In late 2017, my wife asked me a simple question – how she should invest $10,000 in various stocks to maximize her profits or, in adverse moments, minimizes her losses. The only answer I had for her was that I know the technology industry, and these five companies are doing well.
“But, how much of each stock should I buy?” she asked with the curiosity of an eager student. I suggested that she research their financials, look at a 52-week stock movement, and invest her funds and monitor closely. As I was about to ask her what her risk-taking capacity was, I realized that she was looking for a comprehensive solution for allocating her funds by investing the right amount in different stocks and building an optimal portfolio. I don’t think, for a retail stock investor, there is any tool available that helps in maximizing profits and minimizing losses or making optimal investment decisions without asking questions about the investor’s risk-taking capacity, i.e., by solely relying on the performance of different stocks.
This whole episode prompted me to discuss this problem with my data scientist friend Dr. Prateek Aggarwal. That’s when we decided to perform an in-depth analysis of standard portfolio theory. Our research helped us arrive at the following conclusions about the existing state around portfolio allocation.
Significant literature exists in the world of portfolio theory and covers research work spanning almost seven decades. With the increasing capabilities of Data Science and Artificial Intelligence, most of the major financial institutions have research groups focused on finding the best allocation solution. Asset allocation is the most critical part of creating and balancing your investment portfolio as it is one of the main factors that significantly influence your overall returns.
Almost all portfolio management advisors aim to help you balance your portfolio by having a mix of stocks, bonds, cash, and real estate based on your desired goals at any point in time. The desired goals are based on your tolerance for risks, and this tolerance is based on your age. The more you are near to retirement – the lesser the tolerance for higher risks.
The research groups of many financial institutions are using modern data analysis tools, including sentiment analysis, to reach the right mix of allocation. With all this allocation theory, we still have few unanswered questions;
- Suppose you had just one risky instrument to invest in, and you wanted to maximize your return. In that case, standard portfolio theory advocates that you invest all your cash in this instrument if it is a profitable investment. This, however, does not intuitively make sense since even for a profitable instrument, investment automatically comes with a risk of loss that one needs to be prepared for by holding some cash back for future investment if the instrument drops in value temporarily.
- The standard portfolio theory includes variables quantifying the standard deviation of the risky instrument in its computations, apparently to account for investment risk. This has often confused investors who claim that the only thing they care about is the average yearly returns from investing in a risky asset.
Despite the general awareness in the investment community about the need for portfolios to be balanced at regular intervals, the standard portfolio theory offers no insight into the advisable length of these intervals or how the allocation could change given a selected rebalancing interval.
After reviewing the literature and fully grasping the issues at hand, we started finding a solution using data science, artificial intelligence, and sentiment analysis. Last month we were delighted to come up with an approach that addresses the above three questions.
Our approach counters the standard portfolio theory and presents an alternative solution that advocates holding back some cash, even if the expected return from the risky investment is positive, and there is just one risky instrument to consider. The current approach is ideal for multi-period allocation problems and a rebalancing portfolio after each period.
The mathematical computation approach is non-trivial and requires using numerical and machine learning principles to solve complex portfolio allocation problems. We validated the results by applying the algorithm to stock market data using a model-train-and-test approach using historical data. We are happy with the test results so far.
With his excellent data science knowledge, my friend Prateek contributed significantly to developing the algorithm for allocation with cash hold back. The data and analytics group of EnSignis Digital gets tremendous benefits from his same experience. I believe the algorithm that got developed can solve some other complex business problems for different industries. For now, we are focusing on bringing this to market so that retail investors can get the allocation to advise, which has some logic behind it.
