DAGs improve local projections for causal effects of uncertainty shocks
A new working paper by Burkhard Raunig of the Oesterreichische Nationalbank demonstrates how Directed Acyclic Graphs (DAGs) can guide local projections (LPs) to estimate causal responses. An application to German industrial production reveals substantial negative effects of uncertainty shocks.
Causal clarity for local projections
Directed acyclic graphs (DAGs) offer a transparent framework for encoding causal structures and identifying causal effects, guiding the specification of local projections (LPs) for estimating causal impulse responses.
While LPs are popular for their simplicity, they often lack explicit causal assumptions, making them 'black boxes' that may fail to identify true causal effects.
This paper demonstrates how DAGs can unpack these black boxes by applying formal graphical rules to select appropriate control variables and instruments.
An empirical application to German industrial production reveals substantial negative effects of uncertainty shocks.
The study highlights that estimated responses vary widely across different LP specifications, and DAGs are crucial for clarifying these differences and diagnosing biases from violated causal assumptions.
A DAG-based instrumental-variable LP specifically suggests pronounced negative effects of U.S. uncertainty shocks on German industrial production.
Unpacking causal patterns
The paper addresses a gap in macroeconomic time series analysis by introducing DAG basics and graphical rules for identifying causal effects.
It explains how DAGs encode qualitative causal assumptions, visualizing relationships through nodes and arrows.
Key causal patterns – forks, chains, and colliders – are detailed, highlighting how conditioning on colliders can introduce spurious associations and bias.
The study extends these concepts to time series, demonstrating how DAGs guide the selection of appropriate control and instrumental variables (IV) for LPs, particularly in scenarios with unobserved variables.
This approach clarifies conditions for instrument validity and helps isolate specific causal pathways, enhancing causal inference.
Beyond the black box
This paper fills a critical gap in macroeconomic causal inference, offering a transparent framework for local projections via DAGs.
It transforms LPs from 'black boxes' into robust tools for identifying true causal effects.
The findings on uncertainty shocks highlight the practical importance, showing how explicit causal assumptions can significantly refine policy-relevant conclusions.
Source: DAG-Based Local Projections
IN: