What Is Machine Learning? Definition, Types, and Examples.

What is Artificial Intelligence (AI)?

In computer science, the field of artificial intelligence as such was launched in 1950 by Alan Turing. As computer hardware advanced in the next few decades, the field of AI grew, with substantial investment from both governments and industry. However, there were significant obstacles along the way and the field went through several contractions and quiet periods. 
 

Further work was done in the 1980s, and in 1997, IBM’s chess computer, Deep Blue, beat chess Grandmaster Gary Kasparov, a milestone in the AI community. In 2016, Google’s AlphaGo beat Go Master, Lee Se-Dol, another important milestone. Other AI advances over the past few decades include the development of robotics and also speech recognition software, which has improved dramatically in recent years. 

Machine Learning and AI in Finance

Both AI and machine learning are of interest in the financial markets and have influenced the evolution of quant finance, in particular.
 

As an example, in a talk held by the CQF Institute on ‘Reinforcement Learning and Hidden Markov Model Based Smart Trading Strategies’, Samit Ahlawat of JP Morgan Chase explained that traditional trading strategies are based on static rules that may not hold across all types of market conditions due to complex correlations between market variables, which can confound such trading rules. However, it is possible to recalibrate the parameters of these rules to adapt to changing market conditions. Timing matters though and the frequency of the recalibration is either entrusted to other rules, or deferred to expert human judgement. Samit stated that artificial intelligence and machine learning are promising tools for addressing this shortcoming in static or semi-static trading strategies.

Types of Machine Learning

There are three main approaches to machine learning: supervised, unsupervised, and reinforcement learning. There are also hybrid approaches including semi-supervised learning, which can be tailored to the problem a researcher is seeking to solve. Each approach has specific strengths and weaknesses, and some techniques are better suited to particular types of problems than others.
 

• In supervised learning, the computer is trained on a set of data inputs and outputs, with a goal of learning a general rule that maps the given inputs to the given outputs. Two main types of supervised learning are: 1) classification, which entails the prediction of a class label, and 2) regression, which entail the prediction of a numerical value.
• In unsupervised learning, the learning algorithm is not given this type of guidance; instead, it works to discover the pattern or structure in the input on its own. Two main types of unsupervised learning are: 1) clustering, which involves discovering groups within the dataset that share similar characteristics, and 2) density estimation, which involves evaluating the statistical distribution of the data set. Unsupervised learning methods also include visualization with the data and projection, which reduces the dimensions of the data, a form of simplification.
• In reinforcement learning, the computer and algorithms will confront a problem in a dynamic environment and as it works to perform a given goal, it will receive feedback (rewards), which will reinforce its learning and goal seeking effort.  The example of AlphaGo is a case of reinforcement learning; reinforcement learning algorithms include Q-learning, temporal-difference learning, and deep reinforcement learning.

Application Examples of Machine Learning

In the financial markets, machine learning is used for automation, portfolio optimization, risk management, and to provide financial advisory services to investors (robo-advisors).
 

For automation in the form of algorithmic trading, human traders will build mathematical models that analyze financial news and trading activities to discern markets trends, including volume, volatility, and possible anomalies. These models will execute trades based on a given set of instructions, enabling activity without direct human involvement once the system is set up and running.
 

For portfolio optimization, machine learning techniques can help in evaluating large amounts of data, determining patterns, and finding solutions for given problems with regard to balancing risk and reward. ML can also help in detecting investment signals and in time-series forecasting.
 

For risk management, machine learning can assist with credit decisions and also with detecting suspicious transactions or behavior, including KYC compliance efforts and prevention of fraud.
 

For financial advisory services, machine learning has supported the shift towards robo-advisors for some types of retail investors, assisting them with their investment and savings goals.

Returning to Samit Ahlawat’s talk delivered in April 2022, specific examples of the use of machine learning in quant finance included: 

• Genetic algorithms for combining simple security trading rules such as buy or sell when price falls below or rises above a certain threshold.
• Finding non-linear patterns in stock prices by training a deep neural network to predict future returns using historical returns.
• Using probabilistic neural networks to identify patterns in security prices.
• Using convolutional neural networks in technical trading rules.
• Applying quantitative models to assess the consistency of the Efficient Market Hypothesis with available market data.

In each of these examples, Samit noted the difficulty of developing profitable strategies, even with the use of AI. After taking transaction costs into account, prior research concluded that many simple AI strategies failed to beat simple trading rules. As he explained, there is much research devoted to the application of new AI/ML tools to build models that can adapt to changing market conditions. Therefore, researchers must have a thorough understanding, not only of the mathematics of quant finance and knowledge of the financial markets, but also strong skills in machine learning and related programming techniques. A solid grounding in all three domains: mathematics, finance, and programming is increasingly important for quants throughout the industry, from large asset managers, to prop trading firms, to hedge funds.

The CQF and Machine Learning in Quantitative Finance

The field of machine learning is of great interest to financial firms today and the demand for professionals who have a deep understanding of data science and programming techniques is high. The Certificate in Quantitative Finance (CQF) provides a deep background on the mathematics and financial knowledge required for a job in quant finance. In addition, the program takes a deep dive into machine learning techniques used within quant finance in Module 4 and Module 5 of the program. 
 

For those interested in gaining valuable skills in machine learning as it relates to quant finance, the CQF program is both rigorous and practical, with outstanding resources and flexibility for delegates from around the world. Download a brochure today to find out how the CQF could enhance your quant finance and machine learning skill set.