Understanding Simultaneous Equation Bias and Its Solutions

Simultaneous equation bias occurs in econometric models when there is a relationship between two or more endogenous variables that are determined simultaneously within the model. In such systems, the explanatory variables in one equation are correlated with the error term of that same equation. This violates a key assumption of the Ordinary Least Squares (OLS) method, leading to biased and inconsistent parameter estimates.

Why OLS Fails in Simultaneous Equation Systems

In a system of simultaneous equations, variables are jointly determined. Consider a simple demand and supply model where price (P) and quantity (Q) are determined simultaneously:

Demand: $Q_d = \alpha_0 + \alpha_1 P + \alpha_2 I + u_d$

Supply: $Q_s = \beta_0 + \beta_1 P + \beta_2 W + u_s$

Where $I$ is income and $W$ is weather. In equilibrium, $Q_d = Q_s = Q$. If we try to estimate the demand equation using OLS, Price ($P$) is an endogenous variable because it is determined jointly with Quantity ($Q$) within the system. $P$ is correlated with the error term $u_d$ (since $u_d$ affects $Q_d$, which affects equilibrium $P$). This correlation makes the OLS estimator for $\alpha_1$ biased and inconsistent.

Solutions to Simultaneous Equation Bias

To obtain consistent estimates in the presence of simultaneous equation bias, alternative estimation methods are required. Let's examine the options provided:

A. OLS method

As explained above, the OLS method is not a suitable solution for simultaneous equation bias because it assumes uncorrelated error terms and explanatory variables, which is violated in simultaneous systems. Using OLS will lead to biased and inconsistent results.

B. Principle Component Method

Principle Component Analysis (PCA) is a dimensionality reduction technique. It is used to transform a set of possibly correlated variables into a set of linearly uncorrelated variables called principal components. While PCA can be useful in other contexts (e.g., dealing with multicollinearity in some cases), it does not directly address the endogeneity problem caused by simultaneous equations and is not a standard method for solving simultaneous equation bias.

C. Two-Stage Least Squares Method (2SLS method)

The Two-Stage Least Squares (2SLS) method is a widely used technique to address simultaneous equation bias. It is an instrumental variable technique. The core idea is to use instrumental variables (exogenous variables from the system that are correlated with the endogenous regressor but uncorrelated with the error term) to create predicted values for the endogenous regressors that are free from correlation with the error term.

• Stage 1: Regress the endogenous explanatory variable(s) on all exogenous variables in the entire system. Obtain the predicted values of the endogenous variable(s).
• Stage 2: Use these predicted values (which are uncorrelated with the error term) as regressors in the original structural equation. Estimate the parameters using OLS in this second stage.

2SLS provides consistent estimates in the presence of simultaneous equation bias, making it a valid solution.

D. Full Information Maximum Likelihood method (FIML)

The Full Information Maximum Likelihood (FIML) method is another approach to estimating parameters in a system of simultaneous equations. Unlike 2SLS (which is a limited information method, estimating one equation at a time), FIML is a full information method that estimates all equations in the system simultaneously. FIML assumes specific distributions for the error terms (typically multivariate normal) and finds the parameter values that maximize the joint probability of observing the sample data. FIML also provides consistent and asymptotically efficient estimates in the presence of simultaneous equation bias, making it a valid solution.

Summary of Methods for Simultaneous Equation Bias

Based on the analysis, both Two-Stage Least Squares (2SLS) and Full Information Maximum Likelihood (FIML) are true solutions to simultaneous equation bias, as they provide methods to obtain consistent parameter estimates in such systems.

Revision Table: Key Concepts in Simultaneous Equation Bias

Additional Information on Addressing Simultaneous Equation Bias

Choosing between 2SLS and FIML depends on several factors, including assumptions about error distributions and the complexity of the model. 2SLS is generally simpler to implement and is a "limited information" method (estimates one equation at a time), while FIML is a "full information" method (estimates all equations simultaneously) and can be more efficient if the model and distributional assumptions are correct, but it is more sensitive to specification errors across the entire system.

Simultaneous equation bias is a critical topic in econometrics, highlighting the need for estimation techniques that go beyond standard OLS when dealing with interdependencies among variables.