ML is the process of training computers to learn and improve from experience without any human intervention or programming. In it, machines are given vast amounts of data to analyze for patterns and learn from using examples. As time passes, the systems automatically make their own decisions and adjust their actions accordingly.
As mentioned earlier, several of the biggest tech companies already use ML for various purposes. But some believe that while machines can be taught to predict outcomes, they cannot crack the stock market. This post tackles the reasons for the contrasting beliefs.
Machine Learning Can’t Crack the Stock Market
One of the arguments as to why ML cannot predict stock market prices has to do with the Random Walk Theory, introduced as far back as the mid-19th century and popularized by Eugene Fama and Burton Malkiel in the mid-20th century.
Simply put, the theory states that no one can predict stock prices because they change randomly. It argues that financial markets are efficient, all traders can see the publicly available information on every stock, and all market participants are rational profit maximizers. That said, even if an anomaly occurs, traders would quickly exploit and remove it, making the market stable again.
The author of this post on Towards Data Science also argues that new news is the only thing that can change a stock’s price, and when that hits cannot be predicted. In effect, stock prices move randomly. The author cited Jim Simmons, an American mathematician and a billionaire hedge fund manager, who used an early form of machine learning to trade stocks, as an example. While Simmons has been said to have solved the stock market, even his firm is only right half the time.
Another post opines that while machine learning may not be able to completely predict stock market prices, it can help traders get an edge, where the systems can tell them how much risk to take. Since financial markets are not stationary driven by political, social, economic, and natural events that are unpredictable, their movement cannot be determined beforehand. In effect, there is no existing pattern for the systems to see.
However more and more ML enthusiasts are saying the opposite and we summed up their thoughts here as well.
Machine Learning Can Crack the Stock Market
On the flipside, some believe that machine learning can crack the stock market code. Wharton researchers published a paper on how their model can let traders buy low and sell high. In the simplest terms, their machine tells investors to buy a stock when analysts’ expectations are too pessimistic about it. And when the opposite occurs, they should sell their stocks as prices are forecast to decline.
Another post suggests using machine learning in the stock market to predict stock market prices. In this sense, you can create a financial model, much like what SARIMAX and Facebook Prophet use. These two time-series models were utilized to forecast Bitcoin prices.
Classification models can also be used to classify stocks based on their performance in quarterly reports. These can help determine investment decisions.
The models identified above can be configured to run three simple trading strategies—buy, sell, and hold. The machine will tell traders to buy a stock when its predicted price target shows a significant increase from the current price, sell it when the predicted price target shows a significant decrease from the current price, or hold or do nothing if the price target shows neither a significant increase or decrease from the current price. The author noted better return performance when the model and stock trading strategy were used. Sudden stock market crashes, however, could affect results.
Based on the arguments presented above, while machine learning can help with stock market trading, current models have yet to get everything right. The systems need more data to work with and time to learn. It seems that for now, machine learning can help with short-term predictions but not long-term ones just yet since political, social, economic, and natural events are hard to foresee. Over time, however, who knows? Maybe machine learning can crack the stock market.