Author: Gayle Kurtzer-Meyers
This post was originally posted on https://medium.com/
An overview of the most common applications of AI in finance—and challenges faced by the industry.
In January 2020, the Cambridge Centre for Alternative Finance (CCAF) released a study on the impact of AI in the finance industry. Known as one of the most comprehensive global surveys in this domain, it comprised around 151 respondents from 33 countries, including incumbent financial institutions and FinTech firms. The study came up with the following findings:
At least 77% of the respondents believe that AI bears high importance to their organization in the next couple of years.
Almost 64% of the respondents intend to earn revenue through AI via client acquisition, customer service, risk management, process automation, and new products.
At the moment, AI is widely used in risk management, having an implementation rate of 56% among firms.
Traditionally, HFT firms and hedge funds were the primary AI practitioners in the finance sector, but lately, FinTech companies, insurance firms, banks, and regulators are also catching up.
In this industry, some of the AI uses include Robo-advisors, backtesting, model validation, portfolio composition and optimization, stress testing, algorithmic trading, and regulatory compliance. Let’s find out more about AI applications in finance.
1. Risk management
AI and machine learning algorithms are gradually revolutionizing financial risk management. AI-driven solutions are providing insights on:
Determining the loan amount to a customer.
Generating warning alerts to traders regarding position risk.
Enhancing compliance and limiting model risk.
To understand why the respondents in the CCAF study listed risk management as their primary focus in the implementation of AI, consider the case of Baidu.
The most prominent search engine in China is Baidu. (since Google is banned there). In 2016, Baidu sought the assistance of ZestFinance — a US-based FinTech company specializing in AI products. Baidu’s objective was to provide small loan offers to retail customers who bought products from its platform.
However, the Chinese lending landscape is in stark contrast to the Western markets — the lending risk in former is considerably high as more than 80% of people don’t have any credit profile or credit rating. Hence, there is no existing method to determine borrower reliability.
ZestFinance tackled this issue by analyzing Baidu’s massive customer datasets, particularly the search and buying histories. In this way, they employed AI to assist Baidu in deciding whether to lend to a customer or not. By 2017, a survey found that Baidu experienced a 150% increase in small-item lending without any noticeable credit losses.
Since ZestFinance processes financial data through proprietary technology, the complete detail of their AI solution is unknown. However, it’s common knowledge that their process uses a blend of two machine learning algorithms: decision trees and clustering.
For instance, if a customer’s search history indicates extensive visits to gambling websites, they would be grouped in a cluster associated with higher risk. On the other hand, if a borrower is responsible for online spending, they would be categorized as having low-risk borrowers. With automation, it would be quite easy for Baidu’s financial staff to review these applications and approve loans to people according to their risks.
2. Algorithmic trading
For an extended period, investment firms used computers to make trades. A large number of hedge funds rely on data scientists to build statistical models. But, there’s a significant limitation with the approach — it only uses historical data, which is mostly static and depends on human intervention. Hence, these computations struggle as the market undergoes any change.
Luckily, modern AI models have made rapid strides through algorithmic trading. These models are different because they don’t only analyze large amounts of data, but are genuinely autonomous — they learn and improve over time, reaching a point where they can rival humans. This “smartness” derives from sophisticated machine learning techniques, such as evolutionary computation (based on genetics) and Bayesian networks.
AI tools collect voluminous amounts of data from global sources, “learn” from it, and make predictions accordingly. This data consumption is exhaustive; it extracts information from financial exchanges, news reports, books, social media platforms (e.g., tweets), and even TV shows like Saturday Night Live.
What’s important is to understand how AI has made deep inroads in this domain; unlike the traditional technological intervention that allows humans to decide the financial strategy, AI is now dictating the game.
One example of these AI-powered trading systems is Aidiya — a Hong Kong-based AI hedge fund that allows users to make all stock trades via AI. It’s worth noting that startups aren’t the only ones interested in AI trading technology. Earlier, prominent names, such as Goldman Sachs, Wells Fargo, Citigroup, Morgan Stanley, Merrill Lynch, Bank of America, and JP Morgan Chase, took an active interest in Kensho — an AI trading platform.
3. Fraud detection
Another application of AI in finance that is rapidly advancing is fraud detection, which is understandable considering the vast sums of money. The cybercrime industry steals around $600 billion or 0.8% of the global GDP from businesses around the world. Cybercriminals have become more sophisticated and smart, leveraging modern technology for nefarious purposes. According to Statista, the comprehensive fraud detection and prevention market expects to grow by more than $40 billion by 2022.
So, how can AI help? Tackling the skill, modern machine learning cybercriminals can use a blend of supervised and independent techniques to build a model with predictive accuracy and capability.
Supervised learning utilizes annotated data — which humans assess and identify as fraud activity — and learn intricate patterns from corporate datasets. Meanwhile, unsupervised learning processes deal with those datasets that are not identified before and infer data structure by itself. Other fraud detection techniques include regression and classification. They can analyze data and determine whether the transaction is fraudulent or not.
The standard supervised algorithms used for addressing these issues include the following:
· Decision Trees help to introduce a set of rules that learn normal customer behavior while being trained with fraud instances so that they can identify anomalies and alert authorities.
· Neural Networks based on the human brain can learn and adapt to customer behavior to detect real-time fraud.
Examples of unsupervised learning algorithms include the following:
· K-means Clustering splits a dataset into a batch of similar data points, known as a cluster, for anomaly detection.
· Local Outlier Factor determines the local density of data points, identifying areas where similar density exists. Data scientists can use locality concept to mark the end having unusually lower density, known as outliers. This application can come in handy to detect fraudulent transactions.
Regulatory compliance is a vital function in finance, particularly during an economic crisis like the current one. Compliance is associated with enterprise risk management and deals with risk functions, such as operational, market, and credit risks.
RegTech is an advanced function of the FinTech domain focused on compliance. Here, AI’s advantage taken when used for continuous monitoring of a firm’s activities. This way, it offers valuable real-time insights and prevents compliance breaches from incurring in the first place. Moreover, this form of monitoring allows firms to free up regulatory capital and leverage automation for decreasing the excessive compliance costs — major financial firms spend $70 billion on compliance every year.
A well-known player in this field is IBM. A while back, IBM acquired Promontory — a RegTech startup consisting of 600 employees. This acquisition has led IBM to promote a multitude of AI-powered solutions for managing financial compliance. For instance, IBM is using its proprietary AI tool, Watson AI, with Promontory’s RegTech expertise to deploy real-time voice conversation analysis for ensuring compliance. Part of this includes the translation of voice-based conversations to text and then using natural language processing for text classification. The aftermath of this process is the formation of categories that detects potential non-compliance.
Other AI applications include the automated reading and interpretation of the regulatory documentation, especially for determining implications. Waymark, a London-based company, is already providing this service to financial firms.
Although there are numerous other applications of AI in finance, there’s a flip side as well. The industry needs to rectify practical issues to enhance AI implementation.
One of the biggest concerns remains the availability of suitable data. Even though R and Python can read any form of data from Excel spreadsheets to SQL/NoSQL datasets, the pace at which AI-driven solutions function is slower than the organizations’ ability to organize their internal data accurately. Typically, data is stored in separate silos in various departments, and often in different systems, where regulatory and internal political dilemmas limit information sharing.
On a similar note, another predicament is the lack of skilled staff who not only commands an expert-level knowledge of AI, machine learning, and data science, but also have experience in building and implementing AI-centric solutions in the finance industry.