PROJECT TITLE :

Detection of Causal Relationships Based on Residual Analysis

ABSTRACT:

The detection of causal interactions between variables from time series knowledge is a crucial problem in several research areas. Granger causality may be a well-known approach that uses prediction error to infer causality. However, the autoregressive models fitted to information usually do not pass model validation tests based mostly on residual analysis, resulting in low causality values that can be inconclusive. The strategy proposed here fits models for paired combination of all variables and inferences regarding causality are provided when performing residual analysis. The model order is increased until the autocorrelation take a look at of residual and cross-correlation check of residuals and input offer a solution about causality. The thresholds to come to a decision the existence of causality are provided directly by the information. Higher order multivariate systems are equally considered and a test to check if causality is direct or indirect is also proposed. The utility of the proposed approach is illustrated by many examples including application on a simulated data set and routine operating knowledge from business for causality analysis.