Portfolio optimization and machine learning… two things not addressed very often outside of academic papers. At ElectrifAi, our focus is to generate real business products using machine learning. That gives us a great view into how machine learning can help optimize portfolios at a business level.
We spoke with an expert on the subject, ElectrifAi’s VP of Engineering, Chris Ma. With an insider’s experienced perspective, we delve into such questions as: can machine learning increase profitability, what data is needed for the machine learning model to function, and how to ensure a portfolio manager’s performance is acceptable. Read on to find out all this and more!
Portfolio optimization is to build your portfolio in such a way that you maximize potential returns from investments while still not exceeding the amount of risk you’re willing to carry. Creating a balanced portfolio with many different investments, such as stocks, bonds, mutual funds, etc., is the best way to spread out assets to maintain a risk-to-reward ratio.
Artificial intelligence (Ai) is not practical without a machine to run algorithms. Ai is the brain by which machine learning works. Systems can use regular data input to learn and improve from over time without human intervention or assistance. Machine learning helps to track thousands of factors and make accurate predictions.
Machine Learning is powerful to extract patterns from vast amounts of data. With thousands of factors applied in machine learning, the computer can tirelessly dig into data and compare patterns one by one. With the help of complex mathematical models and neural networks, machine learning can quickly extract, transform, and load data to the database, applying complex algorithms and seeking the data signal until reaching the specific goal. Machine learning can also self-learn and continuously improve its capabilities.
Machine learning provides a whole new perspective to optimize financial portfolios. Traditional optimization focuses on quantitative analysis and hedging mechanisms. First, machine learning can process a huge amount of data and extract patterns from the data, which greatly exceed the limits of a traditional mathematical approach. Second, machine learning can easily build a non-linear relationship and reduce dimensionality (normally impossible in any other way). Third, the complex relationship between risk and return, which could end up in thousands of factors, can be processed and identified inside a machine learning algorithm. In the end, reinforcement learning can make the machine learn and continuously improve that no human being can beat.
Any company looking for return on investment (ROI) should use machine learning to improve many different aspects, such as their spending, revenue, spend pattern, inventory, portfolios, etc. We use Statistical Analysis patterns to analyze the current portfolio, searching for ways to continually improve. Machine learning also analyzes actual risk. It’s a versatile tool if you have big data to support the learning capabilities.
Machine learning works purely off data. It doesn’t try to find logical reasoning or the driving force behind the algorithms. It sees the data pattern. If a stock has been consistent in the market, then machine learning can figure that out. Machine learning picks up the pattern and tries to fit it on the time horizon. If it is constantly proved by historical data, then the stock is recommended – which could be dramatically different from what we would normally see. Since machine learning works purely based on data output, it can spot the nuances a human might miss.
Yes. If you analyze the stock from the current traditional way, we can analyze the company strength, industry position, past performance, and the distribution among all the investors. Using machine learning, the data can find the bigger connection and try to filter out all the noise to extract some key information. That’s the way to predict expected stock returns. Sometimes the predictions are ridiculous, but after a benchmark is placed it normally makes sense.
While quantitative analysis can provide a complete risk exposure and market scenarios, machine learning is powerful in its prediction. Equipped with both quantitative analysis and machine learning, the profitability can be significantly increased.
In our software, PortfolioAi, data primarily comes from four areas:
These four sources of data are the basis of the machine learning algorithm. From them, we can extract the market signal, setup optimal trading strategies and hedging mechanisms, build up the machine learning prediction algorithm, and build reinforcement learning procedures so the portfolio can continuously improve itself. We can also use PortfolioAi to compare the manager’s performance and evaluate the manager by thousands of factors.
This is where PortfolioAi shines. PortfolioAi is not for real trading purposes. The data is used to evaluate the portfolio manager’s performance from the day before. In PortfolioAi, the manager’s performance is analyzed in many different aspects based on the Profit and Loss (P&L) data and risk data.
From risk and return calculations, we can determine the manager’s Sharpe ratio and compare it to the benchmark. The beta value (which measures the return based on riding the market) is also calculated. The Sharpe ratio and beta value are calculated on many different levels, such as industry, asset class, country, market sector, etc.
With the large amount of data in our database, we can compare the covariance between any two managers and between the manager and the market. In this way, we can use principal component analysis to determine the equivalent portfolio based on a sub-group of managers.
We can also build a virtual portfolio, which best reflects the investor’s expectation, and compare it with the manager’s performance. For example, investors in retirement funds normally expect a 20+ years horizon. We can build a virtual portfolio with a group of companies with high growth potential and use it to evaluate how good the manager’s portfolio is. For educational funds, investors normally expect a 5 to 10 years holding period.
PortfolioAi, one of ElectrifAi’s most successful products, has been running on a fund of funds with AUM worth over $100 billion for 8 years. The software has also just been selected to be used at another top investment bank in New York. Our powerful ETL engine can easily process multiple terabytes (TB) of data daily. Our data scientists have delivered numerous machine learning models running in all industries, from supply chain optimization, computer vision to complicated reinforcement learning. We can quickly deliver machine learning models needed by clients with prebuilt models ready to go.
Contact us today for a custom demo!