Deep Financial Portfolios :: A Hybrid Structure

Work of Tapasya Pratap Singh & Devang Upadhyay under the guidance of Prof. Abhijeet Chandra, when the former two were working as Research Assistants in the Vinod Gupta School of Management, IIT Kharagpur

MPT :: Reconstructing the Markovian Efficient Frontier

Modern Portfolio Theory has :

• The idea of risk-averse investors
• The idea to maximize profits and minimize risk

Here the following has been done to generate the Efficient Frontier

• 5 Securities of NASDAQ[data from Quandl] have been used in the portfolio
• Portfolio weights have been randomized and generated to give a total of 50,000 portfolios
• Data is from 01-01-2014 to 32-12-2016 [Average of 250 Trading days/year]
• Sharpe-Ratio has been included as a measure of Return/Risk

AutoEncoder :: Outputs

• The AutoEncoder works like a black box model and learns a complex and better representation of the market data
• The encoded data is difficult for us to comprehend. Some of the securities show a lot of variance with the original, while others, close to zero

Re-Tracing the IBB Index ::

Outperforming the IBB Index ::

• In a seperate instance of the model used to re-trace the index, data was augmented in the caliberation phase to replace all the returns smaller than 5% by 5%, this ensures anti-correlation in the periods of large drawdowns
• It is clearly visible that the portfolios of 65 securtites outperform the original Index

Unique Observation :: Spikes in the performance of varying portfolio size

License

MIT