# Intelligent Systematic Investment Agent: an ensemble of deep learning and evolutionary strategies

**Authors:** Prasang Gupta, Shaz Hoda, Anand Rao  
**arXiv:** [2203.13125](https://arxiv.org/abs/2203.13125)  
**Date:** 2022-03-24  
**Summary page:** https://www.summarizepaper.com/en/paper/2203.13125v1/

## Key points

- Machine learning driven trading strategies have gained significant interest, particularly in short-term trading strategies.
- Limited consensus exists on the ideal approach for developing such strategies, with little attention given to long-term wealth building strategies.
- The paper proposes a novel approach using an ensemble of evolutionary algorithms and a deep learning model for long-term investment strategies focusing on Exchange Traded Funds (ETFs).
- The methodology aims to improve systematic investment planning (SIP) decisions over time to build wealth by exploiting short-term volatility instead of averaging it out.
- The objective is to improve annual returns by 10% through making marginally better daily buying decisions over a lifetime of investing.
- The algorithm breaks down the infinite investment horizon into finite windows of 30 trading days, offering flexibility and potential for improved performance over conventional daily SIPs.
- By balancing buy-twice and not-buy actions over multiple 30-day periods, the algorithm aims to converge towards daily SIP trade counts in the long run.
- The new algorithm, GADLE, serves as an alternative to traditional reinforcement learning techniques and has shown promising results in both out-of-sample tests and live production environments.

## Summary

Machine learning driven trading strategies have gained significant interest in recent years, with a focus on short-term trading strategies. However, there is limited consensus on the ideal approach for developing such strategies, and little attention has been given to long-term wealth building strategies. This paper proposes a novel approach for developing long-term investment strategies using an ensemble of evolutionary algorithms and a deep learning model to make short-term purchase decisions. The methodology aims to improve systematic investment planning (SIP) decisions on Exchange Traded Funds (ETFs) over time to build wealth. The problem statement addresses the challenge of making marginally better daily buying decisions over a lifetime of investing to achieve higher returns than traditional SIPs. By exploiting short-term volatility instead of averaging it out, the objective is to improve annual returns by 10% (from 8.8-12.1%). The study breaks down the infinite investment horizon into finite windows of 30 trading days, allowing investors to decide whether to invest twice or not at all in the chosen ETF each day. This approach offers flexibility and potential for improved performance over conventional daily SIPs. The algorithm assumes sufficient liquidity for double the regular monthly investment target, enabling it to make larger investments when needed. By balancing buy-twice and not-buy actions over multiple 30-day periods, the algorithm aims to converge towards daily SIP trade counts in the long run. The new algorithm, GADLE, serves as an alternative to traditional reinforcement learning techniques and has shown promising results in both out-of-sample tests and live production environments. Overall, this paper introduces an innovative approach for developing long-term investment strategies that focus on wealth building through improved short-term purchase decisions. By combining evolutionary algorithms and deep learning models, the proposed methodology demonstrates superior performance compared to traditional daily systematic investment practices on ETFs, offering potential for increased returns and compounding effects in the long run.