I see myself as a part of the “Martyr” archetype. There are times where my parents find me passed out among my plethora of obscenely colored candy hordes and Mt. Dew Voltages (both of which do not actually manage to keep me awake at all times.) I’d like to think that I’m not the “Martyr” all the time, but I admit that when I have projects that I become borderline (or actually) obsessed about, it might be a few days before anyone sees me leave my desk.
Other times, I can’t sit still for more than an hour or two working on the same project, but sometimes, the spark of inspiration becomes more of a roaring bonfire of captivation that burns away everything (homework included, sometimes) but my work. I think that I’d like a bit more “Code Cowboy” in me in the sense that a person like that can always relax in situations where I would probably break into a possessed Visual Studios zombie and refuse to leave my computer for days. I’m pretty proud to admit, however, that I at least manage to pry myself away from the computer long enough to manage personal hygiene and grab more Mt. Dew.
Neural Networks and Their Use in Modeling Optimal Stock Market Operations
Stanford EPGY: 2013
Out of the many ways that people have devised to use artificial intelligence and computer science in order to predict that best stocks to buy, the foremost method has been through the use of NN—neural networks—to support machines and fuzzy systems to forecast the stock market price indexes based on past performances and recorded datum. Time series prediction plays a huge role in that it is used to predict everything from tree ring growth to sunspot appearance to when the best time to sell your AAPL stock is. Neural networks—otherwise known as Parallel Distributed Processing networks are systems that have the roots of their anatomy based upon the neural synaptic connections between dendrite in the human brain. This provides for optimal processing power and nearly limitless parallel processing capabilities given the right specs. A plethora of varying calculations can be made at the same time—thus providing for a powerful tool to make and cross-check hundreds if not thousands of possible paths simultaneously.
Neural networks can learn from their experience in order to improve their performances and adjust their behavior to fit the needs and demands on their surrounding environment. They can deal with data that other systems may find impossible to read as well as deal with human rules in a situations where the rules are blurred and not clearly defined.
- Paper Abstract/Summary
The purpose of this project is to provide a somewhat accurate system that utilizes neural networks in order to predict stock market fluctuations and values for any given day relative to a collection of data to other days that can be used as reference. The model follows one similar used by finance companies and hedge funds to put on tip sheets for their traders. Of course, one must use caution when letting a computer decide how they should invest their funds because no machine is perfect and as of such, human intervention and insight is often recommended if not necessary to make the most of the program.
The application of my project would be simple—to provide a coded neural networks that can accurate predict the optimal stocks to buy on any given day, similarly to how big investment corporations make the use of ‘tip sheets’. The network would be privy to many advantages while avoiding the many fallbacks of human stock market usage. People are impulsive and sometimes do not exercise an appropriate level of caution, computers function almost solely on logical decisions based on previous occurances. Neural networks are trained to operate and make decisions based on a multitude of databases of compiled data from decades of stock market operation. Thus, they have a better handle on the possible patterns that can occur and can better cross reference in order to check their decisions. They are also completely unbiased, which is beneficial in a line of work where people often let emotions rule their actions through unlucky impulses.
- Previous Work
There already exists prototypes such as stock100 and Piefers’s stock simulator but I believe that each has their drawbacks, for example, there isn’t one currently in full-time operation that accounts for time series cross references to regular non-biased datum. Stock100 uses:
- Current Problems in the Area
Admittedly, this project has many merits, but like any project, it has lots of drawbacks and problems as well. The model that I plan to develop shows possibilities of promise, but no computer can perfectly model a dynamic process such as the stock market all the time. Furthermore, abnormalities always exist and these abnormalities would cause the data to be heavily skewed and or incorrect altogether. Events such as natural disaster, crashes, and corporate actions are sometimes hard to factor in. The system only predicts within a relatively short amount of time with marginally diminishing levels of accuracy as people move further away from the last reference point of the amounts in the stock market.
- Proposed Solutions
I plan to combine Time Series analysis with Birgul Egeli’s algorithm in order to make a more optimized Neural network stock market predictor. I plan to use matrices containing values of stock up to 20 days prior to present days’ values and have the algorithm constant recheck values. Once certain patterns are verified, ‘neural’ connections between certain nodes would connect more often and bolster repetitions of paths along the network that allow for the highest accuracy.
Prediction of the stock market is of paramount importance to the field of finance and it would be downright irresponsible not to use to their full potential. In the future, more and more things that are currently done by people behind desks will be done by an artificially intelligent computer in cyberspace. The sooner people can learn to harness the utility of machines to make processes more efficient, the sooner people can learn to use their full potential in the coming age of computers.
- <Carter-Greaves, Laura. “TIME SERIES PREDICTION WITH FEED-FORWARD NEURAL NETWORKS: A Beginners Guide and Tutorial for Neuroph”
- <Dayan, Peter. “Theoretical Neuroscience: Computational and Mathematical Modeling of Neural Systems”. 2000.>
- <D. Ciresan, A. Giusti, L. Gambardella, J. Schmidhuber. Deep Neural Networks Segment Neuronal Membranes in Electron Microscopy Images. In Advances in Neural Information Processing Systems (NIPS 2012), Lake Tahoe, 2012.>
- <Gurney, K. (1997) An Introduction to Neural Networks London: Routledge.?
- Touretzky. Dave. “Artificial Neural Networks.”
- <Valentin Steinhauer. “STOCK MARKET PREDICTION USING NEURAL NETWORKS”.Neuroph.> <http://neuroph.sourceforge.net/tutorials/StockMarketPredictionTutorial.html>
My Preferred Book for Research:
Languages I Speak: Chinese (fluent), English (fluent), Spanish (fluent)
Languages I Program: C++ (fluent), C (fluent), Python (fluent), C♯ (semi-fluent), BASIC (fluent), ParallaxBASIC (fluent), LATEX (semi-fluent), Java (learning), LISP (semi-fluent)
Musical Instruments: Piano, Flute, Vocals
Sports: Tennis, Running, Archery
Clubs and Organizations: Mock Trial, Model United Nations, Speech and Debate, Pre-Medical Association, National Honor Society, Science Bowl, Spanish Club,
Fine Arts and Dance: Instructor (at Sun Century Art Academy: 2011-2013 present), Student (at Sun Century Art Academy: 2008- present), Pianist (Laurel Hills Retirement Home 2011-2013)
Jobs: Peer Tutor (2010- present), Art Instructor (2011- present), Computer Tech Support (2012- present), Computer Lab Assistant (2013-present)
Skills: Programming, Problem Solving, Public Speaking, Leadership, Systems Knowledge, Planning and Organizing, Team player and self directed.
Leadership Experience: Mock Trial (Captain), Model United Nations (Board Member, Ambassador), Speech and Debate (Senior Member, Team Co-captain), Pre-Medical Association (Board Member, Treasurer), Tennis (Varsity Captain), Art (Instructor)
Volunteer Experience: Librarian (2010-present), National Honor Society Coordinated Volunteer Activities (2012-present), Art Instructor (2010-present),
Hobbies and Interests: Building circuits, programming, tinkering with computers, building computers, sports, music, singing, online courses (Coursera Student 2012-present),
Past Summer Camps: Artificial Intelligence: EPGY at Stanford- Summer 2013
Recent Major Engineering/Computer Science Achievements:
• BHSE Science Fair (2012-2013)
o o ISEF Qualifier Top 5
o o Yale Best of Show Project (Regionals)
o o Best of Show Engineering and Computer Science Project
• Oregon State Science Fair (2013)
o o ISEF Qualifier Top 12 Finalist
o o IEEE Best Engineering and Computer Science Project
o o IEEE Best Medical Technology Project
o o Yale Best of Show Project (State)
o o Air Force Best Use of Innovative Technology
o o Navy Best Engineering Project
o o NWSE Best of Show Project
Projects (Note that since I am at camp at the moment, the only projects I have access to are the ones that we worked on in class D: ) :
o https://thegreatacatalepsy.wordpress.com/2013/07/25/ science-fair-project-circuit-board