3 Ways Machine Learning Supercharges Investing
Modern financial markets are marked with the massive growth of information and the rapid speed at which data is collected and processed, especially with the emerging alternative data as new sources of edge for investment management. This calls for new methods and algorithms that can adapt to big data with high efficiency and better results.
by Isabell Sheang, CCO, Kavout
“90% of the data in the world has been created in the last two years.” That was reported by IBM in 2013, and also the time Big Data became a thing that everyone started talking about. Imagine what that is like today?
Finance professionals are finding harder to simply rely on traditional fundamental or technical analysis and old-school statistical methods to find an edge in today’s market. Therefore, financial machine learning is gaining momentum in recent years to become the mainstream tool to tackle many problems in investing.
Financial Machine Learning (ML) sits in the intersection of mathematics, statistics, and computer science. It is a branch of artificial intelligence (AI) that can automate statistical models and data analyses to learn from data and identify patterns. It also involves developing algorithms to optimize decisions over time with minimal human intervention.
To be clear, financial ML does not mean completely discarding fundamental finance theory or statistical methods. Instead, it super-charges them, not only in terms of efficiency but also scale. There are many reasons why financial ML approach is superior to statistical models in solving complex and new challenges in today’s capital market. We are going to explore three here.
1. Big, Complex and Inhomogeneous Data
Big data comes in several dimensions — volume, velocity, variety and comprehensiveness. Without new computing power, efficient algorithms, and advanced analytical capabilities, it simply is inconceivable to process new information at the speed it requires daily or real time, let-alone discovering insights from it.
Grounded in sound data curation, classification and indexing practice, ML can process an enormous amount of data from various sources, and turn unstructured data, such as text, sentiment, and eCommerce transaction, into structured data, which are more amendable to analysis. Investment managers can then layer these alternative datasets to a company’s fundamental financials to discover actionable signals.
Unlike statistical methods that are limited to a few input factors, there isn’t a limit on factors or features for ML. ML can handle data types that are wide (high number of attributes) and deep (high number of observations). Especially with the advancement of deep learning in recent years, new algorithms drastically improve the capability and effectiveness to capture and model useful information from massive unstructured data.
Many hedge funds and quantitative asset management firms have already adopted machine learning in many aspects of their business from research, to forecasting to trading.
2. High-Dimensionality and Nonlinearity
While traditional statistical methods are useful, it’s based on mathematical models and is often limited to clean and definite data, or small number of pre-selected variables or factors. Moreover, statistical models require the modeler to understand or hypothesize the relations among variables in advance, such as linear relation between independent and dependent variable.
What if the assumptions were wrong? One may ask.
Machine learning, on the other hand, learns patterns in a high-dimensional space without being “fed” with pre-determined features. The advantage of ML algorithms is that it doesn’t assume anything before occurrence of events. It takes in a large amount of data sets, look at signals from different places including non-linear patterns, learns, tries to find patterns hidden in the data, and make predictions not conceivable with statistical models.
Within ML, deep learning techniques have a high level of tolerance or forgiveness for data (asynchronous and categorical), making it favorable for practitioners who are working with a large number of datasets and running multiple algorithms simultaneously on a given day.
Deep learning techniques can be used for price prediction for stocks and other asset types. Models evolve as more datasets come in, which means the posterior probability will keep evolving once it sees more data or information.
3. Adaptability and Continuous Improvement
Traditional statistical or econometric models do not update model choices, parameters, or outputs as more data are observed, so it is not clear if the estimated model, algorithms, or strategies are still bringing the user closer to the goals as time goes by.
A branch of ML, called Reinforcement Learning (RL), involves developing goal-oriented algorithms that dynamically drive the optimal behavior towards the objectives. A key advantage of RL is that over time it automatically incorporates new data, and self-evaluates past actions, and optimizes decisions. While a good RL algorithm can adapt to the changing environment, many researchers monitor these algorithms and the data closely to ensure rewards are maximized.
There is little doubt that financial ML will continue to play an essential role in the modern finance and investment space, from algorithmic trading to price forecast to portfolio management. Hence, acquiring the right set of ML tools will be a crucial step for companies and investors to adapt to the rapidly changing market and stay competitive among peers.
Firms that are looking for assistance in either setting up a ML practice internally or establishing new ML capabilities can work with external consultants to get started.
Author: Isabell Shaeng, Chief Commercial Officer at Kavout, a global InvesTech company empowering institutions and investors with augmented intelligence to find edge, manage wealth and do more with less.