DYNAMIC MODELING OF NIGERIAN ELECTION OUTCOMES: A BAYESIANTIME-VARYING CO-EFFICIENT APPROACH

Accurately modeling and forecasting electoral outcomes in Nigeria remains challenging due to the dynamic nature of political, economic, and security conditions, which are inadequately captured by static models. This study addresses this limitation by applying a Bayesian Time-Varying Coefficient (TVC) framework to analyze Nigerian presidential election outcomes from 1999-2023. The study variables include incumbency, voter turnout, inflation rate, and security incidents. Descriptive analysis shows an average vote share of 52.34 % with notable variability; correlation results indicate that security incidents have a strong negative relationship with vote share (-0.52) and incumbency exhibits a positive association (0.48). The Bayesian TVC model reveals that incumbency has a positive effect (posterior mean =6.45), while voter turnout contributes moderately (0.32). In contrast, inflation (-0.41) and security incidents (-0.85) exert negative effects, with security emerging as the most influential factor. Time-varying coefficient plots further highlight structural changes across election cycles. In terms of predictive performance, TVC model outperforms benchmark models such as Ordinary Least Squares and ARIMA, achieving lower forecast errors, with RMSE reduced to 3.87, compared to 6.82 for OLS. Scenario-based forecasts for the 2027 presidential election indicate that vote share could range from 63.4 % under favorable economic and security conditions to 48.6% under adverse conditions, and as low as 44.2 % in the absence of incumbency advantage. The study concludes that electoral outcomes in Nigeria are time-dependent and strongly influenced by evolving economic and security conditions. The Bayesian TVC model provides a flexible framework that enhances both explanatory power and forecasting accuracy.