Robert P. Schumaker

Textual analysis of stock market prediction using breaking financial news: The AZFin text system

Our research examines a predictive machine learning approach for financial news articles analysis using several different textual representations: bag of words, noun phrases, and named entities. Through this approach, we investigated 9,211 financial news articles and 10,259,042 stock quotes covering the S&P 500 stocks during a five week period. We applied our analysis to estimate a discrete stock price twenty minutes after a news article was released. Using a support vector machine (SVM) derivative specially tailored for discrete numeric prediction and models containing different stock-specific variables, we show that the model containing both article terms and stock price at the time of article release had the best performance in closeness to the actual future stock price (MSE 0.04261), the same direction of price movement as the future price (57.1% directional accuracy) and the highest return using a simulated trading engine (2.06% return). We further investigated the different textual representations and found that a Proper Noun scheme performs better than the de facto standard of Bag of Words in all three metrics.

Evaluating sentiment in financial news articles

Can the choice of words and tone used by the authors of financial news articles correlate to measurable stock price movements? If so, can the magnitude of price movement be predicted using these same variables? We investigate these questions using the Arizona Financial Text (AZFinText) system, a financial news article prediction system, and pair it with a sentiment analysis tool. Through our analysis, we found that subjective news articles were easier to predict in price direction (59.0% versus 50.0% of chance alone) and using a simple trading engine, subjective articles garnered a 3.30% return. Looking further into the role of author tone in financial news articles, we found that articles with a negative sentiment were easiest to predict in price direction (50.9% versus 50.0% of chance alone) and a 3.04% trading return. Investigating negative sentiment further, we found that our system was able to predict price decreases in articles of a positive sentiment 53.5% of the time, and price increases in articles of a negative sentiment 52.4% of the time. We believe that perhaps this result can be attributable to market traders behaving in a contrarian manner, e.g., see good news, sell; see bad news, buy.

A quantitative stock prediction system based on financial news

We examine the problem of discrete stock price prediction using a synthesis of linguistic, financial and statistical techniques to create the Arizona Financial Text System (AZFinText). The research within this paper seeks to contribute to the AZFinText system by comparing AZFinTexts predictions against existing quantitative funds and human stock pricing experts. We approach this line of research using textual representation and statistical machine learning methods on financial news articles partitioned by similar industry and sector groupings. Through our research, we discovered that stocks partitioned by Sectors were most predictable in measures of Closeness, Mean Squared Error (MSE) score of 0.1954, predicted Directional Accuracy of 71.18% and a Simulated Trading return of 8.50% (compared to 5.62% for the S&P 500 index). In direct comparisons to existing market experts and quantitative mutual funds, our systems trading return of 8.50% outperformed well-known trading experts. Our system also performed well against the top 10 quantitative mutual funds of 2005, where our system would have placed fifth. When comparing AZFinText against only those quantitative funds that monitor the same securities, AZFinText had a 2% higher return than the best performing quant fund.

A Discrete Stock Price Prediction Engine Based on Financial News

The Arizona Financial Text system leverages statistical learning to make trading decisions based on numeric price predictions. Research demonstrates that AZFinText outperforms the market average and performs well against existing quant funds.

Evaluating mass knowledge acquisition using the ALICE chatterbot: the AZ-ALICE dialog system

In this paper, we evaluate mass knowledge acquisition using modified ALICE chatterbots. In particular we investigate the potential of allowing subjects to modify chatterbot responses to see if distributed learning from a web environment can succeed. This experiment looks at dividing knowledge into general conversation and domain specific categories for which we have selected telecommunications. It was found that subject participation in knowledge acquisition can contribute a significant improvement to both the conversational and telecommunications knowledge bases. We further found that participants were more satisfied with domain-specific responses rather than general conversation.

An evaluation of the chat and knowledge delivery components of a low-level dialog system: The AZ-ALICE experiment

An effective networked knowledge delivery platform is one of the Holy Grails of Web computing. Knowledge delivery approaches range from the heavy and narrow to the light and broad. This paper explores a lightweight and flexible dialog framework based on the ALICE system, and evaluates its performance in chat and knowledge delivery using both a conversational setting and a specific telecommunications knowledge domain. Metrics for evaluation are presented, and the evaluations of three experimental systems (a pure dialog system, a domain knowledge system, and a hybrid system combining dialog and domain knowledge) are presented and discussed. Our study of 257 subjects shows approximately a 20% user correction rate on system responses. Certain error classes (such as nonsense replies) were particular to the dialog system, while others (such as mistaking opinion questions for definition questions) were particular to the domain system. A third type of error, wordy and awkward responses, is a basic system property and spans all three experimental systems. We also show that the highest response satisfaction results are obtained when coupling domain-specific knowledge together with conversational dialog.

Interaction Analysis of the ALICE Chatterbot: A Two-Study Investigation of Dialog and Domain Questioning

This paper analyzes and compares the data gathered from two previously conducted artificial linguistic Internet chat entity (ALICE) chatterbot studies that were focused on response accuracy and user satisfaction measures for six chatterbots. These chatterbots were further loaded with varying degrees of conversational, telecommunications, and terrorism knowledge. From our prior experiments using 347 participants, we obtained 33 446 human/chatterbot interactions. It was found that asking the ALICE chatterbots ¿are¿ and ¿where¿ questions resulted in higher response satisfaction levels, as compared to other interrogative-style inputs because of their acceptability to vague, binary, or clichE¿d chatterbot responses. We also found a relationship between the length of a query and the users perceived satisfaction of the chatterbot response, where shorter queries led to more satisfying responses.

Evaluating the efficacy of a terrorism question/answer system

The TARA Project examined how a trio of modified chatterbots could be used to disseminate terrorism-related information to the general public.

Machine learning the harness track: Crowdsourcing and varying race history

Racing prediction schemes have been with mankind a long time. From following crowd wisdom and betting on favorites to mathematical methods like the Dr. Z System, we introduce a different class of prediction system, the S&C Racing system that derives from machine learning. We demonstrate the S&C Racing system using Support Vector Regression (SVR) to predict finishes and analyzed it on fifteen months of harness racing data from Northfield Park, Ohio. We found that within the domain of harness racing, our system outperforms crowds and Dr. Z Bettors in returns per dollar wagered on seven of the most frequently used wagers: Win

Sports knowledge management and data mining

1.08 return, Place