Presented at QuantCon Singapore 2016, Quantopian's quantitative finance and algorithmic trading conference, November 11th.
In recent years, many funds have moved towards machine learning, where artificially intelligent systems can analyze large amounts of data at speed and improve themselves through such analysis. At the same time, the introduction of cloud and mobile technology has meant that there are now more market participants than ever before in history. Combining these two powerful trends may lead to a breakthrough in quantitative trading that can consistently outperform human managers. Daniel Chia, ex-hedge fund and sovereign wealth systematic trading manager, and now, Co-Founder of fintech start-up Call Levels shares his thoughts on this.
Similar to "Artificial Intelligence Powered By Crowdsourcing - The Next Evolution in Quantitative Trading?" by Daniel Chia, Co-Founder of Call Levels (20)
2. Speaking Today
Daniel Chia
Technical Co-
Founder
Call Levels
Cambridge (BA)
Harvard (AM)
Math & Statistics
VARMA TSA, Bayesian Networks SL, MCMC RL
R, .Net, Java, C++
Hedge Fund and Sovereign Wealth
Quantitative Portfolio Manager
Previously Head of FX Quantitative
Trading Team at GIC & Hedge Fund
Portfolio Manager
Disclaimer – Views and Research Expressed are My
Own
4. Real-Time Machine Learning
Industrial Application
of Machine Learning
Techniques that
Predict the Future
New Data Arrives
Constantly, Non
Stationary Correlations,
unsuitable for batch
training
Predictions have HARD
Decision Deadlines
5. Icons designed by Freepik and distributed by Flaticon
Real-Time Machine Learning is NOT Time
Series Prediction
Incremental,
Potentially Non
Stationary Data
Distributions.
Regression is NOT the
Norm
With Businesses,
Government Policy and
Central Banks do take
action incrementally
Examples:
Japanese Monetary Policy
Quantitative Easing
6. Icons designed by Freepik and distributed by Flaticon
An Example
Japanese Monetary Policy
Jul 14, 2006
BoJ Raises STIR from 0
2009 – 2011
USDJPY and US 10 Year Yields have
an Intraday 5min Delta Correlation of
>90%
2012
Delta Correlation is Negative
Similar Occurrences in
AUD vs S&P Futures
Oct 31, 2008
BoJ Cuts STIR
Dec 2012
Abenomics
8. We are Moving to a Cloud Based Economy
Why is Real-Time Machine Learning
Critical for Financial Markets?
9. Approaches to
Real-Time Machine Learning
Building Incremental
Algorithms that update itself
with new data
Increased Frequent
Retraining of Batch
Algorithms
Cloud Distributed
Training Structure
10. CHALLENGES
Incremental Data
needs to be Labelled.
And creating labelled
data is the slowest and
most expensive process
in ML
Data Relevance Assumptions: needs
to be weighted according to data
horizon and seasonality
Convergence: May not
be ready before the
next hard prediction
deadline!