Warren Buffet Clone: Predicting the Stock Market
by Corey Scamman, Jarek Gozdieski, George Mazzeo, and Joe Elvin
Wall Street has always been a fast paced environment with insane amounts of capital moving left and right. Wall Street, seemingly the physical manifestation of the US stock market, has always been about making money. With the advent of AI and its increasing applications, AI is starting to firmly plant itself on Wall Street. AI and machine learning have a lot of applications on Wall street ranging anywhere from automated high volume trading on cent profits to stock predictors trying to guess trends in the market. We will be focusing on the latter. AI based stock market predictors are becoming normalized nowadays as there are tutorials and articles about how to build one online. Hedge funds and investment banks are now putting more resources into using computer science and AI to help improve their investments and risk management. However, predicting stock prices is extremely complex and difficult and so there is still much left to be desired for market predictors.
Stocks are complex financial assets. Stock prices are determined by the companies earnings, potential to grow, the political climate, the state of the economy, public perception, and a number of other factors. For this reason, predicting them is a very tall order. For AI stock predictors to be effective and even possibly successful, they have to be extensive. In our implementation of our AI stock predictor, we shoot to incorporate as many of these factors as we could.
For our final project, we attempted to make an AI market stock predictor using neural networks, natural language processing, and rule based systems to produce a more holistic model and incorporate many of the factors that go into stock prices. We are working with stock data on Apple taken from Yahoo Finance. Our data is continuous from January 2000 to November 2019 and tells us the open, close, high, low, and volume. We pass this information into a LSTM neural network. We use an LSTM model because these models build off the previous iteration and since our data is continuous stock info it made the most sense. We wanted to use natural language processing to utilize sentiment analysis and analyze articles online. With stocks, articles and speculation from highly respected investors can sway stock prices and so by processing these articles our goal was to account for this speculation. With our rules based system, we used separate states to mimic states of the economy. For example, one state would be when the economy is in a bull market. When the economy is in a bull market, the market is doing well and people expect higher consumerism and more spending. This would result in stock prices going up. For a more in depth explanation of our implementation and results, see our final research report.
Stock predictors are revolutionizing Wall Street, but they come with some inherent flaws that stem from the very way they work. All machine learning learns to do things by iterating through massive sets of data and determining patterns within the data. This is no different for stock predictors. Stock predictors must be trained individually on each stock, as not all stocks are created equal. Observing the difference in trajectories between Amazon stock and Snapchat stock proves this point with ease. Furthermore, all stock predictors are not trained on data from 2008 to the early 2010s because of the housing market crash. Predictors trained on this data would return some skewed and incorrect predictions. They are susceptible to biases in data and the intentional feeding of misinformation as well can damage the integrity of stock predictors. The most frightening flaw of stock predictors are their ability to crash the economy and create financial bubbles. An example of this are mass buy-outs and sell-outs. In addition to what I mentioned earlier, stocks are also priced based off the participants in the market. Should many players in the market have stock predictors that all predict that stock x is going to go down, they will all sell stock x therefore making stock x go down regardless of other factors. If enough people do this, they can crash the stock.
When it comes to science and making revolutionary technological advancements, we often ask ourselves if we can do it instead of whether or not we should do it. There is so much financial and societal gain that can occur from automation we rarely think of the consequences. Stock predictors, even in their adolescence now, have the potential to disrupt and negatively affect peoples’ lives. Automation can put people out of jobs, and in this case it would be low level investment banking and stock trading jobs. Additionally, stock predictors can widen the already massive income gap in our society. The stock market is a zero sum game. For someone to make money off of a stock, someone else must be losing money off that very same stock. Those with the resources to have stock predictors will be at an advantage compared to those without stock predictors, and if stock predictors work as we hope them to, those with the predictors will the majority of the time be making money off of those without them, thus widening the income gap. This can snowball into a number of scenarios, such as the emergence of even more powerful and large hedge funds than today.
In many other Utopian and dystopian machine learning scenarios, there is a distinct good and bad. For example, utopia for automated driving would result in no traffic, no more accidents, and a fleet of autonomous vehicles that bend to the will of society. The dystopia would be a fleet of autonomous cars that are hacked and wreak havoc on society. For stock predictors, there is only dystopia as even the intended utopia would result in dystopia. If we are to think of 100% stock predictor accuracy to be the intended utopia, with further thought we see that it would be the end of the stock market altogether. There are two scenarios and it is contingent on information equality. If stock predictors have 100% trend accuracy and only the select hedge funds have access to that info, they would make money off of every trade and everyone else outside of those organizations would have no chance of making money and stop altogether. If everyone has access to the accurate stock predictions, people would stop buying and selling. If someone has stock x and the prediction is that it will go up and everyone knows this, no one will buy the stock. In both scenarios, all trading in the stock market comes to a halt. No money would be made and the stock market would crash ultimately crashing the entire economy. For the sake of all us, we hope that stock predictors don’t get too accurate.
References
Opinion: Why we should be worried about artificial intelligence on Wall Street. (2019, November 1). Retrieved from https://www.latimes.com/opinion/story/2019-11-01/artificial-intelligence-ai-wall-street.
