Hi everyone,
I’m working on a project where I aim to understand the impact of subjective inflation uncertainty on expected wage growth. The regression I want to estimate is the following:
{Expected Wages}_i = a_1*{Expected Inflation}_i + a_2* {Inflation Uncertainty}_i + \text{Controls} + \e_i
Both Expected Inflation and Inflation Uncertainty are likely endogenous:
• Expected Inflation could be correlated with unobserved determinants of wage expectations.
• Inflation Uncertainty might be influenced by similar unobservables or by respondents’ information sets in ways that bias its relationship with wages.
My goal is to estimate the causal effect of Inflation Uncertainty on Expected Wages (i.e., to identify \alpha_2). However, I understand that in IV estimation, if multiple regressors are endogenous, all endogenous variables should be instrumented to avoid omitted variable bias. So, this leads me to two endogenous regressors that I plan to instrument:
• Inflation Uncertainty: I propose using lagged uncertainty about expected food and gas prices as instruments.
• Expected Inflation: I plan to use survey responses about expected food and gas prices (i.e., levels, not uncertainty) as instruments.
My Questions:
1. Can I instrument both endogenous variables using different combinations of food and gas price-related survey questions?
• That is, food/gas uncertainty for inflation uncertainty, and food/gas levels for expected inflation.
• These instruments are conceptually related but not identical to the endogenous variables. Is that acceptable?
2. Is this specification valid and interpretable?
• I understand that for the IV to be valid, instruments must satisfy relevance (strong correlation with the endogenous regressor) and exclusion (affect the outcome only through the endogenous variable).
• Given that both sets of instruments come from the same survey but target different aspects (level vs. uncertainty), is it still legitimate to use them in a system with multiple endogenous regressors?
- Is there any way to interpret these findings? Is it correct to do thiswith multiple endogenous variables and multiple IV
3. If not, what are the alternatives?
Any guidance on proper identification strategy or interpretation in the context of multiple endogenous regressors and overlapping (but distinct) instruments would be greatly appreciated!
I’m working on a project where I aim to understand the impact of subjective inflation uncertainty on expected wage growth. The regression I want to estimate is the following:
{Expected Wages}_i = a_1*{Expected Inflation}_i + a_2* {Inflation Uncertainty}_i + \text{Controls} + \e_i
Both Expected Inflation and Inflation Uncertainty are likely endogenous:
• Expected Inflation could be correlated with unobserved determinants of wage expectations.
• Inflation Uncertainty might be influenced by similar unobservables or by respondents’ information sets in ways that bias its relationship with wages.
My goal is to estimate the causal effect of Inflation Uncertainty on Expected Wages (i.e., to identify \alpha_2). However, I understand that in IV estimation, if multiple regressors are endogenous, all endogenous variables should be instrumented to avoid omitted variable bias. So, this leads me to two endogenous regressors that I plan to instrument:
• Inflation Uncertainty: I propose using lagged uncertainty about expected food and gas prices as instruments.
• Expected Inflation: I plan to use survey responses about expected food and gas prices (i.e., levels, not uncertainty) as instruments.
My Questions:
1. Can I instrument both endogenous variables using different combinations of food and gas price-related survey questions?
• That is, food/gas uncertainty for inflation uncertainty, and food/gas levels for expected inflation.
• These instruments are conceptually related but not identical to the endogenous variables. Is that acceptable?
2. Is this specification valid and interpretable?
• I understand that for the IV to be valid, instruments must satisfy relevance (strong correlation with the endogenous regressor) and exclusion (affect the outcome only through the endogenous variable).
• Given that both sets of instruments come from the same survey but target different aspects (level vs. uncertainty), is it still legitimate to use them in a system with multiple endogenous regressors?
- Is there any way to interpret these findings? Is it correct to do thiswith multiple endogenous variables and multiple IV
3. If not, what are the alternatives?
Any guidance on proper identification strategy or interpretation in the context of multiple endogenous regressors and overlapping (but distinct) instruments would be greatly appreciated!
