Glossario IA
Il dizionario completo dell'Intelligenza Artificiale
Causal Inference in Time Series
Statistical discipline aimed at identifying and quantifying cause-and-effect relationships between time-evolving variables, distinguishing correlation from causality. It combines econometric methods and machine learning to analyze temporal dependencies and delayed effects.
Dynamic Causal Graph
Graphical representation of causal relationships evolving over time, where nodes correspond to variables at different times and edges indicate temporal causal influences. This structure allows visualization and analysis of causal mechanisms with delays and feedback loops.
Causal VAR Model
Extension of Vector Autoregressive models incorporating structural constraints based on causal assumptions to identify economic shocks and their propagation effects. These models allow distinguishing exogenous shocks from endogenous movements in multivariate time series.
Causal Impulse Response Function
Analytical tool measuring the temporal effect of a unit causal shock on a variable of interest, controlling for other influences in the system. It quantifies the dynamics and persistence of causal effects at different time horizons.
Temporal Causality Test
Formal statistical procedure to evaluate the presence of causal relationships between time series, including Granger tests, Sims tests, and modern Bayesian approaches. These tests allow validating or rejecting hypotheses about the existence and direction of causal effects.
Temporal Counterfactual
Hypothetical scenario describing what would have happened if a causal intervention had occurred at a specific moment in the past, in the absence of this intervention. Counterfactual analysis in time series allows quantitatively estimating the impact of specific policies or events.
Causal Variance Decomposition
Method analyzing the proportion of forecast error variance of a variable attributable to causal shocks from other variables in the system. It quantifies the relative importance of different causal transmission channels at different time horizons.
Temporal Instrumental Variable
Exogenous lagged variable used to identify causal effects in the presence of confounding factors in time series, satisfying relevance and exclusion conditions. This approach allows overcoming endogeneity biases in the estimation of dynamic causal relationships.
Causal Transmission Mechanism
Detailed description of the process by which a cause produces its effect over time, including mediating variables and propagation delays. The identification of mechanisms allows understanding how and when causal effects manifest in time series.
Temporal Causal Effects Heterogeneity
Phenomenon where causal effects vary according to the timing of intervention or system state, requiring flexible models to capture these non-stationary dynamics. This heterogeneity may reflect structural changes or temporal context dependencies.
High-Frequency Causal Inference
Application of causal methods to data observed at very short intervals, posing specific challenges of microstructure and measurement noise. These approaches allow capturing quasi-instantaneous causal mechanisms in financial markets or physical systems.
Non-linear Temporal Causality
Extension of causality tests to systems where relationships between variables follow non-linear dynamics, requiring kernel-based approaches, neural networks or entropy methods. These methods detect complex causal dependencies invisible to traditional linear approaches.
Causal Change Point
Moment in time when the structure of causal relationships between variables changes significantly, requiring adaptive detection and modeling. The identification of these points is crucial for understanding regime transitions and structural breaks.
Temporal Causal Synthesis
Machine learning method generating realistic counterfactuals for time series, respecting causal constraints and temporal dependencies. This approach allows simulating alternative scenarios and evaluating impacts of past or future interventions.