machine learning & big data in financial...

© Man 2016

Machine Learning & Big Data in Financial Services

Anthony Ledford – Chief Scientist, Man AHL Research Laboratory, Oxford

Fund Forum International: June 2016

For investment professionals only. Not for public distribution.

www.man.com

The value of an investment and any income derived from it can go down as well as up and investors may not get back their original amount invested. Alternative investments can involve significant additional risks.

This material is for information purposes only and does not constitute an offer or invitation to invest in any product for which any Man Group plc affiliate provides investment advisory or any other services. The content is not intended to constitute advice of any nature nor an investment recommendation or opinion regarding the appropriateness or suitability of any investment or strategy and does not consider the particular circumstances specific to any individual recipient to whom this material has been sent.

Opinions expressed are those of the author as of the date of their publication, and are subject to change.

Some statements contained in these materials concerning goals, strategies, outlook or other non-historical matters may be “forward-looking statements” and are based on current indicators and expectations at the date of their publication. We undertake no obligation to update or revise them. Forward-looking statements are subject to risks and uncertainties that may cause actual results to differ materially from those implied in the statements.

Unless stated otherwise the source of all information is Man Group plc and its affiliates as of the date on the first page of this material.

This material was prepared by AHL Partners LLP (company number OC380907) which is registered in England and Wales at Riverbank House, 2 Swan Lane, London, EC4R 3AD. Authorised and regulated in the UK by the Financial Conduct Authority. This material is distributed pursuant to global distribution and advisory agreements by subsidiaries of Man Group plc. Specifically, in the following jurisdictions:

Germany: To the extent this material is used in Germany, the communicating entity is Man (Europe) AG, which is authorised and regulated by the Liechtenstein Financial Market Authority (FMA). Man (Europe) AG is registered in the Principality of Liechtenstein no. FL-0002.420.371-2. Man (Europe) AG is an associated participant in the investor compensation scheme, which is operated by the Deposit Guarantee and Investor Compensation Foundation PCC (FL-0002.039.614-1) and corresponds with EU law. Further information is available on the Foundation's website under www.eas-liechtenstein.li. This material is of a promotional nature.

This material is not suitable for US persons.

This material is proprietary information and may not be reproduced or otherwise disseminated in whole or in part without prior written consent. Any data services and information available from public sources used in the creation of this material are believed to be reliable. However accuracy is not warranted or guaranteed. © Man 2016 P/16/0914/O/I/C

Important information

2 © Man 2016

Machine Learning – what is it?

– Algorithms that identify and act on repeatable persistent patterns in observed data

– Not explicitly instructed what to look for

Big Data - powerful techniques are needed for dealing with

– Large data-sets from multiple sources and in different formats (not just “a lot of data”)

– Complex relationships eg. nonlinear interactions

How can Machine Learning and Big Data add value?

– Better modelling of relationships between existing factors

– Discover new factors in existing data

– Discover and incorporate new factors in new data sets

Challenges:

Finance data is not the typical domain for these algorithms

Low signal-to-noise ratio, non-stationarity

Black box perception, algorithm aversion 3

Machine Learning and Big Data in Finance

© Man 2016

4

Machine Learning – multiple approaches & paradigms

© Man 2016

Neighbourhood

Smoothing

Support Vector

Machines

Bayesian

Nonparametrics

Deep Neural

Networks

A maturing hybrid discipline involving statistics, computer

science, engineering, mathematics … and always data

Also called Supervised Learning and Function Approximation

5 © Man 2016

The Prediction Problem

f Inputs Data streams

Predictors

Features

Independent

variables

Output Response

Target

Dependent

variable

Signal

Output = f(Inputs) + Noise

Learn f(.) from observed noisy data, then use it with new data

How can a machine recognise handwritten digits?

Use handcrafted rules / heuristics?

leads to a proliferation of rules / exceptions

Use a machine learning approach

• Pre-process data into a regular format

• Train model on a vast “training set”

• Apply to as yet unseen data

State-of-the-art systems have near perfect accuracy

6

Example: Recognising hand-written digits

© Man 2016

Challenge 1: Signal-to-noise ratio

7

Machine Learning in Finance

© Man 2016

Hand written digits

Finance

Challenge 2: Non-stationarity

8


© Man 2016

Task: Estimate the signal within a noisy data cloud

Source: Man Group database. Illustrative example - for information only.

10

Modelling Financial Market Data

© Man 2016

Linear model representation

Source: Man Group database. Illustrative example - for information only.

11


© Man 2016

𝑦 = 0.8 − 0.66𝑥1 + 0.43𝑥2

Non-linear machine learning model representation

Source: Man Group database.

Illustrative example - for information only.

12


© Man 2016

𝑦 = 𝑓 𝑥1, 𝑥2

Powerful algorithms for modelling large and highly complex data-sets

Can unlock non-linear relationships and new data-sets

Enables principled inference that combines multiple data sources

Man AHL – successful track record of developing and trading machine learning systems

Powerful tools that come with their own challenges – In my experience nothing simply works out of the box – Need to shape the problem to enable Machine Learning to work for you Careful matching of data with a suitable modelling approach In depth understanding of both is needed

Conclusions

13 © Man 2016


machine learning & big data in financial...

Documents