The modern day question of whether Artificial Intelligence contains the missing technological link that can help predict stock prices, has captivated the investment community. Using AI to predict short-term (e.g. next day or two) or long-term (e.g. over six months) stock prices has been researched extensively because of the allure of windfall gains.
The research relating to investing and trading endeavours often focus on two key principles
1) Selecting the best assets that yield the highest returns (This is also known as seeking Alpha). Timing is also a critical part of this decision making process.
2) Managing risk which includes minimizing volatility and maximizing Sharpe Ratio by building portfolios
In the area of stock price prediction, state of the art solutions and our experiments on next day (or two, five) price movement were able to achieve
1) 55% to 65% Accuracy when predicting for 2 classes (e.g. stock price gain vs. drop)
2) 45% to 55% Accuracy when predicting for 3 classes (e.g. moderate stock price changes, above moderate stock price gain, below moderate stock price drop)
Given the above observations we at Semantic Brain decided to ask different questions and take a unique approach that aligned with our Intelligence Augmentation vision (Human + Machine > Human or Machine). We embarked on a journey of creating Intelligent Micro Portfolios, that
Are built using Semantic Brain AI that have achieved 75% to 85% Accuracy on select predictions that aid the process of building effective Micro Portfolios. For example, use Semantic Brain AI to predict volatility and price movement to spot opportunities and / or identify risk.
Combine Human Intelligence with Semantic Brain AI to increase Accuracy to 80% to 95% in future (also increase Success Rate and Sharpe Ratio). For example, AI to explain predictions to Humans, and complement that info with technical charts and fundamental metrics.
Note: This article is based on analysing, learning and predicting 10 years of end of day historical data, and we plan to incorporate Intraday Trading data in the future.
What are "Intelligent Micro Portfolios"?
The following attributes define Intelligent Micro Portfolios
"Micro Portfolios"
Consist of 2 to 5 assets
Current Asset Classes: Stocks, ETFs, Stock / ETF Options
Future Asset Classes: Crypto, Forex, Commodities
Semantic Brain AI trained using US equities and ETFs
Have a window of 2 to 20 trading periods (trading period = trading day)
"Intelligent" refers to the process of combining Semantic Brain AI, Financial Engineering and Investor / Trader intelligence.
What are the Use Cases?
The following are a list of potential use cases, where Intelligent Micro Portfolios can yield high returns and mitigate risk
1) Standalone Investments
Combine Call and Put options into Micro Portfolios to achieve high success rates and Sharpe Ratios
Combine Long and Short positions into Micro Portfolios to achieve high success rates and Sharpe Ratios
Build Micro Portfolios including either Long Positions or Short Positions
2) Portfolio Augmentation
Use Micro Portfolios to mitigate long-term investment (aka secular investment) risk
Use Micro Portfolios to increase return and Sharpe Ratio of long-term investments
3) Non-Financial Industry
Hedge against Forex Rate and Commodity Price change risks
Why do Intelligent Micro Portfolios work?
In 1973 the Princeton University professor Burton Malkiel claimed in his bestselling book, A Random Walk Down Wall Street, that “A blindfolded monkey throwing darts at a newspaper’s financial pages could select a portfolio that would do just as well as one carefully selected by experts.” Malkiel's random walk hypothesis suggests that stock market prices evolve according to a random walk (so price changes are random) and therefore cannot be predicted.
It was later found out that random selection worked better as more funds were allocated to higher growth small and mid cap equities. Professional investors in the other hand had invested more heavily is slower growth large cap stocks.
However, the markets are not random, they are event driven. Depending on nature of events the repercussions can be relatively accurately predicted. It is also possible (and probable) to optimize Portfolios and predict their aggregate outcome more accurately.
Math, Stats and AI
Predicting price movement on subsets of equities during specific windows is a different problem compared to predicting price of all equities at any time.
When predicting price movement of a portfolio containing two or more equities then the law of averages work in your favour
AI is better than humans at recognizing some multi-dimensional patterns, constituting raw numbers, that can't be visually represented
Semantic Brain's intellectual property enables us to filter a lot of noise, and apply innovative deep learning architectures and loss functions to achieve superior results
Financial Engineering
Combining Calls and Puts, or Long and Short positions can help cancel out general market affecting event (e.g. terrorist attack) impact.
Volatility predictions can be leveraged to secure higher returns via options trading
Human Aspects
Humans have a better understanding of general business and longer term trends, which can be combined with AI to achieve better results
Boosting Predictions, Explainability and Results
Standard Approaches
Predicting using Technical Indicators: Popular technical indicators such as RSI (Relative Strength Indicator) have very little predictive power.
Predicting based on Price, Volume and Technical Indicators: Modern AI Architectures are able to more predict using Price, Volume and Technical Indicators over 2 to 5 trading days.
Predicting based on Fundamentals: Fundamentals enable long-term predictability
Integrating Social & Media
Semantic Brain product Asperios is built with the ability to scan
Social Media (e.g. Reddit, Twitter)
News
Security Filings (e.g. EDGAR)
These scans could be used to predict volatility and stock price movements. Below are some charts relating to Reddit/wallstreetbets scan and possible interpretations.
Activity peaked in early June
AMC, Blackberry, Clean Energy and GameStop have dominated the conversation
Conversations relating to AMC had a significant uptick mid June.
Conversations relating to Telsa had a huge uptick in mid June.
Derivatives Trading
Volatility of underlying assets (i.e. stock) often lead to greater fluctuations in derivative prices. Thus ability to predict volatility can be leveraged to secure higher yields by trading in derivatives.
AMC prices moved up mid June (aligning with Reddit activity), and are falling as of early and mid July. Prices have fallen below Bollinger Band threshold, however, trade volumes remain relatively low.
The above chart and further drill down show considerably greater put activity on AMC options.
Tesla prices also gained some upward price momentum following Reddit activity in mid June.
However, majority of Tesla option trades are puts are close to its current stock price.
1000x Investment Analysis
Semantic Brain's Asperios also enables thorough and fast due diligence, via our conversational style interface - https://youtu.be/fHAQckaTLvE
Where are we heading?
Our company, Semantic Brain, is a forward thinking organization that believes in the power of streamlined Artificial Intelligence tools that reduce trading risk and maximize returns. We do not claim to predict stock prices, but rather we have developed an Application that scans the internet for any situational factors ( global or local ) that may affect specific groups of stocks. We package that "heated and distilled" information to the investment consumer who is then equipped to make a better decisions.
The analogy that we use to explain our technology is that it is a "seismograph for stocks." Seismographs are tools used to measure minute changes in the movement of the earth that can lead researchers to warn citizens of an impending earthquake. Similarly, multiple data sources can be mined using AI to alert investors on stocks that may be impacted (opportunities and risks) by preceding events ( e.g. Trading patterns, Earnings, Media Trends & Sentiment, Oil Prices, Political Instability).
Armed with this insight investors can rapidly perform additional due diligence, and then construct and / or dismantle one or many Micro Portfolios that result in higher return and lower risk.
Intraday Data and Real-time Alerts
All of the above is based on EOD (End Of Day) Historical Data. We at Semantic Brain plan to help investors be more successful by adding Intraday Day Historical Data for Predictions and Real-time Alerts.
Adding Crypto, Forex, Commodities
Semantic Brain has immediate plans add Crypto Alerts, Analysis and Prediction capabilities (already working on this capability). Forex, Commodities and related Derivatives will be added subsequently.
Marketing and Other Applications
The Social & Media Data, Alerts and Predictive Models developed for investing, can be reused in areas such as Marketing & Ad spend optimization. We intend to use Transfer Learning to improve investing and marketing performance simultaneously.
References
Articles
Human + Machine > Human or Machine - https://www.semanticbrain.net/post/human-machine-human-or-machine-in-chess-investments-etc
Videos
Delivering 1000x Investment Analysis - https://youtu.be/fHAQckaTLvE