Abstract:Instrumental Variable (IV) provides a source of treatment randomization that is conditionally independent of the outcomes, responding to the challenges of counterfactual and confounding biases. In finance, IV construction typically relies on pre-designed synthetic IVs, with effectiveness measured by specific algorithms. This classic paradigm cannot be generalized to address broader issues that require more and specific IVs. Therefore, we propose an expertise-driven model (ETE-FinCa) to optimize the source of expertise, instantiate IVs by the expertise concept, and interpret the cause-effect relationship by integrating concept with real economic data. The results show that the feature selection based on causal knowledge graphs improves the classification performance than others, with up to a 11.7% increase in accuracy and a 23.0% increase in F1-score. Furthermore, the high-quality IVs we defined can identify causal relationships between the treatment and outcome variables in the Two-Stage Least Squares Regression model with statistical significance.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to optimize the identification and interpretation of instrumental variables (IV) through a professional - knowledge - driven method in the financial field. Specifically, existing methods for constructing instrumental variables usually rely on pre - designed synthetic instrumental variables, and their effectiveness is evaluated by specific algorithms. However, this method cannot be widely applied to problems that require more specific instrumental variables. Therefore, the paper proposes a professional - knowledge - driven model (ExperTise - driven modEl for Financial Causal variables identification and interpretation, abbreviated as ETE - FinCa), aiming to directly identify instrumental variables from textual professional knowledge, instantiate instrumental variables through professional - knowledge concepts, and interpret causal relationships in combination with actual economic data. ### Main problems and solutions: 1. **Limitations of instrumental variable identification**: Traditional methods have limitations in the selection of instrumental variables, especially when more specific instrumental variables need to be selected for specific problems or situations. The ETE - FinCa model improves the diversity and applicability of instrumental variables by extracting them from professional knowledge. 2. **Interpretation of causal relationships**: Traditional instrumental variable methods often lack transparency and interpretability when explaining causal relationships. The ETE - FinCa model enhances the ability to interpret causal relationships by combining causal knowledge graphs and actual economic data. ### Main contributions of the paper: - **Optimizing the sources of instrumental variables**: Through a professional - knowledge - driven method, the sources of instrumental variables are optimized to better meet the needs of actual financial problems. - **Instantiating instrumental variables**: Instrumental variables are instantiated through professional - knowledge concepts, which improves the pertinence and effectiveness of instrumental variables. - **Interpreting causal relationships**: By combining causal knowledge graphs and actual economic data, a new method for interpreting causal relationships is provided, which enhances the interpretability and transparency of the model. ### Experimental results: - **Classification tasks**: The feature - selection method based on professional knowledge significantly improves classification performance, with an accuracy rate that can be increased by up to 11.7% and an F1 score that can be increased by up to 23.0%. - **Instrumental variable mining**: Through the depth - first search (DFS) algorithm, a large number of eligible instrumental variable patterns are mined from the financial causal knowledge graph. - **Quality classification of instrumental variables**: The mined instrumental variables are classified by quality, high - quality instrumental variables are identified, and the effectiveness of these instrumental variables is verified through the 2SLS regression model. In conclusion, this paper proposes a new professional - knowledge - driven method, which solves the challenges of identifying and interpreting instrumental variables in the financial field and provides new ideas and tools for financial causal inference.

Causal Inference in Finance: An Expertise-Driven Model for Instrument Variables Identification and Interpretation

Sequential Underspecified Instrument Selection for Cause-Effect Estimation

Instrumental Variables in Causal Inference and Machine Learning: A Survey

Valid Causal Inference with (Some) Invalid Instruments

Instrumental Variable Model Average With Applications in Nonlinear Causal Inference

Disentangled Representation Learning for Causal Inference with Instruments

BudgetIV: Optimal Partial Identification of Causal Effects with Mostly Invalid Instruments

Bayesian variable selection in linear regression models with instrumental variables

Ivy: Instrumental Variable Synthesis for Causal Inference

Causal Inference with Non-IID Data under Model Uncertainty

Instrumental variable analysis in the presence of unmeasured confounding

Causal Inference under Interference and Model Uncertainty

Active Causal Effect Identification with Expert Knowledge

Discovering Ancestral Instrumental Variables for Causal Inference From Observational Data

Counterfactual and Synthetic Control Method: Causal Inference with Instrumented Principal Component Analysis

Discovering Causal Models with Optimization: Confounders, Cycles, and Instrument Validity

Leaning Time-Varying Instruments for Identifying Causal Effects in Time-Series Data

Auto IV: Counterfactual Prediction via Automatic Instrumental Variable Decomposition

Control Function Instrumental Variable Estimation of Nonlinear Causal Effect Models

Achieving Reliable Causal Inference with Data-Mined Variables: A Random Forest Approach to the Measurement Error Problem

Counterfactual Prediction with Deep Instrumental Variables Networks