FinGPT - Democratising AI Trading!!!

if you want to use large language models to do Financial trading like for example you want algorithmic based trading or Robo advisor then this project fin GPT is exactly giving you that I'm not endorsing this project and this entire videos purely for academic purpose this is not a recommendation to trade real money so use your common sense and always make sure that you consult a professional first given that that we have already put the disclaimer aside now what is fin GPT so fin GPT is an open source initiative by a group called AI for financial foundation and what they have done is they have done two things they have given us how they have collected data and how you can collect data and they've made it available second they've also made couple of models available which they are calling it fin GPT so what is fin GPT a data Centric open source when GPD open source for finance and the first line the very first line that it says let us do not expect Wall Street to open source llms nor apis so the point here is that you do not expect any big corporations or Wall Street to open source anything related to finance and that's why fin GPT exists and fin GPT has its own paper and you can go read about water spin GPT but in this video I'm not going to show you how to execute anything here because there is model weight does not exist you have to train your own model in itself but I'm going to show you everything that they've given about Fin GPT and how you can make use of it to start with let's quickly look at the fin GPT architecture in itself now fin GPT as a system has got four layers the first one is a data source layer the second one is a data engineering layer the third one is an llm layer and the fourth one is an application layer the application layer is where you build applications like Robo advisor and quantitative trading for which they have already given the code so you can go to the repository that they have got on GitHub and if you go there inside fin GPT you can see the code for robo advisor and you can also see the code for um sorry you can see the code for robo advisor and you can see the code for trading so you can see the code for both of these and you also as part of this if you go there you see that they have also used large language models they've done fine tuning so that is also available as part of the code that they shared so you can go here in the same repository and you can go inside the fin GPT repository and you can see okay how they have used fin GPT to train their own charge gpt-like model with the finance which which how they are doing it they're doing it using pre-trained llms and Laura so Laura is a technique that helps you fine-tune models with a lesser resource and just a part of the model in itself so you can see that how do you prepare the data how do you make the data how do you do the fine tuning and how do you do the inference the entire code has been given to you for you to train your own model in itself like they said so if you go back to the architecture the architecture is application you have got the code for you to use it as a robo advisor based on the news or you can use it as a quantitative trading application and you have also got the code to do the fine tuning part or use charge GPT API the layer below that is the data layer and you've got the data engineering and then finally I've got the data source which is one of the main components so if you see any Finance related trading board one of the most important things that they might have is the news data to understand what is happening in the news and the preempt if the stock price will go up or go down and that's exactly what they're doing here now you can go into particular section and try to understand how they are getting all these details and where are they getting the details from starting from the data source in itself from where it is coming but one of the interesting thing is they have also actually mentioned what kind of models that they're using one part of that is they are actually using the aps which is like gpd4 APA the other part is like they're using pre-trained models that can be trained using Lora and the reinforcement learning on stock prices which they are calling as rlsp so the entire detail has been given here and then the final demo is like how if you have got a robo advisor how it would do the robotic advice like automated advice where it could read the Stock's news and then give you an advice whether this company's Trend stock Trend could be on the positive side or the negative side and what kind of reasons could it be and that is available and also if you want to build a quantitative trading application system how do you go about it and they've also shared The Notebook on GitHub reposit you can go there and then see the notebook this is the notebook and they've you know you can say trade with charge EBT click the notebook it will take you to the notebook file The jupyter Notebook file and you can see what kind of things that they're doing like they're loading the file loading the data and all the information I'm not a financial expert so I do not know how much of this is actually good actually bad actually if love I honestly do not have any information but what I found entirely interesting this is this is not anything completely new it's not like Bloomberg GPT it's not like they're building their own model from scratch it's it's not like a rocket science I mean I completely respect what they have done um I respect the efforts that they've gone inside it but it is not a completely radically new approach but what I like is how they've used all the existing resources and put together a very brilliant architecture that can ultimately actually become a good algorithmic trading border Financial engine that can help you leverage large language models for financial trading or financial advice again I do not know what is accuracy effort there is no back testing data available here on all these things are not there but at the foundation level this makes sense it's a very good attempt about combining existing parts of large language models trying to build what you want for a financial Niche here and um I really respect that and I also respect the fact that they have open source as much as what they can open source the model related details The Notebook related details are open source the data related details like if you go to the fin NLP the data related details are open source so they've got their own data how they use and that information is open source and they have also got a very good documentation along with the paper that we just saw so overall this is an exciting space so if you're somebody who is trying to use large language models for finance then you have got an open source project that you can leverage starting today provided you have enough computation and enough data set let me know in the comment section what do you feel about this and I would like to hear from you if you are going to end up using this kind of solution for algorithmic trading or putting your money on it stay safe take care happy prompting

Channel: 1littlecoder

Views: 253

Keywords: ai, machine learning, artificial intelligence

Id: CH3BdIvWxrA

Length: 6min 46sec (406 seconds)

Published: Sun Jun 18 2023