Platform for decision support in the field of finance
One of the project's results will be a cloud-based online platform that will be used as an instrument for creating decision support systems in the field of finance. The project's partner will be Bank Saint Petersburg PJSC. The platform will be publicly available, and will become the basis for different services used by banks and other financial structures.
The use of machine learning, artificial intelligence and financial mathematics will make it possible to model various financial processes at different particularization levels. For instance, it will be able to assess the credit risks of giving loans, as well as create optimal interest terms, forecast the possibility of debt loss, develop effective strategies and define the demand for particular financial products in different regions and among different groups of customers. The services created on the platform's basis will also be able to forecast the consequences of other decisions in financial field, like which stocks to invest in, how to develop new rates or optimize business processes of financial organizations.
Office of Bank Saint-Petersburg PJSC. Credit: bspb.ru
"Most of the existing models are developed for particular situations and customers. As for us, we plan to use a systemic approach: combine the classical formalism of describing financial processes (which can be done using the banks' own data) and psychosocial models that describe financial and general behavior of particular individuals. This will make it possible to combine confidential data (i.e. transactions) and open data derived from different sources - from cameras and mobile phone tracking to social networks," comments Alexander Boukhanovsky, Head of the School of Translational Information Technologies.
The new platform will become an all-purpose instrument with a basic set of models that are to be optimized with regards to the specifics of particular business processes and the peculiarities of the available data.
A public service for assessing financial products
Another of the project's results will be a public online service where users will have the opportunity to rate different financial products based on individual criteria. For instance, there already are services that allow to keep a history of mobile calls from a certain cellphone where algorithms help choose the best service plan for its owner. The new service will help users choose loans in a similar way, and also protect them from deceptive advertising by letting them know of any additional conditions and risks.
Practice-oriented tasks and methods to solving them
As of now, the level of computerization of financial processes as well as their speed has increased considerably. What's more, it is not just the amount of financial information in the world that's constantly growing, but also the number of market players, most of which use their own strategies. Due to the increasing amount of financial data, analysts find it hard to assess the rapid changes in financial systems. This is why decision support algorithms are to be based on quantitative data that is automatically gathered and processed. The question here is whether the system based on transient data only can be reliable - as such data can't be used for predicting rare occurrences, force majeure events and other critical situations.
"There are no causal models in psychosocial processes, hence one has to work directly with data derived by using machine learning methods. So, the models are to be developed based on data. Still, our project combines both data analysis methods and predictive modeling. We are used to making conclusions about a particular situation based on some retrospective data we've accumulated. Yet, this doesn't work in the field of finance. First of all, financial processes constantly evolve and change. Secondly, the fundamental aspect of financial behavior is that most people aim for the best situation, meaning most do their planning with regard to their idea of the best outcome and not their former experiences and mistakes. Thus, retrospective data can only be used to assess the current situation, and decision support calls for models that allow to make predictions based on the interrelation of data. So, we base our work on good data plus reliable models," explains Janusz Holyst.
What's more, predicting social, political and other factors can be very difficult. That is why predicting models can't be used for full-automated control, though they are still good for decision support systems.
Research tasks and development of scientific community in the field of financial modeling.
Among the Russian Science Foundation's tasks is the development of the prioritized trends of Russia's scientific and technical development, as well as support of scientific communities that work in corresponding areas. This is why the project's team has to not just create a practice-oriented solution for banks and clients, but also give rise to the development of new methods of mathematical modeling of financial processes.
The work on the project will be conducted on the basis of the Institute of Cyber Technologies for Financial Systems (ITMO.Fintech) that launches two new Master's programs in this field in September - Mathematical Support and Software for Global Financial Systems and Big Data in Financial Technologies.
"It's most important for us that the Institute develops educational and research activities in a unified manner. This will allow to get a synergetic effect from the participation of experts in the educational process and involvement of Master's students in solving this most serious and relevant task," notes Maria Sigova, Director of ITMO.FinTech Institute.
The development of a common open decision support platform for the financial sector will allow to form the scientific community for this research field. Scientists and users will be able to use the platform for improving their own solutions.
"The current problem is that specialists in the fields of financial mathematics and data analysis often work independently from each other. Thus, the former lack the knowledge and experience in defining the properties of their models, and the latter create models based on their personal experience, which greatly limits their use. We want to unite these efforts, which is why we focus on developing not just some particular model for financial experts or banks, but a general-purpose platform (methodological, algorithmic, software-based) which will serve as a basis for applied research and inventions in this field," comments Alexander Boukhanovsky.
As part of the project, the research team will develop machine learning algorithms for creating and identifying predictive models for financial processes based on a data-driven approach. They will also develop and research multiscale modelling methods for global financial systems with regard to their scale. The researchers plan to work out high-performance computational technology and the infrastructure for gathering, storing and processing large amounts of data, as well as solve other tasks, including those in the field of theoretical research that have to do with creating new classes of machine learning algorithms. According to Alexander Boukhanovsky, the project's most difficult part will be developing the mathematical formalism of the models, and the most resource-consuming part will be coding, troubleshooting and experimental research.
"Here at ITMO University we have vast experience of social and behavioral modeling, and Janusz Holyst, the project's head, will add his results from the field of financial mathematics and econophysics. Yet, the process of combining the competencies from the fields of financial mathematics and data analysis is most difficult, as we will be modeling "financial behavior". As for the infrastructure, the logic of such "digital platform" has already been developed for the cloud-based software-based system CLAVIRE. We plan on using it for creating our own platform," adds the Head of the School of Translational Information Technologies.
Fintech solutions create additional opportunities for the financial sector. Yet, with these opportunities come new challenges.
"As the digital economy develops, the contradictions that have existed for some time already become more evident, and new factors that restrict the development processes emerge. Among those are the limitations set by existing infrastructure, access to information, the costs of technology and outdated legislation. Digital technologies and the economic activity fueled by them develops at a speed that exceeds the adaptive abilities of traditional legislation. The experience of economic reforms in countries that succeeded in maintaining the high rates of economic growth tells us that the increase of the importance of technologies in economic development has to come with the changes in the mechanisms of government regulation. Fin.Tech's main goal is to increase the transparency of the financial sector and to turn financial institutes into interrelated elements of a common ecosystem, and we believe that using a scientific approach will become an essential argument speaking in favor of correcting the current government regulation in the Fin.Tech field," says Maria Sigova.