Krishna Prasad

Algorithmic Trader & Director of operations

Dive Into the New Age of Algorithmic Trading

Our Algorithm


The Beginning

We came across an open source project called the 'RL Trader'- a reinforcement learning based algorithm training and evaluation framework for cryptocurrencies. We don't fancy crypto currencies due to its low reliability nature, thus we started working on bettering the framework altogether to make it work on general forex and stocks. Failure taught us its' not only hard but also near impossible for a team of 2-3 to accomplish such huge task. 



Another version of 'RL Trader' was uploaded on github; a new open source project with 1000s of contributors and huge funding. This was the project TensorTrade1.0 which is a reinforcement learning framework to build, evaluate and algorithms for trading, unlike the 'RL trader' this framework was vast and could support any asset, forex pairs, stock, cryptos, Indian stocks and more. We picked this up and started building an algorithm.


The Algo

We built a reinforcement agent with an advanced reward scheme, this agent's job was to check through the outputs of various other deep learning models, evaluate their relevance to the market data and perform feature selection. Under this agent we placed 7 individual algorithms: RCNN, DQN, NLU-Ensemble, GARCH-LSTM, XGB-SVM, Disco-GAN and raw-RL namely, each of which perform a different kind of analysis and provide input for the final RL agent post feature selection and feature engineering process.



Adding a ton of algorithms usually cause a ton of problems, and though the framework was huge enough to handle complexity we still had to deal with unnecessary noise in the input data post feature selection before predicting the final trade decision. In the Sept 2020 Google announced  their RigL algorithm paper which was a sparse encoding algorithm for neural networks, we integrated a modified version of the same and finally built the ultimate ST.V1.4 Algorithm with 8.5/10 back testing score on past 3 years of market data on 50 top stocks, 8 top performing forex pairs and 50+ other financial assets. and we use a tactical momentum asset allocation strategy to pick the best asset during a trade session.



We understand the importance of cloud integration. To be able to run an algorithm with continuous input of new data 24/7 is exhausting especially in terms of hardware. We hosted our complete algorithm on Amazon's Sagemaker, which is specifically designed for AI and Machine Learning projects.



Although the algorithm is self sufficient, enabling it to trade on multiple broker platforms as per our requirement was tough as each one requires a new api and the multi-thread processing. Thus, we wrote one Api which can communicate model's decisions realtime and with that we are also able to use metatrader charting tool to directly get output from the algo and monitor it visually. 



One advantage with algorithmic trading is that it doesn't require a monitoring 24/7, still as a feature of security at the startup stage we do monitor the algorithm's decisions throughout the session using meta trader as said above. An additional advantage of using the metatrader is the 'Autotrade' feature which enable us to connect to any broker platform we use; directly by logging in to the broker's server. This way the need to have an individual api for each broker platform is unnecessary.

Welcome to the ocean of safe investment


Investment Plans

Annual plan

Duration: 12 Months
Returns: +40%

Mode: Hedge investment

Markets: Foreign Exchange & International Stocks

Long Term Plan

Duration: Minimum 2 years

Returns: +45% compounded annually

Mode: Hedge Investment

Markets: Foreign Exchange & International Stocks