Predicting market uncertainty with machine learning – Vlasios Voudouris

To find out how ML is helping predict commodities markets jump to 12.39

Vlasios is the Chief Data Officer at Argus Media

With a Ph.D. in informatics, Vlasios started his career working in academia. His Ph.D. led him to specialise in managing data sets over geographical space and time.     

When he left university he worked on several consultancy projects and could boast that the UK government was a client. After this Vlasios spent some time setting up his own company where he successfully commercialised a lot of the research he did over his academic career. 

From there he worked in the banking sector where he set up a commercial banks data science capabilities and now he’s at Argus Media as their first Chief Data Officer.    

How is machine learning has changed the commodities trading market

‘Data has been democratised in recent years. This means data sets are readily available to large numbers of market participants. What this means in technical terms is that the Alpha you extract from a data set is reduced. 

The idea of alternative data and pro-creatory machine learning tools have also come to the fore in recent years. A company that successfully manages both of these at the same time can extract a lot of value from their data sets.’ – Vlasios

What are alternative data sets? 

Alternative data refers to data used by investors to evaluate a company or investment that is not within their traditional data sources (financial statements, SEC filings, management presentations, press releases, etc.). Alternative data helps investors get more accurate, faster, or more granular insights and metrics into company performance than traditional data sources. Vlasios explained how Argus uses alternative data sets in the podcast.

‘To get the benefits from alternative data you need to have a data set that’s unique. Argus Media, for example, have a data set called deals data. It’s a data set they use to record the volume and prices that commodities are traded at. 

You can then actually take a data science algorithm and customise it based on the way we understand the market works. To do that you start putting constraints on the algorithm.

Through a combination of market intuition, imposing specific constraints on to algorithms to satisfy market conditions and using a pro-creatory data set that are trained using public data set models you can create alternate data sets.’ – Vlasios 

Further reading

Data Science boom in commodities

Stanford statistical learning course

Show Notes:

01.05 Vlasios’s career

01.50 A CDO’s first year

02.49 CDO challenges

03.59 Developing new products at Argus

05.11 Creating a data culture

06.23 The journey from data to insights

07.37 Data science and financial markets 

08.38 Machine learning’s struggle with commodities

10.14 Solving problems that aren’t i.i.d’s

10.59 The role of feature engineering in commodities markets

12.39 Three ways machine learning has changed the commodities trading market

14.19 AI, ML & commodities in the next 15 years

15.09 Market intuition in data science 

17.13 Predicting market uncertainty

17.57 Further reading recommendations