authors
Nagendra Marisetty

abstract
This article focuses on addressing violations of regression assumptions in the analysis of 20 years of monthly price data for nine international currency pairs. The Ordinary Least Squares (OLS) model, while achieving a high Adjusted R-squared, suffers from significant assumption violations, including residuals heteroscedasticity, autocorrelation, non-linearity, and multicollinearity. To address multicollinearity, removes variables with excessive variance inflation factors, improving coefficient reliability. However, issues like residual heteroscedasticity, autocorrelation and nonlinearity persist, indicating the need for further refinement. Transformation-based approaches, such as First Difference (FD) and Log Difference (LD), significantly improve assumption compliance by reducing residual heteroscedasticity, autocorrelation, non-linearity, and ARCH effects. Among these, the OLS Log Difference (OLS LD) model demonstrates the most effective correction of diagnostic issues, achieving compliance with key assumptions while minimizing standard errors and Akaia criterion (AIC). Although Weighted Lease Square (WLS) and Heteroscedasticity-Corrected (HSC) models also address some violations, their limited success in mitigating residual autocorrelation, nonlinearity, and complexity reduces their practicality. Overall, the OLS LD model emerges as the most effective approach, balancing assumption compliance and precision while providing reliable insights into the dynamics of currency pair price behaviours over the study period.