. Businesses need to consider for efficient prediction of an individual's purchase behaviour. Personalized recommendation systems have been studied with machine learning algorithms. However, the price-aware personalized recommendation has received little attention. In this paper, we attempt to capture insightful economic results considered in the marketing and economics disciplines by employing modern machine learning architecture for predicting customer choice in a large-scale supermarket context. We extract personalized price sensitivities and examine their importance in consumer behaviour. The employed data collected from a supermarket chain in Germany consists of implicit feedback based on customer-product interactions and the price of every interaction. We propose a two-pathway matrix factorization (2way-MF) model that is price-aware and tries to memorize customer-product interaction's implicit feedback. The proposed models achieve better model performance than standard Matrix Factorization models widely used in the industry. The approach was re-validated with data from supermarket chain in Taiwan. Other industries can adopt the proposed framework of modeling customer's preferences based on price sensitivity. We suggest that further research and analyses could help understand the cross-price elasticities.     

“双创”时代高校创客空间的复杂适应系统分析及优化策略

“大众创业、万众创新”时代，创客空间建设已成为高校扶持大学生创业的主流模式和重要举措。基于复杂适应系统（CAS）理论，阐释了高校创客空间的开放性、非线性、自组织性和涌现性等系统特征，指出高校创客空间的演进过程包括接触和交互两个阶段，在其演进过程中，标识机制和积木机制扮演着重要角色。最后，从CAS视角提出促进高校创客空间发展和系统优化的建议。     

Incorporating textual information in customer churn prediction models based on a convolutional neural network

This study investigates the value added by incorporating textual data into customer churn prediction (CCP) models. It extends the previous literature by benchmarking convolutional neural networks (CNNs) against current best practices for analyzing textual data in CCP, and, using real life data from a European financial services provider, validates a framework that explains how textual data can be incorporated in a predictive model. First, the results confirm previous research showing that the inclusion of textual data in a CCP model improves its predictive performance. Second, CNNs outperform current best practices for text mining in CCP. Third, textual data are an important source of data for CCP, but unstructured textual data alone cannot create churn prediction models that are competitive with models that use traditional structured data. A calculation of the additional profit obtained from a customer retention campaign through the inclusion of textual information can be used by practitioners directly to help them make more informed decisions on whether to invest in text mining.     

The role of text-extracted investor sentiment in Chinese stock price prediction with the enhancement of deep learning

Whether investor sentiment affects stock prices is an issue of long-standing interest for economists. We conduct a comprehensive study of the predictability of investor sentiment, which is measured directly by extracting expectations from online user-generated content (UGC) on the stock message board of Eastmoney.com in the Chinese stock market. We consider the influential factors in prediction, including the selections of different text classification algorithms, price forecasting models, time horizons, and information update schemes. Using comparisons of the long short-term memory (LSTM) model, logistic regression, support vector machine, and Naïve Bayes model, the results show that daily investor sentiment contains predictive information only for open prices, while the hourly sentiment has two hours of leading predictability for closing prices. Investors do update their expectations during trading hours. Moreover, our results reveal that advanced models, such as LSTM, can provide more predictive power with investor sentiment only if the inputs of a model contain predictive information.