01

AI ETF

크래프트는 펀드매니저가 지시한 컴플라이언스 정책, 리스크정책, 투자유니버스, 투자방법을 완벽하게 준수하면서 최적 마켓타이밍과 매매전략을 학습할 수 있는 AI ETF 운용시스템을 공급하고 있습니다. 


Qraft offers an AI ETF operating system that has the ability to learn optimal market timing and trading strategies while fully complying with the fund manager's compliance policies, risk policies, investment universes, and investment styles.


 

case study:
horizons AI Global Equity ETF

Horizons ETFs 는 크래프트의 AI ETF 솔루션을 통하여 세계최초의 Global AI ETF 를 Toronto Stock Exchange 에 상장하였습니다.  (Ticker: MIND)  


Horizons ETFs was listed as the world's first Global Equity AI ETF on the Toronto Stock Exchange through Qraft's AI ETF solution. (Ticker: MIND)

mmind.png


 

case study:
Qraft aI enhanced US Large Cap ETF

크래프트와 ETC는 크래프트의 딥러닝 기술이 적용되어 100% AI에 의해 운용되는 ETF 2종의 뉴욕증권거래소(NYSE) 상장을 앞두고 있으며 이는 국내 AI 자산운용 분야 기술 수출의 첫 사례입니다.


Qraft Technologies and ETC will launch on the New York Stock Exchange (NYSE) for two AI-driven ETFs using Qraft’s deep-learning technology. It is the first of its kind in Korea to export AI Asset management technology.

case_study_etc.jpg

02

QraftAI Factor Rotation Engine

Introduction

As we all know, factor matters to stock return. Factor strategy generates alpha on average, but NOT ALWAYS…

스크린샷 2019-03-09 오후 11.20.50.png
 

What matters to factors that matter to stock return

스크린샷 2019-03-09 오후 11.28.26.png

Theoretical background

스크린샷 2019-03-09 오후 11.30.22.png
 

So we had AI learn to rotate factor weights

스크린샷 2019-03-09 오후 11.31.46.png
 

Then how does AI learn ?

스크린샷 2019-03-09 오후 11.37.26.png

Qraft Deep Factor Rotation Engine

1. Reward targeted Asynchronous Multi Network Learning Framework

      - Multi networks compete to maximize reward function asynchronously

      - Our own framework to get optimal portfolio strategy while avoiding overfitting as much as possible.

2. Multi-channel Attention Deep Residual Network 

      - Deep network modeloptimized for finance data to simultaneously capture both time series feature and cross sectional feature

3. Qraft Database 

      - We consolidated the database to feed the data to the network intact

 

Empirical Result

Simulation with train data set (1980-2005) during training

스크린샷 2019-03-09 오후 11.45.27.png

Simulation with test data set (2006-2018) during training

스크린샷 2019-03-09 오후 11.48.33.png

Simulation with test data set after 20000 iterations

스크린샷 2019-03-09 오후 11.53.15.png

How did our AI model make decisions during the financial crisis?

스크린샷 2019-02-27 오후 1.21.21.png

>Reasonable Factor Allocation