Kattintson ide a magyar verzióért.
Surely many of us have had the feeling when browsing through recommendations on the web that the system can read our mind. It is almost terrifying how well some of these online recommendation engines have gotten to know us, be it the ones used by YouTube, Netflix or certain online marketplaces. This is no coincidence, since the technology that sits behind these solutions is rapidly evolving. But what do we mean by “recommendation engine” precisely?
Recommendation engines are software that make personalized recommendations based on the data fed to them. This data can be any information about our behavior and preferences, such as our search history, the device we are browsing on, our geolocation, our online interactions, or more personal information such as our age or gender. Nowadays it is almost impossible to avoid recommendation software while browsing the Internet as they are now an essential part of social media, online entertainment and e-commerce, and in recent years they have started conquering the investment management industry.
In our current blog post, we first look at how recommendation engines are utilized in the world of investments, then we proudly present our own investment recommendation engine developed here, at Dorsum. Let’s dive in!
It comes as no surprise that the “hyper personalization” achieved through the use of recommendation engines had made it to the world of investments. If these solutions are able to generate product recommendations from our browsing habits or movie recommendations from our viewing habits, why couldn’t we use them to provide investment recommendations by looking at financial behavior? The recipe is the same: products and services are recommended to users based on information gathered from large data sets using algorithms. The main difference is that in the case of investments, data sets consist of financial and capital market data that are used to recommend investment products and services. We should note that this required financial institutions to start storing and organizing these already available data in a way that makes them easily processible by recommendation systems.
When recommending investment products, however, we cannot just copy the general formula used in e-commerce due to the unique features of the market. Because while the logic behind the two is basically the same, the characteristics and environment of the recommended products are very different. For example, clients of financial institutions do not rate products explicitly. Plus, certain characteristics of financial products (e.g., the price of stocks) might change radically from one minute to the next which can have a huge financial impact on investors. Which is entirely different from buying a shoe for example, where we can leave a customer rating that is actually relevant, as the next customer will buy the same product in (mostly) the same condition. The price of the product won’t plummet either after a negative comment.
Due to these features, a client’s risk profile and their investment objectives (MiFID), for example, are crucial pieces of information for accurate investment-related recommendations. Then, investors are profiled based on past transactions or certain traits and characteristics (e.g., social and psychological factors). Finally, we can further refine these recommendations with the use of AI technology, which seems to have become a staple in cutting edge recommendation engine software, since it is surprisingly good at uncovering personality traits and characteristics that we do not perceive as outside observers.
Dorsum’s Recommendation Engine is also a sophisticated, AI-based investment recommendation solution. It uses a form of deep learning AI technology called a “neural network”, which is a hardware- or software-implemented information processing system that is capable of parallel, distributed operation. It consists of an interconnected network of neurons (processing elements) modeled after those of the human brain. It has a learning algorithm, and a retrieval algorithm that utilizes the learned information.
Recommendation Engine has the ability to learn what instruments can be recommended to investors by organizing them into groups based on patterns it reveals in data sets consisting of historical transactions and other customer data, practically creating its own ruleset. This is called the learning phase. The retrieval phase typically takes much less time, during which the data of a customer is analyzed against the patterns learned by the system. At the end of the retrieval phase, it outputs the set of investment products that fit the client’s profile and can be recommended to them.
A so-called preprocess is carried out before data is fed to the neural network. We can think of this as pre-filtering: we can tell the system to ignore certain information (e.g., if we know beforehand that the residents of Budapest do not conform to the pattern), or we can pre-define groups (e.g., we can separate metropolitan, city, small-town, rural etc. residents).
An additional postprocess allows for data manipulation after the data has been processed by the neural network. In doing so, we can influence the results so that they align with business or marketing decisions. We can change the weighting of a particular product, increasing its chances of being among the recommended set of products or we can even fix certain products to make sure they are included in the final recommendation.
The potential behind Recommendation Engine is nigh impossible to assess, as the accuracy of recommendations can in theory be increased indefinitely by carrying out learning processes on new data sets or by fine-tuning the operating parameters. Improving data quality should also increase efficiency, and even new types of data may appear in the future that are not currently available or do not exist.
AI-based systems are without a doubt still in their infancy, but it is already clear that they will have a huge impact on the future of humanity. With Recommendation Engine we want to be a part of this breakthrough and bring the technology to the investment market.
Dorsum is committed to further R&D on the future of profiling and recommendation engines as part of tender 018-1.3.1-VKE-2018-00007 „Automated and verified evaluation and categorisation of financial products and service providers for secure, personalised investment services”.