The financial services industry has always been about working with large volumes of data, especially when it comes to asset management. The abundance of data is low hanging fruit for building data-driven and machine learning based solutions to personalize user experience, improve risk management, and enable efficient fraud detection.
But for the most part, traditional wealth management companies remain late adopters, still seeking ways to become data-driven. More than 70 percent of New York fund managers marked Big Data in asset management as an important investment, according to CRISIL Global Research & Analytics.
Over the last decade, data analytics technology has made several leaps. While the early algorithms used structured data only, modern machine learning based solutions can yield insights even from highly unstructured records. Sentiment analysis and image recognition are now employed to assume possible peaks and valleys in the stock market. For instance, collecting and analyzing social media buzz around brands helps the trader anticipate whether a company’s stock prices will rise or fall.
"Access to new types of data, along with the ability to capture and process that data quickly, has given us new ways to capture investment themes such as momentum, value, profitability, and sentiment." - Takashi Suwabe, Portfolio Manager, Quantitative Investment Strategies at Goldman Sachs Asset Management.
So, let's have a look at the main operations that can be enhanced with a data-driven approach.
Personalization and smart advisors reshape customer experience in asset management
Customers desire personalization across all industries, and asset management is no exception. Today, smart advisors (or robo-advisors) are the hottest personalization trend in the industry. The algorithms consider various customer data – risk tolerance, behavior, legal benchmarks, preferences – and make recommendations based on this data.
By combining multiple data sources, you can increase the dimensionality of models and solve complex optimization problems that account for hundreds of individual portfolio factors. This allows portfolio managers to suggest tailored investment plans to their clients both in B2B and B2C operations.
The proliferation of robo-advisors created a buzz over the past year. Although we’re far from their wide adoption, the technology is predicted to have a great future. According to AT Kearney, nearly $2.2 trillion in assets will be managed by robo-advisors by 2020 versus $0.3 trillion in 2016.
Source: AT Kearney
The early use cases of robo-advisors came from Morgan Stanley and UBS from 2013 to 2015. Morgan Stanley launched 3D Insights, a data sorting tool, which analyzes hundreds of incoming reports, filters through them, and provides advisors with only relevant information. UBS made a step further. The company launched UBS Advice that analyzes real time strategic decisions being made at the chief investment officer (CIO) level and directly provides UBS clients’ advisors with tailored recommendations on portfolio movements. Wealth tech companies, like WiseBanyan, Betterment, and Third Financial, also made their contribution to personalization and robo-advisors.
Source: CB Insights
Smart advisors provide two-way benefits: They ensure great customer experience and, on the other hand, enable financial planning insights based on a client’s behavior. Managers can make more accurate predictions on cash inflows and outflows.
Fraud detection powered by neural networks
One of the emerging trends here is anti-money laundering models and fraud detection systems that help identify suspicious activities. The systems are trained to track and assess the behavior of individuals involved in the investment management process. How do these systems work?
Usually, they apply deep neural networks, a machine learning method, to detect fraudulent actions by analyzing various structured and unstructured data, including pdf files, social media activities, CRM records, and other web footprints. Neural networks are very strong in detecting implicit links between customer or employee behavior and the likelihood of fraud. The Capgemini insights, for instance, show the following fraud-detection opportunities:
50-90 percent increase in revealing scams;
Up to 90 percent fraud-detection accuracy improvement;
Investigation time reduction up to 70 percent;
Real-time fraud detection;
Neural networks can be continuously improved by learning from new data and the history of successful/unsuccessful detection cases.
Source: Capgemini
Predictive analytics boost performance of investment companies
The performance of investment companies directly depends on predicting the future movements of the wealth market. Stocks, bonds, futures, options, and rates movements form the stream of billions of deal records every day. All these are sets of non-stationary time series data, the ones which don't have apparent seasonality and trends to base predictions on. Non-stationary records are a complex problem for financial analysts because conventional statistical methods fall short both in terms of prediction accuracy and speed. We’ve recently covered how machine learning based methods combat non-stationarity in time series data. In a nutshell, there are three main approaches:
Traditional machine learning methods: Models are trained on short-term historical data and yield predictions based on it.
Stream learning: A predictive model is constantly updated by each new incoming record, which allows for better accuracy in a constantly changing market environment.
Ensemble models: Multiple machine learning models with their own strengths and weaknesses analyze incoming data, and the predictions are based on consolidated forecasting results.
Another field for data science is natural language processing and sentiment analysis in social media. Such solutions process through posts and tweets analyzing audience reactions to one or another brand. Wealth managers can see which companies are just gaining their share of buzz and can proactively invest in assets with high growth potential before the stock prices go through the roof. And catching early dissatisfaction allows for selling assets before social media anger drops the stocks. A great example is Crimson Hexagon, the analytics firm, which has accumulated about a trillion records from Twitter, Tumblr, and Facebook to forecast stock prices. Stephanie Newby, Crimson Hexagon CEO, shares that they had been able to predict the fall of Samsung stock prices before the media storm around Samsung Note 7 reached its peak. People weren’t talking that much about the phone catching fire right after the launch. The official claims that if the investors had sold stocks at the right moment, they would get great profits.
Scenario-based analytics improve risk management at wealth companies
Risk management is an important part of the asset management, especially, concerning managing the risks connected with volatility. Poor execution is harmful for both portfolio manager and their customer. The former loses commissions and the latter, money.
A traditional approach to risk assessment is to calculate a standard deviation of share prices using Excel spreadsheets. While this method is widely used, it provides fairly blurred understanding of risks as it doesn’t account for all possible market variables.
The growth of computing power and new data processing packages open new opportunities to build stress models both for company operations and stock market performance in general. Today, you can test millions of scenarios accounting for hundreds of unique market conditions. This allows for anticipating and analyzing highly specific events.
The models like Monte Carlo help to imitate random variables with respect to selected distribution of many variables: Financial and macroeconomic indicators, stock prices, and even customer behaviors can be processed via a scenario-based stress testing and reveal potential threats in the what-if format. Eventually, portfolio managers can better understand the probability of investment failure and make predictions on company performance.
How to jumpstart your data strategy?
We’ve mentioned that data science solutions have been in use by the leading players like Goldman Sacks in asset management for about ten years now. Not only does this indicate opportunities, it also defines the growing competition. The question isn’t whether you are going to put data software in your pipeline, but rather how fast you’ll be able to do this. Here are the major steps to jumpstart your data science initiative:
Define the problem and match it with opportunities. Consider the most troubling issues, shortlist them, and match them with available opportunities. Surveying or discussing your initiatives with frontline employees is a great tactic for turning your shortlisted ideas into a roadmap. You’ll be able to define what kind of data you need and narrow the problem down to a specific prediction or anomaly detection task.
Assess your data. You certainly have data, but the question is whether it converts into an ML-friendly dataset. Dataset preparation is a critical step in any data science operation. While you may need expert data scientists to process your raw data, there are simple recommendations to make your data better without employing data science talent. The main criteria at this initial stage is to ensure that you won’t need much by way of additional data gathering operations.
Acquire the right talent. You should decide whether to deliver the project with an in-house team or engage an outside consultants. Hiring or training an in-house data science team is expensive and time-consuming. However, it ensures better data security. On the other hand, you can consider a technology consulting firm with a good reputation. Consultants with deep technology and domain expertise will streamline both data preprocessing, modeling, and setting an infrastructure.
Develop models and prepare infrastructure. Modeling and testing entails creating and benchmarking multiple machine learning methods to choose the best performers in terms of prediction accuracy. Processing datasets and training models can be done on office laptops. Things change when your model is compiled into a production version and about to face the stream of real time data. Processing data on an internal server infrastructure is one of the options, but it requires some server maintenance. Another approach is to use cloud solutions like Azure, Google Cloud, or Amazon.
Integrate the solution with your IT infrastructure. Integrating the analytics solution with your working product will consume additional engineering effort to connect architectural components through APIs and align interface elements. This task may be limited to a new section in your online product with data visualization or stretch to more complex things. For instance, you may plan to adopt fraud detection among your managers that operate a web-environment to control assets allocation and track their performance. These solutions require an extra level of interactions tracking, like capturing copy and paste operations or analyzing common and deviating behaviors with your software.
Onboard and train your employees. You should allocate time and people in charge of mentoring your employees to start using new features and tracking analytics. Otherwise, complex and expensive software may remain merely a gimmick for most of your staff. The final step also considers a cultural shift in your organization. Data-driven decision making has a proven adoption challenge: Executives that are used to making decisions based on their experience tend to stay skeptical about recommendations that data supports. The problem becomes evident with complex machine learning solutions that detect hidden relations between different data records that humans can’t subjectively notice. So, evangelizing the change is no less critical than the change itself.
Final Word
Most wealth management organizations are among the late adopters of data-driven technologies in fintech. But if you look at the industry leaders, you’ll see that they launched their first data science solutions over a decade ago. And now it seems the time has come for the rest of the players.
We believe that robo-advisory and personalization define the hottest direction of data science implementation in wealth management so far. These types of technologies, in turn, will intensify the interest in semantic analysis, ML-based time series forecasting, and scenario-based modeling. Due to a relatively late transformation compared to the financial services industry in general, the smart move today is to seek a partnership among tech consultancies and fintech startups to avoid reinventing the wheel.
