Unveiling the Pitfalls of Endogeneity in Forecasting: An Insight from Monte Carlo Simulation

Abstract

This paper investigates how endogeneity impairs the forecasting performance of cointegration-adjusted regression models. We employ a Monte Carlo simulation framework to generate data under three scenarios. In this regard, we have estimated three model variants: a correctly specified benchmark model ( M1), a model with the issue of endogeneity (M2), and a endogeneity corrected model (M3). Endogeneity is addressed using simulated instrumental variables that satisfy the relevance, exogeneity and exclusion restriction conditions. Forecast accuracy is evaluated across three models using various error metrics. The results reveal that models subject to endogeneity yield systematically poorer forecasts, while endogeneity correction through valid instruments significantly improves predictive performance of the model. The paper emphasizes that failure to address endogeneity in cointegration-adjusted regression models undermines structural validity, leading to biased forecasts and potentially flawed economic or financial decisions. Accurate model specification is thus essential for credible and policy-relevant forecasting.