The world of business is filled with an abundance of information. Meaningful insights are often based on lengthy, tedious calculations, requiring hours of data collection, number-crunching, and spreadsheet grinding. Students and analysts should be able to focus on more significant and meaningful tasks, such as generating insights and financial decision-making.
DeepFinalyst collects data from several sources to evaluate companies based on current market values and future projections. Using a deep learning algorithm and collected data, DeepFinalyst analyzes industry patterns and market trends to enhance cash flow predictions, calculate key financial indices, and ultimately provide valuations for specific companies compared to their market values.
DeepFinalyst was constructed in two distinct components. This helped segregate segments of the project with drastically distinct functionalities.
Language: Python, Flask, and HTML. Modules and APIs: Tensorflow, Numpy, Goldman Sachs Marquee API, and Financial Modelling Prep API.
Logic
The core logic behind the financial analysis process was developed using Python. By combining Tensorflow's neural network implementations with several financial data APIs, such as Marquee, we developed and trained a deep learning algorithm to recognize current industry trends and patterns. This was used in conjunction with existing financial modelling procedures to calculate important financial indices and conduct a DCF analysis given a specific company.
Interface
The web interface was built using a combination of HTML, CSS, and JavaScript. We used Flask to integrate the interface with our Python scripts, as well as build and deploy our project. Furthermore, JSON was used to model and facilitate data communications between processes.
Finding the necessary data for DeepFinalyst proved exceptionally difficult, as we had to collect and combine data from several different APIs to create a comprehensive analysis. Additionally, the deep learning algorithm is relatively unstable, as historical training data is only relevant until a certain point, after which it becomes too archaic, leaving very little training data to work with.
It works!!! ML Model within 5%-10% accuracy.
Don't watch anime while working, especially if it's a good one.
There are quite a few optimizations we'd like to add on DeepFinalyst.
Financial Modelling
Improve our financial models by accounting more investment factors if granted further access to advance and detailed datasets.
Neural Network
With more time, we can train a further complex neural network model - e.g. a convolutional neural network. Such changes would improve the effectiveness of training our deep learning algorithm, as it would better identify and adapt to patterns in market data.