1) The document discusses using recurrent neural networks (RNNs), social media data, and cloud computing for financial forecasting.
2) It mentions how financial forecasting has evolved from oracle bones and box-jenkins models to modern machine learning frameworks.
3) The document demonstrates a hybrid RNN and Holt-Winters model for generating 14-day stock price forecasts based on historical daily data with high accuracy.
2. # DataShowDown
Lucas W
Cloud Lead
Sydney
Yun Zhi Lin
Head of Engineering
Sydney
Raymond Au
Data Engineer
Sydney
Team Quantino Global
Sami Raines
Data Engineer
Melbourne
Ira Cohen
CoFounder Anodot
Israel
3. Evolution of (Financial) Forecasting
Oracle Bones Box and Jenkins
ARIMA
Excel Machine Learning
Frameworks
4. Social Media and Trump Driven Data
https://www.bloomberg.com/features/trump-tweets-market/
$1.3 billion wipe out overnight
10. Training and Algorithms Under the Hood
Inputs:
High
Low
Volume
Closing
Model: Hybrid RNN & Holt-Winters
11. Results
● Rolling 14 day forecasts based on historical daily High, Low, Adjusted Closing Price and Volume
● Symmetric Mean Absolute Percent Error (SMAPE) of < 0.01 % for big four banks, from Dec 29th
2017 to Nov 21st 2018
● Presented at NAB 700 on Nov 7th 2018 as part of Data Showdown meetup event