Combining Causal Discovery Networks with VAR for Market Forecasting

Summary

This article outlines a proposed trading system that represents instruments and market variables as a network, applies causal discovery, and uses a Vector Autoregression model to forecast multiple related time series. It describes FCI as the causal discovery method used in the example and contrasts it with PC: FCI can account for latent confounders and selection bias, but is more computationally demanding and may leave edge directions unresolved. VAR is presented as a linear model in which each variable depends on past values of itself and the other variables; lag and variable selection are described as optimization tasks.

The article gives conceptual explanations, model equations, and MQL5-oriented implementation sketches, then reports qualitative performance changes after adding symbols to the network. It also acknowledges that some performance metrics declined and that results depend on how well the network represents market dynamics. The displayed material does not provide enough methodological detail or controlled evidence to establish that causal discovery or VAR improves future trading outcomes; its claimed benefits should be treated as hypotheses requiring careful validation.

Key ideas

• The proposed workflow builds a network of market variables, searches for causal links, and feeds relationships into a VAR forecasting model.
• FCI is presented as useful when latent confounders or selection bias may exist, with greater computation and less decisive edge directions than PC.
• VAR models linear interactions among multiple time series using their lagged values.
• Adding symbols is reported to improve several trade statistics while some other performance measures decline.
• The article offers implementation sketches and qualitative claims, but insufficient detail to establish predictive or trading efficacy.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.