Banking & Finances
❯
Financial Analysis & Trading
❯Finance advising and asset management with AI
Forecasting prices of commodities
For:
Trading company, manufacturers, suppliers,
Scope:
Build a neural network to forecast the price of base metal commodities
Goal:
Other
Finance advising and asset management with AI
Scope:
Financial advising and portfolio management for financial institutions and
consumers
Goal:
Other
Finance advising and asset management with AI
Goal:
OtherProblem addressed
Designed to manage exchange-traded securities portfolios of conservative
investors in real time, using asset price data and macroeconomic data, to make
the most accurate decisions at a given yield and moderate risk. Prediction of
significant depreciation of exchange-traded asset prices as a result of a sharp
monetary contraction (financial crisis).
Scope of use case
Financial advising and portfolio management for financial institutions and
consumers
Description
For the purposes of realizing efficiency that cannot be
achieved by competitors, the project uses more complex
technologies than other standard solutions for building
neural systems. All algorithms of the basic core of the project
are developed by the creators themselves.
The idea that neural systems are absolute, impenetrable
"black boxes" is mythologized. Therefore, by understanding
exactly what technologies are used to achieve analysis goals,
overloaded "boxed" solutions can be optimized. This was
done in the project.
The algorithm of simple regression analysis of prices (model
William Sharpe/Harry Markowitz - Nobel laureates) does
not lead to the required efficiency. Therefore, the project
uses the "complex" model when weighting factors and the
algorithms of simple regression analysis of prices change
depending on the "field," formed by the regression
assessment of other economic parameters.
The William Sharpe/Harry Markowitz model is unacceptably
simplified precisely because it is very resource-intensive.
This is particularly true when it comes to the hundreds of
asset names around the world for the diversification needed
in this model. If we consider applying a straight-line
approach to the assessment of dozens or even hundreds of
additional macroeconomic parameters of each of the dozens
of different countries (and today it is clear that the world
economy is interrelated), we are talking either about
supercomputers and very expensive neural models, or about
building a fundamentally new economic model for the AI
core.
In this project, the regression evaluation of higher-order
macroeconomic indicators "guides" all subsequent lower-
order models. Resource issue resolved.