Deep Portfolio Theory

Deep Learning in Finance Summit , London, September 2016

Keywords: Deep Learning, Machine Learning, Big Data, Artificial Intelligence, Finance, Asset Pricing, Volatility, Deep Frontier

Transcript

1. h"p://greymaths.technology Deep Por6olio Theory Deep Learning in Finance Summit London, September 2016
2. Source: Jeff Dean, h0p://www.wsdm-conference.org/2016/slides/WSDM2016-Jeff-Dean.pdf h"p://greymaths.technology
3. h"p://greymaths.technology A.I. has been very successful, but finance is a bit different
4. h"p://greymaths.technology regardless of the data available, any two individuals will struggle to agree on the topic of investments
5. h"p://greymaths.technology Source: h0p://awealthofcommonsense.com/2016/05/the-sp-500-is-the-worlds-largest-momentum-strategy/ 40% of all stocks generated no return. Over the period considered, the S&P500 was up almost 1200%.
6. h"p://greymaths.technology Source: h0p://awealthofcommonsense.com/2016/05/the-sp-500-is-the-worlds-largest-momentum-strategy/ 40% of all stocks generated no return. Over the period considered, the S&P500 was up almost 1200%. the top 10% of stocks achieve 90% of returns, and alpha is a needle in a haystack
7. h"p://greymaths.technology Deep Learning Tracking the IBB Biotechnology Index
8. h"p://greymaths.technology Source: h"p://ssrn.com/abstract=2838013 RHS shows no. of stocks used on x-axis, and 2-norm validaVon error on y-axis.
9. h"p://greymaths.technology Deep Learning Outperforming the IBB Biotechnology Index
10. h"p://greymaths.technology Call OpVon Payoff AcVvaVon FuncVon (ReLU) the payoff funcVon of a call opVon is a recVfied linear unit (ReLU)
11. h"p://greymaths.technology the hidden layers of a deep learning rouVne are composiVons of financial put and call opVons on the underlying assets
12. h"p://greymaths.technology Source: h"p://ssrn.com/abstract=2838013 For the amended calibraVon target, we replace all returns <-5% by exactly 5%.
13. h"p://greymaths.technology Source: h"p://ssrn.com/abstract=2838013 RHS shows no. of stocks used on x-axis, and 2-norm validaVon error on y-axis.
14. h"p://greymaths.technology deep por6olios extend Markowitz‘s (1952) linear por6olio design to achieve previously impossible targets