Business Challenges
A key necessity in financial decision-making is the ability to predict how prices of assets and securities change over time. In financial trading, volatility refers to the extent to which the prices of a given financial asset rise or fall compared to the mean price. Volatility is an important measure of the risk associated with an asset and is therefore a key determiner of its price. When predicting the returns from a portfolio it is very helpful to get a sense of how the volatility of the underlying assets will change in the future. This leads to the need for precise and efficient volatility forecasting.
Conventional approaches to volatility modelling include various statistical approaches. However, artificial intelligence-based approaches are able to quickly explore many alternative solutions to a problem and uncover more complex relationships within the data, making volatility prediction faster and more precise.
The Company
TurinTech collaborated with a regional tier-one bank that was looking to better understand the direction of volatility in selected equities. The company was looking to make predictions of the implied volatility, i.e. the market estimate for the future volatility of the price of an asset, and they particularly wanted to make predictions one day as well as three days into the future.
Data
The organisation had historical data on implied volatility, which was used in the model building process. Variables such as the last price, benchmark rate, and daily returns for a given stock were also taken into consideration.
State of the existing system
The organisation had a statistical model that they were using to make predictions, but the model was no longer able to accurately capture rapidly changing market conditions. The data science team of the bank was attempting to deploy machine learning-based solutions. However, there were no stable solutions implemented, despite trialling solutions for over 6 months.
Understanding bottlenecks to ML implementation
The complexity of the business problem: The biggest challenge faced by the bank was the complexity of the business problem. Market volatility predictions required working with a set of fast-moving variables. Capturing these intricate variables and their impact on volatility with high precision was much-needed but a difficult task to achieve.
Time to deployment: Stemming from issues of complexity, the internal data science team faced significant challenges in taking models from conceptualisation to deployment. Previous attempts took as long as 6 months for developing and evaluating machine learning pipelines, with no viable solutions deployed.
Costs: Due to time, personnel, and energy demands, implementing a machine learning solution required significant financial investment. However, the return on investment did not seem worthwhile, thereby leading to scepticism on the value-add brought by machine learning-based solutions.
Leveraging evoML for improved volatility forecasting
Implementing production-ready machine learning models using evoML only required a few steps.
Addressing the bottlenecks and deploying effective ML solutions in weeks
evoML automated machine learning workflow for faster deployment: evoML brings the entire data science pipeline onto a single platform, and automates the machine learning model development process, reducing the time it takes to go from conceptualisation to deployment. evoML provided the data scientists at the bank with a valuable set offunctionalities and featuresto easily draw useful insights from their existing data.
With the use of evoML, the financial institution was able to develop, evaluate, and deploy a set of classification, regression, and forecasting models for time-series data. The conceptualisation to deployment timeline for the entire task was two weeks. The team at the bank noted that evoML-based implied volatility predictions demonstrated an accuracy of 70%, which was an 11% increase from the existing system’s predictions. The bank particularly noted that they found evoML’s pre-processing functionalities to significantly reduce the time and effort that go into preparing data for machine learning model development.
Code ownership for customisability: the bank was provided with the machine learning code developed on evoML to mitigate complications stemming from a plethora of rapidly changing variables. This allowed data scientists at the financial institution to tailor models to their requirements. It also enabled them to better understand and improve models, as there were greater transparency and explainability of models. The TurinTech team, along with the bank’s data science team, combined the models built by evoML to generate a final meta-model that could better predict market movements.
Optimisation for cost-cutting: evoML’s built-in ML model code optimisation features easily pick up inefficient lines of code and optimise them with minimum changes, leading to further reductions in costs related to testing, prediction, and deploying on the cloud. Optimisation also led to improving the speed of predictions by 5X, which ultimately translated into significant increases in profits. When the optimised code was evaluated in an Azure-based cloud environment, cost savings of around 30% per hour were observed, for the virtual machine size used.
Our blog article When AI is Supercharged discusses in detail the unique features and services offered by evoML.
A key aspect of executing a machine learning solution is ensuring that the solution is sustainable and scalable. Due to evoML’s modular design, the bank was able to easily integrate the platform into their evolving tech stacks.
“We appreciate that evoML is portable, making it simple to deploy on-premises,” says a spokesperson for the bank. “The TurinTech team took the time to understand our budget allocations and team capacity, ensuring that the implemented solutions could be sustained for the long term. This level of customisation has allowed us to seamlessly integrate evoML into our existing workflows, resulting in improved efficiency and accuracy across the board.”
To stay ahead in a rapidly evolving financial landscape, companies will need to embrace machine learning-based solutions before it is too late. For more information on how artificial intelligence can help financial decision-making, see our blog: Artificial intelligence for Hedge Funds: How Can Machine Learning and Code Optimisation Generate Greater Alpha?
