Comparative analysis of GEN-A.I. and machine learning models in predictive financial modelling techniques

By: Contributor(s): Material type: Computer fileComputer filePublication details: Ahmedabad Indian Institute of Management 2024Description: 39 p. : ill. ; includes referenceSubject(s): DDC classification:
  • SP2024/3884 SP003884
Summary: Financial modelling and forecasting cashflows is a key aspect of assessing any business opportunity. This task often involves multiple assumptions and judgment calls based on market, industry, and company data. This project proposes to leverage machine learning techniques, including exploring the efficiency of Generative Pre-trained Transformer (GPT) models in building financial models. In this study, we have assessed existing techniques and best practices in discounted cash flow forecasting. We have also conducted primary research to understand the application of GPT-based tools for financial applications from industry experts. We then built our tool: ‘Your Friendly Neighborhood Finbro’ available on the GPT store. This was done by creating a workflow and acceptable output formats for the GPT trainer. This tool was also reviewed with potential users as part of our primary research. We have also presented a comparative analysis of this tool with traditional time series and machine learning models. Financial modelling is very standardized in its steps and the formulae applicable: CAPM, WACC, FCF, and NPV. The complexity comes from the assumptions and the absence of data points in many cases. These are usually resolved with assumptions and industry averages. Both these require deep financial knowledge and experience. A large language model, like GPT, has 3 benefits: internet access to the latest market data, natural language processing, and contextualizing capabilities. These provide it the potential to be best suited to perform the task of assumption-making. There is a need for a GPT tool that complies with the standard practices of the financial industry. Our study suggests that the GPT tools need to be used along with the traditional tools to arrive at the most accurate results, as each tool excels at a different aspect of forecasting.
Tags from this library: No tags from this library for this title. Log in to add tags.
Star ratings
    Average rating: 0.0 (0 votes)
Holdings
Item type Current library Collection Shelving location Call number Status Date due Barcode
Student Project Vikram Sarabhai Library Reference Students Project SP2024/3884 (Browse shelf(Opens below)) e-Book - Digital Access SP003884

Submitted by Shashwat Kumar Sahoo
Anup Itale

Financial modelling and forecasting cashflows is a key aspect of assessing any business opportunity. This task often involves multiple assumptions and judgment calls based on market, industry, and company data. This project proposes to leverage machine learning techniques, including exploring the efficiency of Generative Pre-trained Transformer (GPT) models in building financial models. In this study, we have assessed existing techniques and best practices in discounted cash flow forecasting. We have also conducted primary research to understand the application of GPT-based tools for financial applications from industry experts. We then built our tool: ‘Your Friendly Neighborhood Finbro’ available on the GPT store. This was done by creating a workflow and acceptable output formats for the GPT trainer. This tool was also reviewed with potential users as part of our primary research. We have also presented a comparative analysis of this tool with traditional time series and machine learning models.

Financial modelling is very standardized in its steps and the formulae applicable: CAPM, WACC, FCF, and NPV. The complexity comes from the assumptions and the absence of data points in many cases. These are usually resolved with assumptions and industry averages. Both these require deep financial knowledge and experience. A large language model, like GPT, has 3 benefits: internet access to the latest market data, natural language processing, and contextualizing capabilities. These provide it the potential to be best suited to perform the task of assumption-making. There is a need for a GPT tool that complies with the standard practices of the financial industry. Our study suggests that the GPT tools need to be used along with the traditional tools to arrive at the most accurate results, as each tool excels at a different aspect of forecasting.

There are no comments on this title.

to post a comment.