Michael Rechenthin

My work in Iowa


Management Science Doctoral Student

Office : S283 Pappajohn Business Building
Hours : Friday 1:30-3:00pm
Cell    : (312) 725-0327
Email  :  michael-rechenthin@uiowa.edu

Welcome to my webpage. My name is Michael Rechenthin and I am a fourth year Ph.D. candidate in Management of Information Systems at the University of Iowa. I specialize in Machine Learning and Data Mining with a particular interest in high-frequency financial data. My advisor is Dr. Nick Street. This webpage highlights some information about my research interests and background.



News (updated February 26, 2012)

I am going to present to the UIowa Data Mining Group on Friday a paper by YeYe He entitled "Crawling Deep Web Entity Pages." This paper deals with algorithms for crawling databases on the Internet, which are often called "deep" or "invisible" or "hidden" webs.  Depending on who you ask, the "deep" web is 5 to 500 times more vast as the "shallow" web.  This makes for an interesting source of knowledge that is mostly untaped.

News (updated October 5, 2011)

This Friday I will present to the Data Mining Group my working paper "Stock chatter: Using stock sentiment to predict price direction."

During this talk, I'll present my research of examining Yahoo Finance message boards for knowledge discovery. First I will talk about how I acquired the data using Java and an open-source library. Next, I'll describe how I used Amazon Turk (crowd sourcing) to manually classify the messages into "positive", "negative", and "unrelated" classes for my training set. And third, I'll review how I used the algorithm to classify posts into appropriate sentiment classes.

News (updated July 27, 2011)

Today I present to the Data Mining Group a paper in which the authors compare different performance measures for binary classification algorithms.

News (updated July 12, 2011)

I finished my empirical study titled "Using conditional probability to identify trends in intra-day high-frequency equity pricing." In this paper, I look at the predictability of trade-by-trade data (up to a 30 minute timespan) using conditional probabilities of price movements. I examine, for example, the probability of an upward movement given two downward movements over different time frames. Is there predictability in high-frequency market data? I answer this question in the paper.

News (updated May 24, 2011)

Summer is here, which allows for focus away from classes and more toward research. I am working on a paper with my advisor, Dr. Nick Street, on using probabilities to show the existence of trend in high-frequency intraday equity data. Demonstrating that upticks versus downticks are systematic, rather than random in nature, should help traders make more informed decisions.

Secondly, I am extending my work on using sentiment analysis to cover multiple stocks.  The plan is to use sentiment to better predict stock volatility.

News (updated April 26, 2011)

35% of hedge funds at the end of 2010 were exploring the use of sentiment analysis in their models, up from 3% just a couple of years before (NY Times Dec 2010).  I examined a widely discussed stock in Yahoo Finance message boards and found that 3% of the users, write 63% of the post.  Furthermore, individuals whose profile accounts have been open for six months or less account for 65% of the user-provided sentiment (Strong buy, buy, hold, sell, strong sell).  I have results and a paper, and will discuss them May 3.

News (updated March 29, 2011)

Thursday, another doctoral student, Fahrettin Cakir, and I will present a paper by Branke et al. on using multi-objective evolutionary algorithms for optimizing portfolios.

News (updated March 1, 2011)

This Friday I will present to the Data Mining Group a paper where the authors use a genetic algorithm to find an ideal combination of technical analysis trading rules to maximize profit in the foreign exchange market.  This paper goes along well with some of my own research within the realm of using genetic algorithms for portfolio optimization.

News (updated Jan 10, 2011)

I completed the preliminary examination November 28 and received great guidance from my advisors. This semester, I continue my research under Dr. Nick Street on dynamic learning models for short-term (under 5 minutes) equity trading.

News (updated Nov 19, 2010)

I am currently scheduled to take my preliminary examination after the Thanksgiving holiday.

I am working on several different projects at the moment. The first project observes the probabilities of specific market conditions over different time-spans. The further that one looks into the future, the less that is known. I show this statistically.

I am also working on the analysis and optimization of a number of trading algorithms. How can we compare one model to another? What are different techniques of maximizing profit?

Finally, I am working with a fellow doctoral student, Si-Chi Chin, on a number of Bayesian statistical models for determining leading and lagging stocks. Correlations are interesting to analyze, but they don't infer causality. If stock A and B and are closely correlation, and both move up, can we determine one is moving a split-second before the other by using Bayesian inference? We examine this question and are building some really cool models to deal with this question.

News (updated Oct 15, 2010)

I just recently presented "Identifying and Exploiting Trends in Stock Market Data" to the Data Mining Group.

News (updated Dec 4, 2009)

Today I gave a talk on "Security Short-Term Trading" to the Data Mining Group here at the University of Iowa. I spoke about my eight years of trading experience at the Chicago Stock Exchange. Topics covered include: What moves markets? What is the difference between technical and fundamental analysis? How does one retrieve data to analyze the market? How can we data mine the market?