Machine Learning and Predictive Analytics in Accounting, Finance, and Management

Permanent URI for this collection

Browse

Recent Submissions

Now showing 1 - 5 of 7
  • Item
    To Treat, or Not to Treat: Reducing Volatility in Uplift Modeling Through Weighted Ensembles
    ( 2021-01-05) Rößler, Jannik ; Tilly, Roman ; Schoder, Detlef
    When conducting direct marketing activities, companies strive to know whom to target with a marketing incentive to maximize the campaign effect. For example, which customer should receive churn prevention incentive to minimize overall churn rate? Uplift modeling is a promising approach to answer such a question. It allows to separate customers who would likely react positively to a treatment from those who would remain neutral or even react negatively. However, while different uplift approaches have been proposed, they usually suffer from high volatility and their performance often depends largely on data set and application context. Thus, it is difficult for practitioners and researchers to apply uplift modeling. To overcome these problems, we propose a weighted ensemble of different uplift modeling approaches to reduce volatility and improve robustness. We evaluate the novel approach against single uplift modeling approaches on multiple data sets and find that the ensemble is indeed more robust.
  • Item
    Predicting stock price and spread movements from news
    ( 2021-01-05) Wistbacka, Pontus ; Rönnqvist, Samuel ; Vozian, Katia ; Sagade, Satchit
    We explore several ways of using news articles and financial data to train neural network machine learning models to predict shock events in high-frequency market data, and aggregated shock episodes. We investigate the use of price movements in this context, and separately at a daily interval as well. We describe in detail how training sets are created from our data sources and how our machine learning models are trained. We find that pairing company-related news text with events or movements in financial time series proves less straight-forward than the literature would indicate. We discuss possible reasons for negative results, especially relating to the combination of minute-level news and millisecond-level market data.
  • Item
    High-Frequency News Sentiment and Its Application to Forex Market Prediction
    ( 2021-01-05) Xing, Frank ; Hoang, Duc-Hong ; Vo, Dinh-Vinh
    Financial news has been identified as an important alternative information source for modeling market dynamics in recent years. While most of the attention goes to stock markets, the foreign exchange (Forex) market, in contrast, is much less studied. Most of the existing text mining research for the Forex market combine news sentiment with other text features, making the contribution of each factor unclear. To this end, we want to study the role of news sentiment exclusively. In particular, we propose a FinBERT-based model to extract high-frequency news sentiment as a 4-dimensional time series. We examine the efficacy of this news sentiment for Forex market prediction without involving any other semantic feature. Experiments show that our model outperforms alternative sentiment analysis approaches and confirm that news sentiment alone may have predictive power for Forex price movements. The sentiment analysis method seems to have a big potential to improve despite that the current predictive power is still weak. The results deepen our understanding of financial text processing systems.
  • Item
    Hierarchical learning for option implied volatility pricing
    ( 2021-01-05) Han, Henry
    Machine learning has been a popular option implied volatility pricing approach. It brings a good generalization in pricing by avoiding building different models for different options. However, it suffers from a relatively low prediction accuracy besides a model selection issue. In this study, we propose a novel hierarchical learning approach to enhance machine learning implied volatility pricing. It is designed for the ‘learning-hard’ problem and boosts different machine learning models’ performance for different option data on behalf of moneyness besides identifying the optimal learning models. In particular, the proposed hierarchical learning can be an excellent way to enhance implied volatility pricing for the option datasets with more noise. In addition, we find out-of-the-money options fit machine learning prediction better than the other options. This pioneering study provides a robust way to enhance implied volatility pricing via machine learning and will inspire similar studies in the future.
  • Item
    Follow the money: Revealing risky nodes in a Ransomware-Bitcoin network
    ( 2021-01-05) Turner, Adam ; Mccombie, Stephen ; Uhlmann, Allon
    This paper demonstrates the use of network analysis to identify core nodes associated with ransomware attacks in cryptocurrency transaction networks. The method helps trace the cyber entities involved in cryptocurrency attacks and supports intelligence efforts to identify and disrupt cryptocurrency networks. A data corpus is built by the unsupervised machine learning graph algorithm ‘DeepWalk’ [1]. DeepWalk evaluates the position of nodes within networks. It compares the relative position of different nodes (similarity) and identifies those whose removal would most affect the network (riskiness). This method helps identify on the blockchain the key nodes that are involved in the execution of a ransomware attack. When applied to the ransomware “cash out” graph, the method derived “riskiness” scores for specific nodes. Analysing the derived “riskiness” at a community level (groups of nodes in the network) provides an enhanced granularity for identifying and targeting influential nodes. Such insight could potentially support both intelligence and forensics investigations.