Nova V1.00

Experimental Deep Learning Algorithm for Trading. Doesn't mimic real market representation so mainly for educational and experimental purposes.

Intuition

People like to think intuitively when viewing trading charts. Many seasons traders cannot explain how or why they make their decisions (something scene with many areas of expertise). We aim to model just that by using 2D chart data in the same fashion as viewing charts.

Features

• Open Price
• Close Price
• MA5
• MA10
• MA20

Initial Labeling Method

For Labeling we perform a Sliding Window Approach

We will model 15 days to create an image, and label accordingly:

if open_price_20th_day > close_of_15th_by_2_percent:
    prediction = "buy" # buy on 15th, sell on 20th

if open_price_20th_day < close_of_15th_by_1_percent:
    prediction = "sell" # sell on 15th, buy on 20th 

else:
    prediction = "hold"

Future Work

We will experiment with different Labeling Methods along with other Features

White Paper

https://arxiv.org/pdf/1801.03018.pdf